Electromagnetic compatibility analysis method based on deep learning

Abstract

The invention discloses an electromagnetic compatibility analysis method based on deep learning, and belongs to the technical field of data analysis, and the method comprises the steps: constructing a training sample through employing sample electromagnetic compatibility data, constructing a training label through a state corresponding to the sample electromagnetic compatibility data, and obtaining training data; then training the deep learning model by adopting a multi-order information fusion optimization algorithm on the basis of the training data to obtain a trained deep learning model; and finally, collecting to-be-analyzed electromagnetic compatibility data, scheduling the trained deep learning model to analyze the to-be-analyzed electromagnetic compatibility data, and determining an electromagnetic compatibility analysis result, thereby reducing the complexity of electromagnetic compatibility analysis, and improving the accuracy, timeliness and operability of electromagnetic compatibility fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of data analysis technology, specifically relating to an electromagnetic compatibility analysis method based on deep learning. Background Technology

[0002] Electromagnetic compatibility (EMC) refers to the ability of a device or system to function normally in its electromagnetic environment without causing unacceptable electromagnetic interference to anything in that environment. As electronic technology advances towards higher frequencies, higher speeds, and higher integration, the electromagnetic environment inside electronic devices is becoming increasingly complex. EMC issues have become a key factor restricting product development cycles, affecting product reliability, and impacting market competitiveness. Currently, EMC analysis mainly relies on analytical or numerical calculations based on electromagnetic theories such as Maxwell's equations. While this method offers high accuracy, for modern electronic devices with complex structures and numerous nonlinear components, the calculation process is extremely complex, computationally intensive, and time-consuming, making rapid iteration difficult in the early stages of product design. Summary of the Invention

[0003] This invention provides a deep learning-based electromagnetic compatibility analysis method to solve the problems of extremely complex calculation processes, huge computational loads, and extremely long processing times in existing technologies.

[0004] This application provides a deep learning-based electromagnetic compatibility analysis method, including: Acquire electromagnetic compatibility data of samples under different states; wherein, the different states include compatible states or incompatible states; Training samples are constructed using the electromagnetic compatibility data of the samples, and training labels are constructed using the states corresponding to the electromagnetic compatibility data of the samples to obtain training data. Based on the training data, a multi-level information fusion optimization algorithm is used to train the deep learning model to obtain the trained deep learning model. Collect electromagnetic compatibility (EMC) data to be analyzed, and schedule the trained deep learning model to analyze the EMC data to determine the EMC analysis results.

[0005] In some possible implementations, the step of constructing training samples from the sample electromagnetic compatibility data, constructing training labels from the states corresponding to the sample electromagnetic compatibility data, and obtaining training data includes: The peak features, envelope features, and / or wavelet packet energy features of the sample electromagnetic compatibility data are extracted to obtain preprocessed sample electromagnetic compatibility data. The preprocessed sample electromagnetic compatibility data is used as training samples, and training labels are constructed from the states corresponding to the sample electromagnetic compatibility data to obtain training data.

[0006] In some possible implementations, the step of training a deep learning model using a multi-order information fusion optimization algorithm based on the training data to obtain the trained deep learning model includes: The training population is initialized based on the hyperparameters of the deep learning model; wherein the training population includes multiple different individuals, and each individual includes all or some of the hyperparameters of the deep learning model. Based on the training data, the root mean square error function value corresponding to each individual in the training population is obtained, and the individual with the smallest root mean square error function value is determined as the first optimal individual. Based on the first optimal individual, a boundary information fusion search strategy is used to perform a first-order search on the individual in the solution space to determine the individual after the first-order search. For individuals after the first-order search, a dual-state information fusion search strategy is used to perform a second-order search on the individuals after the first-order search in the solution space to determine the individuals after the second-order search. For individuals after the second-order search, a population information fusion search strategy is used to perform a third-order search on the individuals after the second-order search in the solution space to determine the individuals after the third-order search. For individuals after the third-order search, a random position fusion search strategy is used to perform a fourth-order search on the individuals after the third-order search to determine the individuals after the fourth-order search. Determine whether the training termination condition is met. If so, determine the second optimal individual based on the individuals after the fourth-order search. Otherwise, return to the step of obtaining the first optimal individual. Based on the second optimal individual, obtain the trained deep learning model.

