A method for separating engine speed sensor electromagnetic pulse coupled signals

Abstract

The application discloses a method for separating electromagnetic pulse coupled signals of an engine speed sensor, and belongs to the field of electromagnetic protection of engines. The specific steps are as follows: step 1: obtaining an electromagnetic pulse coupled signal data set through equivalent injection test; step 2: performing short-time Fourier transform to transform the data set to a time-frequency domain to meet the sparsity requirement; step 3: training the transformed data set by using IWE clustering integration to obtain a consensus cluster, and obtaining an estimated confusion matrix according to the consensus cluster result obtained by the clustering integration; step 4: obtaining an electromagnetic pulse coupled signal separation mathematical expression according to the estimated confusion matrix combined with an SKWMP reconstruction algorithm; and step 5: performing inverse short-time Fourier transform to convert the time-frequency domain signal into a time domain signal. The application obtains a mathematical model for separating electromagnetic pulse coupled signals through the IWE clustering integration algorithm and the SKWMP reconstruction algorithm, and provides a reference basis for electromagnetic pulse protection of the engine speed sensor.

Description

Technical Field

[0001] This invention belongs to the field of electromagnetic protection for engine speed sensors, and relates to a method for separating electromagnetic pulse coupling signals of engine speed sensors. Background Technology

[0002] Engine speed is a critical parameter of the engine electronic control system. When the speed sensor is subjected to strong electromagnetic pulse interference, the electromagnetic pulse couples with the normal speed signal to form an electromagnetic pulse coupling signal. This electromagnetic pulse coupling signal is an underdetermined, transient, mixed signal with an unknown number of source signals. Separating the electromagnetic pulse-sensitive frequency points from the sensor signal is crucial for electromagnetic pulse protection of the speed sensor. Traditional filter methods are flexible but cannot adaptively filter out multiple electromagnetic pulse frequencies. Single-stage filters struggle to cover the wideband electromagnetic pulse signal, making it difficult to eliminate interference across all frequency bands. Wideband filter design can lead to excessive redundancy in non-sensitive frequency bands, increasing filter design cost and size. Currently, most filter designs and constructions require prior information about the signal. However, any change in electromagnetic pulse conditions alters the sensitive frequency points of the electromagnetic pulse coupling signal, resulting in a relative lack of prior information. In summary, traditional filter methods are not ideal for separating electromagnetic pulse coupling signals. Therefore, proposing an effective method for separating electromagnetic pulse coupling signals from engine speed sensors is of great significance for electromagnetic pulse protection and active fault tolerance of engine speed sensors. Summary of the Invention

[0003] This invention aims to overcome the shortcomings of existing technologies by providing a method for separating electromagnetic pulse coupling signals from engine speed sensors. This method combines an internal weighted evaluation (IWE) clustering ensemble algorithm with an atom adaptive matching sparse reconstruction (SKWMP) algorithm to separate the sensor reconstruction signal from the electromagnetic pulse coupling signal, thus providing a foundation for electromagnetic pulse protection and active fault tolerance of engine speed sensors.

[0004] This invention is achieved through the following technical solution:

[0005] A method for separating electromagnetic pulse coupling signals from an engine speed sensor includes the following steps:

[0006] Electromagnetic pulse coupling signals were collected through experiments, and a raw dataset was created.

[0007] Time-frequency analysis was performed on the raw dataset of electromagnetic pulse coupling signals to obtain information about the electromagnetic pulse coupling signals in the time-frequency domain.

[0008] The internal weighted evaluation (IWE) clustering ensemble algorithm is used to cluster the time-frequency domain data of electromagnetic pulse coupling signals to obtain a consensus cluster of electromagnetic pulse coupling signals;

[0009] Based on the consensus clustering results of the obtained electromagnetic pulse coupling signals, the aliasing matrix of the electromagnetic pulse coupling signals is estimated.

[0010] The Atomic Adaptive Matching Sparse Reconstruction (SKWMP) algorithm is used to reconstruct the source signals of each component in the aliasing matrix of electromagnetic pulse coupling signals, obtaining the frequency domain source-separated signal. The SKWMP algorithm is an optimization of the Matching Pursuit (MP) algorithm. Considering the unknown sparsity in electromagnetic pulse coupling signal separation and the degradation of reconstruction results due to noise in engineering applications, an atomic adaptive matching sparsity algorithm is proposed based on the Matching Pursuit algorithm. The atomic adaptive matching sparsity algorithm includes constructing singular value decomposition measurement equations, constructing Kalman filter equations, and an atomic adaptive matching strategy.

