Circuit board signal transmission optimization method based on deep learning

Abstract

The invention relates to the field of circuit board signal transmission, and discloses a circuit board signal transmission optimization method based on deep learning, and the method comprises the following steps: S10, collecting circuit board signal data in real time in circuit board signal transmission; s20, preprocessing the collected circuit board signal data; s30, extracting signal features from the preprocessed signal data by using a deep learning algorithm; s40, modeling by using a deep learning algorithm, and inputting training data to train a deep learning model; s50, parameter tuning is carried out; s60, evaluating the performance of the model obtained by training; and S70, based on the feature extraction capability and generalization capability of the existing model, applying the feature extraction capability and generalization capability to the current task through transfer learning. And various parameters and configurations in the signal transmission process are optimized, so that the signal transmission quality is improved. Signal features are analyzed and optimized, noise interference can be reduced, the signal distortion degree can be reduced, key indexes such as signal strength and signal-to-noise ratio can be improved, and therefore reliable transmission of signals is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit board signal transmission, and specifically to a circuit board signal transmission optimization method based on deep learning. Background Art

[0002] Circuit boards are the main components of electronic products, and their design often directly affects the reliability and stability of signal transmission in the electronic products used. Therefore, the design of circuit boards plays an important role in the quality of electronic products.

[0003] Traditional circuit board signal transmission optimization methods usually rely on manual adjustments and empirical rules to optimize signal transmission. The optimization ability of this method is limited by the limitations of human intervention and expert knowledge, and it cannot fully capture the complex relationship between signal characteristics and optimization strategies. Therefore, traditional methods often cannot effectively solve complex signal transmission problems, limiting the room for improvement in signal transmission optimization performance.

[0004] Based on the above problems, a circuit board signal transmission optimization method based on deep learning is proposed to improve the problems arising in the prior art. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a circuit board signal transmission optimization method based on deep learning, which solves the problem that the existing circuit board signal transmission optimization method is more dependent on manual experience.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a circuit board signal transmission optimization method based on deep learning, comprising the following steps:

[0007] S10, collecting circuit board signal data in real time during circuit board signal transmission;

[0008] S20, preprocessing the collected circuit board signal data;

[0009] S30, extracting signal features from the preprocessed signal data using a deep learning algorithm;

[0010] S40, using a deep learning algorithm to build a model, and inputting training data to train the deep learning model;

[0011] S50, perform parameter tuning;

[0012] S60, evaluating the performance of the trained model;

[0013] S70. Based on the feature extraction and generalization capabilities of the existing model, apply it to the current task through transfer learning;

[0014] S80. Use incremental learning to further train the model to adapt to new situations based on task changes or the addition of new data;

[0015] S90. Optimize and adjust the model based on the evaluation results.

[0016] Preferably, the pretreatment step in step S20 is:

[0017] S21, processing missing values, using one or more of the above calculations including interpolation, mean filling or similarity-based methods;

[0018] S22, outlier processing, detect and process outliers, and use statistical methods for calculation;

[0019] S23, feature normalization, use normalization processing to eliminate the dimensional differences of different features.

[0020] Preferably, the deep learning algorithm in steps S30 and S40 is a convolutional neural network, and the convolutional neural network includes multiple parallel convolutional layers, pooling layers and fully connected layers.

[0021] Preferably, in step S50, the number of network layers, the number of nodes, the learning rate, and the regularization parameters are tuned by grid search or random search.

[0022] Preferably, in step S60, the test data is input to obtain the calculated evaluation index, and the evaluation is completed by comparing one or more of the accuracy rate, error rate, and signal-to-noise ratio.

[0023] Preferably, the function expressions of the convolutional layer, pooling layer and fully connected layer are:

[0024] First, given an input image x, the size is (W, H, C), where W and H represent the width and height of the image, and C represents the number of channels of the image;

[0025] Convolutional Layer:

[0026] h(i,j,k)=σ(∑(m,n,c)(x(i+m,j+n,c)*w(m,n,c,k))+b(k))

[0027] Where h(i,j,k) represents the kth feature at position (i,j) in the output feature map; σ represents the activation function; x(i+m,j+n,c) represents the pixel value of the cth channel at position (i+m,j+n) in the input image; w(m,n,c,k) represents the weight of the convolution kernel (filter); b(k) represents the bias term;

[0028] Pooling layer: Downsample the feature map through maximum pooling to reduce the size and number of parameters of the feature map while retaining the main features;

[0029] Fully connected layer:

[0030] z=σ(W*h+b)

[0031] Among them, z represents the output of the fully connected layer; W represents the weight matrix; h represents the feature vector output by the pooling layer; and b represents the bias term.

