A method for predicting limiter damage effects based on machine learning
Through machine learning methods, a fully connected neural network was constructed based on the characteristics of electromagnetic pulses, which solved the problem of rapid prediction of limiter damage effects and achieved efficient electromagnetic pulse protection for the RF front-end of electronic equipment.
Patent Information
- Application Number
- CN202211511622.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-29
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2042-11-29
AI Technical Summary
Existing technologies make it difficult to quickly and accurately predict the damage effects of limiters under the action of electromagnetic pulses. Traditional simulation methods are time-consuming and require detailed parameters, and each parameter change requires recalculation.
A machine learning method is used to separate broadband and narrowband electromagnetic pulses, extract characteristic parameters, and construct a fully connected neural network to predict the limiter damage effect. The electromagnetic pulse characteristics are used as input and the limiter damage is used as output to construct a data set and train the neural network.
It achieves fast and effective prediction of limiter damage effects, solves the time-consuming and inaccurate problems of traditional simulation methods, and provides support for electromagnetic pulse protection of the RF front-end of electronic equipment.
Smart Images
Figure CN115877100B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of electromagnetic effect analysis, and in particular relates to a method for predicting limiter damage effects based on machine learning. Background Art
[0002] When exposed to electromagnetic pulses, the primary energy coupling channel for electronic devices is the "front door," where the pulse energy is injected into the RF front-end components of the device via the antenna. As a crucial protection module for the RF front-end of electronic devices, the limiter limits the output voltage amplitude to a certain range, protecting back-end components from damage. However, under the influence of extremely high-energy electromagnetic pulses, the limiter also faces the risk of degradation of its limiting capability, burnout, and loss of its limiting function, which can damage the back-end electronic components of the device.
[0003] Therefore, predicting the effects of limiters is an essential foundation for analyzing electromagnetic effects on the RF front-end of electronic equipment. Traditionally, simulation software has been used to calculate the output voltage and insertion loss of the limiter under electromagnetic pulses and compare these values with those during normal operation to determine if the limiter is damaged. However, accurate modeling is difficult and time-consuming. For example, modeling a PIN limiter requires inputting a detailed set of parameters, such as device structural parameters and doping concentration distribution, which are difficult for manufacturers to access. Furthermore, every change to any parameter, such as the electromagnetic pulse frequency, pulse width, repetition rate, amplitude, and cumulative exposure time, requires re-simulation, which is extremely time-consuming. Therefore, a rapid and effective method for predicting limiter damage effects is urgently needed to support analysis of the RF front-end effects of electronic equipment and provide a basis for protective reinforcement. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for predicting the damage effects of limiters based on machine learning, which can predict the damage effects of limiters of different types and models under different electromagnetic pulse environments. This method is suitable for analyzing the RF front-end effects of electronic equipment and can be applied to the design of electromagnetic pulse protection for electronic equipment.
[0005] The technical solutions adopted in the present invention are as follows:
[0006] A method for predicting limiter impairment effects based on machine learning, comprising the following steps:
[0007] Electromagnetic pulses are divided into two categories: broadband electromagnetic pulses and narrowband electromagnetic pulses. For broadband electromagnetic pulses, amplitude, pulse width, repetition frequency, application time and rise time are extracted as features; for narrowband electromagnetic pulses, amplitude, pulse width, frequency, repetition frequency and application time are extracted as features.
[0008] The limiter's limited output power and insertion loss are collected when it is in normal working condition without electromagnetic pulses, which serve as a benchmark for determining whether the limiter is damaged. When the electromagnetic pulse is changed, the limiter's limited output power and insertion loss at the output end of the limiter are collected and compared with the benchmark values. If they exceed the allowable error range, the limiter is considered damaged, otherwise it is not damaged.
[0009] A data set is constructed with the characteristics of electromagnetic pulses as input and whether the limiter is damaged as output. The data set is divided according to the limiter model and electromagnetic pulse classification. For each sub-dataset, a neural network is constructed and trained, and finally the limiter damage effect is predicted.
[0010] Furthermore, electromagnetic pulses are divided into two categories according to the signal bandwidth: broadband electromagnetic pulses and narrowband electromagnetic pulses.
[0011] Furthermore, a bandwidth greater than 25% is a broadband electromagnetic pulse, and a bandwidth less than 10% is a narrowband electromagnetic pulse.
