A composite material stability performance prediction method and related device
Through the integrated recurrent neural network model and attention mechanism, combined with time series analysis and adaptive feature weight matrix, the limitations of composite material stability performance prediction methods in high-dimensional and nonlinear environments are solved, and the stable performance of composite material is achieved. Efficient and accurate predictions enhance the adaptability of the prediction model.
Patent Information
- Application Number
- CN202411296104.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing composite material stability performance prediction methods have limitations when dealing with the relationship between high-dimensional, nonlinear and dynamic environmental factors and material attribute characteristics, and cannot achieve efficient and accurate prediction effects, especially it is difficult to capture the real-time stability performance of environmental factors on the material. Influence.
The integrated recurrent neural network model is used to combine attention mechanism and model weighting strategy to generate input feature vectors through time series analysis and adaptive feature weighting matrix, capture the service environment impact factors and their dynamic impact on the stable performance of materials, and use feedback optimization steps. Adjust model parameters to improve prediction accuracy.
It realizes efficient and accurate prediction of the stability performance of composite materials, can effectively capture the dynamic interaction between environmental factors and material properties, and improves the adaptability and accuracy of the prediction model, especially in the face of complex service environments.
Smart Images

Figure CN118824438B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of computational material science and technology, and particularly relates to a composite material stability performance prediction method and a related device. Background Art
[0002] In today's materials science and engineering field, composite materials are widely used in a variety of industries such as aerospace, automobile manufacturing, energy equipment and infrastructure construction due to their unique physical and chemical properties. Composite materials usually use one material as a matrix and one or more other materials as reinforcements to form materials with new properties through physical or chemical methods. With the continuous expansion of the scope of application, ensuring the long-term stability and reliability of composite materials in complex service environments has become a research focus. However, the stability performance of composite materials is affected by many factors, including but not limited to environmental factors and material property characteristics. Therefore, accurately predicting the stability performance of composite materials in a specific service environment is crucial to improving the reliability of product design and extending service life.
[0003] At present, the stability performance prediction methods for composite materials mainly include: analysis methods based on traditional mechanical models, finite element simulation methods and simple statistical learning methods, etc.; the above schemes have significant limitations when dealing with the relationship between high-dimensional, nonlinear and dynamically changing environmental factors and material stability performance, and cannot achieve efficient and accurate prediction effects; specifically as follows: existing prediction methods often ignore the dynamic interaction between environmental factors and material property characteristics, and it is difficult to accurately capture the immediate impact of subtle changes in environmental factors on material stability performance; secondly, traditional prediction models have limited capabilities in processing large data sets and complex time series data, especially in the face of the fusion process of real-time monitoring data and long-term trend forecasts. Summary of the invention
[0004] In view of the technical problems existing in the prior art, the present invention provides a method and related device for predicting the stability performance of composite materials to solve the technical problem that the existing prediction methods have significant limitations and cannot achieve efficient and accurate prediction effects.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is:
[0006] The present invention provides a method for predicting the stability performance of a composite material, comprising:
[0007] Collect service environment data and property data of the material to be tested to obtain an initial data set;
[0008] Based on the time series analysis method, the service environment data of the tested material that affects the stability performance of the tested material is extracted from the initial data set to obtain a set of key service environment influencing factors;
[0009] Generate an input feature vector based on the initial data set and the set of key service environment impact factors, combined with a pre-built adaptive feature weight matrix related to environmental impact adaptability;
[0010] The input feature vector is input into a pre-built composite material stability performance prediction model, and the stability performance prediction result of the material to be tested is output; wherein, the pre-built composite material stability performance prediction model adopts an integrated recurrent neural network model, and introduces an attention mechanism and a model weight strategy.
[0011] Furthermore, the service environment data of the material to be tested includes the temperature, humidity, pressure and vibration frequency in the service environment of the material to be tested; the property data of the material to be tested includes the microstructure parameters and composition characteristic parameters of the material to be tested;
[0012] Among them, the microstructure parameters of the material to be tested include the diameter of the reinforcement in the material to be tested, the arrangement structure parameters of the reinforcement, and the interface characteristic parameters between the reinforcement and the matrix; the component characteristic parameters of the material to be tested include the density, porosity and chemical composition ratio of the material to be tested.
[0013] Furthermore, according to the initial data set and the set of key service environment influencing factors, combined with the pre-built adaptive feature weight matrix related to environmental impact adaptability, the process of generating the input feature vector is as follows:
[0014] Select the property data of the material to be tested from the initial data set and construct the property feature matrix of the material to be tested;
[0015] Perform dimension reduction processing on the characteristic matrix of the property of the material to be tested to obtain the characteristic matrix after dimension reduction;
[0016] Based on the reduced dimension feature matrix and the key service environment influencing factor set, an extended feature matrix is constructed;
[0017] Sparse representation learning is performed on the extended feature matrix to obtain a sparse feature weight matrix; wherein the sparse feature weight matrix is specifically:
[0018]
[0019]
[0020] in, is the sparse feature weight matrix; It is the minimum angle regression analysis algorithm; The L2 norm squared is a measure of the difference between the expanded feature matrix and the product of the expanded feature matrix and the sparse feature weight matrix; Calculate the L1 norm of the sum of the absolute values of vector elements in the sparse feature weight matrix; is the regularization parameter; is the extended feature matrix;
[0021] Based on the set of key service environment influencing factors, the sparse feature weight matrix is optimized to obtain a pre-constructed adaptive feature weight matrix related to environmental impact adaptability;
[0022] The pre-constructed adaptive feature weight matrix related to the adaptability of environmental impact is used to perform weighted reorganization on the reduced-dimensional feature matrix to obtain the input feature vector.
[0023] Furthermore, based on the set of key service environment influencing factors, the sparse feature weight matrix is optimized to obtain a pre-built adaptive feature weight matrix related to environmental impact adaptability. The specific steps are as follows:
[0024] Based on the key service environment influencing factor set, the feature weight elements in the sparse feature weight matrix are optimized to obtain the adaptive feature weight elements related to the environmental impact adaptability;
[0025] Combining the adaptive feature weight elements related to the environmental impact adaptability to obtain a pre-constructed adaptive feature weight matrix related to the environmental impact adaptability;
[0026] Among them, the adaptive feature weight elements related to environmental impact adaptability are:
[0027]
[0028] in, is the adaptive feature weight element related to the adaptability of environmental impact; is the feature weight element in the sparse feature weight matrix; It is the first key service environment influencing factor. elements; is the mean value of the elements in the key service environment influencing factors; is the standard deviation of the key service environment influencing factors; is a natural exponential function.
[0029] Furthermore, the integrated recurrent neural network model includes a number of LSTM sub-models, and the number of LSTM sub-models is the same as the number of feature sequences in the input feature vector; the LSTM sub-model is used to capture the hidden state of the feature sequence in the input feature vector as it steps over time;
[0030] Among them, the process of capturing the hidden state of the feature sequence in the input feature vector over time is specifically:
[0031]
[0032] in, is the first feature vector in the input The feature sequence at time step Hidden state when It is the long short-term memory network algorithm; is the first feature vector in the input feature sequence; is the first feature vector in the input The feature sequence at time step Hidden state when
[0033] Attention mechanism, specifically:
[0034]
[0035] in, is the first feature vector in the input The feature sequence at time step The attention weight at time ; is the number of feature sequences in the input feature vector;
[0036] Model weight strategy, specifically:
[0037]
[0038] in, After the environmental impact adaptation adjustment The weights of the LSTM sub-models; Assigning functions for adaptive weights of preset environmental impacts; For the The weights of the LSTM sub-models before adjustment; is a set of key service environment influencing factors; is the standard deviation of the key service environment influencing factors set; is the balance coefficient; is the nonlinear adjustment term index.
