Resistance detection method for planar rectifier diode
By performing segmented differential processing and sparse encoding of current and voltage data, combined with gradient enhancement regression tree and random forest algorithm, the problems of large errors and high error rate of resistance detection in the prior art are solved, and high accuracy and adaptability of resistance detection are achieved.
Patent Information
- Application Number
- CN202510424486.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-03
- Publication Date
- 2025-07-25
AI Technical Summary
In the face of changes in resistance value caused by slight fluctuations or environmental changes, the prior art has large errors and high misjudgment rates, and lacks dynamic analysis and real-time adjustment of resistance data trends, which affects the accuracy of component screening and the stability of circuits.
By performing segmented differential processing on current and voltage data, a local differential matrix is generated, and important eigenvalues are extracted in combination with sparse coding and gradient enhancement regression tree, a random forest algorithm is used to predict trends, and the measurement parameters are dynamically adjusted when the resistance value changes trend abnormally.
It improves the accuracy and adaptability of resistance detection, reduces errors, enhances the accuracy of determining whether the resistance is qualified, and adapts to high concurrency requirements.
Smart Images
Figure CN120370123A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of resistance detection, and particularly to a method for detecting the resistance of a planar rectifier diode. Background Art
[0002] The technical field of resistance detection aims to evaluate performance, identify faults, and ensure quality control by accurately measuring and analyzing the resistance characteristics of electronic components, circuits, or systems. By accurately measuring the resistance value, the on-off state, effectiveness, and reliability of components can be quickly judged.
[0003] The purpose of the method for detecting the resistance of a planar rectifier diode is to accurately discriminate whether there are failure modes such as breakdown, open circuit, and leakage in the diode. Without damaging the device, it can be determined whether the performance of the diode meets the usage standards, improve the component screening efficiency, ensure the reliability of the circuit operation, and facilitate mass automatic detection or on-site maintenance judgment.
[0004] The existing technology will produce large errors when facing small fluctuations or resistance value changes caused by environmental changes, resulting in inaccurate judgments. Moreover, it lacks long-term analysis of the data trend by effectively using the dynamic changes in the resistance data, causing misjudgments when the resistance value is close to the critical standard, affecting the accuracy of component screening and the stability of the circuit. It lacks dynamic prediction and real-time adjustment of the resistance change trend, resulting in a slow measurement speed and inability to adapt to high-concurrency requirements. Summary of the Invention
[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art, and a method for detecting the resistance of a planar rectifier diode is proposed.
[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A method for detecting the resistance of a planar rectifier diode, comprising the following steps: Step 1: Detect current and voltage data through a sensor, perform segmented difference processing, calculate the difference gradient between each current segment and voltage segment, retain key information according to the gradient change, and generate a local difference matrix; Step 2: Perform sparse coding on the local difference matrix, use a gradient boosting regression tree to extract important eigenvalue in the matrix, optimize the calculation process, remove redundant information and compress the gradient data, generate a low-dimensional feature vector, and adjust the gradient weight to strengthen the effective information to obtain a sparse feature vector group; Step 3: Based on the sparse feature vector group, perform fast parallel calculation, optimize the matrix multiplication process, compare with the existing judgment rules in combination with the feature vector, and judge whether the resistance is qualified by judging the similarity between the feature vector and the rule to generate a judgment result; Step 4: Based on the determination result, use random forest, compare with historical resistance data, combine with the current measured value, conduct trend prediction, calculate the potential trend of future changes in the resistance value, and when the trend is abnormal, compare with existing data for dynamic adjustment to generate a resistance change prediction result; Step 5: Based on the resistance change prediction result, compare its deviation from the current measured value. When the deviation exceeds the set threshold, adjust the measurement parameters to generate a measurement parameter adjustment plan.
[0007] As a further solution of the present invention, the specific steps for generating the local difference matrix are as follows: Detect current and voltage data through sensors, segment the data at a predetermined time interval, extract the corresponding current and voltage values within each time period, calculate the difference between the current and voltage within the time period, obtain the difference value between each current segment and voltage segment, and generate segmented difference data; Based on the segmented difference data, perform difference gradient calculation, compare the change amplitudes of each current segment and voltage segment, determine the parts with larger changes and retain them, eliminate low-frequency noise and irrelevant data, and generate a gradient change matrix; Based on the gradient change matrix, screen out the regions with relatively stable changes, remove redundant gradient information, and retain the data that has a key impact on the relationship between current and voltage to generate a local difference matrix.
