Optimized trace analysis method based on deep learning

Abstract

The invention discloses an optimized trace analysis method based on deep learning, and relates to the field of instrument back noise analysis, and the method comprises the following steps: collecting original data after sample trace analysis, and making a data set; selecting an LSTM model based on pooling; extracting features in the spectral signals through a convolutional layer and a pooling layer of the model; main features of the data are reserved through a maximum pooling layer, and redundant information is removed; adopting a gradient descent method to cooperate with the training sample set to perform concentration fitting regression training on the model; analyzing the relationship between the model output and the target value, and evaluating the accuracy rate and the decision coefficient of the model; and finally outputting a trace analysis result of the sample. According to the method, the collected LSTM model sample set for trace analysis is subjected to preprocessing, screening and data adjustment, and a data analysis result with higher accuracy is obtained.

Description

Technical Field

[0001] The present invention relates to the field of instrument background noise analysis, and in particular to an optimized trace analysis method based on deep learning. Background Art

[0002] Due to the high background noise of the instrument itself, the high blank and parallel phenomenon of elements in trace and ultra-trace analysis is more obvious; the existing solutions are as follows: Electrochemical noise analysis method: Electrochemical noise (EN) technology can effectively monitor the corrosion process of materials. By statistically analyzing EN data, a specific spectrum is obtained to characterize the initiation and development of corrosion. This method can be used to monitor the corrosion behavior of X70 steel and its welds in Cl-containing high pH solutions, where the amplitude of the current noise reflects the initiation and development of local corrosion.

[0003] Signal processing methods: In laser trace gas monitors, fast scanning, integration, and high-precision fitting of absorption spectra can be used to identify and reduce measurement noise through high-performance software control. The principle of spectral information analysis is to load the spectrum into an array and process it using filters, Fourier analysis, multivariate fitting, and component analysis to reduce noise.

[0004] ICEEMDAN method: ICEEMDAN (Empirical Mode Decomposition with Adaptive Noise) is a new adaptive signal decomposition method that can handle nonlinear and non-stationary signals and has good local time-frequency analysis capabilities. This method combines empirical mode decomposition (EMD) with adaptive noise control technology, and introduces adaptive mean shift (AEMD) and adaptive extreme value large jump (AEMD-H) on the basis of EMD to achieve signal decomposition and reduce the impact of noise.

[0005] The present invention uses deep learning to improve optimization and solve the problem of high blank values ​​and parallelism of elements in trace analysis. Summary of the invention

[0006] The present invention provides an optimized trace analysis method based on deep learning to solve the problems of the prior art.

[0007] In a first aspect, the present invention provides an optimized trace analysis method based on deep learning, comprising: Collect raw data after trace analysis and create data sets, including training sample sets and test sample sets; Select the pooling-based LSTM model, including input layer, convolution layer, pooling layer, and fully connected layer; The convolution layer and pooling layer of the model are used to extract features from the spectral signal. The one-dimensional convolution layer is used to extract the characteristic relationship between sequence features, sequence boundary information and gas concentration. The main features of the data are retained through the maximum pooling layer, redundant information is removed, and the complexity of the neural network is simplified; The model is trained for concentration fitting regression using a gradient descent method in conjunction with a training sample set; Optimize training results based on parameter adjustment; Perform performance tests on the trained model based on the test sample set and calculate the model accuracy; Analyze the relationship between model output and target value, and evaluate the accuracy and determination coefficient of the model; Finally, the trace analysis results of the sample are output.

[0008] Furthermore, the raw data after the sample trace analysis is collected to produce a data set, including a training sample set and a test sample set, which specifically includes processing the raw data corresponding to the sample data set as follows: Construct element blank analysis model to correct raw data; The element blank analysis model is used to: Input the numerical type in the numbered input layer and calculate the variance of each type of parameter in the same batch; The collected raw data are classified and fitted to construct the extreme value neighborhood. The expected values ​​of multiple categories in the extreme value neighborhood are calculated and the percentage difference between them and the extreme value is analyzed.

[0009] Furthermore, after the element blank analysis model calculates the values, the values ​​are statistically analyzed: Eliminate extreme points and adjacent values ​​whose extreme value difference percentage is greater than the preset percentage.

[0010] Furthermore, the step of screening out extreme value points and adjacent values ​​whose extreme value difference percentage is greater than a preset percentage specifically includes: Set the window and use the window shift to select multiple adjacent values ​​in the neighborhood of the fitted extreme value; Step A1, adjusting the window position, when the extreme value difference percentage is less than the preset percentage, sliding the window toward the extreme value point, and continuously calculating the extreme value difference percentage of the window circle until the screening judgment condition is met, and the screening judgment condition is: the screening extreme value difference percentage is greater than the preset percentage; Step A2, adjusting the window size, enlarging the window size at the window position that meets the screening judgment condition, and continuing to execute step A1; Step A1 and step A2 are executed repeatedly until the window slides to cover all the value areas.