[0032] Preferably, the transfer learning in step S70 uses a frozen feature extractor method.

[0033] Preferably, the frozen feature extractor function expression is:

[0034] z=σ(W*h+b)

[0035] Among them, z represents the output of the classifier; W represents the weight matrix of the newly added classifier; h represents the output feature vector of the frozen feature extractor; and b represents the bias term of the newly added classifier.

[0036] Preferably, the circuit board signal data in step S10 includes different signal types, noise levels and transmission distances.

[0037] Preferably, the signal characteristics in step S30 include signal-to-noise ratio, signal strength, signal distortion, spectrum distribution, and bit error rate.

[0038] The present invention provides a circuit board signal transmission optimization method based on deep learning. It has the following beneficial effects:

[0039] 1. The present invention can optimize various parameters and configurations in the signal transmission process through a circuit board signal transmission optimization method based on deep learning, thereby improving the quality of signal transmission. By analyzing and optimizing signal characteristics, noise interference can be reduced, signal distortion can be reduced, and key indicators such as signal strength and signal-to-noise ratio can be improved, thereby ensuring reliable signal transmission.

[0040] 2. The present invention optimizes the signal transmission process and can effectively improve the performance and stability of the entire circuit board system. Through the prediction and optimization of the deep learning model, the bit error rate can be reduced, the spectrum distribution of the signal can be improved, and the adaptability of the optimization method to different types of signals and transmission distances can be improved, which will help improve the working efficiency of the optimization method, reduce transmission errors, and improve overall performance and reliability.

[0041] 3. The circuit board signal transmission optimization method based on deep learning of the present invention can automatically optimize signal transmission. By training the deep learning model, the optimization method can automatically learn signal characteristics and optimization strategies without human intervention and manual adjustment, which will reduce the workload of manual optimization, improve optimization efficiency, and make the optimization process more intelligent and sustainable.

[0042] 4. Since the collected circuit board signal data of the present invention contains samples of different signal types, noise levels and transmission distances, the deep learning-based method has strong generalization ability and adaptability, can adapt to diverse signals and environmental conditions, and optimize for different situations. This enables the optimization method to cope with different application scenarios and actual needs, and provide a flexible and efficient signal transmission solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the process structure of the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Example:

[0046] Please refer to the attached Figure 1 , an embodiment of the present invention provides a circuit board signal transmission optimization method based on deep learning, comprising the following steps:

[0047] S10. During the circuit board signal transmission, the circuit board signal data is collected in real time. During the circuit board signal transmission process, appropriate sensors or instruments are needed to collect the circuit board signal data in real time. These sensors are temperature sensors, pressure sensors, and acceleration sensors.

[0048] S20, preprocessing the collected circuit board signal data; the preprocessing is to improve data quality and reduce the impact of noise on subsequent analysis, so as to further optimize the data.

[0049] S30. Use deep learning algorithms to extract signal features from preprocessed signal data. The purpose of feature extraction is to convert the original signal into a more expressive and discriminative feature representation for subsequent modeling and optimization.

[0050] S40. Use deep learning algorithms to build models, and input training data to train the deep learning models. Deep learning algorithms are applied to circuit board signal transmission optimization tasks, and a large amount of pre-processed signal data is used for model training. By inputting training data, the model will automatically learn the complex relationship between signals, and model and predict the signals.

[0051] S50, perform parameter tuning; parameter tuning can be achieved through grid search, random search, and Bayesian optimization methods. The ultimate goal is to find the best hyperparameter combination to improve model performance and generalization ability. By training and evaluating different hyperparameter combinations, their performance on the validation set is compared, and the hyperparameter combination with the best performance is selected as the final model configuration.

[0052] S60. Evaluate the performance of the trained model; the evaluation results can help determine the performance and effect of the model and provide a reference for subsequent optimization.

[0053] S70. Based on the feature extraction and generalization capabilities of the existing model, apply it to the current task through transfer learning; through transfer learning, use the features and knowledge learned by the trained model on other related tasks to accelerate the training process of the current task and improve performance. By fixing part or all of the parameters of the model, only some parameters are updated to adapt to the characteristics and data of the current task. This can reduce the dependence on a large amount of labeled data and achieve faster convergence and optimization of the model.