[0008] As a further solution of the present invention, the specific steps for generating the sparse feature vector group are as follows: Based on the local difference matrix, use the gradient boosting regression tree to compress the data in the matrix, analyze the importance of each data segment one by one, weight-adjust the features of each data segment, remove redundant data, and generate a compressed data set; Based on the compressed data set, through weighted adjustment, adjust the feature weights in the data according to the importance of each data segment, strengthen the effective information therein, and generate adjusted feature data; Based on the adjusted feature data, extract the low-dimensional features in the data, compress the high-dimensional feature information to low dimensions, and generate a sparse feature vector group.
[0009] As a further solution of the present invention, the gradient boosting regression tree is calculated according to the formula:
[0010] Where: is the predicted result of the resistance value of the th sample in the th round of iteration, is the predicted result of the resistance value of the th sample in the previous round of iteration, is the learning rate, For the After rounds of training, based on the input data The predicted output is is the data distribution coefficient, is the feature importance weighting term, is the noise suppression coefficient, is the noise measure, is the fitting accuracy coefficient, is the error term in each round of training.
[0011] As a further solution of the present invention, the low-dimensional features in the extracted data, the low-dimensional feature extraction process converts the original high-dimensional data into a set of irrelevant principal components through principal component analysis, and selects the number of principal components to be retained according to the variance contribution, and evaluates the importance of each feature, selects the features that have a significant impact on the prediction results, and screens the features at the same time to obtain a compressed data set.
[0012] As a further solution of the present invention, the specific steps of generating the determination result are: Based on the sparse feature vector group, feature grouping is performed, the feature vector is divided into multiple calculation units according to the dimension, each unit calculates the matching degree respectively, and the features in each group are compared with the judgment rules one by one to generate a group matching data set; Based on the group matching data set, perform element-by-element multiplication on each group of data, calculate the product of the matching degree and the feature weight, obtain a weighted sum, sum each calculation unit, and generate a weighted matching result; Based on the weighted matching result, a threshold judgment is performed, and the comparison result is compared with a set qualified range. If the result exceeds the threshold, it is marked as unqualified, and a judgment result is generated.
[0013] As a further solution of the present invention, the specific steps of generating the resistance change prediction result are: Based on the determination result, extract the current resistance value and compare it with the historical resistance data, calculate the difference between the current measured value and the historical data by fitting the current resistance value and the historical data through random forest, and generate resistance change comparison data; Based on the resistance change comparison data, a trend curve analysis is performed to compare the change trends of the historical data with the current data, calculate the increase or decrease amplitude of the trend change, and generate a resistance change trend trajectory; Based on the resistance change trend trajectory, the difference between the predicted point and the historical data is compared to determine whether it exceeds the preset error range. If it exceeds, the detection parameters are adjusted and the trend level is redefined to generate a resistance change prediction result.
[0014] As a further solution of the present invention, for the random forest, according to the formula:
[0015] Where: is the th sample, is the total number of decision trees, is the th decision tree's predicted value for the th sample, is the th sample's input feature, is the data distribution coefficient, is the feature importance weighting term, is the noise suppression coefficient, is the noise measure, is the fitting accuracy coefficient, is the error term in the round training.
[0016] As a further solution of the present invention, the specific steps for generating the measurement parameter adjustment scheme are as follows: Based on the predicted result of the resistance change, calculate the deviation between the current measurement value and the predicted resistance value, compare the gap between the current measurement value and the predicted value one by one, and generate measurement deviation data; Based on the measurement deviation data, determine whether the deviation exceeds the set threshold range. If it exceeds the range, mark it as abnormal and generate a deviation over-limit judgment result; Based on the deviation over-limit judgment result, adjust the measurement parameters according to the over-limit judgment, modify the measurement setting parameters respectively, and generate a measurement parameter adjustment scheme.
[0017] As a further solution of the present invention, the measurement setting parameters include measurement current, measurement frequency, measurement period, sampling rate, ambient temperature, signal filtering setting, measurement mode, and test voltage range.