[0012] Furthermore, before constructing the dataset, the collected data is preprocessed, which includes data cleaning, outlier removal and data standardization.
[0013] Furthermore, the neural network is a fully connected neural network.
[0014] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0015] The present invention proposes a method for predicting the damage effect of a limiter under the action of an electromagnetic pulse based on a machine learning algorithm. The electromagnetic pulses are divided into two categories: broadband electromagnetic pulses and narrowband electromagnetic pulses. For broadband electromagnetic pulses, the amplitude, pulse width, repetition frequency, application time and rise time are extracted as features. For narrowband electromagnetic pulses, the amplitude, pulse width, frequency, repetition frequency and application time are extracted as features. A machine learning model for quickly predicting the damage effect of the limiter based on historical data is established, which solves the problems that commonly used simulation methods are difficult to accurately fit and time-consuming, and provides a basis for the effect analysis and protection reinforcement of the radio frequency front end of electronic equipment under electromagnetic pulses. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 This is a flow chart of electromagnetic effect data collection of the limiter of the present invention under the action of electromagnetic pulses;
[0017] Figure 2 This is a schematic diagram of data set classification of the present invention. DETAILED DESCRIPTION
[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
[0019] This paper proposes a method for predicting limiter damage effects based on machine learning. The key technical points are as follows:
[0020] (1) Modeling of electromagnetic pulse characteristics and collection and construction of electromagnetic effect data sets of limiters under electromagnetic pulses
[0021] By comprehensively considering the types and characteristics of electromagnetic pulses, electromagnetic pulses are divided into two categories: broadband electromagnetic pulses and narrowband electromagnetic pulses, and features are extracted to construct models for each category. By changing the test conditions, the electromagnetic effect data of the limiter under the action of electromagnetic pulses is obtained. Then, data preprocessing is completed through steps such as data cleaning, outlier removal and data standardization. The damage of the limiter corresponding to the preprocessed electromagnetic pulse characteristic data is matched, and a data set is constructed for algorithm learning and training.
[0022] (2) Network structure design based on machine learning to predict limiter damage effects
[0023] Predicting whether a limiter is damaged is essentially a classification problem. The present invention utilizes the advantages of fully connected neural networks in machine learning in terms of nonlinear fitting ability and algorithm robustness to predict whether a limiter is damaged under the action of electromagnetic pulses. The main steps include data set division, network input and output node design, network hyperparameter selection, network training and prediction.
[0024] The method for predicting limiter damage effects based on machine learning in an embodiment of the present invention includes:
[0025] 1. Modeling of electromagnetic pulse characteristics and collection and construction of electromagnetic effect data sets of limiters under electromagnetic pulses
[0026] The data collection process of electromagnetic effect of limiter under electromagnetic pulse is as follows: Figure 1 Electromagnetic pulses can be generally divided into broadband electromagnetic pulses (bandwidth greater than 25%) and narrowband electromagnetic pulses (bandwidth less than 10%) according to the signal bandwidth.
[0027] Broadband electromagnetic pulses have the characteristics of a steep rising edge (in the order of ps), a narrow pulse width (in the order of ns-ps), and a wide radiation spectrum. The present invention extracts the amplitude, pulse width, repetition frequency, application time, and rise time of broadband electromagnetic pulses as features. The present invention constructs a data set based on experimental data of damage to a limiter caused by broadband electromagnetic pulses. First, the limiter's limited output power and insertion loss are tested in a normal operating state without the action of the electromagnetic pulse, serving as a benchmark for determining whether the limiter is damaged. The broadband electromagnetic pulse source is then activated, and the amplitude, pulse width, repetition frequency, application time, and rise time of the broadband electromagnetic pulse are collected at the limiter input. The limited output power and insertion loss at the limiter output are also collected and compared with the benchmark values. If the values exceed the allowable error range, the limiter is considered damaged; otherwise, it is considered undamaged. By changing the test conditions, the amplitude, pulse width, repetition frequency, application time and rise time will change accordingly. The amplitude, pulse width, repetition frequency, application time and rise time of the broadband electromagnetic pulse will be collected again at the input end of the limiter. The limited output power and insertion loss at the output end of the limiter will also be collected and compared with the benchmark values to determine whether the limiter is damaged, until the required amount of data is collected.