[0007] In some possible implementations, initializing the training population based on the hyperparameters of the deep learning model includes: The hyperparameters of the deep learning model are initialized using a random initialization method or a chaotic mapping initialization method to obtain the training population.

[0008] In some possible implementations, the step of using the first optimal individual as a basis and employing a boundary information fusion search strategy to perform a first-order search on the individual in the solution space to determine the individual after the first-order search includes: Determine the current training iteration count, and based on the current training iteration count, determine the adaptive information fusion factor as follows: in, Represents the adaptive information fusion factor. This represents the maximum value of the adaptive information fusion factor. This represents the minimum value of the adaptive information fusion factor. This indicates the preset maximum number of training iterations. t Indicates the current number of training iterations. Represents the first random number between (0,1). Represented by natural constant e An exponential function with base 0. Indicates control parameters, and ln represents the logarithmic function; Based on the adaptive information fusion factor and the first optimal individual, a first-order search is performed on the individual in the solution space to determine the individual after the first-order search: in, Indicates the first t During the training process, the first m Individual, m =1,2,…,M, where M represents the total number of individuals in the training population. Indicates the first m The individual after a first-order search; Represents (0,2) π Random angles between ) π Represents pi (π). This represents the upper limit individual, that is, the individual composed of the upper limit of each hyperparameter dimension; This represents the lower bound individual, that is, the individual composed of the lower bounds of each hyperparameter dimension; This represents the second random number between (0,1). Represents the sine function. Represents the cosine function. This represents a third random number between (0,1). This represents the fourth random number between (0,1). This represents the optimal individual.

[0009] In some possible implementations, the step of employing a dual-state information fusion search strategy to perform a second-order search on the individuals after the first-order search within the solution space to determine the individuals after the second-order search includes: The control factor for obtaining the two-state search is: in, Indicates the first t Dual-state search control factor during the training process. Indicates the first t The two-state search control factor during training iteration +1, initially set to (0, 2...). π Random angles between ) Based on the dual-state search control factor, a second-order search is performed on the individuals after the first-order search in the solution space to determine the individuals after the second-order search as follows: in, Indicates the first t During the training process, the first n The individual after a first-order search, n =1,2,…,M, Indicates the first n The individual after a second-order search, Represents an individual The corresponding historical optimal state.

[0010] In some possible implementations, the step of using a population information fusion search strategy to perform a third-order search on the individuals after the second-order search within the solution space to determine the individuals after the third-order search includes: Determine the current training iterations, and based on the current training iterations, determine the adaptive population fusion factor as follows: in, Represents the adaptive population fusion factor. This represents the maximum value of the adaptive population fusion factor. This represents the minimum value of the adaptive population fusion factor; For the individual obtained after the second-order search, N other individuals are randomly matched to obtain the information fusion individual corresponding to the individual obtained after the second-order search; where N is less than M. Based on the adaptive population fusion factor and the information fusion individuals, a third-order search is performed on the individuals after the second-order search in the solution space to determine the individuals after the third-order search as follows: in, Indicates the first t During the training process, the first i The individual after the second-order search d dimensional hyperparameters, i =1,2,…,M, d =1,2,…,D, where D represents the total dimension of hyperparameters in an individual. Indicates the first i The individual after the third-order search d dimensional hyperparameters, The lower limit individual's first d dimensional hyperparameters, The upper limit of the individual's number d dimensional hyperparameters, This indicates the control parameters for the first information fusion range. This indicates the control parameter for the second information fusion range. Indicates the first t During the training process, the first i The individual after the second-order search and its corresponding first-order search h Euclidean distance between individuals involved in information fusion Indicates the first t During the training process, the first i The individual corresponding to the second-order search h The first information fusion individual d dimensional hyperparameters, The first element representing the optimal individual d Dimensional hyperparameters.