[0011] Furthermore, the initial value of the window setting is the minimum window size, and each window selection point includes at least three numerical points.

[0012] Furthermore, the window is initially set to have an origin of the first quadrant of the fitting data, and the components of the initial sliding direction move along the fitting curve in the positive direction of the x-axis and the positive direction of the y-axis.

[0013] Furthermore, after the element blank analysis model screens out the extreme value points that meet the judgment conditions, each batch of data is screened out of the extreme value points that meet the judgment conditions and then parallel processing is performed: Fit and calculate expected values ​​for different batches of the same type of data; Sort the data of the same type according to the expected value and construct the probability density function curve of Gaussian distribution; Select the Gaussian distribution empirical rule in the 68% data-95% data segment, change the expectation of the selected data toward the adjacent 68% data segment, and calculate the expected difference; The expected difference is accumulated for each value corresponding to the selected data.

[0014] Preferably, the calculation of expected values ​​for fitting different batches of the same type of data specifically includes: The data is divided into each continuous segment, and the expectation is calculated for the data of each segment; Set the variance of each cut-out segment data to be within the same preset fluctuation range; Sort each selected segment by segment number, and align segment numbers of the same type of data in different batches; The numerical expectation of each corresponding segment number is uniformly adjusted to the expectation of the numerical expectations of all selected segments corresponding to the segment number.

[0015] Preferably, the calculation of expected values ​​for fitting different batches of the same type of data further includes: The data is divided into each continuous segment, and the expectation is calculated for the data of each segment; Set the variance of each cut-out segment data to be within the same preset fluctuation range; For each selected segment, align the number of segments according to the same type of data in different batches; The variance fluctuation of the expected value of each selected segment of the aligned data is within the selected segment variance threshold range; This includes one serial number segment from one batch corresponding to serial number segments from multiple different batches, or one serial number segment from one batch corresponding to multiple segments from another batch; The processing expectation for the data segment that meets the variance fluctuation within the segment variance threshold range is the expectation of the data segment expectation.

[0016] The optimized trace analysis method based on deep learning provided by the present invention uses data selection and fitting extreme value screening and judgment to solve data anomalies caused by high blank values ​​and non-parallelism, improve the accuracy of data detection, and has low calculation amount, which is beneficial to the final detection result of trace analysis samples.

[0017] The present invention preprocesses the collected LSTM model sample set for trace analysis to screen out and adjust the data to obtain data analysis results with higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the present invention, and do not constitute a limitation of the embodiments of the present invention. In the drawings: Figure 1 A flow chart of an optimized trace analysis method based on deep learning provided as an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0019] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0020] The present invention considers that in the process of trace analysis, due to the influence of various elements, different types of blank values ​​are generated. It may be that the purity of chemical reagents is insufficient, or it may be that the vessel is contaminated or the operator is contaminated. The trace analysis blank is high or non-parallel phenomenon occurs due to various reasons; The blank value in trace analysis refers to the result obtained by conducting the test under the same conditions and procedures as the component to be tested without adding the component to be tested; The present invention matches different batches of data selections with different expected values ​​based on extreme value screening judgment conditions and data selections, greatly reduces the impact of erroneous data, performs model analysis on the data after screening data and expected adjustment data, and outputs sample trace analysis results with improved accuracy.

[0021] The optimized trace analysis method based on deep learning provided by the present invention is intended to solve the above technical problems of the prior art.

[0022] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0023] Embodiment 1: like Figure 1 As shown, this embodiment provides an optimized trace analysis method based on deep learning, including the following processing steps: Step 1: Collect raw data after sample trace analysis; Step 2: construct an element blank analysis model to correct the original data; Step 3: After the element blank analysis model calculates the values, statistical analysis is performed on the values ​​to screen out extreme value points and adjacent values ​​whose extreme value difference percentage is greater than a preset percentage; Step 4: After the extreme value points that meet the judgment conditions are screened out in the element blank analysis model, the extreme value points that meet the judgment conditions are screened out for each batch of data, and then parallel processing is performed; Step 5: Use the original data after parallel processing to create a data set, including a training sample set and a test sample set; Step 6: Select the pooled LSTM model, including the input layer, convolution layer, pooling layer, and fully connected layer; Step 7: Extract features from the spectral signal through the convolution layer and pooling layer of the model. The one-dimensional convolution layer is used to extract the characteristic relationship between sequence features, sequence boundary information and gas concentration; Step 8: Main features of the data are retained through the maximum pooling layer, redundant information is removed, and the gradient descent method is used in conjunction with the training sample set to perform concentration fitting regression training on the model; Step 9: Based on the parameter optimization training results, the trained model is tested based on the test sample set to calculate the model accuracy; Step 10: Analyze the relationship between the model output and the target value, evaluate the accuracy and determination coefficient of the model, and finally output the trace analysis results of the sample.

[0024] In this embodiment, the core processes the sampled raw data before making the data set, and pre-processes the collected LSTM model sample set for trace analysis to filter out and adjust the data to obtain a data analysis result with higher accuracy.

