TWI865976B - Error data processing method of electrochemical detection - Google Patents

Abstract

An error data processing method of electrochemical detection includes following steps: training data obtaining, characterization processing, and characteristics inducing and learning. The present invention obtains the voltage and current variation curves of test pieces manufactured in different batches through the training data obtaining step, and establishes a batch error database. Through the characterization processing step, the voltage and current variation curve undergoes the various characterization processes, including voltage, current, and peak value curve area. Through the characteristics inducing and learning step, the characteristics are induced and learned, and a correction database is established. Therefore, the present invention is able to perform an automatic error calibration on new test pieces of unknown batches, so as to achieve the purpose of improving measurement accuracy.

Description

Electrochemical detection error data processing method

The present invention relates to a method for processing error data, and in particular to a method for processing electrochemical detection error data.

Electrochemical analysis is a fast and simple testing technology. Electrochemical analysis test strips are placed on the test strip and electrochemical measurements are performed. The value of the test strip is determined by the obtained electrochemical change curve. For example, blood sugar testing and concentration testing of specific components in liquids can be performed.

In order to avoid contamination caused by repeated use, most test strips are used in a disposable manner. The manufacturing method is usually to form a circuit structure on an insulating substrate by screen printing, thereby achieving the advantages of large-scale manufacturing and reducing costs. However, after multiple screen printings, due to the wear of the physical structure of the screen itself, the printed circuit structure will differ from the original design in thickness and printing area. That is, the test strips manufactured at different times and batches will have different degrees of errors. That is, the impedance of the electrode and the degree of chemical reaction between the electrode and the liquid to be tested will be different due to the above manufacturing differences. As a result, when the test strips from different batches are subjected to electrochemical analysis, the measured results will have different degrees of errors, affecting the accuracy of the test.

The main purpose of this case is to solve the problem of inaccurate electrochemical test results caused by different degrees of manufacturing errors in test strips manufactured in different batches.

To achieve the above object, the present invention provides an electrochemical detection error data processing method, which includes a training data acquisition step, a feature processing step and a feature induction learning step.

In the training data acquisition step, multiple voltage-current variation curves of test specimens manufactured in multiple different batches are obtained by electrochemical voltammetry, and a batch error database is established corresponding to the multiple batches of known test error data of the test specimens; and in the characterization processing step, after the training data acquisition step, the voltage-current variation curve is further characterized to obtain multiple training data corresponding to the test specimens, and the training data includes a peak voltage data, a peak The present invention relates to a feature induction learning step, wherein the training data and the batch error database are input into a feature learning correction model after the characterization processing step. The feature learning correction model uses the batch error database to perform feature induction learning on the training data to build a correction database. The correction database includes a plurality of feature data that have been summarized corresponding to different batches and a plurality of correction data corresponding to the feature data.

Thus, the feature learning correction model established by the method of the present invention can further determine the batch manufacturing source of the test strip and the corresponding correction value required for the test strip according to the features provided by the training data and the batch test error data, thereby achieving the purpose of improving the measurement accuracy of the electrochemical test strip.

In order to facilitate the explanation of the central idea of the present invention in the above-mentioned creative content column, a specific embodiment is used for expression. It is necessary to first explain that various objects in the embodiment are drawn in a proportion suitable for enumeration and description, rather than in a proportion of actual elements.

Please refer to FIG. 1 to FIG. 8 , which discloses an electrochemical detection error data processing method 100 of an embodiment of the present invention, which includes a training data acquisition step S1, a characterization processing step S2, and a feature induction learning step S3, thereby establishing a feature learning correction model, and correcting the batch error of the detection test strip used in the electrochemical analysis through the feature learning correction model. Among them, in the preferred embodiment of the present invention, it further includes a model verification step S4 and an actual test strip correction step S5.

Training data acquisition step S1: obtaining a plurality of voltage-current variation curves of a plurality of test strips manufactured in different batches by electrochemical voltammetry, and establishing a batch error database corresponding to a plurality of known batch error data of the test strips. In this embodiment, the test strip is used to detect the caffeine content or chlorogenic acid content in a liquid; in this embodiment, the electrochemical voltammetry can be differential pulse voltammetry (DPV) or linear sweep voltammetry (LSV), and the voltage-current variation curve corresponding to the differential pulse voltammetry and the linear sweep voltammetry is divided into a plurality of first voltage-current variation curves and a plurality of second voltage-current variation curves (as shown in FIG. 3 and FIG. 4 ).

In this embodiment, if the test strip is used to detect the caffeine content in the liquid, then in the training data acquisition step S1, the first voltage-current variation curve (as shown in FIG3 ) is obtained by the differential pulse voltammetry. As shown in FIG3 , the first voltage-current variation curve has two peak curves L1 and L2.