[0011] In some possible implementations, the step of using a random position fusion search strategy to perform a fourth-order search on individuals after the third-order search, and determining individuals after the fourth-order search, includes: Based on the adaptive population fusion factor, the first global search control factor is determined as follows: in, Indicates the first global search control factor; Based on the first global search control factor, a fourth-order search is performed on the individuals after the third-order search to determine the individuals after the fourth-order search as follows: in, Indicates the first t During the training process, the first j The individual after the third-order search d dimensional hyperparameters, j =1,2,…,M, Indicates the first j The individual after the fourth-order search d dimensional hyperparameters, This represents the second global search control factor, which is set as a constant term. Determine whether the root mean square error function value corresponding to the individual after the fourth-order search decreases. If so, accept the fourth-order search; otherwise, reject the fourth-order search.

[0012] In some possible implementations, obtaining the trained deep learning model based on the second optimal individual includes: The hyperparameters of the second optimal individual are used as the target hyperparameters of the deep learning model, and the target hyperparameters are applied to the deep learning model to obtain the trained deep learning model.

[0013] In some possible implementations, the step of collecting electromagnetic compatibility (EMC) data to be analyzed and scheduling the trained deep learning model to analyze the EMC data to determine the EMC analysis results includes: Collect the electromagnetic compatibility data to be analyzed, and extract the peak features, envelope features and / or wavelet packet energy features of the electromagnetic compatibility data to be analyzed to obtain the preprocessed electromagnetic compatibility data to be analyzed; The preprocessed electromagnetic compatibility data to be analyzed is input into the trained deep learning model for identification, and electromagnetic compatibility analysis results are obtained.

[0014] This invention provides a deep learning-based electromagnetic compatibility (EMC) analysis method. First, training samples are constructed using sample EMC data. Training labels are then built based on the states corresponding to the sample EMC data to obtain training data. Next, a multi-level information fusion optimization algorithm is used to train a deep learning model based on the training data to obtain the trained deep learning model. Finally, the EMC data to be analyzed is collected, and the trained deep learning model is used to analyze the data to determine the EMC analysis results. This method reduces the complexity of EMC analysis and improves the accuracy, timeliness, and ease of operation of EMC fault diagnosis. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0016] Figure 1 This is a flowchart of an electromagnetic compatibility analysis method based on deep learning, provided as an embodiment of the present invention.

[0017] Figure 2 This is a flowchart for obtaining a trained deep learning model, provided as an embodiment of the present invention.

[0018] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0019] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0020] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] like Figure 1 As shown, this embodiment of the invention provides an electromagnetic compatibility analysis method based on deep learning, including: S101. Obtain electromagnetic compatibility data of samples under different states; wherein, different states include compatible states or incompatible states.

[0022] Electromagnetic compatibility (EMC) testing can check whether the electromagnetic signals emitted into space by an antenna port in the 25Hz~30MHz frequency band comply with specifications, or whether the electromagnetic signals radiated into space in the 25Hz~40GHz frequency band comply with specifications. Therefore, sample EMC data can be the electromagnetic signals emitted by the antenna port. Because EMC testing is highly susceptible to environmental influences, related tests are usually performed in a shielded room to avoid interference from external factors. Therefore, sample EMC data can be obtained in a shielded room under both compatible and incompatible conditions.

[0023] S102. Construct training samples using sample electromagnetic compatibility data, construct training labels using the states corresponding to the sample electromagnetic compatibility data, and obtain training data.

[0024] In the process of constructing training samples using sample electromagnetic compatibility data, the sample electromagnetic compatibility data can also be preprocessed to improve the accuracy of data recognition.

[0025] S103. Based on the training data, a multi-level information fusion optimization algorithm is used to train the deep learning model to obtain the trained deep learning model.

[0026] In existing technologies, gradient descent and backpropagation are commonly used to train deep learning models. While these methods can complete training to some extent, they can also lead to local optima, causing the trained deep learning model to be unable to accurately identify electromagnetic compatibility data. Therefore, this application provides a multi-level information fusion optimization algorithm for training deep learning models, which solves the problem of local optima in existing technologies and improves the accuracy of the trained deep learning model in identifying electromagnetic compatibility data.

[0027] S104. Collect the electromagnetic compatibility data to be analyzed, and schedule the trained deep learning model to analyze the electromagnetic compatibility data to determine the electromagnetic compatibility analysis results.