[0054] S80. Use incremental learning to further train the model to adapt to new situations based on task changes or the addition of new data. In practical applications, circuit board signal transmission may face task changes or the addition of new data. In order to adapt to these changes, use incremental learning, that is, further train the model based on the original model based on the new situation. This can reduce the need to retrain all data and better adapt to new situations and changes.

[0055] S90. Optimize and adjust the model based on the evaluation results. Based on the evaluation results, optimize and adjust the model, including adjusting the model structure, changing hyperparameters, adding regularization terms, etc., to improve the performance, generalization ability and robustness of the model. The optimization process can be carried out through repeated experiments and verification until the expected performance and effect are achieved.

[0056] The preprocessing steps in step S20 are:

[0057] S21. Process missing values ​​using one or more of the above calculations including interpolation, mean filling or similarity-based methods; in the actual circuit board signal transmission process, there may be missing values ​​in the circuit board signal data, that is, the values ​​of certain data points or features are missing. Different processing methods can be used for missing values: interpolation methods can perform interpolation calculations based on existing data points, mean filling can use the average value of the feature to fill the missing values, and similarity-based methods can use the information of other samples or features to estimate the missing values.

[0058] S22. Outlier processing: detect and process outliers, and use statistical methods to calculate; in the circuit board signal data, there are some outliers, that is, extreme values ​​or error values ​​that are significantly different from other observed values. Outliers may have an adverse effect on subsequent analysis and modeling, so outlier detection and processing are required.

[0059] S23, feature normalization, use normalization processing to eliminate the dimensional differences of different features. Through feature normalization, the influence weights of different features on the model are more balanced, thereby improving the stability and convergence speed of model training.

[0060] The deep learning algorithm in steps S30 and S40 is a convolutional neural network, which includes multiple parallel convolutional layers, pooling layers, and fully connected layers.

[0061] The function expressions of convolutional layer, pooling layer and fully connected layer are:

[0062] First, given an input image x, the size is (W, H, C), where W and H represent the width and height of the image, and C represents the number of channels of the image;

[0063] Convolutional Layer:

[0064] h(i,j,k)=σ(∑(m,n,c)(x(i+m,j+n,c)*w(m,n,c,k))+b(k))

[0065] Where h(i,j,k) represents the kth feature at position (i,j) in the output feature map; σ represents the activation function; x(i+m,j+n,c) represents the pixel value of the cth channel at position (i+m,j+n) in the input image; w(m,n,c,k) represents the weight of the convolution kernel (filter); b(k) represents the bias term;

[0066] Pooling layer: Downsample the feature map through maximum pooling to reduce the size and number of parameters of the feature map while retaining the main features;

[0067] Fully connected layer:

[0068] z=σ(W*h+b)

[0069] Among them, z represents the output of the fully connected layer; W represents the weight matrix; h represents the feature vector output by the pooling layer; and b represents the bias term.

[0070] In step S70, transfer learning uses the frozen feature extractor method.

[0071] The frozen feature extractor function expression is:

[0072] z=σ(W*h+b)

[0073] Among them, z represents the output of the classifier; W represents the weight matrix of the newly added classifier; h represents the output feature vector of the frozen feature extractor; and b represents the bias term of the newly added classifier.

[0074] In step S50, the number of network layers, number of nodes, learning rate, and regularization parameters are tuned by grid search and random search. In the process of parameter tuning, grid search or random search is used to traverse the combination space of hyperparameters. Grid search will traverse the specified parameter combination, while random search will perform random sampling within the given parameter range. By training and evaluating different hyperparameter combinations, their performance on the validation set is compared, and the hyperparameter combination with the best performance is selected as the final model configuration.

[0075] In step S60, the test data is input to obtain the calculated evaluation index, and the evaluation is completed by comparing one or more of the accuracy rate, error rate, and signal-to-noise ratio. By comparing the evaluation indicators such as accuracy rate, error rate, and signal-to-noise ratio, the performance of the deep learning model in the circuit board signal transmission optimization task can be comprehensively evaluated. These indicators can help judge the model's performance in terms of predictive ability, classification accuracy, and signal quality. Based on the evaluation results, the advantages and disadvantages of the model are further analyzed, and subsequent optimization and adjustment are carried out.