[0039] Furthermore, after the input feature vector is input into a pre-built composite material stability performance prediction model and the stability performance prediction result of the material to be tested is output, a feedback optimization step is also included;
[0040] Among them, the feedback optimization steps are as follows:
[0041] Based on the prediction results of the stability performance of the tested materials and the actual monitoring results of the stability performance of the tested materials, a global sensitivity analysis is performed on the reduced-dimensional feature matrix and the set of key service environment influencing factors to obtain the prediction result error analysis results;
[0042] According to the prediction result error analysis results, a dynamic weight update function is constructed;
[0043] The extended feature matrix is updated by using a dynamic weight update function to obtain an updated extended feature matrix; wherein the updated extended feature matrix is specifically:
[0044]
[0045] in, is the updated extended feature matrix; is the first feature matrix after dimensionality reduction row feature sequence; is the row index of the feature matrix after dimensionality reduction; It is the first key service environment influencing factor. elements, is the element index of the key service environment influencing factor set; is the time step The first The first Dynamic weight of each element;
[0046] Based on the updated extended feature matrix, an updated input feature vector is generated;
[0047] Based on the updated input feature vector, the model parameters of the integrated recurrent neural network model and the characteristic parameters of the model weight strategy are adjusted to obtain an optimized composite material stability performance prediction model; wherein, based on the updated input feature vector, the model parameters of the integrated recurrent neural network model and the characteristic parameters of the model weight strategy are adjusted to obtain an optimized composite material stability performance prediction model, and the specific steps are as follows:
[0048] Inputting the updated input feature vector into a pre-built composite material stability performance prediction model, and outputting a new stability performance prediction result of the material to be tested;
[0049] The mean square error calculation method is used to calculate the mean square error between the new prediction result of the stability performance of the material to be tested and the actual monitoring result of the stability performance of the material to be tested, so as to obtain the mean square error of the new prediction result;
[0050] Based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters;
[0051] Among them, based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters. The specific process is as follows:
[0052]
[0053] in, are the optimized model parameters; To adjust the model parameters of the front-end integrated recurrent neural network model; is the learning rate; is the gradient vector of the model parameters of the integrated recurrent neural network model; is the mean square error of the new prediction result; The decay coefficient for the intensity of model parameter update;
[0054] Based on the updated extended feature matrix and the mean square error of the new prediction results, an environmental impact adaptability loss function is constructed; the environmental impact adaptability loss function is specifically:
[0055]
[0056] in, is the adaptive loss function of environmental impact; is the adjustment coefficient of environmental factors; is the proportion of environmental impact factors in the loss function;
[0057] According to the environmental impact adaptability loss function and the optimized model parameters, an environmental impact adaptability function is constructed; using the environmental impact adaptability function, characteristic parameters of the model weight strategy are adjusted to obtain optimized characteristic parameters; wherein the characteristic parameters of the model weight strategy include a balance coefficient and an adjustment term index of nonlinearity;
[0058] Specifically, the characteristic parameters of the model weight strategy are adjusted using the environmental impact adaptability function to obtain the implementation steps of the optimized characteristic parameters as follows:
[0059]
[0060]
[0061] in, is the optimized balance coefficient; is the optimized nonlinear adjustment term index; Stabilize performance prediction results for new materials to be tested; Actual monitoring results of the stability performance of the material to be tested; is the fitness function for environmental impact;
[0062] Substituting the optimized model parameters, the optimized balance coefficient and the optimized nonlinear adjustment term index into the pre-built composite material stability performance prediction model to obtain an optimized composite material stability performance prediction model;
[0063] The updated input feature vector is re-input into the optimized composite material stability performance prediction model, and the optimal stability performance prediction result of the material to be tested is output.
[0064] Furthermore, the dynamic weight update function is as follows:
[0065]
[0066] in, is the time step The first The first Dynamic weight of each element; is the time step The first The first Dynamic weight of each element; is the learning rate; is the sign function that determines the weight change; is the first feature matrix after dimensionality reduction The first The contribution of each element; It is the average contribution of the feature sequence in the feature matrix after dimensionality reduction and the elements in the key service environment influencing factor set.
[0067] The present invention also provides a composite material stability performance prediction system, comprising:
[0068] A data acquisition module is used to collect service environment data and property data of the material to be tested to obtain an initial data set;
[0069] A key environmental factor extraction module is used to extract service environment data of the material to be tested that affects the stability performance of the material to be tested from the initial data set based on a time series analysis method, and obtain a set of key service environment influencing factors;
[0070] An input feature vector generation module is used to generate an input feature vector based on an initial data set and a set of key service environment impact factors, combined with a pre-built adaptive feature weight matrix related to environmental impact adaptability;
[0071] The performance prediction module is used to input the input feature vector into the pre-built composite material stability performance prediction model, and output the stability performance prediction result of the material to be tested; wherein, the pre-built composite material stability performance prediction model adopts an integrated recurrent neural network model, and introduces an attention mechanism and a model weight strategy.
[0072] The present invention also provides a composite material stability performance prediction device, comprising:
[0073] Memory for storing computer programs;
[0074] A processor is used to implement the steps of the composite material stability performance prediction method when executing the computer program.
[0075] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the composite material stability performance prediction method are implemented.
[0076] Compared with the prior art, the present invention has the following beneficial effects:
[0077] The composite material stability performance prediction method provided by the present invention uses the service environment data of the material to be tested and the property data of the material to be tested to construct an initial data set, thereby achieving comprehensive coverage of multi-dimensional parameters in the actual working environment of the material to be tested; the initial data set is analyzed and processed using a time series analysis method to obtain a set of key service environment influencing factors, which can effectively capture the changing patterns of the service environment influencing factors over time and their possible dynamic impacts on the material stability performance; the initial data set is combined with the set of key service environment influencing factors, and based on a pre-constructed adaptive feature weight matrix related to environmental impact adaptability, an optimized feature vector is generated, thereby ensuring the accuracy and efficiency of the prediction model; the pre-constructed composite material stability performance prediction model adopts an integrated recurrent neural network model, and introduces an attention mechanism and a model weight strategy, which can effectively learn and simulate the complex nonlinear relationship between the material stability performance and environmental factors; the method described in the present invention effectively considers the dynamic interaction between environmental factors and material properties, thereby accurately capturing the immediate impact of environmental factors on the material stability performance, and thereby achieving efficient and accurate prediction of the composite material stability performance.
[0078] The composite material stability performance prediction system, composite material stability performance prediction device and computer-readable storage medium provided by the present invention have all the advantages of the above-mentioned composite material stability performance prediction method. BRIEF DESCRIPTION OF THE DRAWINGS
[0079] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below; obviously, the drawings described below are only some embodiments of the present application; for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0080] Figure 1A flow chart of the method for predicting the stability performance of a composite material provided in Example 1;
[0081] Figure 2 A structural block diagram of a composite material stability performance prediction system provided in Example 2;
[0082] Figure 3 This is a structural block diagram of the composite material stability performance prediction device provided in Example 3. DETAILED DESCRIPTION
[0083] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood, the present invention is further described in detail in the following specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0084] Example 1
[0085] As attached Figure 1 As shown, this embodiment 1 provides a method for predicting the stability performance of a composite material, comprising the following steps:
[0086] Step 1: Collect the service environment data and property data of the material to be tested to obtain an initial data set. The service environment data of the material to be tested includes the temperature, humidity, pressure and vibration frequency in the service environment of the material to be tested; the property data of the material to be tested includes the microstructure parameters and component characteristic parameters of the material to be tested; specifically, the microstructure parameters of the material to be tested include the diameter of the reinforcement in the material to be tested, the arrangement structure parameters of the reinforcement and the interface characteristic parameters between the reinforcement and the matrix, and the component characteristic parameters of the material to be tested include the density, porosity and chemical composition ratio of the material to be tested.
[0087] Specifically, the process of collecting the service environment data of the material to be tested is as follows:
[0088] Step 11: Use the predetermined computational fluid dynamics software to predict the distribution characteristics of temperature, humidity, pressure and vibration frequency in the service environment of the material to be tested, and obtain the prediction results of the service environment characteristics. The predetermined computational fluid dynamics software is used to simulate the process of complex fluid flow and heat transfer to predict the distribution characteristics of temperature, humidity, pressure and vibration frequency in the service environment of the material to be tested; preferably, the predetermined computational fluid dynamics software is, for example: ANSYS Fluent (a general computational fluid dynamics software developed by ANSYS) or Open FOAM (Open Field Operation and Manipulation, an open source computational fluid dynamics software).