[0028] For narrowband electromagnetic pulses, the present invention extracts amplitude, pulse width, frequency, repetition rate, and application time as features. A dataset is constructed based on data from narrowband electromagnetic pulse damage tests on a limiter. First, the limiter's limited output power and insertion loss are measured during normal operation without the electromagnetic pulse, serving as a benchmark for determining whether the limiter is damaged. The narrowband electromagnetic pulse signal source is then activated, and the amplitude, pulse width, frequency, repetition rate, and application time of the narrowband electromagnetic pulse are collected at the limiter input. The limited output power and insertion loss at the limiter output are also collected and compared with the benchmark values. If the values exceed the allowable error range, the limiter is considered damaged; otherwise, they are considered undamaged. The test conditions are then modified, with the amplitude, pulse width, frequency, repetition rate, and application time being altered accordingly. The amplitude, pulse width, frequency, repetition rate, and application time of the narrowband electromagnetic pulse are again collected at the limiter input. The limited output power and insertion loss at the limiter output are then collected and compared with the benchmark values to determine whether the limiter is damaged. This continues until a sufficient amount of data has been collected.
[0029] The collected data is preprocessed through steps such as data cleaning, outlier removal and data standardization, and then a data set is constructed using the preprocessed data for algorithm learning and training.
[0030] 2. Network structure design for predicting limiter damage effects based on machine learning
[0031] The constructed data set is classified according to the limiter model and electromagnetic pulse environment. The classification diagram is shown in the following figure. Figure 2For each subset of data, a fully connected neural network is constructed and trained to predict whether the limiter is damaged when given an input. Taking broadband electromagnetic pulses as an example, the key points for constructing and training the fully connected neural network are as follows:
[0032] (1) The characteristics of broadband electromagnetic pulses include amplitude, pulse width, repetition frequency, application time, and rise time. After considering the bias term, the number of input nodes of the corresponding fully connected neural network is 6;
[0033] (2) The network output is whether the limiter is damaged. Two nodes are used to represent the probability of damage or non-damage, respectively. The value range is [0, 1], and the sum of the probabilities is 1.
[0034] (3) There is a hidden layer between the input layer and the output layer. The number of hidden layers and the number of nodes in each layer are the hyperparameters of the fully connected neural network. The choice of hyperparameters directly affects the performance of the network.
[0035] (4) The sub-dataset is divided into a training set, a cross-validation set, and a test set. For several preset sets of hyperparameters, the training set is first used for training, and then the cross-validation set is used to calculate the prediction error under different hyperparameter combinations. The hyperparameters that perform best on the cross-validation set are selected as the network structure parameters.
[0036] (5) Use the trained fully connected neural network to predict the effect on the test set and calculate the test set error. When the data is balanced and the data volume is sufficient, the error of the test set can represent the performance of the fully connected neural network on the real data.
[0037] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.
[0038] It will be easily understood by those skilled in the art that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements 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 method for predicting limiter damage effects based on machine learning, characterized in that: The following steps are involved: Electromagnetic pulses are divided into two categories: broadband electromagnetic pulses and narrowband electromagnetic pulses. For broadband electromagnetic pulses, amplitude, pulse width, repetition frequency, application time and rise time are extracted as features; for narrowband electromagnetic pulses, amplitude, pulse width, frequency, repetition frequency and application time are extracted as features. The limiter's limited output power and insertion loss are collected when it is in normal working state without electromagnetic pulses, which serve as a benchmark for determining whether the limiter is damaged. When the electromagnetic pulses are changed, the limiter's limited output power and insertion loss at the output end are collected under the action of broadband and narrowband electromagnetic pulses, and compared with the benchmark values. If the values exceed the allowable error range, the limiter is considered damaged, otherwise it is considered undamaged. A data set is constructed with the characteristics of electromagnetic pulses as input and whether the limiter is damaged as output. The data set is divided according to the limiter model and electromagnetic pulse classification. For each sub-dataset, a neural network is constructed and trained, and finally the limiter damage effect is predicted.
2. The method for predicting limiter damage effects based on machine learning according to claim 1, characterized in that: A broadband electromagnetic pulse has a bandwidth greater than 25%, and a narrowband electromagnetic pulse has a bandwidth less than 10%.
3. The method for predicting limiter damage effects based on machine learning according to claim 1, characterized in that: Before constructing the data set, the collected data is preprocessed, which includes data cleaning, outlier removal and data standardization.
4. The method for predicting limiter damage effects based on machine learning according to claim 1, characterized in that: The neural network is a fully connected neural network.