[0028] This application provides an electromagnetic compatibility (EMC) analysis method based on deep learning. By using deep learning technology to perform EMC analysis on electromagnetic signals, the complexity of EMC analysis is reduced, and the accuracy, timeliness, and ease of operation of EMC fault diagnosis are improved.

[0029] In some possible implementations, training samples are constructed using sample electromagnetic compatibility data, and training labels are constructed using the states corresponding to the sample electromagnetic compatibility data to obtain training data, including: Peak features, envelope features, and / or wavelet packet energy features of the sample electromagnetic compatibility data are extracted to obtain preprocessed sample electromagnetic compatibility data. The preprocessed sample electromagnetic compatibility data is then used as training samples, and training labels are constructed from the states corresponding to the sample electromagnetic compatibility data to obtain training data.

[0030] The preprocessed sample electromagnetic compatibility data can be constructed by using peak features, envelope features, and wavelet packet energy features together, or by using only one feature, or by using all features together.

[0031] like Figure 2 As shown, based on the training data, a multi-order information fusion optimization algorithm is used to train the deep learning model, obtaining the trained deep learning model, including: S201. Initialize the training population based on the hyperparameters of the deep learning model; wherein, the training population includes multiple different individuals, and each individual includes all or some of the hyperparameters of the deep learning model. As can be seen, the multi-order information fusion optimization algorithm provided in this application can be used to train all the hyperparameters of the deep learning model, or it can be used to train some of the hyperparameters of the deep learning model.

[0032] S202. Based on the training data, obtain the root mean square error function value for each individual in the training population, and determine the individual with the smallest root mean square error function value as the first optimal individual. S203. Based on the first optimal individual, a boundary information fusion search strategy is used to perform a first-order search on the individual in the solution space to determine the individual after the first-order search. S204. For individuals after the first-order search, a dual-state information fusion search strategy is adopted to perform a second-order search on the individuals after the first-order search in the solution space to determine the individuals after the second-order search. S205. For individuals after the second-order search, a population information fusion search strategy is used to perform a third-order search on the individuals after the second-order search in the solution space to determine the individuals after the third-order search. S206. For individuals after the third-order search, a random position fusion search strategy is used to perform a fourth-order search on the individuals after the third-order search to determine the individuals after the fourth-order search. S207. Determine whether the training termination condition is met. If yes, determine the second optimal individual based on the individuals after the fourth-order search. Otherwise, return to the step of obtaining the first optimal individual. Optionally, the training termination condition can be defined as follows: the training termination condition is met when the current number of training iterations is greater than or equal to the maximum number of training iterations.

[0033] S208. Based on the second best individual, obtain the trained deep learning model.

[0034] Optionally, after each search of an individual, out-of-bounds processing can be performed to ensure that the individual is always valid, thereby enabling the trained deep learning model to perform data recognition tasks normally.

[0035] Existing technologies typically employ gradient descent and backpropagation to train deep learning models. While these methods can complete training to some extent, they can also lead to local optima, causing the trained deep learning model to be unable to accurately identify electromagnetic compatibility (EMC) data. Therefore, this application provides a multi-level information fusion optimization algorithm for training deep learning models, addressing the problem of existing technologies easily getting trapped in local optima and improving the accuracy of the trained deep learning model in identifying EMC data.

[0036] In some possible implementations, the training population is initialized based on the hyperparameters of the deep learning model, including: The hyperparameters of the deep learning model are initialized using random initialization or chaotic mapping initialization methods to obtain the training population.

[0037] Optionally, the deep learning model can be set to a convolutional neural network.

[0038] The random initialization method can be as follows: perform random initialization between the upper and lower bounds of the hyperparameters of the deep learning model, encode the initialized hyperparameters into vectors to obtain individuals, and repeat the initialization and encoding multiple times to obtain a training population composed of multiple individuals.

[0039] The initialization method for chaotic mapping can be as follows: first, randomly generate an individual, and then, based on the randomly generated individual, use a chaotic sequence to obtain multiple other individuals to obtain the training population.

