CN119087075B - Electromagnetic interference source identification method based on electromagnetic principle optimization neural network - Google Patents
Electromagnetic interference source identification method based on electromagnetic principle optimization neural network Download PDFInfo
- Publication number
- CN119087075B CN119087075B CN202411182507.3A CN202411182507A CN119087075B CN 119087075 B CN119087075 B CN 119087075B CN 202411182507 A CN202411182507 A CN 202411182507A CN 119087075 B CN119087075 B CN 119087075B
- Authority
- CN
- China
- Prior art keywords
- electromagnetic interference
- electromagnetic
- neural network
- signals
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/001—Measuring interference from external sources to, or emission from, the device under test, e.g. EMC, EMI, EMP or ESD testing
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0499—Feedforward networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Electromagnetism (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an electromagnetic interference source identification method based on an electromagnetic principle optimization neural network, which comprises the following steps of S1, constructing a combined electromagnetic interference signal set under electromagnetic environments with different noise intensities, carrying out frequency conversion, S2, carrying out multidimensional feature extraction on the constructed combined electromagnetic interference signal set, respectively extracting typical feature indexes in a time domain, a frequency domain and an energy domain, preprocessing feature index data, S3, randomly selecting the feature index data extracted in the S2 to form a small sample set, carrying out feature selection on the processed data in the small sample set, reducing feature dimension, preventing an over-fitting phenomenon from occurring in the process of establishing an identification model, effectively reducing the operation cost of the model, S4, constructing an electromagnetic interference source identification model of the electromagnetic principle optimization neural network, carrying out training based on the training set, and S5, carrying out electromagnetic interference source identification by utilizing the electromagnetic interference source identification model obtained by training.
Description
Technical Field
The invention relates to the field of electromagnetism, in particular to an electromagnetic interference source identification method based on an electromagnetic principle optimization neural network.
Background
With the rapid development of electronic technology, electronic and electric devices in a system are increasingly different in variety, and the electronic and electric devices are used as electromagnetic interference sources to form a complex electromagnetic environment, so that the identification work of the electromagnetic interference sources is particularly important for analyzing the electromagnetic compatibility of all electronic and electric devices in the system. The existing electromagnetic interference source identification technology mainly comprises a matched filtering method, a time-frequency analysis method, an algorithm based on machine learning and the like, and the technology can achieve a certain identification accuracy in a certain occasion, but also has the problems of difficult identification of large sample signals, high algorithm noise sensitivity, insufficient generalization capability, low adaptability to unknown signals and the like. In summary, existing electromagnetic interference source identification techniques still have significant shortcomings in processing complex electromagnetic environments and unknown signals. The improvement of the real-time performance, the robustness and the generalization capability of the identification method to better adapt to complex and changeable electromagnetic environments is a key problem to be researched.
Neural networks present significant advantages in the work of electromagnetic interference source identification. The neural network not only has excellent performance in processing complex signals and nonlinear signals, but also has good adaptability and generalization capability, and can keep high recognition accuracy in the face of diversified and complex-change electromagnetic signals. However, the performance of the neural network is highly dependent on the selection of network parameters, including weights, biases, and network structures, and a neural network model with optimized network parameters needs to be built to be able to search for the globally optimal solution of the network parameters effectively.
At present, in the recognition research of electromagnetic interference sources, there is a case of using an optimized neural network as a recognition model, but the corresponding optimization algorithm still has the problems that convergence cannot be achieved within limited iteration times and the like.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides an electromagnetic interference source identification method based on an electromagnetic principle to optimize a neural network, the method comprises the steps of adding background noise to a combined interference signal to simulate a complex electromagnetic environment in actual industrial engineering, and identifying a combined electromagnetic interference signal sample set by using an electromagnetic interference source identification model based on an electromagnetic principle.
The invention aims at realizing the technical scheme that the electromagnetic interference source identification method based on the electromagnetic principle for optimizing the neural network comprises the following steps:
S1, constructing a combined electromagnetic interference signal set under electromagnetic environments with different interference intensities, and performing frequency conversion;
S2, extracting multidimensional features of the constructed combined electromagnetic interference signal set, extracting typical feature indexes in a time domain, a frequency domain and an energy domain respectively, and preprocessing feature index data;
S3, randomly selecting the characteristic index data extracted in the S2 to form a small sample set, carrying out characteristic selection on the processed data in the small sample set, reducing characteristic dimension, preventing the occurrence of over-fitting phenomenon in the process of establishing the recognition model and effectively reducing the operation cost of the model;
s4, constructing an electromagnetic interference source identification model of the optimized neural network based on an electromagnetic principle, and training based on a training set;
s5, performing electromagnetic interference source identification by using the electromagnetic interference source identification model obtained through training.