[0018] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. In the present invention, by performing segmented difference processing on the current and voltage data, calculating the difference gradient between each current segment and voltage segment, extracting the key information in the data and generating a local difference matrix, important signal features are effectively retained, and the subtle differences that may be missed in conventional measurements are avoided; 2. In the present invention, by combining sparse coding with gradient boosting regression trees, important eigenvalue in the matrix is extracted and optimized calculation is performed. By removing redundant information and compressing gradient data, the calculation process is streamlined, the data processing efficiency and accuracy are enhanced, the feature representation is made more compact and effective, and the determination accuracy of whether the resistance is qualified can be improved; 3. In the present invention, the historical resistance data and the current measurement value are compared through the random forest algorithm to conduct trend prediction and dynamic adjustment. When the change trend of the resistance value is abnormal, the measurement parameters are adjusted in a timely manner, the resistance detection scheme is optimized, and the adaptability and accuracy of the detection process are improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 is a schematic diagram of the main steps of the present invention; Figure 2 is a detailed schematic diagram of S1 of the present invention; Figure 3 is a detailed schematic diagram of S2 of the present invention; Figure 4 is a detailed schematic diagram of S3 of the present invention; Figure 5 is a detailed schematic diagram of S4 of the present invention; Figure 6 is a detailed schematic diagram of S5 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0020] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0021] Please refer to Figure 1 , the present invention provides a technical solution: a resistance detection method for a planar rectifier diode, including the following steps: S1: The current and voltage data are detected through a sensor, segmented difference processing is performed, the difference gradient between each current segment and voltage segment is calculated, key information is retained according to the gradient change, and a local difference matrix is generated; S2: Through sparse coding of the local difference matrix, a gradient boosting regression tree is used to extract important eigenvalue in the matrix, the calculation process is optimized, redundant information is removed and the gradient data is compressed, a low-dimensional feature vector is generated, and the gradient weight is adjusted to strengthen the effective information to obtain a sparse feature vector group; S3: Based on the sparse feature vector group, fast parallel calculation is performed, the matrix multiplication process is optimized, the comparison is made by combining the existing determination rules and the feature vector, and whether the resistance is qualified is judged by judging the similarity between the feature vector and the rule, and a determination result is generated; S4: Based on the determination result, a random forest is used to compare with the historical resistance data, combined with the current measurement value, trend prediction is performed, the potential trend of the future change of the resistance value is calculated, and when the trend is abnormal, dynamic adjustment is performed by comparing with the existing data to generate a resistance change prediction result; S5: Based on the prediction result of the resistance change, compare its deviation from the current measurement value. When the deviation exceeds the set threshold, adjust the measurement parameters to generate a measurement parameter adjustment plan.
[0022] Please refer to Figure 2 , and the specific steps to generate the local difference matrix are as follows: S101: Detect the current and voltage data through sensors, segment the data at a predetermined time interval, extract the corresponding current and voltage values within each time period, calculate the difference between the current and voltage within the time period, obtain the difference values between each current segment and voltage segment, and generate segmented difference data; S102: Based on the segmented difference data, calculate the difference gradient, compare the change amplitudes of each current segment and voltage segment, determine the parts with larger changes and retain them, eliminate low-frequency noise and irrelevant data, and generate a gradient change matrix; S103: Based on the gradient change matrix, screen out the regions with relatively stable changes, remove redundant gradient information, retain the data that has a key impact on the relationship between current and voltage, and generate a local difference matrix; S101: Based on the current and voltage data detected by the sensor, segment the data, extract the corresponding current and voltage values within each time period, calculate the difference between the current and voltage within the time period, perform correlation calculation, obtain the difference values between each current segment and voltage segment, generate an array of difference data, process the data by setting the time interval to 100 milliseconds, generate segmented difference data, ensure that the intervals of each data segment are consistent, and use the pandas library for time series data storage and processing to ensure the integrity and accuracy of the data; S102: Based on the segmented difference data, use the gradient calculation formula to calculate the difference gradient of each current segment and voltage segment, extract the significant peaks of the change of the difference value, identify the change of the difference value by setting the threshold to 0.5, eliminate low-frequency noise and remove irrelevant data, design a low-pass filter, set the cut-off frequency to 1Hz, filter the signal, and generate a gradient change matrix to ensure that the low-frequency noise is effectively removed while maintaining the high-frequency effective signal; S103: Based on the gradient change matrix, use the K-means clustering algorithm for data partitioning, set the number of clusters to 3, screen out the regions with relatively stable changes, perform clustering of the smooth regions, set 300 iterations of calculation, minimize the difference values of the data within the cluster, remove redundant gradient information, and perform feature selection, retain the data that has a key impact on the relationship between current and voltage, use a feature number of k = 5, and generate a local difference matrix to ensure that the selected data can effectively reflect the main relationship between current and voltage and remove irrelevant data.