[0089] It should be noted that the distribution characteristics of temperature, humidity, pressure and vibration frequency in the service environment of the material to be tested are important factors in evaluating the stability performance degradation of composite materials in the service environment; among them, temperature is used to characterize the thermal state of the environment, and abnormal temperature may cause thermal expansion of the composite material, thereby accelerating the aging of the composite material; humidity is used to characterize the water vapor content in the environment, which can affect the hygroscopicity of the composite material, thereby causing hydrolysis or corrosion of the composite material; pressure is used to characterize the environmental air pressure state, which can affect the strength and toughness of the composite material; vibration frequency is used to characterize the frequency characteristics of the composite material under external dynamic loads, which is related to the fatigue life of the composite material.
[0090] Step 12: Determine the planning results of environmental monitoring points based on the prediction results of the service environment characteristics. The planning results of environmental monitoring points include the location and number of environmental monitoring points. It should be noted that obtaining the prediction results of service environment characteristics can help identify key areas in the service environment, such as hot spots, wet areas, pressure concentration areas, or high-vibration areas. Planning the location and number of environmental monitoring points based on the prediction results of service environment characteristics can achieve effective monitoring of key areas in the service environment and reduce the cost of data monitoring and collection.
[0091] Step 13: Install environmental monitoring sensors according to the planning results of the environmental monitoring points. The environmental monitoring sensors include temperature and humidity sensors, pressure sensors and acceleration sensors. Specifically, the temperature and humidity sensors are used to monitor the temperature data and humidity data in the service environment of the material to be tested, the pressure sensors are used to monitor the pressure data in the service environment of the material to be tested, and the acceleration sensors are used to monitor the vibration frequency data in the service environment of the material to be tested.
[0092] Step 14: Collect monitoring data from environmental monitoring sensors to obtain an environmental data stream; perform data cleaning and standardization on the environmental data stream to obtain standardized environmental parameter time series data, that is, obtain the service environment data of the material to be tested.
[0093] The process of cleaning and standardizing the environmental data stream to obtain standardized environmental parameter time series data is as follows:
[0094] The Z-score (Standard Score) method or the IQR (Interquartile Range) method is used to remove outliers from the environmental data stream to obtain an environmental data stream without outliers. The linear interpolation method or the nearest neighbor method is used to fill in the missing values in the environmental data stream without outliers to obtain a cleaned environmental data stream. The Z-score method is used to perform data standardization on the cleaned environmental data stream to obtain standardized environmental parameter time series data; wherein, the data standardization of the cleaned environmental data stream is performed to eliminate the impact of the dimension, so that different types of data can be directly compared.
[0095] Specifically, the process of collecting the microstructure parameters of the material to be tested is as follows:
[0096] The microstructure of the material to be tested is observed using a scanning electron microscope or a transmission electron microscope, and the microstructure image of the material to be tested is obtained by combining the energy scattering spectrum analysis method; the microstructure image of the material to be tested is feature extracted using a preset image processing algorithm to obtain the microstructure parameters of the material to be tested.
[0097] Among them, the process of using a preset image processing algorithm to extract features from the microstructure image of the material to be tested and obtain the microstructure parameters of the material to be tested is specifically as follows: using an image segmentation algorithm to segment the microstructure image of the material to be tested to distinguish between the matrix and the reinforcement in the material to be tested and obtain an image segmentation result; using a preset image analysis software to quantify the image segmentation result to extract the microstructure features of the material to be tested from the image segmentation result, and output it in numerical form to obtain a high-dimensional digital data set, that is, to obtain the microstructure parameters of the material to be tested; it should be noted that the Hough transform algorithm is used to obtain the diameter of the reinforcement in the material to be tested, and the directional gradient histogram method is used to obtain the arrangement structure parameters of the reinforcement in the material to be tested.
[0098] It should be noted that the initial data set can be obtained by integrating the standardized environmental parameter time series data and the high-dimensional digital data set; that is, the initial data set contains high-dimensional characteristic data of the material to be tested, so as to realize the conversion of information in the physical world into a form that can be processed by a computer and provide data input for subsequent steps.
[0099] Step 2: Based on the time series analysis method, the service environment data of the material to be tested that affects the stability performance of the material to be tested is extracted from the initial data set to obtain a set of key service environment influencing factors.
[0100] Specifically, the process of obtaining the key service environment impact factor set is as follows:
[0101] Step 21: Use the sliding window method to perform statistical analysis on the service environment data of the material to be tested in the initial data set to obtain the statistical characteristics of the service environment data, wherein the statistical characteristics of the service environment data include the mean, standard deviation and extreme value of the service environment data of the material to be tested.
[0102] It should be noted that the specific implementation steps of using the sliding window method to perform statistical analysis on the service environment data of the material to be tested in the initial data set and obtain the statistical characteristics of the service environment data are as follows:
[0103] According to the time resolution of the service environment data of the material to be tested in the initial data set and the pre-captured statistical feature types, the characteristic parameters of the sliding window are determined to obtain a predetermined sliding window; wherein the characteristic parameters of the sliding window include the window size and the sliding step size; the predetermined sliding window is slid on the service environment data of the material to be tested in the initial data set, and during the sliding process, statistical values of the data in the predetermined sliding window are calculated to obtain the trend, discrete degree and boundary conditions of the service environment data of the material to be tested in the initial data set, and obtain the statistical characteristics of the service environment data.
[0104] Step 22: Use the fast Fourier transform algorithm to perform spectrum analysis on the service environment data of the material to be tested in the initial data set to obtain the periodic characteristics of the service environment data, such as seasonal variation characteristics, daily variation characteristics, working condition variation characteristics, and short-term fluctuation characteristics in long-term trends.
[0105] It should be noted that the process of using the fast Fourier transform algorithm to perform spectral analysis on the service environment data of the material to be tested in the initial data set to obtain the periodic characteristics of the service environment data is specifically: using the fast Fourier transform algorithm to convert the time series of the service environment data of the material to be tested in the initial data set into the frequency domain to identify the periodic components and obtain the periodic characteristics of the service environment data.
[0106] Step 23: According to the type of service environment data of the materials to be tested in the initial data set, an ARIMA (Autoregressive Integrated Moving Average Model) model is configured; according to the configured ARIMA model, and in combination with the statistical characteristics of the service environment data and the periodic characteristics of the service environment data, parameter prediction is performed on the service environment data of each type of the materials to be tested to obtain a prediction sequence of service environment influencing factors.
[0107] It should be noted that when configuring the ARIMA model, the ARIMA model is configured separately for each type of service environment data of the material to be tested, so as to perform parameter prediction on the service environment data of different types of materials to be tested respectively; the order of the ARIMA model includes the order of the autoregressive term, the order of the difference term and the order of the moving average term; the order of the ARIMA model is solved using the ACF diagram (Autocorrelation Function Plot) or the PACF (Partial Autocorrelation Function Plot); the parameter prediction is performed on the service environment data of each type of material to be tested by the maximum likelihood estimation method, and the prediction sequence of service environment influencing factors is obtained; wherein, the prediction sequence of service environment influencing factors includes several prediction values.
[0108] Step 24: Based on the prediction sequence of service environment influencing factors, a comprehensive environmental response function is constructed by weighted combination. Specifically, firstly, an influence weight is assigned to the prediction sequence of each type of service environment influencing factors; then, the influence weight assigned to the prediction sequence of each type of service environment influencing factors is used to perform weighted summation on the predicted values in the prediction sequence of service environment influencing factors to obtain the comprehensive environmental response function.
[0109] Step 25: Based on the comprehensive environmental response function, the prediction sequence of service environment influencing factors is subjected to Z-score standardization to obtain a standardized environmental influencing factor prediction sequence.
[0110] Step 26: Analyze and identify the mutation points and extreme value points in the standardized environmental impact factor prediction sequence to obtain the mutation points and extreme value points in the standardized environmental impact factor prediction sequence; determine the type of key service environment impact factors based on the mutation points and extreme value points in the standardized environmental impact factor prediction sequence.
[0111] Step 27: According to the type of key service environment influencing factors, data with the same type as the key service environment influencing factors are extracted from the initial data set to obtain a set of key service environment influencing factors.
[0112] Step 3: Generate an input feature vector based on the initial data set and the set of key service environment impact factors, combined with a pre-built adaptive feature weight matrix related to environmental impact adaptability.