[0040] In some possible implementations, based on the first optimal individual, a boundary information fusion search strategy is used to perform a first-order search on the individual in the solution space to determine the individuals after the first-order search, including: Determine the current training iteration count, and based on the current training iteration count, determine the adaptive information fusion factor as follows: in, Represents the adaptive information fusion factor. This represents the maximum value of the adaptive information fusion factor, which can be set to 1; This represents the minimum value of the adaptive information fusion factor, which can be set to 0.1; This indicates the preset maximum number of training iterations. t Indicates the current number of training iterations. Represents the first random number between (0,1). Represented by natural constant e An exponential function with base 0. Indicates control parameters, and ln represents the logarithmic function; Based on the adaptive information fusion factor and the first optimal individual, a first-order search is performed on the individual in the solution space to determine the individual after the first-order search: in, Indicates the first t During the training process, the first m Individual, m =1,2,…,M, where M represents the total number of individuals in the training population. Indicates the first m The individual after a first-order search; Represents (0,2) π Random angles between ) π Represents pi (π). This represents the upper limit individual, that is, the individual composed of the upper limit of each hyperparameter dimension; This represents the lower bound individual, that is, the individual composed of the lower bounds of each hyperparameter dimension; This represents the second random number between (0,1). Represents the sine function. Represents the cosine function. This represents a third random number between (0,1). This represents the fourth random number between (0,1). This represents the optimal individual.

[0041] The boundary information fusion search strategy provided in this application can combine boundary information to achieve neighborhood search. In the early stage of the algorithm, random information is used for search, while in the later stage of the algorithm, the information of the first optimal individual is more inclined to be used for search. This helps to improve the random search capability in the early stage of the algorithm and the convergence capability in the later stage of the algorithm.

[0042] In some possible implementations, for individuals after the first-order search, a dual-state information fusion search strategy is used to perform a second-order search on these individuals in the solution space to determine the individuals after the second-order search, including: The two-state search control factor is obtained as follows: in, Indicates the first t Dual-state search control factor during the training process. Indicates the first t The two-state search control factor during training iteration +1, initially set to (0, 2...). π Random angles between ) Based on the two-state search control factor, a second-order search is performed on the individuals after the first-order search in the solution space to determine the individuals after the second-order search as follows: in, Indicates the first t During the training process, the first n The individual after a first-order search, n =1,2,…,M, Indicates the first n The individual after a second-order search, Represents an individual The corresponding historical optimal state, i.e., the individual The state in which the root mean square error function value is minimized during the historical training process.

[0043] The dual-state information fusion search strategy provided in this application embodiment can also effectively balance global search and local search, while learning the information of the historical best state and the first best individual, and searching with irregular paths and step sizes, which has stronger search capabilities and can effectively avoid the algorithm getting stuck in local optima. At the same time, the algorithm has better convergence ability in the later stage, which further enhances the local search capability.

[0044] In some possible implementations, for individuals obtained after the second-order search, a population information fusion search strategy is used to perform a third-order search on these individuals in the solution space to determine the individuals obtained after the third-order search, including: Determine the current training iterations, and based on the current training iterations, determine the adaptive population fusion factor as follows: in, Represents the adaptive population fusion factor. This represents the maximum value of the adaptive population fusion factor, which can be set to 1; This represents the minimum value of the adaptive population fusion factor, which can be set to 0.001; For the individual obtained after the second-order search, N other individuals are randomly matched to obtain the information fusion individual corresponding to the individual obtained after the second-order search; where N is less than M. Based on the adaptive population fusion factor and the information fusion individuals, a third-order search is performed on the individuals after the second-order search in the solution space to determine the individuals after the third-order search as follows: in, Indicates the first t During the training process, the first i The individual after the second-order search d dimensional hyperparameters, i =1,2,…,M, d =1,2,…,D, where D represents the total dimension of hyperparameters in an individual. Indicates the first i The individual after the third-order search d dimensional hyperparameters, The lower limit individual's first d dimensional hyperparameters, The upper limit of the individual's number d dimensional hyperparameters, This represents the control parameter for the first information fusion range, which can be set to a constant between (-1, 1) (e.g., -0.9 or 0.9). This represents the control parameter for the second information fusion range, which can be set to a constant between (-1, 1) (e.g., -0.5 or 0.5). Indicates the first t During the training process, the first i The individual after the second-order search and its corresponding first-order search h Euclidean distance between individuals involved in information fusion Indicates the first t During the training process, the first i The individual corresponding to the second-order search h The first information fusion individual d dimensional hyperparameters, The first element representing the optimal individual d Dimensional hyperparameters.