[0025] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0026] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0027] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0028] Those skilled in the art will appreciate that the embodiments of the present invention may be provided as methods or systems. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0029] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0030] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

[0031] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention is indicated by the claims above.

[0032] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An optimized trace analysis method based on deep learning, characterized in that: include: Collect raw data after sample trace analysis and create data sets, including training sample sets and test sample sets; Select the pooling-based LSTM model, including input layer, convolution layer, pooling layer, and fully connected layer; The convolution layer and pooling layer of the model are used to extract features from the spectral signal. The one-dimensional convolution layer is used to extract the characteristic relationship between sequence features, sequence boundary information and gas concentration. The main features of the data are retained through the maximum pooling layer to remove redundant information; The model is trained for concentration fitting regression using a gradient descent method in conjunction with a training sample set; Optimize training results based on parameter adjustment; Perform performance tests on the trained model based on the test sample set and calculate the model accuracy; Analyze the relationship between model output and target value, and evaluate the accuracy and determination coefficient of the model; Finally, the trace analysis results of the sample are output.

2. The optimized trace analysis method based on deep learning according to claim 1, characterized in that: The raw data after the sample trace analysis is collected to produce a data set, including a training sample set and a test sample set, specifically including processing the raw data corresponding to the sample data set as follows: Construct element blank analysis model to correct raw data; The element blank analysis model is used to: Input the numerical type in the numbered input layer and calculate the variance of each type of parameter in the same batch; The collected raw data are classified and fitted to construct the extreme value neighborhood. The expected values ​​of multiple categories in the extreme value neighborhood are calculated and the percentage difference between them and the extreme value is analyzed.

3. The optimized trace analysis method based on deep learning according to claim 2, characterized in that: After the element blank analysis model calculates the values, perform statistical analysis on the values: Eliminate extreme points and adjacent values ​​whose extreme value difference percentage is greater than the preset percentage.

4. The optimized trace analysis method based on deep learning according to claim 3, characterized in that: The step of screening out extreme value points and adjacent values ​​whose extreme value difference percentage is greater than a preset percentage specifically includes: Set the window and use the window shift to select multiple adjacent values ​​in the neighborhood of the fitted extreme value; Step A1, adjusting the window position, when the extreme value difference percentage is less than the preset percentage, sliding the window toward the extreme value point, and continuously calculating the extreme value difference percentage of the window circle until the screening judgment condition is met, and the screening judgment condition is: the screening extreme value difference percentage is greater than the preset percentage; Step A2, adjusting the window size, enlarging the window size at the window position that meets the screening judgment condition, and continuing to execute step A1; Step A1 and step A2 are executed repeatedly until the window slides to cover all the value areas.

5. The optimized trace analysis method based on deep learning according to claim 4, characterized in that: The initial value of the window setting is the minimum window size, and each window selection point includes at least three numerical points.

6. The optimized trace analysis method based on deep learning according to claim 5, characterized in that: The window is initially set to an origin of the first quadrant of the fitting data, and the components of the initial sliding direction move along the fitting curve in the positive direction of the x-axis and the positive direction of the y-axis.

7. The optimized trace analysis method based on deep learning according to claim 6, characterized in that: After the element blank analysis model screens out the extreme value points that meet the judgment conditions, each batch of data is screened out of the extreme value points that meet the judgment conditions and then parallel processing is performed: Fit and calculate expected values ​​for different batches of the same type of data; Sort the data of the same type according to the expected value and construct the probability density function curve of Gaussian distribution; Select the Gaussian distribution empirical rule in the 68% data-95% data segment, change the expectation of the selected data toward the adjacent 68% data segment, and calculate the expected difference; The expected difference is accumulated for each value corresponding to the selected data.

8. The optimized trace analysis method based on deep learning according to claim 7, characterized in that: The calculation of expected values ​​for fitting different batches of the same type of data specifically includes: The data is divided into each continuous segment, and the expectation is calculated for the data of each segment; Set the variance of each cut-out segment data to be within the same preset fluctuation range; Sort each selected segment by segment number, and align segment numbers of the same type of data in different batches; The numerical expectation of each corresponding segment number is uniformly adjusted to the expectation of the numerical expectations of all selected segments corresponding to the segment number.

9. The optimized trace analysis method based on deep learning according to claim 7, characterized in that: The calculation of expected values ​​for fitting data of the same type in different batches specifically includes: The data is divided into each continuous segment, and the expectation is calculated for the data of each segment; Set the variance of each cut-out segment data to be within the same preset fluctuation range; For each selected segment, align the number of segments according to the same type of data in different batches; The variance fluctuation of the expected value of each selected segment of the aligned data is within the selected segment variance threshold range; This includes one serial number segment from one batch corresponding to serial number segments from multiple different batches, or one serial number segment from one batch corresponding to multiple segments from another batch; The processing expectation for the data segment whose variance fluctuation is within the segment variance threshold range is the expectation of the data segment expectation.

Images