[0113] The steps of generating the input feature vector are as follows:
[0114] Step 31: Select the property data of the material to be tested from the initial data set to obtain the property characteristic matrix of the material to be tested. The property characteristic matrix of the material to be tested is specifically:
[0115]
[0116] in, is the property characteristic matrix of the material to be tested; The material to be tested material property data, ; is the number of types of material property data to be tested.
[0117] Step 32: Perform dimensionality reduction processing on the characteristic rectangle of the material property to be tested, so as to reduce the data dimension while retaining as much variability as possible, and obtain the characteristic rectangle after dimensionality reduction. Preferably, the PCA (Principal Component Analysis) method is used to reduce the dimension of the characteristic matrix of the material property to be tested, and obtain the characteristic matrix after dimensionality reduction; wherein, the characteristic matrix after dimensionality reduction is specifically:
[0118]
[0119] in, is the feature matrix after dimensionality reduction; It is the principal component analysis algorithm.
[0120] Step 33: Based on the reduced dimension feature matrix and the key service environment influencing factor set, an extended feature matrix is constructed. The extended feature matrix is specifically:
[0121]
[0122] in, is the extended feature matrix; is the first feature matrix after dimensionality reduction row feature sequence; is the row index of the feature matrix after dimensionality reduction; It is the first key service environment influencing factor. elements, It is the element index of the key service environment influencing factor set.
[0123] Step 34: Use the LARS (Least Angle Regression) algorithm to perform sparse representation learning on the extended feature matrix to minimize the reconstruction error and maintain the sparsity of the weights, thereby obtaining a sparse feature weight matrix.
[0124] Among them, the sparse feature weight matrix is specifically:
[0125]
[0126]
[0127] in, is the sparse feature weight matrix; It is the minimum angle regression analysis algorithm; The L2 norm squared is a measure of the difference between the expanded feature matrix and the product of the expanded feature matrix and the sparse feature weight matrix; Calculate the L1 norm of the sum of the absolute values of vector elements in the sparse feature weight matrix; is the regularization parameter.
[0128] Step 35: Based on the key service environment impact factor set, the sparse feature weight matrix is optimized to obtain a pre-constructed adaptive feature weight matrix related to environmental impact adaptability. Specifically, the specific implementation steps for obtaining the pre-constructed adaptive feature weight matrix related to environmental impact adaptability are as follows:
[0129] Step 351: Based on the key service environment impact factor set, the feature weight elements in the sparse feature weight matrix are optimized to introduce environmental impact adaptability features and obtain adaptive feature weight elements related to environmental impact adaptability.
[0130] Among them, the adaptive feature weight elements related to environmental impact adaptability are:
[0131]
[0132] in, is the weight element of adaptive features related to the adaptability of environmental impact; is the feature weight element in the sparse feature weight matrix; is the mean value of the elements in the key service environment influencing factors; is the standard deviation of the key service environment influencing factors; is a natural exponential function.
[0133] Step 352: Combine the adaptive feature weight elements related to environmental impact adaptability to obtain a pre-constructed adaptive feature weight matrix related to environmental impact adaptability.
[0134] It should be noted that after optimizing each feature weight element in the sparse feature weight matrix based on the set of key service environment influencing factors, an adaptive feature weight element related to environmental impact adaptability corresponding to the weight element in the sparse feature weight matrix is obtained respectively. By combining all the adaptive feature weight elements related to environmental impact adaptability, a pre-constructed adaptive feature weight matrix related to environmental impact adaptability can be obtained.
[0135] Step 36: Use the pre-built adaptive feature weight matrix related to environmental impact adaptability to perform weighted reorganization on the reduced-dimensional feature matrix to obtain an input feature vector. The input feature vector is specifically:
[0136]
[0137] in, is the input feature vector; is a pre-built adaptive feature weight matrix related to environmental impact adaptability.
[0138] It should be noted that the input feature vector incorporates the original service environment data of the material to be tested, the influence of environmental factors and the environmental influence adaptability weight; by optimizing and adjusting the feature weight elements in the sparse feature weight matrix to introduce the environmental influence adaptability characteristics, the pre-constructed adaptive feature weight matrix related to environmental influence adaptability can better match the current service environment of the material to be tested, thereby improving the predictive ability and robustness of the stability performance of the material to be tested.
[0139] Step 4: Input the input feature vector into the pre-built composite material stability performance prediction model, and output the stability performance prediction result of the material to be tested. The pre-built composite material stability performance prediction model adopts an integrated recurrent neural network model, and introduces an attention mechanism and a model weight strategy; preferably, the integrated recurrent neural network model includes several LSTM (Long Short-Term Memory) sub-models, and the number of LSTM sub-models is the same as the number of feature sequences in the input feature vector; wherein the LSTM sub-model is used to capture the hidden state of the feature sequence in the input feature vector as it steps over time.
[0140] Specifically, the input feature vector is input into the pre-built composite material stability performance prediction model, and the implementation steps of outputting the stability performance prediction result of the material to be tested are as follows:
[0141] Step 41: Based on each feature sequence in the input feature vector, construct an independent LSTM sub-model respectively; use the constructed LSTM sub-model to capture the hidden state of each feature sequence in the input feature vector as time steps progress, and obtain the hidden state of the feature sequence.
[0142] Among them, the process of capturing the hidden state of the feature sequence in the input feature vector over time is specifically:
[0143]
[0144] in, is the first feature vector in the input The feature sequence at time step Hidden state when It is the long short-term memory network algorithm; is the first feature vector in the input feature sequence; is the first feature vector in the input The feature sequence at time step The hidden state of .
[0145] Step 42: Based on the hidden state of the feature sequence, an attention mechanism is applied to assign weights to the importance of each feature sequence in the input feature vector at each time step to emphasize the contribution of important time steps, and obtain the attention weight assignment results of the feature sequence at each time step. The attention mechanism is specifically:
[0146]
[0147] in, is the first feature vector in the input The feature sequence at time step The attention weight at time ; is the number of feature sequences in the input feature vector.
[0148] It should be noted that in the process of applying the attention mechanism to assign weights to the importance of each feature sequence in the input feature vector at each time step, the natural exponential function is used to calculate and normalize the operation to ensure that the sum of the attention weights of all time steps is 1, so that the integrated recurrent neural network model can dynamically focus on the most critical feature sequences and time points.
[0149] Step 43: According to the attention weight distribution results of the feature sequence at each time step, combined with the model weight strategy, the weights of each LSTM sub-model are dynamically adjusted to obtain the weights of the LSTM sub-models after environmental impact adaptability adjustment. Among them, by introducing the environmental impact adaptability weight distribution function, it is ensured that the influence of each LSTM sub-model can be flexibly adjusted under different environmental conditions, thereby improving the accuracy of the prediction results.
[0150] Among them, the model weight strategy is specifically:
[0151]
[0152] in, After the environmental impact adaptation adjustment The weights of the LSTM sub-models; Assigning functions for adaptive weights of preset environmental impacts; For the The weights of the LSTM sub-models before adjustment; is a set of key service environment influencing factors; is the standard deviation of the key service environment influencing factors set; is the balance coefficient; is the nonlinear adjustment term index.
[0153] It should be noted that The weights of the LSTM sub-models before adjustment , balance coefficient and nonlinear adjustment index All are preset based on experience; after environmental impact adaptability adjustment, The weights of the LSTM sub-models Using the preset environmental impact adaptive weight distribution function For the pre-set The weights of the LSTM sub-models before adjustment Adaptive adjustment is performed.
[0154] Step 44, using an integrated learning strategy, based on the weights of the LSTM sub-models that have been adjusted for environmental impact adaptability, the prediction outputs of all LSTM sub-models are combined using a weighted average method to obtain a prediction result of the stability performance of the material to be tested. Among them, combining the prediction outputs of all LSTM sub-models using a weighted average method can reduce the overfitting risk of a single sub-model and improve the overall prediction performance.
[0155] Among them, the stability performance prediction results of the tested materials are as follows:
[0156]
[0157] in, Predict the results of the stability performance of the material to be tested; For the The LSTM sub-model is at time step The predicted output at .
