Electrochemical multi-component simultaneous quantitative detection system and method based on one-dimensional convolutional neural network

Abstract

The application discloses an electrochemical multi-component simultaneous quantitative detection system and method based on a one-dimensional convolutional neural network, and belongs to the technical field of electrochemical analysis. The method first collects square wave voltammetry (SWV) signals of a mixed sample; then, the signals are subjected to baseline correction, filtering and normalization pretreatment; then, a one-dimensional convolutional neural network (1D CNN) regression model is constructed and trained, the model can automatically learn and extract deep features related to the concentration of each component from the pretreated complex overlapping signals; finally, the trained model is used to predict the sample to be measured, and the concentration of each component is output. The application innovatively combines 1D CNN with SWV technology, effectively solves the problem of multi-component signal overlap, and significantly improves the detection accuracy, sensitivity and automation degree. The system is suitable for simultaneous rapid detection of various components such as antioxidants, and has wide popularization and application value.

Description

Technical Field

[0001] This invention relates to the field of electrochemical analysis technology, specifically to a method and system for simultaneous quantitative detection of multiple electrochemical components using deep learning algorithms, which is particularly suitable for the rapid simultaneous detection of multiple components such as antioxidants. Background Technology

[0002] Electrochemical analysis offers advantages such as ease of operation, rapid response, and low cost, making it suitable for rapid on-site detection. Its core principle is to measure the changes in current, potential, or charge generated by the redox reaction of the analyte on the electrode surface to achieve qualitative and quantitative analysis of the target substance. However, due to the narrow electrochemical potential window, redox peaks tend to overlap when multiple components coexist, leading to difficulties in signal identification and quantification. Existing technologies often employ chemometric methods such as partial least squares (PLS) for processing, but their ability to resolve weak signals and complex overlapping signals is limited, and their model generalization ability is weak. Furthermore, while traditional chromatography offers high accuracy, its complex pretreatment and long processing time make it unsuitable for rapid on-site detection.

[0003] Therefore, improving the accuracy, sensitivity, and automation of electrochemical multi-component detection has become a pressing technical problem in this field. In recent years, artificial intelligence technologies, represented by deep learning, have brought revolutionary opportunities to analytical chemistry data analysis. Deep learning models, especially convolutional neural networks (CNNs), possess powerful capabilities for automatically extracting and abstracting nonlinear features, enabling them to learn complex mapping relationships from raw or slightly preprocessed data without relying on manually designed features or strict physicochemical assumptions. Applying this powerful feature learning capability of CNNs to one-dimensional electrochemical spectral signal analysis opens up a completely new technical path for solving the classic problem of overlapping peaks in multi-component systems.

[0004] Therefore, developing a new method and system that can deeply integrate advanced electrochemical sensing technology with intelligent deep learning algorithms to achieve high-precision, rapid, automated, and simultaneous quantitative detection of multi-component targets in complex samples is of great scientific significance and practical application value. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes a simultaneous quantitative detection system and method for multiple electrochemical components based on a one-dimensional convolutional neural network. Based on one-dimensional convolutional neural network (1D CNN) technology, and combined with a multi-component electrochemical quantitative detection method, this method utilizes deep neural networks to process and analyze electrochemical signals, thereby improving the accuracy and sensitivity of multi-component electrochemical quantitative detection.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: On the one hand, this invention proposes a method for simultaneous quantitative detection of multiple electrochemical components based on a one-dimensional convolutional neural network, comprising the following process steps: S1: The square wave voltammetry method is used to scan the mixed sample containing multiple target analytes and acquire a one-dimensional mixed current-potential signal; S2: Preprocess the hybrid current-potential signal; S3: Construct a one-dimensional convolutional neural network regression model, using the preprocessed signal as input and the concentration vector of each target analyte as output label, and train the one-dimensional convolutional neural network regression model; S4: Input the signal obtained after processing the mixed sample to be tested through steps S1 and S2 into the trained one-dimensional convolutional neural network regression model, and the output is the predicted concentration of each target analyte.

[0007] Furthermore, the preprocessing includes at least baseline correction, filtering and smoothing, and normalization; the filtering and smoothing uses Savitzky-Golay filtering; the baseline correction uses a polynomial fitting method; and the normalization scales the signal amplitude to the [0, 1] interval.

[0008] Furthermore, the preprocessing also includes signal enhancement and noise filtering steps to improve the quality of the model input data.

[0009] Furthermore, the one-dimensional convolutional neural network regression model includes at least an input layer, a one-dimensional convolutional layer, an activation function layer, a pooling layer, and a fully connected output layer connected in sequence. The core of the CNN model is the convolutional layer, which extracts features through multiple convolutional and pooling layers, and maps these features to the output through a fully connected layer. In studies involving simultaneous quantitative detection of multiple components in electrochemistry, one-dimensional convolutional layers can be used to process electrochemical signals.