The method has the beneficial effects that the parameters of the electromagnetic interference identification model are optimized through the optimization algorithm based on the electromagnetic principle, so that the accuracy and the stability of the electromagnetic interference source identification model are improved. The invention also establishes a new emission signal set of the electromagnetic interference source, provides an identification sample for the identification model through a data processing algorithm and a feature extraction algorithm, and more intuitively reflects the features of the time-frequency energy domain of electromagnetic interference.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a time domain waveform diagram of a sample signal with a power ratio of 3dB of an interference signal to background noise;
FIG. 3 is a two-dimensional time-frequency plot of a sample signal with a power ratio of 3dB of an interference signal to background noise;
Fig. 4 is a schematic diagram of a BP neural network topology;
Fig. 5 is a flowchart of electromagnetic interference source recognition algorithm model establishment based on an electromagnetic optimization neural network.
Detailed Description
The technical solution of the present invention will be described in further detail with reference to the accompanying drawings, but the scope of the present invention is not limited to the following description.
Maxwell derived optimization is an optimization algorithm based on the electromagnetic principle, which shows excellent performance in solving the parameter optimization problem, so maxwell derived optimization can be used as an algorithm for optimizing the parameters of the neural network to construct an electromagnetic interference source identification model. The invention discloses an electromagnetic interference source identification technology based on an electromagnetic principle optimized neural network, wherein the identification flow of the technology for an electromagnetic interference emission source is shown in a figure 1, and the electromagnetic interference source identification method based on the electromagnetic principle optimized neural network comprises the following steps:
s1, constructing a combined electromagnetic interference signal set under electromagnetic environments with different noise intensities, and performing frequency conversion;
the invention firstly constructs an electromagnetic interference basic signal set, takes four basic electromagnetic emission signals as individuals of combined signals, numbers the four signals with the number of 1-4 according to the sequence of pulse signals, mismatch signals, analog signals and digital signals, extracts any one, two, three and all four signals with the number of 1-4, carries out random convex combination on the extracted signals, finally constructs the combined signals, and finally constructs the basic signal set of the combined electromagnetic interference signals.
Any one, two, three and all four types of basic signals with the number of 1-4 are extracted, and the extracted single signals and a plurality of signals are subjected to random convex combination to form a combined signal, and the combined signal is constructed by the following specific method through random convex combination:
taking four basic electromagnetic emission signals as an individual of the combined signals, and numbering the four basic signals with numbers 1-4 according to the sequence of pulse signals, mismatch signals, analog signals and digital signals;
any one, two, three and all four types of basic signals with the number of 1-4 are extracted, and the extracted single signals and a plurality of signals are subjected to random convex combination to form a combined signal, and the combined signal is constructed by the following specific method through random convex combination:
First, for the basic signals of the category 1-4, respectively expressed as { h i (t) }, i=1, 2,3,4, a real number u i satisfies the condition 0.2.ltoreq.u i.ltoreq.0.8 and U i is randomly valued under the condition that the condition is satisfied, wherein n is the number of signal categories of a sample vector { w m (t) }, m=1, 2,..15 is a combined signal, and the total of 15 kinds of combined signals are combined signals, and the expression of the basic combined signal is as follows:
wherein p, q=1, 2,3,4, and the number n=q-p+1 of categories of the basic signals contained in the combined signal, the 15 kinds of combined signals constructed are respectively numbered 1-15, and the numbers correspond to sample tag values of each category of sample signals.
Taking the combined signal of the sample class containing all four basic signals as an example, p=1, q=4, m=15 at this time, the basic combined signal expression constructed by random convex combination is as follows:
w15(t)＝u1h1(t)+u2h2(t)+u3h3(t)+u4h4(t).
And (3) performing signal generation on the basic interference signal set, and constructing a combined electromagnetic interference signal x m (t) containing background noise by using Gaussian white noise as environmental background noise epsilon m during signal generation, wherein m=1, 2, 15 has the following expression:
xm(t)＝wm(t)+εm
The effective power of the basic combined signal w m (t) is recorded as P n, the effective power of the environmental background noise epsilon m is recorded as P s, the power ratio of the interference signal to the background noise is S, and an electromagnetic interference signal sample set of an interference signal forming model when S is 3dB, 6dB and 10dB is selected, wherein the calculation formula is as follows:
S=10lg(Ps/Pn)。
Notably, for the basic identification sample, the value of S is infinity. Fig. 2 is a time domain waveform diagram of the electromagnetic interference sample signal set, taking S as an example of 3dB, and fig. 3 is a two-dimensional time-frequency diagram of the electromagnetic interference sample signal set, taking S as an example of 3 dB. The construction of the electromagnetic interference signal sample set provides an input signal for the subsequent optimization of the neural network model based on the electromagnetic principle.