In this embodiment, if the test strip is used to detect the chlorogenic acid content in the liquid, in the training data acquisition step S1, the first voltage-current variation curve (as shown in FIG. 3 ) is first obtained by the differential pulse voltammetry, and then the second voltage-current variation curve (as shown in FIG. 4 ) is obtained by the linear scanning voltammetry. As shown in FIG. 4 , the second voltage-current variation curve has a peak curve L3.

Characterization processing step S2: After the training data acquisition step S1, the plurality of voltage-current variation curves are further characterized to obtain a plurality of training data respectively corresponding to the test specimens, wherein the training data includes a peak voltage data, a peak current data and a peak curve area data.

In this embodiment, if the test strip is used to detect the caffeine content in the liquid, then in the characterization processing step S2, the first voltage-current variation curve is characterized to obtain the training data, and the training data includes a first peak voltage data, a first peak current data and a first peak curve area data obtained from the peak curve L1, and a second peak voltage data, a second peak current data and a second peak curve area data obtained from the peak curve L2 (as shown in FIG. 3 ).

As shown in FIG. 3 , in this embodiment, the peak curve L1 has a peak point P1 and two end points E1 and E2, the peak curve L2 has a peak point P2 and two end points E3 and E4, the first peak voltage data and the first peak current data are the voltage value and the current value of the peak point P1, the first peak curve area data are a peak area A1 roughly enclosed by the peak point P1, the end point E1 and the end point E2, the second peak voltage data and the second peak current data are the voltage value and the current value of the peak point P2, and the second peak curve area data are a peak area A2 roughly enclosed by the peak point P2, the end point E3 and the end point E4.

In this embodiment, if the test strip is used to detect the chlorogenic acid content in the liquid, then in the characterization step S2, the first voltage-current variation curve and the second voltage-current variation curve are characterized to obtain the training data, and the training data includes the first peak voltage data, the first peak current data and the first peak curve area data obtained from the peak curve L1 (as shown in FIG. 3 ), and a third peak voltage data, a third peak current data and a third peak curve area data obtained from the peak curve L3 (as shown in FIG. 4 ).

As shown in FIG. 4 , in this embodiment, the peak curve L3 has a peak point P3 and two end points E5 and E6, the third peak voltage data and the third peak current data are the voltage value and the current value of the peak point P3, and the third peak curve area data is a peak area A3 roughly enclosed by the peak point P3, the end point E5 and the end point E6.

In another embodiment of the present invention, the training data may further include a peak slope data, a peak displacement data and a peak difference data, etc. It is conceivable that the more training data there are, the more accurate the judgment can be. However, too much training data will also cause the training time to be extended or the subsequent detection and judgment time to be extended. Among them, as shown in Figure 3, taking the first voltage-current variation curve as an example, the peak slope data includes the slope value from the end point E1 to the peak point P1 and the slope value from the peak point P1 to the end point E2; the peak displacement data is the lateral displacement of the same peak point in any two of the voltage-current variation curves; as shown in Figure 3, taking the first voltage-current variation curve as an example, when there are two peak curves in one voltage-current variation curve at the same time, there will be the peak gap data, and the peak gap data is the vertical gap between the two peak points P1 and P2.

Furthermore, the batch detection error data also corresponds to the peak slope data and has a batch of peak slope error data, and the batch peak slope error data is that the test pieces of different batches have different slope error interval values, for example, the slope error interval value of the test pieces of the second batch is ±3% to ±5% (that is, ±3% to ±5% of the slope value of the test pieces of the first batch); the batch detection error data also corresponds to the peak displacement data and has a batch of peak displacement error data, and the batch peak displacement error data is that the test pieces of different batches have different peak displacement error interval values. For example, the peak displacement error interval of the second batch of the test pieces is ±3% to ±5% (i.e., ±3% to ±5% of the peak lateral displacement of the first batch of the test pieces); the batch test error data also corresponds to a batch of peak difference error data corresponding to the peak difference data, and the batch peak difference error data means that different batches of the test pieces have different peak difference error intervals, for example, the peak difference error interval of the second batch of the test pieces is ±3% to ±5% (i.e., ±3% to ±5% of the vertical difference between the two peak points of the first batch of the test pieces).

Feature induction learning step S3: After the feature processing step S2, the training data and the batch error database are input into the feature learning correction model. The feature learning correction model uses the batch error database to perform feature induction learning on the training data to build a correction database. The correction database includes a plurality of feature data that have been inferred corresponding to different batches and a plurality of correction data that respectively correspond to the feature data. In this embodiment, the feature learning correction model performs feature induction learning through an artificial neural network (ANN).