[0023] Please refer to Figure 3, the specific steps for generating the sparse feature vector group are as follows: S201: Based on the local difference matrix, use the gradient boosting regression tree to compress the data in the matrix, analyze the importance of each data segment segment by segment, weight-adjust the features of each data segment, remove redundant data, and generate a compressed data set; S202: Based on the compressed data set, through weight adjustment, according to the importance of each data segment, adjust the feature weights in the data, strengthen the effective information therein, and generate adjusted feature data; S203: Based on the adjusted feature data, extract the low-dimensional features in the data, compress the high-dimensional feature information to low dimensions, and generate a sparse feature vector group; S201: Based on the local difference matrix, use the gradient boosting regression tree, specifically the XGBoost algorithm, set the parameter iteration number to 100, the learning rate to 0.01, and the maximum depth to 3. Compress the data in the matrix, analyze the importance of each data segment segment by segment, use the fitting function of XGBoost to train the data, and obtain the feature importance score of each data segment through the feature importance method. According to the score, weight-adjust the features of each data segment, remove redundant data, perform standardization processing on the data to ensure that the data values vary between 0 and 1, generate a compressed data set, and use the prediction method to output the predicted value to obtain the compressed data result; S202: Based on the compressed data set, through weight adjustment, according to the importance of each data segment, use a linear regression model to adjust the data feature weights, set whether to include the intercept as True and whether to perform normalization as False, extract the weight coefficient of each feature through the coefficient method of the linear regression model, weight-adjust the features, strengthen the effective information therein, perform data normalization using min-max scaling to ensure that the adjusted data remains within the standard interval, generate adjusted feature data, and further optimize the contribution of each feature through the coefficient and intercept; S203: Based on the adjusted feature data, perform principal component analysis for feature dimensionality reduction, set the dimensionality after dimensionality reduction to 5, compress the high-dimensional feature information to low dimensions, use the fitting transformation method of principal component analysis to perform dimensionality reduction processing on the data, generate a sparse feature vector group, and ensure that the data after dimensionality reduction has an even distribution by setting the whitening option to True.
[0024] The gradient boosting regression tree, according to the formula:
[0025] where: is the predicted result of the resistance value of the th sample in the th round of iteration, is the predicted result of the resistance value of the th sample in the previous iteration, is the learning rate, is the th round of predicted output based on the input data after training, is the data distribution coefficient, is the feature importance weighting term, is the noise suppression coefficient, is the noise measure, is the fitting accuracy coefficient, is the error term in each round of training process; Execution process: First, represents the predicted value of the th sample in the previous iteration, is the learning rate, which controls the step size of model update in each round. In each iteration, based on the current input data , using the th round of training-generated predicted resistance value, the preliminary prediction result of the sample is obtained. Next, in order to further improve the accuracy of the model, considering the distribution characteristics of the current-voltage response data and the influence of features, data distribution coefficient is added to adjust the processing weight of the data segment to make the data segment receive more attention. At the same time, is the feature weighting term. By weighting each data segment according to the influence of the feature, it ensures that the influence of important features on the prediction result is fully reflected. At the same time, noise suppression coefficient and noise measure are added. By controlling the suppression of noise and the error in the measurement process, the prediction accuracy of the model for resistance is improved. as the fitting accuracy coefficient, controls the acceptable range of error in each iteration process to ensure that the model can fit the resistance change trend and is appropriately adjusted according to the error size. Finally, is generated, optimizing the accuracy and robustness of resistance detection.
[0026] Extract low-dimensional features from the data. The low-dimensional feature extraction process uses principal component analysis to convert the original high-dimensional data into a set of uncorrelated principal components, selects the number of principal components to be retained according to the variance contribution degree, evaluates the importance of each feature, selects the features that have an obvious impact on the prediction result, and at the same time performs feature screening to obtain the compressed data set.