S2, extracting multidimensional features of the combined electromagnetic interference signals in the constructed signal set, extracting typical feature indexes in the time domain, the frequency domain and the energy domain respectively, and preprocessing feature index data;
The method comprises the steps of extracting multi-domain features of electromagnetic interference, extracting peak features, trend features, margin features and pulse features in a time domain by adopting a time domain analysis method, extracting trend features, peak features, center of gravity frequency features, mean square frequency features and frequency variance features in a frequency domain by adopting a fast Fourier transform and wavelet transform method in the process of extracting the frequency domain, wherein the frequency domain peak features comprise main peak features and secondary peak features, the frequency domain trend features comprise main trend features and trend position features, and extracting power spectrum entropy features and energy entropy features in an energy domain by adopting an entropy analysis method in the process of extracting the energy domain features. The extracted l features were noted as X j, j=1, 2,.. the total characteristic number l=14 of the electromagnetic interference signal set in the invention.
According to the invention, the extracted characteristic index data is preprocessed by using a standard score detection method, the outlier is judged based on the deviation between the data points and the mean value, and the outlier is removed, so that the accuracy and the robustness of the model training process are effectively improved, and optimized data is provided for the subsequent characteristic selection process. In the invention, each kind of electromagnetic interference combined signal in the combined electromagnetic interference signal set has 500 samples, and the total signal set contains 7500 samples.
S3, randomly selecting the characteristic index data extracted in the S2 to form a small sample set, carrying out characteristic selection on the processed data in the small sample set, reducing characteristic dimension, preventing the occurrence of over-fitting phenomenon in the process of establishing the recognition model and effectively reducing the operation cost of the model;
in the step S3, sample data of an electromagnetic interference source recognition technology based on an electromagnetic principle is processed, wherein the characteristic value of each sample in the sample data is the high-dimensional characteristic value extracted in the step S2, the label value of the sample is the sample number specified in the step S101, the sample data is required to be subjected to characteristic dimension reduction, a random forest is selected as a characteristic dimension reduction model, the sample characteristics are subjected to dimension reduction, firstly, partial samples in the sample characteristics are required to be extracted, then the extracted samples are taken as a small sample set, and the network model is pre-trained to remove redundant characteristics, so that the phenomenon of fitting in the training process of the optimized network model is prevented.
The feature selection algorithm is based on a random forest model principle, the random forest improves the accuracy and stability of the model by combining the prediction results of a plurality of decision trees, and reduces the overfitting by introducing sample randomness and feature randomness. In the feature selection of the present invention, random forests use two main methods to evaluate feature importance, namely, keni importance I MDI and substitution importance I MDA. The keni importance is the calculation of the sum of the purity reductions that l brings when splitting the nodes in all trees, I MDI, of the characteristics X j, j=1, 2, in the electromagnetic interference signal sample set, as follows:
where N t is the total number of decision trees in the random forest and N t,j is the set of all nodes in the t-th tree split using feature X j. Δg n is the purity reduction that occurs when feature X j at node n splits.
The replacement importance is measured by disturbing the feature value of l in the electromagnetic interference signal sample set, I j, j=1, 2, and the feature importance is measured by observing the model performance change, I MDA has the following calculation formula:
where a t is the classification accuracy or error rate of the t-th tree on the benchmark test set, Is the classification accuracy or error rate on the test set after the use of feature X j in the t-th tree is disrupted.
The method integrates the characteristics to evaluate the importance of each characteristic in the high-dimensional characteristics, selects the characteristic with higher importance score to construct a simplified and effective model, and performs a specific characteristic selection flow by using a random forest model as follows:
(1) A combined electromagnetic interference signal set subjected to data preprocessing in S2 is prepared, 750 samples out of 7500 samples of the combined electromagnetic interference signal set are randomly selected and used as a small sample set X j for feature selection, j=1, 2.
(2) And constructing a feature selection model based on a random forest algorithm, calculating the importance degree of each feature of a small sample set with the feature number of l by taking I MDA and I MDI as measurement standards, drawing an importance histogram, comparing the importance degree of each feature, and evaluating and sequencing the importance degree of each feature.
(3) And (3) pre-training an electromagnetic interference source identification model by using the small sample data to obtain the identification accuracy P m of the electromagnetic interference signal small sample set with the characteristic number of l.
(4) And setting an importance threshold value, removing the characteristics with importance degrees lower than the threshold value in the small sample set, and taking the small sample containing the residual characteristics as pre-training data of the electromagnetic interference source identification model.
(5) Training is carried out by utilizing pre-training data, whether the importance threshold set in the step (4) is reasonable or not is judged by the recognition condition of the small sample set, the step (4) is repeated after the importance threshold is adjusted if the recognition accuracy is far lower than P m, and the cycle is ended until the recognition accuracy reaches P m.