[0158] It should be noted that by using the integrated recurrent neural network model as the framework of the pre-built composite material stability performance prediction model and introducing the attention mechanism and model weight strategy, it can fully consider the dynamic changes in the importance of timing characteristics and features, as well as the influence of service environment factors, thereby improving the practicality and generalization ability of the prediction model and realizing dynamic and accurate prediction of the stability performance of composite materials.
[0159] Step 5: Based on the prediction results of the stability performance of the materials to be tested and the actual monitoring results of the stability performance of the materials to be tested, a global sensitivity analysis is performed on the reduced-dimensional feature matrix and the set of key service environment influencing factors to obtain the prediction result error analysis results.
[0160] The process of obtaining the prediction result error analysis results is as follows:
[0161] Step 51, obtaining actual monitoring data of stability performance characterization indicators of the material to be tested, wherein the stability performance characterization indicators of the material to be tested include material stress, material strain and fatigue life of the service device of the material to be tested.
[0162] Step 52: The actual monitoring data of the stability performance characterization index of the material to be tested is processed by outlier elimination and missing value filling to eliminate the monitoring values beyond the normal range, so as to ensure the accuracy of data analysis and the integrity of data, and obtain the actual monitoring results of the stability performance of the material to be tested.
[0163] Step 53: using a mean square error calculation method, calculate the mean square error between the predicted result of the stability performance of the material to be tested and the actual monitoring result of the stability performance of the material to be tested, and obtain the mean square error of the predicted result. The mean square error of the predicted result is displayed in the form of a time series graph.
[0164] It should be noted that the time series diagram is generated based on the comparison between the predicted results of the stability performance of the material to be tested and the actual monitored results of the stability performance of the material to be tested, and can intuitively show the performance of the prediction model; wherein, the time series diagram uses time as the horizontal axis and the mean square error between the predicted results of the stability performance of the material to be tested and the actual monitored results of the stability performance of the material to be tested as the vertical axis, so that the difference between the predicted results of the stability performance of the material to be tested and the actual monitored results of the stability performance of the material to be tested can be clearly obtained.
[0165] Step 54: Obtain an error period according to the mean square error of the prediction result. The error period includes an error time point or an error time interval. Specifically, the process of obtaining the error period is as follows:
[0166] The mean square error of the prediction result is compared with the preset error threshold, and the time point or time interval exceeding the preset error threshold is determined to obtain the error period.
[0167] Step 55: Use the preset sensitivity analysis method to perform global sensitivity feature analysis on the reduced dimension feature matrix and the key service environment influencing factor set within the error period to obtain the error source, that is, to obtain the prediction error analysis result. Among them, the preset global sensitivity analysis method adopts the Sobol sensitivity analysis method, the Morris screening method or the meta-model-based global sensitivity analysis method. Specifically, the prediction error analysis result includes the first The first The contribution of each element and the average contribution of the feature sequence in the feature matrix after dimensionality reduction and the elements in the key service environment influencing factor set are calculated.
[0168] It should be noted that Sobol sensitivity analysis is a sensitivity analysis method based on variance, Morris screening method is a sensitivity analysis method based on local changes, and the meta-model-based global sensitivity analysis method evaluates the sensitivity of input variables by constructing an approximate model between input variables and output variables; preferably, the approximate model between input variables and output variables adopts a random forest model or a support vector machine model.
[0169] It should also be noted that the source of error refers to the factors that cause the deviation between the predicted results of the stability performance of the material to be tested by the pre-constructed composite material stability performance prediction model within the error period and the actual monitoring results of the stability performance of the material to be tested; by performing global sensitive feature analysis on the reduced-dimensional feature matrix and the set of key service environment influencing factors within the error period, it is possible to quantify the contribution of the input variables to the error, thereby identifying the factors that have the greatest impact on the prediction results.
[0170] Step 6: Construct a dynamic weight update function based on the prediction result error analysis results. The dynamic weight update function is specifically:
[0171]
[0172] in, is the time step The first The first Dynamic weight of each element; is the time step The first The first Dynamic weight of each element; is the learning rate; is the sign function that determines the weight change; is the first feature matrix after dimensionality reduction The first The contribution of each element; It is the average contribution of the feature sequence in the feature matrix after dimensionality reduction and the elements in the key service environment influencing factor set.
[0173] It should be noted that the first The first The contribution of each element The global sensitive characteristics are obtained by performing a preset sensitivity analysis method on the reduced-dimensional feature matrix and the set of key service environment influencing factors within the error period.
[0174] Step 7: Based on the dynamic weight update function, the extended feature matrix is updated to obtain an updated extended feature matrix. The updated extended feature matrix is specifically:
[0175]
[0176] in, is the updated extended feature matrix.
[0177] Step 8: Generate an updated input feature vector based on the updated extended feature matrix. Specifically, the implementation steps of generating an updated input feature vector based on the updated extended feature matrix are as follows:
[0178] Step 81: Use the LARS (Least Angle Regression) algorithm to perform sparse representation learning on the updated extended feature matrix to obtain an updated sparse feature weight matrix.
[0179] Step 82: Based on the key service environment impact factor set, the updated sparse feature weight matrix is optimized to obtain an updated adaptive feature weight matrix related to environmental impact adaptability.
[0180] Step 83: Use the updated adaptive feature weight matrix related to environmental impact adaptability to perform weighted reorganization on the reduced-dimensional feature matrix to obtain an updated input feature vector.
[0181] It should be noted that the principles and specific operations of steps 81-83 are basically the same as those of the above steps 34-36, and will not be repeated here.
[0182] Step 9: Input the updated input feature vector into the pre-built composite material stability performance prediction model, and output a new stability performance prediction result of the material to be tested.
[0183] Step 10: Using a mean square error calculation method, calculate the mean square error between the new predicted result of the stability performance of the material to be tested and the actual monitored result of the stability performance of the material to be tested, and obtain the mean square error of the new predicted result.
[0184] Step 11: Based on the mean square error of the new prediction result, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters.
[0185] Among them, based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters. The specific process is as follows:
[0186]
[0187] in, are the optimized model parameters; To adjust the model parameters of the front-end integrated recurrent neural network model; is the gradient vector of the model parameters of the integrated recurrent neural network model; is the mean square error of the new prediction result; Decay coefficient for the strength of model parameter updates.
[0188] Step 12: Based on the updated extended feature matrix and the mean square error of the new prediction result, construct an environmental impact adaptability loss function. The environmental impact adaptability loss function is specifically:
[0189]
[0190] in, is the adaptive loss function of environmental impact; is the adjustment coefficient of environmental factors; is the proportion of environmental impact factors in the loss function.
[0191] Step 13: construct an environmental impact adaptability function based on the environmental impact adaptability loss function and the optimized model parameters; use the environmental impact adaptability function to adjust the characteristic parameters of the model weight strategy to obtain the optimized characteristic parameters. The characteristic parameters of the model weight strategy include the balance coefficient and the adjustment term index of the nonlinearity.
[0192] Specifically, the characteristic parameters of the model weight strategy are adjusted using the environmental impact adaptability function to obtain the implementation steps of the optimized characteristic parameters as follows:
[0193]
[0194]
[0195] in, is the optimized balance coefficient; is the optimized nonlinear adjustment term index; Stabilize performance prediction results for new materials to be tested; Actual monitoring results of the stability performance of the material to be tested; is the environmental impact fitness function.
[0196] Step 14: Substitute the optimized model parameters, the optimized balance coefficient and the optimized nonlinear adjustment term index into the pre-built composite material stability performance prediction model to obtain the optimized composite material stability performance prediction model.
[0197] Step 15: re-input the updated input feature vector into the optimized composite material stability performance prediction model, and output the optimal stability performance prediction result of the material to be tested.
[0198] The composite material stability performance prediction method described in Example 1 uses an integrated recurrent neural network model as the framework of a pre-built composite material stability performance prediction model, and introduces an attention mechanism and a model weight strategy to fully consider the dynamic transformation of timing characteristics, the importance of features, and the influence of service environment factors, thereby significantly improving the adaptability and accuracy of the prediction model in various complex service environments, and achieving accurate prediction of the stability performance of composite materials; especially when facing environmental mutation points and extreme conditions, the model can more accurately identify the influence of key factors and reduce prediction errors, thereby providing more reliable technical support for performance evaluation and life prediction of composite materials.