In this embodiment, the training data is first normalized by a normalization function and then input into the feature learning correction model. Thus, the normalization process can significantly reduce the amount of data to be processed by the feature learning correction model during the feature inductive learning process. Therefore, the normalization process can improve the feature inductive learning speed of the feature learning correction model. In this embodiment, the normalization function is: .

Model verification step S4: After the feature induction learning step S3, a test data is input into the feature learning correction model. The test data is the test specimen of a specific batch known to the tester, but the feature learning correction model is unaware of its batch source. Therefore, it can be used to test whether the feature learning correction model can accurately determine the result required for correction of the aforementioned test specimen. Therefore, after the feature learning correction model obtains the test data, the feature learning correction model corrects the test data through the correction database to output a test correction data, and compares the test correction data with an actual correction data, wherein, if the error between the test correction data and the actual correction data is lower than or equal to an error standard, it means that the data of the correction database has completed the correction training. In this embodiment, the error standard is set to ±10% of the actual correction data. In this way, the feature learning correction model can verify the correctness of the correction database through the model verification step S4.

For example, as shown in Figures 5 to 8, if the points presented on the graph are more concentrated in the vertical direction and closer to the standard line, it means that the error between the test calibration data and the actual calibration data is smaller. As shown in Figures 5 and 6, they are batch error graphs obtained by performing batch error correction on the test strips used to detect caffeine using the learning correction method and the feature learning correction model, respectively. It can be seen that the points presented in Figure 6 are more concentrated and closer to the standard line than the points presented in Figure 5 (that is, the errors between most of the test calibration data and the actual calibration data are lower than or equal to the error standard). Therefore, it can be seen that the data in the calibration database has indeed completed the calibration training, and the present invention does have a higher calibration accuracy than the learning calibration method; As shown in FIG. 7 and FIG. 8 , they are batch error diagrams obtained by performing batch error correction on the test strip for detecting chlorogenic acid using the learning correction method and the feature learning correction model, respectively. It can be seen that the points presented in FIG. 8 are more concentrated and closer to the standard line than the points presented in FIG. 7 (i.e., the errors between most of the test correction data and the actual correction data are lower than or equal to the error standard). Therefore, it can be seen that the data in the correction database has indeed completed the correction training, and the present invention does have a higher correction accuracy than the learning correction method.

Actual test strip calibration step S5: When actual test strip testing is required, the user titrates the test sample onto a new test strip of an unknown batch. The present invention obtains a new voltage-current variation curve of the new test strip through the electrochemical voltammetry, and performs the characterization processing on the new voltage-current variation curve to obtain a corresponding feature comparison data, and inputs the feature comparison data into the feature learning calibration model. The feature learning calibration model uses the calibration database to compare the feature comparison data to detect the batch source of the new test strip, and further calibrates the feature comparison data to obtain a final concentration data. It should also be noted that the model validation step S4 is not necessarily performed every time. If the manufacturing-related parameters and environmental variability are small, it is possible to directly perform the actual sample calibration step S5 after the feature induction learning step S3.

For example, if there is a new test strip with a manufacturing batch of the second batch and used to detect the caffeine content, and the batch detection error data of the second batch is an error value of 3%, then in the actual test strip calibration step S5, the new voltage-current variation curve of the new test strip is first obtained by the differential pulse voltammetry, and the characteristic comparison data corresponding to the new test strip (including the first peak voltage data, the first peak current data, the first peak curve area data, the second peak voltage data, the second peak current data and the second peak curve area data corresponding to the new test strip) is obtained from the new voltage-current variation curve by the characterization processing.

The feature learning correction model can use the correction database to compare the feature comparison data to find out that the manufacturing batch of the new test strip is the second batch and the correction value required for the new test strip is 3%. The feature learning correction model can further calibrate the feature comparison data and obtain the final concentration data. The final concentration data is the correct caffeine concentration that the new test strip should detect.

Thus, the present invention has the following advantages:

1. The feature learning calibration model established by the method of the present invention can obtain the manufacturing batch corresponding to each new test strip and the manufacturing error of the corresponding manufacturing batch based on the calibration database, and calibrate the new test strip accordingly to achieve the purpose of improving measurement accuracy.

2. The present invention uses the batch error database obtained in the training data acquisition step S1 and the training data including the peak voltage data, the peak current data and the peak curve area data obtained in the characterization processing step S2 to enable the feature learning correction model to perform fast and accurate feature induction learning in the shortest time to obtain the correct correction database.

Although the present invention is described with a preferred embodiment, those skilled in the art can make various changes without departing from the spirit and scope of the present invention. The above embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. All modifications or changes that do not violate the spirit of the present invention are within the scope of the patent application of the present invention.