(6) K features reaching the ideal recognition effect are left, and the feature selection flow is finished. The flow realizes the feature compression, compresses the feature number from l to k, eliminates redundant features, prevents the occurrence of the over-fitting phenomenon, and effectively reduces the operation cost of the model.
The invention completes the process of feature selection by pre-training the extracted small samples, compresses the l=14-dimensional features to the k=8-dimensional features by utilizing a feature extraction model based on random forests, effectively reduces the dimensions of the electromagnetic interference signal sample set while ensuring the recognition accuracy, and eliminates redundant features.
S4, constructing an electromagnetic interference source identification model of the optimized neural network based on an electromagnetic principle, and training based on a training set;
the invention establishes an electromagnetic interference source identification model based on an electromagnetic principle optimization neural network, selects the BP neural network as a basic model for identifying an electromagnetic interference emission source, and has strong function approximation and classification capability, good self-organization, self-adaptability and fault tolerance due to the characteristics of a special massive parallel structure, distributed storage and the like of the BP neural network, and the BP neural network has strong capability in identifying the electromagnetic interference source. As an implementation method of the feedforward neural network, the BP algorithm is a back propagation algorithm based on the feedforward neural network structure, and is used for performing parameter training of the neural network, and forward transmission is performed by calculating the weighted sum of each layer of neurons and the activation function output thereof, and the forward transmission is performed from an input layer to an output layer. Fig. 4 presents a schematic view of the neural network topology. The BP network is first analyzed for the functional relationship of the hidden layer, where each neuron contains an activation function that functions to convert the input of the neuron into output, enabling the neural network to capture more complex patterns and relationships.
The back propagation process of the neural network is described below with reference to fig. 4. Let the input of the neural network be x 1,x2,…,xn and the output be y, whereFor the output of the ith neuron in the hidden layer of the r layer, m is the number of neurons of the hidden layer,Weights of input variables to the jth neuron of the next layer are set for the ith,The bias of the jth neuron corresponding to the layer is hidden for the r layer, and b y is the bias of the output layer. The neural network adopted by the invention is a neural network with a single hidden layer, and r=1 and m=q exist for the neural network with the single hidden layer, and at the moment, the forward propagation formula of the neural network model is as follows:
Wherein g (x) is an activation function, since the step function is a discontinuous function, a plurality of troubles are brought to the calculation process when complex mathematical processing is performed, a neural network using the continuous function as an excitation function starts to appear, and the selection of the excitation function becomes various along with the further development of the neural network technology, and the non-linear characteristic is introduced to improve the expression capability of the model. The input of each neuron can be represented by the output of the last neuron, which is a linear combination of the outputs of the last layer. The weight items and the bias items are respectively represented by w and b, b u、by represents the bias items of the hidden layer and the output layer, and w u、wy represents the weight items of the hidden layer and the output layer. The optimization network model selects the neural network with a single hidden layer, wherein an activation function utilized by the hidden layer is a Sigmoid function, and a function utilized by an output layer is a linear function, wherein the expression of the Sigmoid function is as follows:
For the back propagation process, the invention uses the gradient descent method to back propagate the error, and the objective function is the predicted output The deviation from the actual output y, defining an objective function as L, is calculated as follows:
Where k represents the sequence number of the object sample, the gradient descent method and error back propagation are commonly used in combination for BP neural network models, which together form the basis of neural network training. In the process of back propagation of the error signal, the error signal starts to propagate forward layer by layer from the output end, the weight of the network is regulated by error feedback, and the actual output of the network is closer to the expected output through continuous correction of the weight. The following represents the process of network parameter optimization using gradient descent:
Wherein the method comprises the steps of AndRespectively representing bias items of the hidden layer and the output layer after the reverse gradient is reduced; And The weight items of the hidden layer and the output layer after the backward gradient is reduced are respectively shown, and the updating of the parameters shows the process of error back propagation each time. Mu is the learning rate, the step length of each parameter update is controlled, meanwhile, the adjustment degree of the parameter in each iteration is determined, the training is slow due to too small parameter, the vibration or the convergence is impossible due to too large parameter, and the mu is required to be adjusted according to the specific problem. The invention utilizes a single hidden layer neural network, and when an activation function utilized by a hidden layer is a Sigmoid function and an output layer is a linear function, a back propagation formula of an optimized network model is as follows:
The optimization process of the initial weight and bias of the BP neural network applies an optimization algorithm based on an electromagnetic principle, namely an optimization algorithm derived from Maxwell equations. The algorithm optimizes the weight and bias components of the basic model of the BP neural network to obtain an electromagnetic interference source identification model with stronger identification accuracy and stability, a specific algorithm model establishment flow chart is shown in fig. 5, and the identification flow is expressed as follows:
(1) Dividing the established combined electromagnetic interference signal sample set into a training set, a test set and a verification set according to the proportion of 8:1:1, normalizing sample data of the training set to be used as input of the BP neural network, taking a sample label of the training set as network output, and determining a network topological structure.