[0045] The population information fusion search strategy provided in this application embodiment can enable individuals to update based on the information of the first optimal individual and other individuals, effectively avoid search path collisions during the search process, and have stronger global search capabilities in the early stage of the algorithm and fine search capabilities in the later stage of the algorithm, thereby improving the algorithm's search capabilities for unexplored areas.

[0046] In some possible implementations, for individuals after the third-order search, a random position fusion search strategy is used to perform a fourth-order search on these individuals to determine the individuals after the fourth-order search, including: Based on the adaptive population fusion factor, the first global search control factor is determined as follows: in, Indicates the first global search control factor; Based on the first global search control factor, a fourth-order search is performed on the individuals after the third-order search to determine the individuals after the fourth-order search as follows: in, Indicates the first t During the training process, the first j The individual after the third-order search d dimensional hyperparameters, j =1,2,…,M, Indicates the first j The individual after the fourth-order search d dimensional hyperparameters, This represents the second global search control factor, which is set as a constant term and can be set to 2. Determine whether the root mean square error function value corresponding to the individual after the fourth-order search decreases. If so, accept the fourth-order search; otherwise, reject the fourth-order search.

[0047] Optionally, the fourth-order search can be controlled by simulated annealing, which can ensure that the algorithm has a strong search capability in the early stage.

[0048] The random location fusion search strategy provided in this application embodiment can effectively improve the global search capability of the algorithm. As training progresses, the global search capability gradually weakens and is converted into a local fine search. The introduction of the greedy strategy can effectively ensure the training speed of the algorithm. Combining the above strategies can effectively improve the training effect of the algorithm.

[0049] This application provides a multi-level information fusion optimization algorithm that improves the search effect of the algorithm layer by layer, comprehensively enhancing the global search capability of the algorithm while ensuring the local search capability, thus guaranteeing the convergence accuracy of the algorithm.

[0050] In some possible implementations, the trained deep learning model is obtained based on the second-best individual, including: The hyperparameters of the second optimal individual are used as the target hyperparameters of the deep learning model, and the target hyperparameters are applied to the deep learning model to obtain the trained deep learning model.

[0051] In some possible implementations, electromagnetic compatibility (EMC) data to be analyzed is collected, and a trained deep learning model is scheduled to analyze the EMC data to determine the EMC analysis results, including: Collect the electromagnetic compatibility data to be analyzed, and extract the peak features, envelope features and / or wavelet packet energy features of the electromagnetic compatibility data to be analyzed to obtain the preprocessed electromagnetic compatibility data to be analyzed; The preprocessed electromagnetic compatibility data to be analyzed is input into the trained deep learning model for identification, and the electromagnetic compatibility analysis results are obtained.

[0052] This invention provides a deep learning-based electromagnetic compatibility (EMC) analysis method. First, training samples are constructed using sample EMC data. Training labels are then built based on the states corresponding to the sample EMC data to obtain training data. Next, a multi-level information fusion optimization algorithm is used to train a deep learning model based on the training data to obtain the trained deep learning model. Finally, the EMC data to be analyzed is collected, and the trained deep learning model is used to analyze the data to determine the EMC analysis results. This method reduces the complexity of EMC analysis and improves the accuracy, timeliness, and ease of operation of EMC fault diagnosis.

[0053] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0057] Those skilled in the art will understand that all or part of the steps in the above facts and methods can be implemented by a program instructing related hardware. The program or the program described therein can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: at this time, the corresponding method steps are introduced. The storage medium can be ROM / RAM, magnetic disk, optical disk, etc.