[0027] Please refer to Figure 4 for the specific steps to generate the judgment result: S301: Based on the sparse feature vector group, perform feature grouping. Divide the feature vectors into multiple computing units according to dimensions. Each unit calculates the matching degree separately, and compare the features within each group with the decision rules one by one to generate a grouped matching data set; S302: Based on the grouped matching data set, perform element-wise multiplication on each group of data, calculate the product of the matching degree and the feature weight to obtain the weighted sum, and sum each computing unit to generate a weighted matching result; S303: Based on the weighted matching result, perform threshold judgment, compare the comparison result with the set qualified range. If the result exceeds the threshold, mark it as unqualified to generate a judgment result; S301: Based on the sparse feature vector group, perform feature grouping. Use the split function in the NumPy library to divide the feature vectors into multiple computing units according to dimensions. Set the axis to 1, that is, group by columns. Each unit calculates the matching degree separately. Use the cosine similarity function in the SciPy library to calculate the matching degree between each group of feature vectors and the decision rules, and compare them one by one to generate a grouped matching data set. Combine the matching results of each group using the concatenate function in NumPy to generate the final grouped matching data set; S302: Based on the grouped matching data set, perform element-wise multiplication on each group of data. Use the multiply function in the NumPy library to perform element-wise multiplication on each group of data, calculate the product of the matching degree and the feature weight, use the sum function in NumPy to sum the products of each group to obtain the weighted sum, and sum the weighted sum results of each computing unit again using the sum function in NumPy to generate the final weighted matching result. Set the axis to 0 to summarize each group of results by column; S303: Based on the weighted matching result, perform threshold judgment. Use the where function in NumPy to compare the comparison result with the set qualified range. Set the qualified threshold to 0.8. If the result exceeds the threshold, mark it as unqualified to generate a judgment result. Use the array function in NumPy to construct an array to ensure that each matching result can be accurately marked as qualified or unqualified to generate a judgment result.
[0028] Please refer to Figure 5 , and the specific steps to generate the predicted result of resistance change are as follows: S401: Based on the judgment result, extract the current resistance value, compare it with the historical resistance data, and use random forest to fit the current resistance value and the historical data to calculate the difference between the current measurement value and the historical data to generate resistance change comparison data; S402: Based on the resistance change comparison data, perform trend curve analysis, compare the change trends of historical data and current data, calculate the increase or decrease amplitude of the trend change, and generate a resistance change trend trajectory; S403: Based on the resistance change trend trajectory, compare the difference between the prediction point and historical data, determine whether it exceeds the preset error range. If it exceeds, adjust the detection parameters and redefine the trend level, and generate a resistance change prediction result; S401: Based on the determination result, extract the current resistance value, compare it with the historical resistance data, use the random forest regression method, use the fit function to fit the current resistance value and historical data, calculate the difference between the current measurement value and historical data, generate resistance change comparison data, ensure reproducibility by setting random_state to 42, use the predict method to generate a prediction result of the resistance value, and obtain the final resistance change comparison data; S402: Based on the resistance change comparison data, perform trend curve analysis, analyze the change trends of historical data and current data, identify significant change points in the data, calculate the change amplitude, generate a resistance change trend trajectory, calculate the increase or decrease amplitude of the change, ensure that the trend change accurately reflects, generate a trend graph of the resistance change, and finally generate a resistance change trend trajectory; S403: Based on the resistance change trend trajectory, compare the difference between the prediction point and historical data, calculate the absolute difference between the prediction point and historical data, determine whether it exceeds the preset error range, set the error threshold to 0.1. If the result exceeds the threshold, adjust the detection parameters, redefine the trend level, calculate the average value of the current trend, and finally generate a resistance change prediction result.
[0029] Random forest, according to the formula:
[0030] Where: is the th sample, is the total number of decision trees, is the th decision tree's prediction value for the th sample, is the th sample's input feature, is the data distribution coefficient, is the feature importance weighting term, is the noise suppression coefficient, is the noise measure, is the fitting accuracy coefficient, is the error term in the round training; Execution process: First, obtain the resistance value of each sample from the measurement system , representing the resistance measurement value of the th planar rectifier diode, is provided as an input feature to the model. Then, using the random forest regression method, each sample is predicted through multiple decision trees. represents the predicted value of the sample by the th tree. The predicted value of each tree will be weighted according to different feature importances. is the feature weighting term, used to enhance the attention to important features. is the noise suppression coefficient, reducing the impact of noise in the prediction. At the same time, is the noise measure, measuring the magnitude of noise in each round of training, helping to further suppress unimportant noise data. is the fitting accuracy coefficient, ensuring that the training process avoids overfitting and underfitting. The error term of each tree reflects the prediction error of the model in the current tree, and the error term will be dynamically adjusted during each round of training to improve the prediction accuracy of the model. Finally, by taking the weighted average of the prediction results of all trees, the final resistance prediction value of the th sample is obtained. , combining the prediction results of multiple trees and the weighted adjustment of features, improves the accuracy and reliability of detection.