[0199] In this embodiment 1, based on the updated input feature vector, the model parameters of the integrated recurrent neural network model and the characteristic parameters of the model weight strategy are adjusted to achieve feedback optimization of the pre-built composite material stability performance prediction model, which can effectively learn and simulate the complex nonlinear relationship between material stability performance and environmental factors, thereby achieving dynamic and accurate prediction of the stability performance of composite materials, further improving the prediction accuracy of the prediction model.
[0200] Example 2
[0201] As attached Figure 2 As shown, this embodiment 2 provides a composite material stability performance prediction system, including a data acquisition module, a key environmental factor extraction module, an input feature vector generation module, a performance prediction module and a feedback optimization module.
[0202] The data acquisition module is used to collect the service environment data and property data of the material to be tested to obtain an initial data set.
[0203] The key environmental factor extraction module is used to extract the service environment data of the material to be tested that affects the stability performance of the material to be tested from the initial data set based on the time series analysis method, and obtain a set of key service environment influencing factors.
[0204] The input feature vector generation module is used to generate an input feature vector based on the initial data set and the set of key service environment impact factors, combined with a pre-built adaptive feature weight matrix related to environmental impact adaptability.
[0205] The performance prediction module is used to input the input feature vector into the pre-built composite material stability performance prediction model, and output the stability performance prediction result of the material to be tested.
[0206] The feedback optimization module is used to perform a global sensitivity analysis on the reduced-dimensional feature matrix and the key service environment influencing factor set based on the stability performance prediction results of the material to be tested and the actual monitoring results of the stability performance of the material to be tested, so as to obtain the prediction result error analysis results; construct a dynamic weight update function based on the prediction result error analysis results; update the extended feature matrix based on the dynamic weight update function to obtain an updated extended feature matrix; generate an updated input feature vector based on the updated extended feature matrix; input the updated input feature vector into the pre-constructed composite material stability performance prediction model, and output a new stability performance prediction result of the material to be tested; use a mean square error calculation method to calculate the mean square error between the new stability performance prediction result of the material to be tested and the actual monitoring results of the stability performance of the material to be tested, so as to obtain the mean square error of the new prediction result. ; Based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters; based on the updated extended feature matrix and the mean square error of the new prediction results, an environmental impact adaptability loss function is constructed; according to the environmental impact adaptability loss function and the optimized model parameters, an environmental impact adaptability function is constructed; using the environmental impact adaptability function, the characteristic parameters of the model weight strategy are adjusted to obtain the optimized characteristic parameters; the optimized model parameters, the optimized balance coefficient and the optimized nonlinear adjustment item index are substituted into the pre-constructed composite material stability performance prediction model to obtain the optimized composite material stability performance prediction model; the updated input feature vector is re-input into the optimized composite material stability performance prediction model, and the optimal stability performance prediction result of the material to be tested is obtained as output.
[0207] Example 3
[0208] As attached Figure 3 As shown, this embodiment 3 provides a composite material stability performance prediction device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a composite material stability performance prediction method program; wherein the attached Figure 3 The communication interface in is used to access external devices to obtain data.
[0209] When the processor executes the computer program, the steps of the above-mentioned composite material stability performance prediction method are implemented, for example: the service environment data of the material to be tested and the property data of the material to be tested are collected to obtain an initial data set. Based on the time series analysis method, the service environment data of the material to be tested that affects the stability performance of the material to be tested is extracted from the initial data set to obtain a set of key service environment influencing factors. According to the initial data set and the set of key service environment influencing factors, and in combination with a pre-constructed adaptive feature weight matrix related to environmental impact adaptability, an input feature vector is generated. The input feature vector is input into a pre-constructed composite material stability performance prediction model, and the stability performance prediction result of the material to be tested is obtained as an output. Based on the stability performance prediction results of the materials to be tested and the actual monitoring results of the stability performance of the materials to be tested, a global sensitivity analysis is performed on the reduced-dimensional feature matrix and the key service environment influencing factor set to obtain the prediction result error analysis results; according to the prediction result error analysis results, a dynamic weight update function is constructed; based on the dynamic weight update function, the extended feature matrix is updated to obtain an updated extended feature matrix; based on the updated extended feature matrix, an updated input feature vector is generated; the updated input feature vector is input into the pre-constructed composite material stability performance prediction model, and a new stability performance prediction result of the materials to be tested is obtained as output; the mean square error calculation method is used to calculate the mean square error between the new stability performance prediction results of the materials to be tested and the actual monitoring results of the stability performance of the materials to be tested, and the mean square error of the new prediction results is obtained; based on the new The mean square error of the prediction results is used to adjust the model parameters of the integrated recurrent neural network model to obtain the optimized model parameters; based on the updated extended feature matrix and the mean square error of the new prediction results, an environmental impact adaptability loss function is constructed; according to the environmental impact adaptability loss function and the optimized model parameters, an environmental impact adaptability function is constructed; using the environmental impact adaptability function, the characteristic parameters of the model weight strategy are adjusted to obtain the optimized characteristic parameters; the optimized model parameters, the optimized balance coefficient and the optimized nonlinear adjustment item index are substituted into the pre-constructed composite material stability performance prediction model to obtain the optimized composite material stability performance prediction model; the updated input feature vector is re-input into the optimized composite material stability performance prediction model, and the optimal stability performance prediction result of the material to be tested is obtained as output.
[0210] Alternatively, when the processor executes the computer program, the functions of each module in the above system are implemented, for example: a data acquisition module is used to collect the service environment data of the material to be tested and the property data of the material to be tested to obtain an initial data set. A key environmental factor extraction module is used to extract the service environment data of the material to be tested that affects the stability performance of the material to be tested from the initial data set based on a time series analysis method to obtain a set of key service environment influencing factors. An input feature vector generation module is used to generate an input feature vector based on the initial data set and the set of key service environment influencing factors, in combination with a pre-built adaptive feature weight matrix related to environmental impact adaptability. A performance prediction module is used to input the input feature vector into a pre-built composite material stability performance prediction model, and output a prediction result of the stability performance of the material to be tested. The feedback optimization module is used to perform a global sensitivity analysis on the reduced-dimensional feature matrix and the key service environment influencing factor set based on the stability performance prediction results of the material to be tested and the actual monitoring results of the stability performance of the material to be tested, so as to obtain the prediction result error analysis results; construct a dynamic weight update function based on the prediction result error analysis results; update the extended feature matrix based on the dynamic weight update function to obtain an updated extended feature matrix; generate an updated input feature vector based on the updated extended feature matrix; input the updated input feature vector into the pre-constructed composite material stability performance prediction model, and output a new stability performance prediction result of the material to be tested; use a mean square error calculation method to calculate the mean square error between the new stability performance prediction result of the material to be tested and the actual monitoring results of the stability performance of the material to be tested, so as to obtain the mean square error of the new prediction result. ; Based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters; based on the updated extended feature matrix and the mean square error of the new prediction results, an environmental impact adaptability loss function is constructed; according to the environmental impact adaptability loss function and the optimized model parameters, an environmental impact adaptability function is constructed; using the environmental impact adaptability function, the characteristic parameters of the model weight strategy are adjusted to obtain the optimized characteristic parameters; the optimized model parameters, the optimized balance coefficient and the optimized nonlinear adjustment item index are substituted into the pre-constructed composite material stability performance prediction model to obtain the optimized composite material stability performance prediction model; the updated input feature vector is re-input into the optimized composite material stability performance prediction model, and the optimal stability performance prediction result of the material to be tested is obtained as output.
[0211] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the composite material stability performance prediction device.