(2) The method comprises the steps of taking the weight and bias of a neural network as individual corresponding parameters, setting the population scale, initializing MEDO population position and speed, and taking the identification accuracy of the neural network as an fitness function value.
(3) And calculating a fitness function, updating optimal parameters of the individual in the neural network, and adjusting the position and speed of the individual.
(4) Judging whether the training error or the termination condition of the maximum iteration number is reached, stopping iteration if the training error or the termination condition of the maximum iteration number is reached, and returning to the step (3) to continue training if the training error or the termination condition of the maximum iteration number is not reached.
(5) According to the weight and bias optimized by MEDO algorithm as the initial value of neural network training, training the network model by using the sample data of the training set to obtain an optimized neural network model, namely the invention.
(6) And taking the characteristic value of the training set as the input of the trained optimized neural network model to obtain the label prediction result of the training set by the network model, comparing the prediction type result of the training set with the actual sample type, calculating to obtain the recognition accuracy, and verifying the electromagnetic interference source recognition model.
S5, performing electromagnetic interference source identification by using the electromagnetic interference source identification model obtained through training.
The invention uses the part to identify the unknown electromagnetic interference source, the invention selects the verification set and the test set in the combined electromagnetic interference signal set as the unknown interference source, uses the characteristic of the unknown interference source as the input of the electromagnetic interference source identification model, obtains the label prediction result through the identification model, compares the prediction category result of the training set with the actual sample category, calculates to obtain the identification accuracy, and completes the identification of the unknown electromagnetic interference source.
The optimization principle of the optimization algorithm based on the electromagnetic principle in the invention for the parameters of the neural network and the process of identifying the unknown electromagnetic interference source are specifically described below:
MEDO is an optimization algorithm derived based on physical laws in electromagnetic principles and has been applied in a number of practical cases of electromagnetic compatibility. This algorithm concludes by analyzing the current distribution of the coaxial model in the time-varying electromagnetic field. The push to the algorithm principle first reduces the coaxial circuit to a parallel circuit and calculates the current on a particular conductor using the law of conservation of charge and faraday's law. The left hand rule is then applied to convert the current in the constant magnetic field into an ampere force on the conductor. In order to move the conductor to the minimum point of the objective function during the operation, the voltage source needs to be set to be a negative gradient of the objective function at the current point, and the approximation calculation is performed through the current difference value of the adjacent points. The iterative formula for speed and position of MEDO is as follows:
x=x+v·Δt
new cur new
Wherein v new and v cur respectively represent the conductor movement speed values before and after each generation of update, x new and x cur respectively represent the position values of the conductor before and after each generation of update, in the process of optimizing the parameters of the network model, the position value x represents the parameters corresponding to the weight and the bias of the network model, and when the value of the objective function f (x) reaches the minimum in the iteration times, the position parameter x corresponding to the conductor is the initial weight and the bias parameter of the optimized target network model. t is the updated time and represents the number of iterations in the optimization process.
In order to obtain the position parameter x new of the update iteration, the velocity parameter v new of the update iteration needs to be calculated, and values of other parameters in the formula, including the gravity acceleration g, the magnetic flux density B 0, the conductor density ρ, the conductor cross-sectional area S bar and the current i GH flowing through the conductor, need to be set according to the physical principle. Wherein the expression of the current i GH is as follows:
For the parameters of the formula in i GH, such as Z 3 and di 2, are calculated from the Max Wei Daochu equation model. In the process of constructing the optimized neural network model based on the electromagnetic principle, the recognition accuracy f p (x) of the electromagnetic interference signal sample set is used as an objective function f (x), and the optimal weight and parameter value of the network model are calculated through iteration, so that the trained optimized neural network model based on the electromagnetic principle is finally obtained. The recognition process of the electromagnetic interference source can be generalized as a classification problem, the parameters of the neural network are trained through the characteristics and the labels of the training set, and finally the optimized model parameters are obtained, so that a recognition model of the electromagnetic interference source is built, and the model is optimized through calculation of classification accuracy in the process of building the model, so that the accuracy of classifying the electromagnetic interference signal sample by the optimization neural network model based on the electromagnetic principle is calculated in the sample of which the classification model is recognized as a positive example, the proportion of the total sample is correctly classified, the classification accuracy is an objective function f p (x), and the calculation formula is as follows:
Wherein P TP represents the number of correctly classified samples for identifying the model as positive examples, P FP represents the number of incorrectly classified samples for identifying the model as positive examples, and f p (x) represents the ratio of the number of correctly classified samples to the total number of samples, namely, the accuracy of classifying the electromagnetic interference sources by the neural network model is optimized based on the electromagnetic principle, and the accuracy is used as an objective function of an optimization algorithm.