[0011] The obtained frequency domain source separation signal is inversely transformed to obtain the time domain separation components of the electromagnetic pulse coupling signal.

[0012] Preferably, electromagnetic pulse coupling signals are collected through experiments. The experiments adopt equivalent injection experiments, and the implementation steps are as follows: a programmable signal generator is used to inject multi-frequency damped sinusoidal signals. In the injection experiment, the magnitude, amplitude and number of frequency points of the damped sinusoidal signals are changed to simulate different electromagnetic pulse signals. The injection target is the engine speed sensor, and the injection path is the cable connecting the speed sensor and the engine digital controller.

[0013] Preferably, the original dataset is transformed into the time-frequency domain by performing a short-time Fourier transform.

[0014] Preferably, the Internal Weighted Evaluation (IWE) clustering ensemble algorithm selects several k-means algorithms with different initial values ​​to perform preliminary partitioning of the electromagnetic pulse coupled signal dataset.

[0015] Preferably, the high-quality basis clustering selection threshold in the Internal Weighted Evaluation (IWE) clustering ensemble algorithm is [value missing]. This represents the total number of base clusters generated.

[0016] Preferably, in the Internal Weighted Evaluation (IWE) clustering ensemble algorithm, M is arbitrarily selected from the high-quality base clustering candidate set. C Each base cluster forms a mutual evaluation group, and the uncertainty of all clusters in the mutual evaluation group is calculated through the "cluster evaluates cluster" strategy.

[0017] Preferably, in the Internal Weighted Evaluation (IWE) clustering ensemble algorithm, the graph cut method is selected as the ensemble strategy to obtain consensus clustering.

[0018] Preferably, the estimated aliasing matrix is ​​obtained through consensus clustering results using the Internal Weighted Evaluation (IWE) clustering ensemble algorithm. The expression for the aliasing matrix is:

[0019] A = {a1, a2, ..., a} n}

[0020] a i Let i represent the consensus cluster center, i = 1, 2, ..., n, where n represents the number of source signals.

[0021] Preferably, the atomic adaptive matching sparse reconstruction (SKWMP) algorithm is used to reconstruct the source signals of each component in the electromagnetic pulse coupling signal into a frequency domain source-separated signal. The specific implementation process is as follows: the electromagnetic pulse coupling signal is represented as y (mK×1), the original dimension of the measurement matrix is ​​mK×nK, the IWE clustering ensemble aliasing matrix estimation method yields the estimated matrix A (m×n), and a dimensional transformation of A constructs the measurement matrix required for the compressed sensing mathematical model; the identity matrix E is used... k Adaptive increment / decrement estimation of elements in confusion matrix A:

[0022] B ij =E k A ij

[0023] Among them, B ij This represents the measurement matrix after the aliasing matrix transformation, from which the mathematical expression for electromagnetic pulse coupling signal separation based on compressed sensing can be obtained:

[0024]

[0025] Where s represents the reconstructed signal with a dimension of nK×1.

[0026] Preferably, the time-domain separated components of the electromagnetic pulse coupling signal are obtained by performing an inverse short-time Fourier transform on the separated components of the time-frequency domain electromagnetic pulse coupling signal.

[0027] Beneficial effects: 1. To address the problem that existing clustering ensemble algorithms do not consider the local diversity of each cluster in the base cluster, an Internal Weighted Evaluation (IWE) clustering ensemble algorithm is proposed. This algorithm considers the local diversity of clusters in the dataset, avoids the impact of low-quality clustering on the consensus clustering effect, and improves the accuracy and robustness of the clustering ensemble algorithm.

[0028] 2. The IWE aliasing matrix estimation method is adopted. A high-quality basis cluster candidate set is used instead of the pre-selection process of extracting single source points. By weighting and evaluating the data points of the high-quality basis cluster, and then using the graph cut method integration strategy, a consensus cluster of sparse data points in the time-frequency domain of electromagnetic pulse coupling signal is obtained, which can effectively estimate the aliasing matrix and has high robustness.

[0029] 3. To address the issues of unknown sparsity and complex measurement noise in electromagnetic pulse environments, an atomic adaptive matching sparse reconstruction algorithm is proposed, which avoids the problem of inaccurate signal reconstruction caused by measurement noise and unknown sparsity in engineering applications. Attached Figure Description

[0030] Figure 1 This is a flowchart of the present invention.

[0031] Figure 2 This is a flowchart of the Internal Weighted Evaluation (IWE) clustering ensemble algorithm.

[0032] Figure 3 This is a schematic diagram of the weighted connections of a cluster. Detailed Implementation

[0033] The present invention will now be described in further detail with reference to the accompanying drawings.