[0212] For example, the computer program can be divided into a data acquisition module, a key environmental factor extraction module, an input feature vector generation module, a performance prediction module and a feedback optimization module, and the specific functions of each module are as follows:
[0213] The data acquisition module is used to collect the service environment data of the material to be tested and the property data of the material to be tested to obtain the initial data set. The key environmental factor extraction module is used to extract the service environment data of the material to be tested that affects the stability performance of the material to be tested from the initial data set based on the time series analysis method to obtain the key service environment influencing factor set. The input feature vector generation module is used to generate the input feature vector based on the initial data set and the key service environment influencing factor set, combined with the pre-built adaptive feature weight matrix related to environmental impact adaptability. The performance prediction module is used to input the input feature vector into the pre-built composite material stability performance prediction model, and output the stability performance prediction result of the material to be tested. The feedback optimization module is used to perform a global sensitivity analysis on the reduced-dimensional feature matrix and the key service environment influencing factor set based on the stability performance prediction results of the material to be tested and the actual monitoring results of the stability performance of the material to be tested, so as to obtain the prediction result error analysis results; construct a dynamic weight update function based on the prediction result error analysis results; update the extended feature matrix based on the dynamic weight update function to obtain an updated extended feature matrix; generate an updated input feature vector based on the updated extended feature matrix; input the updated input feature vector into the pre-constructed composite material stability performance prediction model, and output a new stability performance prediction result of the material to be tested; use a mean square error calculation method to calculate the mean square error between the new stability performance prediction result of the material to be tested and the actual monitoring results of the stability performance of the material to be tested, so as to obtain the mean square error of the new prediction result. ; Based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters; based on the updated extended feature matrix and the mean square error of the new prediction results, an environmental impact adaptability loss function is constructed; according to the environmental impact adaptability loss function and the optimized model parameters, an environmental impact adaptability function is constructed; using the environmental impact adaptability function, the characteristic parameters of the model weight strategy are adjusted to obtain the optimized characteristic parameters; the optimized model parameters, the optimized balance coefficient and the optimized nonlinear adjustment item index are substituted into the pre-constructed composite material stability performance prediction model to obtain the optimized composite material stability performance prediction model; the updated input feature vector is re-input into the optimized composite material stability performance prediction model, and the optimal stability performance prediction result of the material to be tested is obtained as output.
[0214] The composite material stability performance prediction device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The composite material stability performance prediction device can include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the attached Figure 3The schematic diagram is merely an example of a composite material stability performance prediction device and does not constitute a limitation of the composite material stability performance prediction device. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the composite material stability performance prediction device may also include input and output devices, network access devices, buses, etc.
[0215] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the composite material stability performance prediction device, and uses various interfaces and lines to connect various parts of the entire composite material stability performance prediction device.
[0216] The memory can be used to store the computer program and / or module, and the processor implements various functions of the composite material stability performance prediction device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.
[0217] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.
[0218] Example 4
[0219] This embodiment 4 also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method for predicting the stability performance of a composite material are implemented.
[0220] If the module / unit integrated in the composite material stability performance prediction device 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.
[0221] Based on this understanding, the present embodiment 4 implements all or part of the processes in the above-mentioned composite material stability performance prediction method, and can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned composite material stability performance prediction method can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc.
[0222] The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.
[0223] It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electrical carrier signals and telecommunication signals.
[0224] The composite material stability performance prediction method provided by the present invention uses an integrated recurrent neural network model as the framework of a pre-built composite material stability performance prediction model, and significantly enhances the accuracy and practicability of composite material stability performance prediction by introducing an attention mechanism and a model weight strategy, effectively solving the limitations of the prior art in processing dynamic interactions between the environment and material microstructure, high-dimensional nonlinear data, and the fusion of real-time monitoring and long-term prediction, enhancing the adaptability of the prediction model to complex environments, achieving high-precision predictions and dynamic optimization and adjustment, promoting the progress of intelligent material management and maintenance, and bringing higher safety and economic benefits to composite material applications.
[0225] The above embodiment is only one of the implementation methods that can realize the technical solution of the present invention. The scope of protection claimed by the present invention is not limited only to this embodiment, but also includes changes, replacements and other implementation methods that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention.
Claims
1. A method for predicting the stability performance of a composite material, characterized in that: include: Collect service environment data and property data of the material to be tested to obtain an initial data set; Based on the time series analysis method, the service environment data of the tested material that affects the stability performance of the tested material is extracted from the initial data set to obtain a set of key service environment influencing factors; Generate an input feature vector based on the initial data set and the set of key service environment impact factors, combined with a pre-built adaptive feature weight matrix related to environmental impact adaptability; Input the input feature vector into a pre-built composite material stability performance prediction model, and output the stability performance prediction result of the material to be tested; wherein the pre-built composite material stability performance prediction model adopts an integrated recurrent neural network model, and introduces an attention mechanism and a model weight strategy; The process of generating the input feature vector based on the initial data set and the set of key service environment impact factors, combined with the pre-built adaptive feature weight matrix related to environmental impact adaptability, is as follows: Select the property data of the material to be tested from the initial data set and construct the property feature matrix of the material to be tested; Perform dimension reduction processing on the characteristic matrix of the property of the material to be tested to obtain the characteristic matrix after dimension reduction; Based on the reduced dimension feature matrix and the key service environment influencing factor set, an extended feature matrix is constructed; Sparse representation learning is performed on the extended feature matrix to obtain a sparse feature weight matrix; wherein the sparse feature weight matrix is specifically: in, is the sparse feature weight matrix; It is the minimum angle regression analysis algorithm; The L2 norm squared is a measure of the difference between the expanded feature matrix and the product of the expanded feature matrix and the sparse feature weight matrix; Calculate the L1 norm of the sum of the absolute values of vector elements in the sparse feature weight matrix; is the regularization parameter; is the extended feature matrix; Based on the set of key service environment influencing factors, the sparse feature weight matrix is optimized to obtain a pre-constructed adaptive feature weight matrix related to environmental impact adaptability; Using the pre-built adaptive feature weight matrix related to environmental impact adaptability, the reduced-dimensional feature matrix is weighted and reorganized to obtain the input feature vector; Based on the set of key service environment influencing factors, the sparse feature weight matrix is optimized to obtain the pre-built adaptive feature weight matrix related to environmental impact adaptability. The specific steps are as follows: Based on the key service environment influencing factor set, the feature weight elements in the sparse feature weight matrix are optimized to obtain adaptive feature weight elements related to environmental impact adaptability; Combining the adaptive feature weight elements related to the environmental impact adaptability to obtain a pre-constructed adaptive feature weight matrix related to the environmental impact adaptability; Among them, the adaptive feature weight elements related to environmental impact adaptability are: in, is the adaptive feature weight element related to the adaptability of environmental impact, is the row index of the feature matrix after dimensionality reduction, is the element index of the key service environment influencing factor set; is the feature weight element in the sparse feature weight matrix; It is the first key service environment influencing factor. elements; is the mean value of the elements in the key service environment influencing factors; is the standard deviation of the key service environment influencing factors; is the natural exponential function; The integrated recurrent neural network model includes several LSTM sub-models, and the number of LSTM sub-models is the same as the number of feature sequences in the input feature vector; the LSTM sub-model is used to capture the hidden state of the feature sequence in the input feature vector as time steps; Among them, the process of capturing the hidden state of the feature sequence in the input feature vector over time is specifically: in, is the first feature vector in the input The feature sequence at time step Hidden state when It is the long short-term memory network algorithm; is the first feature vector in the input feature sequence; is the first feature vector in the input The feature sequence at time step Hidden state when Attention mechanism, specifically: in, is the first feature vector in the input The feature sequence at time step The attention weight at time ; is the number of feature sequences in the input feature vector; Model weight strategy, specifically: in, After the environmental impact adaptation adjustment The weights of the LSTM sub-models; Assigning functions for adaptive weights of preset environmental impacts; For the The weights of the LSTM sub-models before adjustment; is a set of key service environment influencing factors; is the standard deviation of the key service environment influencing factors set; is the balance coefficient; is the nonlinear adjustment term index.
2. A composite material stability performance prediction method according to claim 1, characterized in that: The service environment data of the material to be tested includes the temperature, humidity, pressure and vibration frequency in the service environment of the material to be tested; the property data of the material to be tested includes the microstructure parameters and composition characteristic parameters of the material to be tested; Among them, the microstructure parameters of the material to be tested include the diameter of the reinforcement in the material to be tested, the arrangement structure parameters of the reinforcement, and the interface characteristic parameters between the reinforcement and the matrix; the component characteristic parameters of the material to be tested include the density, porosity and chemical composition ratio of the material to be tested.