The performance evaluation of the electromagnetic interference source identification model based on the electromagnetic principle is based on the identification effect of the electromagnetic interference source. The algorithm model can obtain the recognition accuracy of the whole electromagnetic interference signal samples, meanwhile, the recognition accuracy of the model for the electromagnetic interference signal samples of each category can be obtained, and the confusion matrix is drawn to obtain the recognition condition of the model for the electromagnetic interference signal samples of all categories. For this multi-classification problem, the confusion matrix of the algorithmic model may exhibit a strong confusion identifying which categories are more serious, helping to improve the classification performance of the model. Meanwhile, the algorithm model can draw ROC curves of various electromagnetic interference signal samples, and calculate the value of AUC corresponding to each curve. The closer the ROC curve is to the upper left corner, the better the performance of the electromagnetic interference source identification model is, the closer the ROC curve is to the diagonal, the poor the performance of the electromagnetic interference source identification model is, and the performance of the electromagnetic interference source identification model can be evaluated through curve trend. Meanwhile, the AUC value is the size of the area under the ROC curve, the maximum value is 1, the method can be used for more intuitively and accurately representing the evaluation of the ROC curve on the performance of the electromagnetic interference source recognition model, and the closer the value is to 1, the better the performance of the classification model is represented, and the closer the value is to 0.5, the poor performance of the classification model is indicated to the prediction effect of the classifier. The recognition accuracy, the confusion matrix, the ROC curve and the AUC value corresponding to the ROC curve of various electromagnetic interference signal samples in the training set, the verification set and the test set are output through the model, the recognition effect of the electromagnetic interference source recognition model can be evaluated in a full-scale multi-dimension mode, and finally a conclusion is obtained, so that the network model has excellent recognition performance.
The electromagnetic interference source identification technology based on the electromagnetic principle improves the identification precision of the electromagnetic interference source, has certain universality and can automatically identify different electromagnetic environments. The technology particularly solves the defects of low accuracy rate, complex identification process and the like of the existing electromagnetic interference source identification, and has the application prospect of engineering industry.
The foregoing is a preferred embodiment of the invention, and it is to be understood that the invention is not limited to the form disclosed herein, but is not to be construed as limited to other embodiments, but is capable of other combinations, modifications and environments and is capable of changes or modifications within the scope of the inventive concept, either as a result of the foregoing teachings or as a result of the knowledge or knowledge of the relevant art. And that modifications and variations which do not depart from the spirit and scope of the invention are intended to be within the scope of the appended claims.
Claims (7)
1. An electromagnetic interference source identification method based on an electromagnetic principle optimization neural network is characterized by comprising the following steps:
s1, constructing a combined electromagnetic interference signal set under electromagnetic environments with different noise intensities, and performing frequency conversion;
S2, extracting multidimensional features of the constructed combined electromagnetic interference signal set, extracting typical feature indexes in a time domain, a frequency domain and an energy domain respectively, and preprocessing feature index data;
s3, randomly selecting the characteristic index data extracted in the S2 to form a small sample set, performing characteristic selection on the processed data in the small sample set, reducing characteristic dimension, and preventing the occurrence of over-fitting phenomenon in the process of establishing the recognition model;
s4, constructing an electromagnetic interference source identification model of the optimized neural network based on the electromagnetic principle, and training based on a training set:
S401, dividing the established electromagnetic interference signal sample set into a training set, a test set and a verification set, normalizing sample data of the training set to be used as input of a neural network model, taking the sample label value of the training set obtained in the S3 as network output, and determining a network topological structure;
S402, setting population scale, initializing MEDO population position and speed by taking the weight and bias of the neural network as individual corresponding parameters, and taking the identification accuracy of the neural network as an fitness function value;
S403, calculating a fitness function, updating optimal parameters of an individual in the neural network, and adjusting the position and speed of the individual;
s404, judging whether a training error or a termination condition of the maximum iteration number is reached, stopping iteration if the training error or the termination condition of the maximum iteration number is reached, and returning to S303 to continue training if the training error or the termination condition of the maximum iteration number is not reached;
S405, training a network model by using sample data of a training set according to the optimal weight and the initial value of the bias to obtain an optimized neural network model;
S406, taking the test set and the verification set as the input of the trained optimized neural network model, obtaining the recognition accuracy of the network model to the test set and the verification set, and verifying the electromagnetic interference source recognition model;
S5, carrying out electromagnetic interference source identification by using the electromagnetic interference source identification model obtained through training:
And taking the characteristics of the unknown interference source as the input of an electromagnetic interference source identification model, and obtaining a label prediction result through the electromagnetic interference source identification model.