[0034] This invention employs a method combining the Internal Weighted Evaluation (IWE) clustering ensemble algorithm with the Atomic Adaptive Matching Sparse Reconstruction (SKWMP) algorithm. The specific working principle is as follows:

[0035] In step 1, electromagnetic pulse coupling signals are acquired through experiments. Specifically, electromagnetic pulse coupling signals are obtained through equivalent injection experiments. A programmable signal generator is used to perform equivalent injection experiments on the engine speed sensor cable. To match the wide spectrum characteristics of real electromagnetic pulses, multi-frequency damped sinusoidal signals are injected. A dataset is created by dividing the acquired electromagnetic pulse coupling signals into training, validation, and test sets in an 8:1:1 ratio.

[0036] In step 2, time-frequency analysis is performed on the original dataset of electromagnetic pulse coupling signals. Specifically, short-time Fourier transform (STFT) is performed on the electromagnetic pulse coupling signals. The stronger the sparsity of each source signal component in the mixed signal, the better the estimation effect of the aliasing matrix. The electromagnetic pulse coupling signals do not meet the sparsity requirements in the time domain, but exhibit good sparsity characteristics in the time-frequency domain.

[0037] Step 3 employs the Internal Weighted Evaluation (IWE) clustering ensemble algorithm to cluster the time-frequency domain data of the electromagnetic pulse coupling signal. The specific implementation process is as follows: Figure 2 As shown, it specifically includes:

[0038] 1) Generating base clusters and forming a high-quality base cluster candidate set: Five k-means algorithms with different initial values ​​are selected to initially partition the electromagnetic pulse coupling signal dataset, generating five different base clusters. High-quality base clusters are selected from the generated base clusters and reorganized into a new base cluster set. The selection criterion for high-quality base clusters is to calculate the Q-value of all base clusters and compare it with a threshold. Base clusters below the threshold are removed, and the remaining base clusters form the high-quality base cluster candidate set. The calculation method is as follows:

[0039]

[0040]

[0041]

[0042] Where, χ a and χ b This represents two unrelated base clusters in the initial base cluster set. Indicates belonging to the basal cluster χ a One of the clusters, Represents the χ² clustering of the base cluster b A cluster in the original dataset contains n samples, n i Each sample belongs to the cluster n j Each sample belongs to the cluster n ij Each sample simultaneously belongs to the cluster and cluster The total number of base clusters generated is given, where Q > 0 and ∑Q = 1. The comparison threshold is set to...

[0043] 2) Evaluate the uncertainty of the cluster: Randomly select M from the high-quality basis clustering candidate set. C Each base cluster forms a mutual evaluation group, and the uncertainty of all clusters in the mutual evaluation group is calculated. A base cluster χ is randomly selected from the high-quality base cluster candidate set. m and a cluster C i C i Relative to χ m The uncertainty is calculated as follows:

[0044]

[0045]

[0046] Where |*| represents the number of data samples contained in *. C i The sum of the uncertainties of all base clusters in a cluster peer review group is the uncertainty of that cluster.

[0047] 3) Consensus clustering is obtained based on the graph cut method strategy. The uncertainty of each cluster is calculated, and the weight of each cluster is also calculated:

[0048]

[0049]

[0050] The graph cut method is a protocol function for obtaining consensus clustering of target data. Based on the obtained weighted connection graph, the weighted connection graph is divided into multiple disjoint sets according to the weighted connection relationship between clusters. Data samples in the same set constitute a new cluster, thus obtaining the final consensus clustering of electromagnetic pulse coupling signals. Figure 3 This is a schematic diagram of the weighted connections of a cluster.

[0051] In step 4, the aliasing matrix of the electromagnetic pulse coupling signal is estimated based on the clustering results. Specifically, in step 3, the IWE clustering ensemble algorithm is used to cluster the time-frequency domain data of the electromagnetic pulse coupling signal. The consensus clustering result obtained from the clustering ensemble represents the number n of source signals. Assume that a consensus cluster center a is obtained. i (i = 1, 2, ..., n), each corresponding to a column vector of the estimated aliasing matrix, thus the estimated aliasing matrix A can be obtained as follows:

[0052] A = {a1, a2, ..., a} n}

[0053] Step 5 utilizes the Atom Adaptive Matching Sparse Reconstruction (SKWMP) algorithm. The source signals of each component in the electromagnetic pulse coupling signal are reconstructed into a frequency domain source-separated signal. Specifically, the electromagnetic pulse coupling signal is represented as y (mK×1), the original dimension of the measurement matrix is ​​mK×nK, and the IWE clustering ensemble aliasing matrix estimation method yields an estimated matrix A (m×n). A dimension transformation of A constructs the measurement matrix required for the compressed sensing mathematical model. The identity matrix E is used. k Adaptive increment / decrement estimation of elements in confusion matrix A:

[0054] B ij =E k A ij

[0055] Among them, B ij This represents the measurement matrix after the aliasing matrix transformation, from which the mathematical expression for electromagnetic pulse coupling signal separation based on compressed sensing can be obtained:

[0056]

[0057] Where s represents the reconstructed signal with a dimension of nK×1.