3. A composite material stability performance prediction method according to claim 1, characterized in that: The input feature vector is input into a pre-built composite material stability performance prediction model, and after the stability performance prediction result of the material to be tested is output, a feedback optimization step is also included; Among them, the feedback optimization steps are as follows: Based on the prediction results of the stability performance of the tested materials and the actual monitoring results of the stability performance of the tested materials, a global sensitivity analysis is performed on the reduced-dimensional feature matrix and the set of key service environment influencing factors to obtain the prediction result error analysis results; According to the prediction result error analysis results, a dynamic weight update function is constructed; The extended feature matrix is updated by using a dynamic weight update function to obtain an updated extended feature matrix; wherein the updated extended feature matrix is specifically: in, is the updated extended feature matrix; is the first feature matrix after dimensionality reduction row feature sequence; is the row index of the feature matrix after dimensionality reduction; It is the first key service environment influencing factor. elements, is the element index of the key service environment influencing factor set; is the time step The first The first Dynamic weight of each element; Based on the updated extended feature matrix, an updated input feature vector is generated; Based on the updated input feature vector, the model parameters of the integrated recurrent neural network model and the characteristic parameters of the model weight strategy are adjusted to obtain an optimized composite material stability performance prediction model; wherein, based on the updated input feature vector, the model parameters of the integrated recurrent neural network model and the characteristic parameters of the model weight strategy are adjusted to obtain an optimized composite material stability performance prediction model, and the specific steps are as follows: Inputting the updated input feature vector into a pre-built composite material stability performance prediction model, and outputting a new stability performance prediction result of the material to be tested; The mean square error calculation method is used to calculate the mean square error between the new prediction result of the stability performance of the material to be tested and the actual monitoring result of the stability performance of the material to be tested, so as to obtain the mean square error of the new prediction result; Based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters; Among them, based on the mean square error of the new prediction results, the model parameters of the integrated recurrent neural network model are adjusted to obtain the optimized model parameters. The specific process is as follows: in, are the optimized model parameters; To adjust the model parameters of the front-end integrated recurrent neural network model; is the learning rate; is the gradient vector of the model parameters of the integrated recurrent neural network model; is the mean square error of the new prediction result; The decay coefficient for the intensity of model parameter update; Based on the updated extended feature matrix and the mean square error of the new prediction results, an environmental impact adaptability loss function is constructed; the environmental impact adaptability loss function is specifically: in, is the adaptive loss function of environmental impact; is the adjustment coefficient of environmental factors; is the proportion of environmental impact factors in the loss function; According to the environmental impact adaptability loss function and the optimized model parameters, an environmental impact adaptability function is constructed; using the environmental impact adaptability function, characteristic parameters of the model weight strategy are adjusted to obtain optimized characteristic parameters; wherein the characteristic parameters of the model weight strategy include a balance coefficient and an adjustment term index of nonlinearity; Specifically, the characteristic parameters of the model weight strategy are adjusted using the environmental impact adaptability function to obtain the implementation steps of the optimized characteristic parameters as follows: in, is the optimized balance coefficient; is the optimized nonlinear adjustment term index; Stabilize performance prediction results for new materials to be tested; Actual monitoring results of the stability performance of the material to be tested; is the fitness function for environmental impact; Substituting the optimized model parameters, the optimized balance coefficient and the optimized nonlinear adjustment term index into the pre-built composite material stability performance prediction model to obtain an optimized composite material stability performance prediction model; The updated input feature vector is re-input into the optimized composite material stability performance prediction model, and the optimal stability performance prediction result of the material to be tested is output.
4. A composite material stability performance prediction method according to claim 3, characterized in that: Dynamic weight update function, specifically: in, is the time step The first The first Dynamic weight of each element; is the time step The first The first Dynamic weight of each element; is the learning rate; is the sign function that determines the weight change; is the first feature matrix after dimensionality reduction The first The contribution of each element; It is the average contribution of the feature sequence in the feature matrix after dimensionality reduction and the elements in the key service environment influencing factor set.
5. A composite material stability performance prediction system, characterized in that: include: A data acquisition module is used to collect service environment data and property data of the material to be tested to obtain an initial data set; A key environmental factor extraction module is used to extract service environment data of the material to be tested that affects the stability performance of the material to be tested from the initial data set based on a time series analysis method, and obtain a set of key service environment influencing factors; An input feature vector generation module is used to generate an input feature vector based on an initial data set and a set of key service environment impact factors, combined with a pre-built adaptive feature weight matrix related to environmental impact adaptability; The performance prediction module is used to input the input feature vector into the pre-built composite material stability performance prediction model, and output the stability performance prediction result of the material to be tested; wherein the pre-built composite material stability performance prediction model adopts an integrated recurrent neural network model, and introduces an attention mechanism and a model weight strategy; The process of generating the input feature vector based on the initial data set and the set of key service environment impact factors, combined with the pre-built adaptive feature weight matrix related to environmental impact adaptability, is as follows: Select the property data of the material to be tested from the initial data set and construct the property feature matrix of the material to be tested; Perform dimension reduction processing on the characteristic matrix of the property of the material to be tested to obtain the characteristic matrix after dimension reduction; Based on the reduced dimension feature matrix and the key service environment influencing factor set, an extended feature matrix is constructed; Sparse representation learning is performed on the extended feature matrix to obtain a sparse feature weight matrix; wherein the sparse feature weight matrix is specifically: in, is the sparse feature weight matrix; It is the minimum angle regression analysis algorithm; The L2 norm squared is a measure of the difference between the expanded feature matrix and the product of the expanded feature matrix and the sparse feature weight matrix; Calculate the L1 norm of the sum of the absolute values of vector elements in the sparse feature weight matrix; is the regularization parameter; is the extended feature matrix; Based on the set of key service environment influencing factors, the sparse feature weight matrix is optimized to obtain a pre-constructed adaptive feature weight matrix related to environmental impact adaptability; Using the pre-built adaptive feature weight matrix related to environmental impact adaptability, the reduced-dimensional feature matrix is weighted and reorganized to obtain the input feature vector; Based on the set of key service environment influencing factors, the sparse feature weight matrix is optimized to obtain the pre-built adaptive feature weight matrix related to environmental impact adaptability. The specific steps are as follows: Based on the key service environment influencing factor set, the feature weight elements in the sparse feature weight matrix are optimized to obtain adaptive feature weight elements related to environmental impact adaptability; Combining the adaptive feature weight elements related to the environmental impact adaptability to obtain a pre-constructed adaptive feature weight matrix related to the environmental impact adaptability; Among them, the adaptive feature weight elements related to environmental impact adaptability are: in, is the adaptive feature weight element related to the adaptability of environmental impact, is the row index of the feature matrix after dimensionality reduction, is the element index of the key service environment influencing factor set; is the feature weight element in the sparse feature weight matrix; It is the first key service environment influencing factor. elements; is the mean value of the elements in the key service environment influencing factors; is the standard deviation of the key service environment influencing factors; is the natural exponential function; The integrated recurrent neural network model includes several LSTM sub-models, and the number of LSTM sub-models is the same as the number of feature sequences in the input feature vector; the LSTM sub-model is used to capture the hidden state of the feature sequence in the input feature vector as time steps; Among them, the process of capturing the hidden state of the feature sequence in the input feature vector over time is specifically: in, is the first feature vector in the input The feature sequence at time step Hidden state when It is the long short-term memory network algorithm; is the first feature vector in the input feature sequence; is the first feature vector in the input The feature sequence at time step Hidden state when Attention mechanism, specifically: in, is the first feature vector in the input The feature sequence at time step The attention weight at time ; is the number of feature sequences in the input feature vector; Model weight strategy, specifically: in, After the environmental impact adaptation adjustment The weights of the LSTM sub-models; Assigning functions for adaptive weights of preset environmental impacts; For the The weights of the LSTM sub-models before adjustment; is a set of key service environment influencing factors; is the standard deviation of the key service environment influencing factors set; is the balance coefficient; is the nonlinear adjustment term index.
6. A composite material stability performance prediction device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the composite material stability performance prediction method as described in any one of claims 1 to 4 when executing the computer program.
7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the stability performance of a composite material according to any one of claims 1 to 4 are implemented.