2. The method for identifying electromagnetic interference sources based on the optimized neural network of the electromagnetic principle according to claim 1, wherein the step S1 comprises the following steps:
S101, carrying out random convex combination on four types of signals to form a basic signal set for combined electromagnetic interference, wherein the four types of basic electromagnetic emission signals comprise pulse signals, mismatch signals, analog signals and digital signals;
S102, adding background noise to a sample signal of a constructed basic signal set of combined electromagnetic interference to obtain a combined electromagnetic interference signal, and using the combined electromagnetic interference signal as an identification signal of an electromagnetic interference source identification technology based on an electromagnetic principle;
S103, carrying out time-frequency domain transformation on the identification sample to obtain a time domain waveform diagram and a two-dimensional time-frequency diagram of the sample set, and showing time domain and frequency domain characteristics of the sample signal.
3. The method for identifying electromagnetic interference sources based on the optimized neural network of the electromagnetic principle according to claim 2, wherein the step S101 comprises the following steps:
taking four basic electromagnetic emission signals as an individual of the combined signals, and numbering the four basic signals with numbers 1-4 according to the sequence of pulse signals, mismatch signals, analog signals and digital signals;
any one, two, three and all four types of basic signals with the number of 1-4 are extracted, and the extracted single signals and a plurality of signals are subjected to random convex combination to form a combined signal, and the combined signal is constructed by the following specific method through random convex combination:
First, for the basic signals of the category 1-4, respectively expressed as { h i (t) }, i=1, 2,3,4, a real number u i satisfies the condition 0.2.ltoreq.u i.ltoreq.0.8 and U i is randomly valued under the condition that the condition is satisfied, wherein n is the number of signal categories of a sample vector { w m (t) }, m=1, 2,..15 is a combined signal, and the total of 15 kinds of combined signals are combined signals, and the expression of the basic combined signal is as follows:
wherein p, q=1, 2,3,4, and the number n=q-p+1 of categories of the basic signals contained in the combined signal, the 15 kinds of combined signals constructed are respectively numbered 1-15, and the numbers correspond to sample tag values of each category of sample signals.
4. The method for identifying electromagnetic interference sources based on the optimized neural network of the electromagnetic principle according to claim 2, wherein the step S102 comprises the following steps:
signal generation is performed according to the basic interference signal set in S101, and a combined electromagnetic interference signal x m (t) containing background noise is constructed by using gaussian white noise as environmental background noise epsilon m in signal generation, and m=1, 2,..15, the expression is as follows:
xm(t)＝wm(t)+εm
The effective power of the basic combined signal w m (t) is recorded as P n, the effective power of the environmental background noise epsilon m is recorded as P s, the power ratio of the interference signal to the background noise is S, and an electromagnetic interference signal sample set of an interference signal forming model when S is 3dB, 6dB and 10dB is selected, wherein the calculation formula is as follows:
S=10lg(Ps/Pn)。
5. The electromagnetic interference source identification method based on the electromagnetic principle optimization neural network of claim 3, wherein in the step S2, a time domain analysis method is adopted in the process of extracting the time domain peak value characteristics, trend characteristics, margin characteristics and pulse characteristics, wherein the time domain peak value characteristics comprise main peak value characteristics and peak value factor characteristics;
Extracting trend features, peak features, center of gravity frequency features, mean square frequency features and frequency variance features in a frequency domain by adopting a fast Fourier transform and wavelet transform method in the process of extracting the frequency domain features, wherein the peak features in the frequency domain comprise main peak features and secondary peak features, and the trend features in the frequency domain comprise main trend features and trend position features;
the method comprises the steps of extracting power spectrum entropy features and energy entropy features in an energy domain by adopting an entropy analysis method in the process of extracting the features of the energy domain;
the extracted l features were noted as X j, j=1, 2.
6. The method for identifying electromagnetic interference sources based on the electromagnetic principle optimization neural network of claim 5, wherein in the step S2, the extracted characteristic index data is preprocessed by using a standard score detection method, outliers are judged based on deviation between data points and a mean value, and the outliers are eliminated.