[0031] Please refer to Figure 6 for the specific steps to generate the measurement parameter adjustment scheme: S501: Based on the prediction result of the resistance change, calculate the deviation between the current measurement value and the predicted resistance value, compare the gap between the current measurement value and the predicted value one by one, and generate measurement deviation data. S502: Based on the measurement deviation data, determine whether the deviation exceeds the set threshold range. If it exceeds the range, mark it as abnormal and generate a deviation overrun judgment result. S503: Based on the deviation overrun judgment result, adjust the measurement parameters according to the overrun judgment, modify the measurement setting parameters respectively, and generate a measurement parameter adjustment scheme. S501: Based on the prediction result of the resistance change, calculate the deviation between the current measurement value and the predicted resistance value, compare the gap between the current measurement value and the predicted value one by one, use the abs function in the NumPy library to calculate the absolute difference between each pair of measurement values and predicted values, obtain the measurement deviation data, calculate the difference between each pair of data to ensure that the gap between each data point is accurately measured, and finally generate the measurement deviation data. S502: Based on the measurement deviation data, determine whether the deviation exceeds the set threshold range. The set threshold is 0.1. Use the where function in NumPy to determine whether each deviation value exceeds the threshold. If it exceeds the range, mark it as abnormal, generate the deviation over-limit judgment result. By setting the threshold to 0.1 for comparison, ensure the accuracy of the deviation judgment, and finally generate the deviation over-limit judgment result; S503: Based on the deviation over-limit judgment result, according to the over-limit judgment, adjust the measurement parameters. Modify the measurement setting parameters respectively. Use the mean and std functions in NumPy to calculate the average value and standard deviation of the current deviation, adjust the measurement parameters respectively, apply the modified settings to the device parameters, and generate the measurement parameter adjustment plan. Update each parameter by adjusting the step size with an increment of 0.05 to generate the measurement parameter adjustment plan.
[0032] The measurement setting parameters include measurement current, measurement frequency, measurement period, sampling rate, ambient temperature, signal filtering setting, measurement mode, and test voltage range.
[0033] The above are only the preferred embodiments of the present invention and do not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.
Claims
1. A method for detecting the resistance of a planar rectifier diode, characterized in that, The following steps are involved: Step 1: Obtain current and voltage data through sensor detection, perform segmented differential processing, calculate the difference gradient between each current segment and voltage segment, retain key information based on the gradient change, and generate a local difference matrix; Step 2: By sparse coding the local difference matrix, using the gradient boosting regression tree, extracting important eigenvalues in the matrix, optimizing the calculation process, removing redundant information and compressing gradient data, generating low-dimensional feature vectors, and adjusting the gradient weights to strengthen effective information, a sparse feature vector group is obtained; Step 3: Based on the sparse feature vector group, perform fast parallel calculation, optimize the matrix multiplication process, compare the feature vector with the existing judgment rules, judge whether the resistor is qualified by judging the similarity between the feature vector and the rule, and generate a judgment result; Step 4: Based on the determination result, random forest is used to compare with historical resistance data, combined with the current measurement value, trend prediction is performed, and the potential trend of future changes in resistance value is calculated. When the trend is abnormal, dynamic adjustment is performed by comparing with the existing data to generate a resistance change prediction result; Step 5: Based on the resistance change prediction result, compare it with the deviation of the current measurement value. When the deviation exceeds the set threshold, adjust the measurement parameters and generate a measurement parameter adjustment plan.
2. The resistance detection method for the planar rectifying diode according to claim 1, wherein The specific steps of generating the local difference matrix are: The current and voltage data are obtained by sensor detection, and the data are segmented according to a predetermined time interval. The corresponding current and voltage values are extracted in each time period, and the difference between the current and voltage in the time period is calculated to obtain the difference value between each current segment and voltage segment, and generate segment difference data; Based on the segmented difference data, difference gradient calculation is performed, the change amplitude of each current segment and voltage segment is compared, the part with larger change is determined and retained, low-frequency noise and irrelevant data are eliminated, and a gradient change matrix is generated; Based on the gradient change matrix, the region with relatively stable changes is screened out, redundant gradient information is removed, data that has a key impact on the relationship between current and voltage is retained, and a local difference matrix is generated.