[0058] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A deep learning-based electromagnetic compatibility analysis method, characterized in that, include: Acquire electromagnetic compatibility data of samples under different states; wherein, the different states include compatible states or incompatible states; Training samples are constructed using the electromagnetic compatibility data of the samples, and training labels are constructed using the states corresponding to the electromagnetic compatibility data of the samples to obtain training data. Based on the training data, a multi-level information fusion optimization algorithm is used to train the deep learning model to obtain the trained deep learning model. Collect electromagnetic compatibility (EMC) data to be analyzed, and schedule the trained deep learning model to analyze the EMC data to determine the EMC analysis results.

2. The electromagnetic compatibility analysis method based on deep learning according to claim 1, characterized in that, The step of constructing training samples using the sample electromagnetic compatibility data, constructing training labels using the states corresponding to the sample electromagnetic compatibility data, and obtaining training data includes: The peak features, envelope features, and / or wavelet packet energy features of the sample electromagnetic compatibility data are extracted to obtain preprocessed sample electromagnetic compatibility data. The preprocessed sample electromagnetic compatibility data is used as training samples, and training labels are constructed from the states corresponding to the sample electromagnetic compatibility data to obtain training data.

3. The electromagnetic compatibility analysis method based on deep learning according to claim 1, characterized in that, The step of training a deep learning model using the training data and employing a multi-level information fusion optimization algorithm to obtain the trained deep learning model includes: The training population is initialized based on the hyperparameters of the deep learning model; wherein the training population includes multiple different individuals, and each individual includes all or some of the hyperparameters of the deep learning model. Based on the training data, the root mean square error function value corresponding to each individual in the training population is obtained, and the individual with the smallest root mean square error function value is determined as the first optimal individual. Based on the first optimal individual, a boundary information fusion search strategy is used to perform a first-order search on the individual in the solution space to determine the individual after the first-order search. For individuals after the first-order search, a dual-state information fusion search strategy is used to perform a second-order search on the individuals after the first-order search in the solution space to determine the individuals after the second-order search. For individuals after the second-order search, a population information fusion search strategy is used to perform a third-order search on the individuals after the second-order search in the solution space to determine the individuals after the third-order search. For individuals after the third-order search, a random position fusion search strategy is used to perform a fourth-order search on the individuals after the third-order search to determine the individuals after the fourth-order search. Determine whether the training termination condition is met. If so, determine the second optimal individual based on the individuals after the fourth-order search. Otherwise, return to the step of obtaining the first optimal individual. Based on the second optimal individual, obtain the trained deep learning model.

4. The electromagnetic compatibility analysis method based on deep learning according to claim 3, characterized in that, The initialization of the training population based on the hyperparameters of the deep learning model includes: The hyperparameters of the deep learning model are initialized using a random initialization method or a chaotic mapping initialization method to obtain the training population.

5. The electromagnetic compatibility analysis method based on deep learning according to claim 3, characterized in that, The step of using the first optimal individual as a basis and employing a boundary information fusion search strategy to perform a first-order search on the individual in the solution space to determine the individual after the first-order search includes: Determine the current training iteration count, and based on the current training iteration count, determine the adaptive information fusion factor as follows: in, Represents the adaptive information fusion factor. This represents the maximum value of the adaptive information fusion factor. This represents the minimum value of the adaptive information fusion factor. This indicates the preset maximum number of training iterations. t Indicates the current number of training iterations. Represents the first random number between (0,1). Represented by natural constant e An exponential function with base 0. Indicates control parameters, and ln represents the logarithmic function; Based on the adaptive information fusion factor and the first optimal individual, a first-order search is performed on the individual in the solution space to determine the individual after the first-order search: in, Indicates the first t During the training process, the first m Individual, m =1,2,…,M, where M represents the total number of individuals in the training population. Indicates the first m The individual after a first-order search; Represents (0,2) π Random angles between ) π Represents pi (π). This represents the upper limit individual, that is, the individual composed of the upper limit of each hyperparameter dimension; This represents the lower bound individual, that is, the individual composed of the lower bounds of each hyperparameter dimension; This represents the second random number between (0,1). Represents the sine function. Represents the cosine function. This represents a third random number between (0,1). This represents the fourth random number between (0,1). This represents the optimal individual.