7. The method for identifying electromagnetic interference sources based on the optimized neural network of the electromagnetic principle according to claim 1, wherein the step S3 comprises the following steps:
Processing sample data of an electromagnetic interference source recognition technology based on an electromagnetic principle, wherein the characteristic value of each sample in the sample data is the characteristic extracted in S2 after pretreatment, the label value of the sample is the sample number specified in S101, characteristic dimension reduction is required to be carried out on the sample data, a random forest is selected as a model for characteristic dimension reduction, and dimension reduction is carried out on the sample characteristics:
Firstly, extracting part of samples in sample characteristics, taking the extracted samples as a small sample set, and pre-training a network model to remove redundant characteristics, so as to prevent the phenomenon of fitting in the training process of the optimized network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411182507.3A CN119087075B (en) | 2024-08-27 | 2024-08-27 | Electromagnetic interference source identification method based on electromagnetic principle optimization neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411182507.3A CN119087075B (en) | 2024-08-27 | 2024-08-27 | Electromagnetic interference source identification method based on electromagnetic principle optimization neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN119087075A CN119087075A (en) | 2024-12-06 |
| CN119087075B true CN119087075B (en) | 2025-04-25 |
Family
ID=93698515
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411182507.3A Active CN119087075B (en) | 2024-08-27 | 2024-08-27 | Electromagnetic interference source identification method based on electromagnetic principle optimization neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119087075B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119986196B (en) * | 2025-01-23 | 2025-11-04 | 艾博标准技术(深圳)有限公司 | Method for identifying and positioning electromagnetic interference source based on multi-band EMI |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113569742A (en) * | 2021-07-29 | 2021-10-29 | 西南交通大学 | Broadband electromagnetic interference source identification method based on convolutional neural network |
| CN115965401A (en) * | 2022-12-14 | 2023-04-14 | 中电投融和资产管理有限公司 | Electricity price prediction method based on neural network and pattern recognition |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108469560B (en) * | 2018-03-16 | 2020-01-14 | 武汉大学 | Electromagnetic interference objective complexity evaluation method based on rapid S-transform time-frequency space model |
| CN112465267B (en) * | 2020-12-14 | 2025-09-16 | 成都易书桥科技有限公司 | Self-adaptive exponential decay method in crowdsourcing task recommendation |
| WO2023097022A1 (en) * | 2021-11-23 | 2023-06-01 | Strong Force Ee Portfolio 2022, Llc | Ai-based energy edge platform, systems, and methods |
| CN118152839A (en) * | 2022-12-05 | 2024-06-07 | 中国石油化工股份有限公司 | Well testing interpretation model identification method based on deep neural network |
-
2024
- 2024-08-27 CN CN202411182507.3A patent/CN119087075B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113569742A (en) * | 2021-07-29 | 2021-10-29 | 西南交通大学 | Broadband electromagnetic interference source identification method based on convolutional neural network |
| CN115965401A (en) * | 2022-12-14 | 2023-04-14 | 中电投融和资产管理有限公司 | Electricity price prediction method based on neural network and pattern recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN119087075A (en) | 2024-12-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117574274B (en) | A PSO-XGBoost system construction method with hybrid feature screening and hyperparameter optimization | |
| Shih et al. | Grey number prediction using the grey modification model with progression technique | |
| CN109581339B (en) | Sonar identification method based on automatic adjustment self-coding network of brainstorming storm | |
| Qiao et al. | A self-organizing RBF neural network based on distance concentration immune algorithm | |
| CN115238749A (en) | A Modulation Recognition Method Based on Transformer Feature Fusion | |
| CN112766537B (en) | A Short-term Electric Load Forecasting Method | |
| CN115561005B (en) | Chemical process fault diagnosis method based on EEMD decomposition and lightweight neural network | |
| Zhao et al. | An efficient and lightweight approach for intrusion detection based on knowledge distillation | |
| CN119087075B (en) | Electromagnetic interference source identification method based on electromagnetic principle optimization neural network | |
| CN113988215A (en) | Power distribution network metering cabinet state detection method and system | |
| CN117574776A (en) | A model self-learning optimization method for task planning | |
| CN118798742B (en) | A quality assessment method, system, device and medium for equipment spare parts | |
| CN111681132A (en) | A typical power consumption pattern extraction method suitable for massive unbalanced load data | |
| CN118483770B (en) | A wind speed prediction method based on graph embedding and GIN-GRU for multiple wind farm features | |
| CN118690245A (en) | Radar individual recognition method based on multi-channel attention mechanism convolution model | |
| CN109901064B (en) | ICA-LVQ-based high-voltage circuit breaker fault diagnosis method | |
| dos Santos Coelho et al. | Nonlinear identification using a B-spline neural network and chaotic immune approaches | |
| CN117217809A (en) | A short-term commodity demand forecasting method based on cross-sequence | |
| CN111090679A (en) | Temporal Data Representation Learning Method Based on Temporal Influence and Graph Embedding | |
| CN120320329A (en) | A distribution network power flow calculation method, device and equipment based on deep learning | |
| Putrada et al. | Nocasc: A novel optimized cost-complexity pruning for adaboost model compression on edge computing-based smart lighting | |
| CN116342958B (en) | A global prototype-based small-sample node classification method based on auxiliary graph enhancement | |
| CN117592825A (en) | Voltage sag assessment method based on graph convolutional neural network | |
| Shi et al. | PCA-based PSO-BP neural network optimization algorithm | |
| CN116527346B (en) | Threat node perception method based on deep learning graph neural network theory |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |