An expert self-organizing mapping weighted vector machine aircraft electrical signal identification method

Abstract

The present application relates to a kind of expert self-organizing mapping weighted vector machine aircraft electric signal identification method.It includes: electric characteristic classifier offline training module, electric characteristic online monitoring identification module.Historical data are input into classifier offline training module for training (102), the module constructs into historical data set after historical data are preprocessed (103), and artificial error correction (104) is carried out to the data set, constructs into expert data set (105), finally using self-organizing mapping method to the feature extraction (106) of expert data set is completed classifier offline training (107).After classifier offline training ends, test data are input into electric characteristic monitoring identification module (202), and the classification (205) of trained electric characteristic classifier is carried out, and the final identification and interpretation result is output.

Description

Technical Field

[0001] This invention belongs to the field of signal recognition technology and relates to an expert self-organizing map weighted vector machine method for identifying electrical signals of aircraft. Background Technology

[0002] With the continuous development of the aviation industry, the monitoring and identification of aircraft electrical characteristics has become increasingly important. Electrical characteristic monitoring and identification refers to the real-time monitoring and identification of various parameters of an aircraft's electronic systems to ensure its safety and reliability. These parameters include voltage, current, temperature, and frequency, and changes in these parameters can lead to aircraft malfunctions or failures, thus requiring timely monitoring and identification. The monitoring and identification of aircraft electrical characteristics is the core of aircraft health management strategies. By assessing the current health status of complex systems, the causes and sources of failures can be effectively identified, and a series of maintenance and support-related suggestions or decisions can be made.

[0003] Existing technologies for monitoring and identifying the electrical characteristics of aircraft mainly include traditional statistical analysis methods and machine learning methods. Traditional statistical analysis methods are primarily based on empirical formulas and data, but these methods suffer from limitations such as small data sets and low model accuracy. Machine learning methods, on the other hand, utilize algorithms to analyze and learn from large amounts of data to derive predictive results. Support Vector Machines (SVMs) are a commonly used machine learning method that can perform classification and regression analysis on data, exhibiting high accuracy and robustness.

[0004] In recent years, Weighted Approximate Support Vector Machines (SVMs) have been widely used in many fields as a novel machine learning method. SVM is an improved algorithm based on Support Vector Machines (SVMs), which enhances the accuracy and robustness of the model by weighting the sample data. Furthermore, it can continuously update the model through online learning, adapting to different environments and data changes.

[0005] An online aircraft electrical characteristic monitoring and identification method based on weighted approximate support vector machines can achieve real-time monitoring and identification of aircraft electronic systems, thereby improving aircraft safety and reliability. This method can achieve accurate monitoring and identification of aircraft electronic systems by continuously updating the model through real-time acquisition and processing of aircraft electrical characteristic data. Furthermore, this method can improve the accuracy and robustness of the model by weighting the data, thus better adapting to different environments and data changes, ultimately safeguarding the safety and reliability of aircraft. Summary of the Invention

[0006] The purpose of this invention is to address the technical deficiency of low recognition accuracy in existing electrical characteristic monitoring and identification methods by proposing an expert self-organizing mapping weighted vector machine method for identifying aircraft electrical signals.

[0007] Beneficial effects:

[0008] The expert self-organizing map weighted vector machine method for identifying aircraft electrical signals, as described in this invention, has the following advantages compared to existing identification methods:

[0009] 1) The method for monitoring and identifying the electrical characteristics of the aircraft has good robustness and universality.

[0010] 2) Compared with traditional classification methods, this method effectively solves the problem of extracting multi-scale features from high-dimensional signals. The method of this invention can identify and classify signals with different scale features, while improving the recognition accuracy and greatly increasing the calculation speed.

[0011] 3) This method significantly improves the accuracy of classifying and identifying the electrical characteristics of aircraft, making an outstanding contribution to the identification of core faults in aircraft health management strategies. Attached Figure Description

[0012] Figure 1 A flowchart of an expert self-organizing map weighted vector machine method for identifying aircraft electrical signals is shown according to an embodiment of the present invention.

[0013] Figure 2 This refers to the results of monitoring and identifying typical electrical characteristics of spacecraft electronic loads during the specific implementation of the expert self-organizing map weighted vector machine-based spacecraft electrical signal identification method. Detailed Implementation

[0014] The present invention will be further described and illustrated below with reference to the accompanying drawings and embodiments.

[0015] like Figure 1 As shown, an expert self-organizing map weighted vector machine method for identifying aircraft electrical signals according to an embodiment of the present invention includes the following steps:

[0016] S1. Perform offline training on the electrical characteristic classifier (101), specifically: input historical data into the electrical characteristic classifier (102) and perform offline training;

[0017] First, the data is preprocessed (103). The original data is filtered and truncated through data preprocessing to obtain the electrical characteristic event dataset.

[0018] Next, the electrical characteristic event dataset is manually corrected (104), and after removing the more obvious errors, the standard events, i.e. the expert training set, are obtained (105).

[0019] Then, offline training (106) is performed using extracted features from the expert training set to obtain the offline trained electrical property classifier;

[0020] S2. Perform online identification of aircraft electrical signals (201), specifically: input test data into the online identification module and perform online identification (202);

[0021] First, the test data is preprocessed (203). The original data is filtered and truncated through data preprocessing to obtain an online dataset of electrical characteristic events.

[0022] Then, the features of the online dataset of electrical characteristic events are extracted (204), and these category features are input into the offline trained electrical characteristic classifier (205). The electrical characteristic classifier updates the model and adjusts the parameters based on the online dataset, and realizes the identification and interpretation of electrical characteristics.

[0023] The expert-designated self-organizing map weighted vector machine (SVM) method for identifying electrical signals in spacecraft consists of two stages: sample training and electrical characteristic identification. Directly introducing high-dimensional electrical characteristic data into this process can negatively impact the classification speed and accuracy of the electrical characteristic classifier. This invention addresses this issue by extracting features from the original data and reducing its dimensionality. The reduced sample features are then used to train a support vector machine (SVM) electrical characteristic classifier, thereby improving recognition accuracy and significantly increasing computational speed during online electrical characteristic identification. Furthermore, a weighted approximate SVM algorithm is employed to address the issue of excessive noise, effectively improving the recognition rate and resulting in better identification results in the spacecraft electrical characteristic classification and identification system.

[0024] Compared to the standard support vector machine, the weighted approximation vector machine transforms the optimization problem into the following form:

[0025]

[0026] Among them, C i This is represented by introducing different penalty factors for positive and negative classes; ξ i ω is the slack variable; d is the number of samples; ω is the normal vector of the optimal hyperplane; b is the displacement. y is a vector in the sample space; i for The category identifier. i This means that an adjustable weight is introduced for each training sample to represent the contribution of that sample to a certain class, 0 ≤ s i If ≤1, then s i This indicates the probability that the sample point belongs to a class, while (1-s) i ) indicates the probability that the sample point belongs to another class.

[0027] To solve the above optimization problem, we introduce Lagrange's theorem. The problem is transformed into the following form:

[0028]

[0029] Where α=(α1,α2,……α) d ) is a Lagrange multiplier. According to Wolfe's definition of duality, we first find the minimum of L(ω,b,ξ,α) with respect to ω,b,ξ,α. By the extremum condition, we get:

[0030]

[0031]

[0032]

[0033]

[0034] Introducing δ i =C i s i Solving equations (2) and (5) yields:

[0035]

[0036] Substituting equation (5) into equation (1) yields the optimal solution to the dual problem. Choosing a positive component of α*, α* > 0, the solution to the original problem can be expressed as:

[0037]

[0038]

[0039] in The kernel function maps the input space to a corresponding high-dimensional feature space. Introducing different kernel functions allows for the evaluation of similarity and its degree using different criteria. The final decision function is:

[0040]

[0041] Since weighted approximation vector machines (JVMs) are two-class classifiers, and there are many different electrical characteristics in the process of electrical characteristic identification, it is difficult to solve the identification problem using only a single two-class classifier. Therefore, a multi-class weighted approximation vector machine classifier is extracted for online electrical characteristic identification. The guiding principle of multi-class identification is to decompose the multi-class problem into a series of two-class problems that can be directly solved by a series of weighted approximation vector machines. The final discrimination result is obtained based on the solution results of this series of weighted approximation vector machines. Experiments are conducted to compare the results using a one-class complement class method and a one-to-one method.

[0042] The steps of the one-class complement method are to construct n binary classifiers (assuming there are n classes in total). The i-th classifier separates the i-th class from the remaining classes. During training, the i-th classifier takes the i-th class as the positive class and the remaining classes as the negative classes in the training set for training. During discrimination, the test electrical characteristic data is processed by the n classifiers to obtain n output values. If only one +1 appears, its corresponding class is the test electrical characteristic data class. In reality, there is always an error. If there is more than one +1 output (more than one class claims it belongs to itself), or if there is no +1 output (i.e., no class claims it belongs to itself), then the output values ​​are compared, and the class corresponding to the largest one is the test electrical characteristic data class.

[0043] The one-to-one method trains a classifier between every two classes. Therefore, for an electrical characteristic problem with n classes, there will be n(n-1) / 2 classifiers. When classifying the test electrical characteristic data, each classifier determines its class and "votes" for it. The class with the most votes is then chosen as the class for that test electrical characteristic data.

[0044] Self-organizing maps (SMOs), also known as Kohonen networks, are a commonly used dimensionality reduction algorithm. They can map high-dimensional data to a low-dimensional space, achieving dimensionality reduction and feature extraction. The basic idea of ​​SMOs is to map high-dimensional data into a grid structure in a low-dimensional space through competitive and adaptive learning, so that similar data are closer together in the low-dimensional space, while dissimilar data are farther apart.

[0045] Distance calculation formula:

[0046]

[0047] Where di,j represents the distance between node i and node j, and wi,k represents the k-th weight value of node i.

[0048] Update weight formula:

[0049] wi,k(t+1)=wi,k(t)+α(t)hi,j(t)(xk-wi,k(t)) (11)

[0050] Where wi,k(t+1) represents the value of the kth weight of node i at time t+1, α(t) represents the learning rate, hi,j(t) represents the neighborhood function between the winning node i and node j, and xk represents the kth feature value of the input data.

[0051] This embodiment illustrates the process of monitoring and identifying experimental data on typical electrical characteristics of spacecraft electronic loads using the present invention.

[0052] A flowchart of a method for classifying and identifying the electrical characteristics of an aircraft according to an embodiment of the present invention is shown in Figure (1), which includes:

[0053] When the electrical characteristics of an aircraft need to be monitored and identified (201),

[0054] First, the electrical characteristic classifier needs to be trained (101), which means training it using a portion of existing data:

[0055] The historical data is then input into the electrical characteristic classifier (102).

[0056] Next, the historical data is preprocessed (103), and the original data is filtered and truncated to obtain multiple electrical signal waveforms containing 1000 data points. These electrical signal waveforms constitute the training dataset. Then, the training dataset is manually corrected (104). After removing the more obvious errors, the standard events, i.e., the expert dataset, are obtained (105). Then, the self-organizing map feature extraction method is used to reduce the dimensionality of the expert dataset, extract the electrical characteristics, and train the electrical characteristic classifier (106).

[0057] After the electrical characteristic classifier is trained offline (107), it can perform online detection and identification of the electrical characteristics of the aircraft (201). First, test data is input (202), and then the test data is preprocessed (203) to obtain multiple electrical signal waveforms containing 1000 data points. These electrical signal waveforms constitute the test dataset. Then, the self-organizing map feature extraction method is used to extract the features of the training dataset (204). These category features are input into the offline trained electrical characteristic classifier to realize online identification and interpretation of electrical characteristics (205).

[0058] Figure 2 This is the result of monitoring and identifying typical electrical characteristics of spacecraft electronic loads using experimental data from the implementation of the expert self-organizing map weighted vector machine-based spacecraft electrical signal identification method. It can be seen that the weighted approximate support vector machine algorithm is highly feasible in the spacecraft electrical characteristic detection and identification system. Specifically, the one-to-one weighted approximate support vector machine classifier combined with the self-organizing map feature extraction algorithm can obtain the most accurate identification results, effectively identifying events and errors occurring in the electrical characteristics of spacecraft and effectively realizing the health status judgment function.

[0059] The above description is merely a preferred embodiment of the present invention, and the present invention should not be limited to the content disclosed in this embodiment and the accompanying drawings. Any equivalent or modified embodiments made without departing from the spirit of the present invention fall within the scope of protection of the present invention.

Claims

1. A method for identifying electrical signals of an aircraft using an expert self-organizing map weighted vector machine, characterized in that... Includes the following steps: S1. Offline training of the electrical characteristic classifier, including: Historical data are input into the electrical property classifier (102) for offline training; The data is preprocessed (103) by filtering and truncating the original data to obtain an electrical characteristic event dataset; Next, the electrical characteristic event dataset is manually corrected (104), and after removing the more obvious errors, the standard events, i.e. the expert training set, are obtained (105). Then, the feature extraction operation of self-organizing map is used to reduce the dimensionality of the expert dataset, extract the electrical properties, and train the electrical property classifier. S2. Perform online identification of aircraft electrical signals, including: Input the test data into the online recognition module for online recognition (202); The test data is preprocessed (203). The original data is filtered and truncated by data preprocessing to obtain an online dataset of electrical characteristic events. Then, the self-organizing map feature extraction operation is used to extract features from the training dataset (204). These category features are then input into the offline trained electrical property classifier (205). The electrical property classifier updates the model and adjusts the parameters based on the online dataset, and realizes the identification and interpretation of electrical properties. in: A weighted approximation support vector machine is used to overcome the problem of excessive noise. Feature extraction and dimensionality reduction are performed on the original data. The dimensionality-reduced sample features are then used to train a support vector electromechanical characteristic classifier, thereby improving recognition accuracy and computation speed in online electrical characteristic identification. The optimization problem can be transformed into the following form using a weighted approximation vector machine: in, This is represented by introducing different penalty factors for positive and negative classes; These are slack variables; The number of samples; is the normal vector of the optimal hyperplane; It is the displacement; For the sample space vector; for The category identifier, This means that an adjustable weight is introduced for each training sample to represent the contribution of that sample to a certain class. ,but This indicates the probability that the sample point belongs to a class, while This indicates the probability that the sample point belongs to another class.

2. The expert self-organizing map weighted vector machine method for aircraft electrical signal recognition according to claim 1, characterized in that: To solve the above optimization problem, we introduce Lagrange's theorem to transform the problem into the following form: in It is a Lagrange multiplier. According to Wolfe's definition of duality, we first find... about The minimum of is obtained from the extremum condition: Introduction Solving equations (2) and (5) yields: Substituting equation (5) into equation (1) yields the optimal solution to the dual problem. ,choose A positive component Then the solution to the original problem can be expressed as: in Let represent the kernel function, used to map the input space to the corresponding high-dimensional feature space. The final decision function is: 。 3. The expert self-organizing map weighted vector machine method for aircraft electrical signal recognition according to claim 2, characterized in that: By introducing different kernel functions, different standards can be used to evaluate similarity and degree of similarity.

Images