[0076] The circuit board signal data in step S10 includes different signal types, noise levels, and transmission distances. By collecting circuit board signal data including different signal types, noise levels, and transmission distances, the actual application scenarios can be better simulated, thereby improving the robustness and generalization ability of the circuit board signal transmission optimization method based on deep learning. Through such calculations, the data set can be used for model training, verification, and testing to evaluate the performance and effectiveness of the model in various situations.

[0077] The signal features in step S30 include signal-to-noise ratio, signal strength, signal distortion, spectrum distribution, and bit error rate. By extracting these signal features related to the deep learning model, we can better understand the characteristics of the circuit board signal and provide more comprehensive information for the optimization method. These features can be used as part of the input data to help the deep learning model learn the pattern and law of the signal, thereby improving the performance and effect of the circuit board signal transmission optimization method.

[0078] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A circuit board signal transmission optimization method based on deep learning, characterized in that: The following steps are involved: S10, collecting circuit board signal data in real time during circuit board signal transmission; S20, preprocessing the collected circuit board signal data; S30, extracting signal features from the preprocessed signal data using a deep learning algorithm; S40, using a deep learning algorithm to build a model, and inputting training data to train the deep learning model; S50, perform parameter tuning; S60, evaluating the performance of the trained model; S70. Based on the feature extraction and generalization capabilities of the existing model, apply it to the current task through transfer learning; S80. Use incremental learning to further train the model to adapt to new situations based on task changes or the addition of new data; S90. Optimize and adjust the model based on the evaluation results.

2. The circuit board signal transmission optimization method based on deep learning according to claim 1 is characterized in that: The pre-processing steps in step S20 are: S21, processing missing values, using one or more of the above calculations including interpolation, mean filling or similarity-based methods; S22, outlier processing, detect and process outliers, and use statistical methods for calculation; S23, feature normalization, use normalization processing to eliminate the dimensional differences of different features.

3. The circuit board signal transmission optimization method based on deep learning according to claim 1 is characterized in that: The deep learning algorithm in steps S30 and S40 is a convolutional neural network, which includes multiple parallel convolutional layers, pooling layers and fully connected layers.

4. The circuit board signal transmission optimization method based on deep learning according to claim 1 is characterized in that: In the step S50, the number of network layers, the number of nodes, the learning rate, and the regularization parameters are optimized by grid search and random search methods.

5. The circuit board signal transmission optimization method based on deep learning according to claim 1 is characterized in that: In the step S60, the test data is input to obtain the calculated evaluation index, and the evaluation is completed by comparing one or more of the accuracy rate, error rate, and signal-to-noise ratio.

6. The circuit board signal transmission optimization method based on deep learning according to claim 3 is characterized in that: The function expressions of the convolutional layer, pooling layer and fully connected layer are: First, given an input image x, the size is (W, H, C), where W and H represent the width and height of the image, and C represents the number of channels of the image; Convolutional Layer: h(i,j,k)=σ(∑(m,n,c)(x(i+m,j+n,c)w(m,n,c,k))+b(k)) Where h(i,j,k) represents the kth feature at position (i,j) in the output feature map; σ represents the activation function; x(i+m,j+n,c) represents the pixel value of the cth channel at position (i+m,j+n) in the input image; w(m,n,c,k) represents the weight of the convolution kernel (filter); b(k) represents the bias term; Pooling layer: Downsample the feature map through maximum pooling to reduce the size and number of parameters of the feature map while retaining the main features; Fully connected layer: z=σ(W*h+b) Among them, z represents the output of the fully connected layer; W represents the weight matrix; h represents the feature vector output by the pooling layer; and b represents the bias term.

7. The circuit board signal transmission optimization method based on deep learning according to claim 1, characterized in that: The transfer learning in step S70 uses a frozen feature extractor method.

8. The circuit board signal transmission optimization method based on deep learning according to claim 7 is characterized in that: The frozen feature extractor function expression is: z=σ(W*h+b) Among them, z represents the output of the classifier; W represents the weight matrix of the newly added classifier; h represents the output feature vector of the frozen feature extractor; and b represents the bias term of the newly added classifier.

9. The circuit board signal transmission optimization method based on deep learning according to claim 1, characterized in that: The circuit board signal data in step S10 includes different signal types, noise levels and transmission distances.

10. The circuit board signal transmission optimization method based on deep learning according to claim 1, characterized in that: The signal characteristics in step S30 include signal-to-noise ratio, signal strength, signal distortion, spectrum distribution, and bit error rate.

Images