[0058] In step 6, the time-frequency separated signal is inversely transformed to obtain the time-domain separated components of the electromagnetic pulse coupling signal. Specifically, the inverse transformation refers to the inverse short-time Fourier transform (ISTFT).

[0059] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for separating electromagnetic pulse coupling signals from an engine speed sensor, characterized in that, Includes the following steps: Electromagnetic pulse coupling signals were collected through experiments, and a raw dataset was created. A short-time Fourier transform is performed on the original dataset of electromagnetic pulse coupling signals to obtain the information of electromagnetic pulse coupling signals in the time-frequency domain; An internal weighted evaluation clustering ensemble algorithm is used to cluster the time-frequency domain data of electromagnetic pulse coupling signals to obtain a consensus cluster of electromagnetic pulse coupling signals. Based on the consensus clustering results of the obtained electromagnetic pulse coupling signals, the aliasing matrix of the electromagnetic pulse coupling signals is estimated. The expression for the aliasing matrix is ​​as follows: , where a i Let i represent the consensus cluster center, i = 1, 2, ..., n, where n represents the number of source signals; The source signals of each component in the aliasing matrix of the electromagnetic pulse coupling signal are reconstructed using the atomic adaptive matching sparse reconstruction algorithm to obtain the frequency domain source separation signal. The specific implementation process is as follows: the electromagnetic pulse coupling signal is represented as y (mK×1), the original dimension of the measurement matrix is ​​mK×nK, the estimated matrix A (m×n) is obtained from the estimated aliasing matrix, and the measurement matrix required for the compressed sensing mathematical model is constructed by performing dimension transformation on A. Using the identity matrix E k Adaptive increment / decrement estimation of elements in aliasing matrix A: B ij This represents the measurement matrix after the aliasing matrix transformation; thus, the mathematical expression for electromagnetic pulse coupling signal separation based on compressed sensing is obtained: , where s represents the reconstructed signal with a dimension of nK×1; The obtained frequency domain source separation signal is inversely transformed to obtain the time domain separation components of the electromagnetic pulse coupling signal.

2. The method for separating the electromagnetic pulse coupling signal of the engine speed sensor according to claim 1, characterized in that, Electromagnetic pulse coupling signals were collected through experiments. The experiments adopted equivalent injection experiments. The implementation steps and methods were as follows: a programmable signal generator was used to inject multi-frequency damped sinusoidal signals. During the injection experiment, the magnitude, amplitude and number of frequency points of the damped sinusoidal signals were changed to simulate different electromagnetic pulse signals. The injection target was the engine speed sensor, and the injection path was the cable connecting the speed sensor and the engine digital controller.

3. The method for separating the electromagnetic pulse coupling signal of the engine speed sensor according to claim 1, characterized in that, The internal weighted evaluation clustering ensemble algorithm selects several k-means algorithms with different initial values ​​to perform preliminary partitioning of the electromagnetic pulse coupled signal dataset.

4. The method for separating the electromagnetic pulse coupling signal of the engine speed sensor according to claim 1, characterized in that, The high-quality basis clustering selection threshold in the internal weighted evaluation clustering ensemble algorithm is: , This represents the total number of base clusters generated.

5. The method for separating the electromagnetic pulse coupling signal of the engine speed sensor according to claim 1, characterized in that, In the internal weighted evaluation clustering ensemble algorithm, M is arbitrarily selected from the high-quality base clustering candidate set. C Each base cluster forms a mutual evaluation group, and the uncertainty of all clusters in the mutual evaluation group is calculated through the "cluster evaluates cluster" strategy.

6. The method for separating the electromagnetic pulse coupling signal of the engine speed sensor according to claim 1, characterized in that, In the internal weighted evaluation clustering ensemble algorithm, the graph cut method is selected as the ensemble strategy to obtain consensus clustering.

7. The method for separating the electromagnetic pulse coupling signal of the engine speed sensor according to claim 1, characterized in that, The time-domain components of the electromagnetic pulse coupling signal are obtained by performing an inverse short-time Fourier transform on the separated components in the time-frequency domain.

Images