6. The electromagnetic compatibility analysis method based on deep learning according to claim 5, characterized in that, For individuals obtained after the first-order search, a dual-state information fusion search strategy is used to perform a second-order search in the solution space to determine individuals obtained after the second-order search, including: The control factor for obtaining the two-state search is: in, Indicates the first t Dual-state search control factor during the training process. Indicates the first t The two-state search control factor during training iteration +1, initially set to (0, 2...). π Random angles between ) Based on the dual-state search control factor, a second-order search is performed on the individuals after the first-order search in the solution space to determine the individuals after the second-order search as follows: in, Indicates the first t During the training process, the first n The individual after a first-order search, n =1,2,…,M, Indicates the first n The individual after a second-order search, Represents an individual The corresponding historical optimal state.

7. The electromagnetic compatibility analysis method based on deep learning according to claim 6, characterized in that, For individuals identified after the second-order search, a population information fusion search strategy is used to perform a third-order search within the solution space to determine the individuals identified after the third-order search, including: Determine the current training iterations, and based on the current training iterations, determine the adaptive population fusion factor as follows: in, Represents the adaptive population fusion factor. This represents the maximum value of the adaptive population fusion factor. This represents the minimum value of the adaptive population fusion factor; For the individual obtained after the second-order search, N other individuals are randomly matched to obtain the information fusion individual corresponding to the individual obtained after the second-order search; where N is less than M. Based on the adaptive population fusion factor and the information fusion individuals, a third-order search is performed on the individuals after the second-order search in the solution space to determine the individuals after the third-order search as follows: in, Indicates the first t During the training process, the first i The individual after the second-order search d dimensional hyperparameters, i =1,2,…,M, d =1,2,…,D, where D represents the total dimension of hyperparameters in an individual. Indicates the first i The individual after the third-order search d dimensional hyperparameters, The lower limit individual's first d dimensional hyperparameters, The upper limit of the individual's number d dimensional hyperparameters, This indicates the control parameters for the first information fusion range. This indicates the control parameter for the second information fusion range. Indicates the first t During the training process, the first i The individual after the second-order search and its corresponding first-order search h Euclidean distance between individuals involved in information fusion Indicates the first t During the training process, the first i The individual corresponding to the second-order search h The first information fusion individual d dimensional hyperparameters, The first element representing the optimal individual d Dimensional hyperparameters.

8. The electromagnetic compatibility analysis method based on deep learning according to claim 7, characterized in that, The process of performing a fourth-order search on individuals after the third-order search using a random position fusion search strategy to determine individuals after the fourth-order search includes: Based on the adaptive population fusion factor, the first global search control factor is determined as follows: in, Indicates the first global search control factor; Based on the first global search control factor, a fourth-order search is performed on the individuals after the third-order search to determine the individuals after the fourth-order search as follows: in, Indicates the first t During the training process, the first j The individual after the third-order search d dimensional hyperparameters, j =1,2,…,M, Indicates the first j The individual after the fourth-order search d dimensional hyperparameters, This represents the second global search control factor, which is set as a constant term. Determine whether the root mean square error function value corresponding to the individual after the fourth-order search decreases. If so, accept the fourth-order search; otherwise, reject the fourth-order search.

9. The electromagnetic compatibility analysis method based on deep learning according to claim 3, characterized in that, The step of obtaining the trained deep learning model based on the second optimal individual includes: The hyperparameters of the second optimal individual are used as the target hyperparameters of the deep learning model, and the target hyperparameters are applied to the deep learning model to obtain the trained deep learning model.

10. The electromagnetic compatibility analysis method based on deep learning according to claim 1, characterized in that, The process of collecting electromagnetic compatibility (EMC) data to be analyzed and scheduling the trained deep learning model to analyze the EMC data to determine the EMC analysis results includes: Collect the electromagnetic compatibility data to be analyzed, and extract the peak features, envelope features and / or wavelet packet energy features of the electromagnetic compatibility data to be analyzed to obtain the preprocessed electromagnetic compatibility data to be analyzed; The preprocessed electromagnetic compatibility data to be analyzed is input into the trained deep learning model for identification, and electromagnetic compatibility analysis results are obtained.

Images