3. The resistance detection method for the planar rectifier diode according to claim 1, wherein The specific steps of generating the sparse feature vector group are: Based on the local difference matrix, a gradient boosting regression tree is used to compress the data in the matrix, the importance of each data segment is analyzed segment by segment, the features of each data segment are weighted and adjusted, redundant data are removed, and a compressed data set is generated; Based on the compressed data set, by weighted adjustment, according to the importance of each data segment, the feature weights in the data are adjusted to strengthen the effective information therein, and the adjusted feature data is generated; Based on the adjusted feature data, low-dimensional features in the data are extracted, high-dimensional feature information is compressed to low dimensions, and a sparse feature vector group is generated.
4. The method for detecting the resistance of the planar rectifying diode according to claim 1, wherein The gradient boosting regression tree is based on the formula: Wherein: is the predicted result of the resistance value of the th sample in the th round of iteration, is the predicted result of the resistance value of the th sample in the previous round of iteration, is the learning rate, is the predicted output based on the input data after the th round of training, is the data distribution coefficient, is the feature importance weighting term, is the noise suppression coefficient, is the noise measure, is the fitting accuracy coefficient, is the error term in each round of training process.
5. The method for detecting the resistance of a planar rectifier diode according to claim 3, wherein The low-dimensional features in the extracted data are converted into a set of irrelevant principal components through principal component analysis, and the number of principal components to be retained is selected according to the variance contribution, and the importance of each feature is evaluated, and the features that have a significant impact on the prediction results are selected. At the same time, the features are screened to obtain a compressed data set.
6. The method for detecting the resistance of the planar rectifier diode according to claim 1, characterized in that The specific steps for generating the determination result are as follows: Based on the sparse feature vector group, perform feature grouping. Divide the feature vectors into multiple calculation units according to dimensions. Each unit calculates the matching degree separately, and compare the features within each group with the determination rules one by one to generate a grouped matching data set; Based on the grouped matching data set, perform element-wise multiplication on each group of data, calculate the product of the matching degree and the feature weight to obtain the weighted sum, and sum each calculation unit to generate a weighted matching result; Based on the weighted matching result, perform threshold judgment, compare the comparison result with the set qualified range. If the result exceeds the threshold, mark it as unqualified to generate the determination result.
7. The resistance detection method for the planar rectifier diode according to claim 1, characterized in that, The specific steps for generating the predicted result of resistance change are as follows: Based on the determination result, extract the current resistance value, and compare it with the historical resistance data. Use a random forest to fit the current resistance value and the historical data to calculate the difference between the current measurement value and the historical data to generate resistance change comparison data; Based on the resistance change comparison data, perform trend curve analysis, compare the change trends of the historical data and the current data, and calculate the increase or decrease amplitude of the trend change to generate a resistance change trend trajectory; Based on the resistance change trend trajectory, compare the difference between the prediction point and the historical data, judge whether it exceeds the preset error range. If it exceeds, adjust the detection parameters and redefine the trend level to generate the predicted result of resistance change.
8. The method for detecting the resistance of the planar rectifying diode according to claim 1, characterized in that, The random forest is calculated according to the formula: Wherein: is the th sample, is the total number of decision trees, is the predicted value of the th decision tree for the th sample, is the input feature of the th sample, is the data distribution coefficient, is the feature importance weighting term, is the noise suppression coefficient, is the noise measure, is the fitting accuracy coefficient, is the error term in the round training.
9. The method for detecting the resistance of the planar rectifying diode according to claim 1, characterized in that The specific steps for generating the measurement parameter adjustment plan are as follows: Based on the predicted result of resistance change, calculate the deviation between the current measurement value and the predicted resistance value, and compare the gap between the current measurement value and the predicted value one by one to generate measurement deviation data; Based on the measurement deviation data, judge whether the deviation exceeds the set threshold range. If it exceeds the range, mark it as abnormal to generate a deviation overrun judgment result; Based on the deviation overrun judgment result, adjust the measurement parameters according to the overrun judgment, and modify the measurement setting parameters respectively to generate a measurement parameter adjustment plan.
10. The method for detecting resistance of a planar rectifying diode according to claim 3, characterized in that, The measurement setting parameters include measurement current, measurement frequency, measurement period, sampling rate, ambient temperature, signal filtering setting, measurement mode, and test voltage range.