100: Electrochemical detection error data processing method S1: Training data acquisition step S2: Characterization processing step S3: Feature induction learning step S4: Model verification step S5: Actual specimen calibration step L1: Peak curve L2: Peak curve L3: Peak curve P1: Peak point P2: Peak point P3: Peak point E1: End point E2: End point E3: End point E4: End point E5: End point E6: End point A1: Peak area A2: Peak area A3: Peak area

[Figure 1] is a block diagram of the electrochemical detection error data processing method of the embodiment of the present invention. [Figure 2] is a block diagram of the electrochemical detection error data processing method of the preferred embodiment of the present invention. [Figure 3] is a first voltage-current variation curve obtained by differential pulse voltammetry. [Figure 4] is a second voltage-current variation curve obtained by linear sweep voltammetry. [Figure 5] is a batch error diagram obtained by performing batch error correction on the test strip for detecting caffeine using the learning correction method. [Figure 6] is a batch error diagram obtained by performing batch error correction on the test strip for detecting caffeine using the feature learning correction model. [Figure 7] is a batch error diagram obtained by correcting the batch error of the test strip for detecting chlorogenic acid using the learning correction method. [Figure 8] is a batch error diagram obtained by correcting the batch error of the test strip for detecting chlorogenic acid using the feature learning correction model.

100: Electrochemical detection error data processing method

S1: Steps to obtain training data

S2: Characterization processing step

S3: Feature induction learning steps

Claims (9)

A method for processing electrochemical detection error data includes: a training data acquisition step: obtaining a plurality of voltage-current curves of a plurality of test specimens manufactured in different batches by electrochemical voltammetry, and establishing a batch error database corresponding to the known multiple batch detection error data of the test specimens, wherein the electrochemical voltammetry is differential pulse voltammetry (DPV) or linear sweep voltammetry (LSV). Voltammetry (LSV), and the voltage-current variation curves corresponding to the differential pulse voltammetry and the linear sweep voltammetry are divided into a plurality of first voltage-current variation curves and a plurality of second voltage-current variation curves; a characterization processing step: after the training data acquisition step, the voltage-current variation curves are further characterized to obtain a plurality of training data corresponding to the test strips, wherein the training data correspond to the first voltage-current variation curves and include a first peak voltage data, a second peak voltage data, a first Peak current data, a second peak current data, a first peak curve area data and a second peak curve area data; and a feature induction learning step: after the characterization processing step, the training data and the batch error database are input into a feature learning correction model, and the feature learning correction model uses the batch error database to perform feature induction learning on the training data to build a correction database, which includes a plurality of feature data that have been inferred corresponding to different batches and a plurality of correction data that respectively correspond to the feature data. The electrochemical detection error data processing method as described in claim 1, wherein in the feature induction learning step, the feature learning correction model performs feature induction learning through an artificial neural network (ANN). The electrochemical detection error data processing method as described in claim 1, wherein the training data further includes a peak slope data, a peak displacement data and a peak difference data. The electrochemical detection error data processing method as described in claim 1, wherein, after the feature induction learning step, there is further included a model verification step: a test data is input into the feature learning correction model, the feature learning correction model corrects the test data through the correction database to output a test correction data, and the test correction data is compared with an actual correction data, wherein, if the error between the test correction data and the actual correction data is lower than or equal to an error standard, it means that the data of the correction database has completed the correction training. The electrochemical detection error data processing method as described in claim 5, wherein after the feature induction learning step or the model verification step, there is further included an actual test strip calibration step: obtaining a new voltage-current variation curve of a new test strip of an unknown batch through the electrochemical voltammetry, and performing the feature processing on the new voltage-current variation curve to obtain a feature comparison data, and inputting the feature comparison data into the feature learning calibration model, the feature learning calibration model uses the calibration database to compare the feature comparison data to detect the batch source of the new test strip, and further calibrates the feature comparison data to obtain a final concentration data. The electrochemical detection error data processing method as described in claim 1, wherein the test strips are used to detect the caffeine content or chlorogenic acid content in a liquid. The electrochemical detection error data processing method as described in claim 6, wherein if the test strips are used to detect the caffeine content in the liquid, then in the training data acquisition step, the first voltage-current variation curves are obtained by the differential pulse voltammetry. As described in claim 7, in the electrochemical detection error data processing method, if the detection test strips are used to detect the chlorogenic acid content in the liquid, then in the training data acquisition step, the first voltage-current variation curves are first obtained by the differential pulse voltammetry, and then the second voltage-current variation curves are obtained by the linear scanning voltammetry. The electrochemical detection error data processing method as described in claim 8, wherein in the characterization processing step, the training data includes the first peak voltage data, the first peak current data and the first peak curve area data corresponding to the first voltage-current variation curves, and includes a third peak voltage data, a third peak current data and a third peak curve area data corresponding to the second voltage-current variation curves.

Images