[0010] Generally, multiple convolutional and pooling layers can be used to progressively extract signal features to achieve higher accuracy and sensitivity. Hyperparameters such as the kernel size and number of convolutional layers, and the pooling size and stride of pooling layers, can be adjusted using methods like cross-validation to obtain optimal model performance. When training a CNN model, the dataset needs to be divided into training, validation, and test sets. The training set is used to train the model, the validation set is used to adjust hyperparameters and prevent overfitting, and the test set is used to evaluate model performance.

[0011] Preferably, the one-dimensional convolutional neural network regression model further includes a regularization layer to prevent overfitting. During training, regularization methods can be used to prevent model overfitting.

[0012] Furthermore, in step S3, the model training is optimized using the Adam optimizer, and the mean squared error (MSE) is used as the loss function.

[0013] On the other hand, the present invention also proposes an electrochemical multi-component simultaneous quantitative detection system based on a one-dimensional convolutional neural network for implementing the method described above, comprising: An electrochemical detection module is used to acquire square wave voltammetric signals from mixed samples; The signal preprocessing module is used to perform baseline correction, filtering, and normalization on the acquired signals; The feature extraction and modeling module has a built-in pre-trained one-dimensional convolutional neural network regression model, which is used to receive the pre-processed signal, extract features, and output the concentration prediction results of each component. The results output module is used to output the concentration of each component and the corresponding evaluation index.

[0014] Furthermore, the electrochemical detection module employs a three-electrode system, with an ITO electrode as the working electrode, a platinum wire electrode as the counter electrode, and a non-aqueous Ag / Ag reference electrode. + electrode.

[0015] Furthermore, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0016] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: This invention overcomes the difficulties in identification and quantification caused by narrow potential windows and severe overlap of redox peaks in traditional electrochemical multi-component analysis, as well as the limitations of traditional methods such as chromatography, which involve complex pretreatment, long processing times, and difficulty in rapid on-site detection. Its advantage lies in its innovative combination of a one-dimensional convolutional neural network and square-wave voltammetry, utilizing the powerful nonlinear feature extraction capabilities of deep learning to effectively analyze overlapping signals. This enables simultaneous, rapid, and high-precision quantitative detection of target analytes with partially overlapping electrochemical signals in complex samples. This method significantly improves the sensitivity, accuracy, and automation level of multi-component analysis in complex systems. Attached Figure Description

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a schematic diagram of the system structure of the present invention; Figure 2 These are square wave voltammetry spectra of five antioxidants; Figure 3These are the chronoamperometry curves and calibration curves for a single antioxidant; in the figure: A~E are the chronoamperometry curves for TBHQ, PG, NDGA, BHT and BHA, and F~J are the corresponding calibration curves; Figure 4 This is a comparison chart of the prediction results of PLS ​​and CNN models; in the chart, A~E are the result charts of the PLS models of BHA, BHT, NDGA, PG and TBHQ, and F~J are the result charts of the corresponding neural networks; Figure 5 This is a graph showing the changes in fit and loss function during the training of a CNN model; in the graph, A~E are the fit graphs of BHA, BHT, NDGA, PG and TBHQ, and F~J are the result graphs of the corresponding loss function. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to specific embodiments.

[0019] Antioxidants are substances used to prevent food from oxidative rancidity and extend its shelf life. Common synthetic antioxidants include butylated hydroxyanisole (BHA), tert-butylhydroquinone (TBHQ), butylated hydroxytoluene (BHT), propyl gallate (PG), and dihydroguaiac acid (NDGA). These phenolic compounds can donate protons to free radicals, thereby inhibiting the oxidation process. However, antioxidants have certain toxicity and carcinogenic effects, and excessive use can harm human health. However, due to the synergistic effect among these antioxidants, two or more antioxidants are often used simultaneously in practical applications to achieve better antioxidant effects. These antioxidants have similar molecular structures and similar redox potentials, making it very difficult to simultaneously quantify multiple antioxidants in actual samples using traditional electrochemical methods.

[0020] Example 1: Construction of a method and system for the simultaneous detection of five antioxidants in edible oils This embodiment uses the simultaneous detection of five common synthetic antioxidants (TBHQ, BHA, BHT, PG, NDGA) in edible oils as an example to illustrate the implementation process of the present invention in detail.

[0021] This invention provides a method for the simultaneous quantitative detection of multiple antioxidants in edible oils using electrochemical methods based on a one-dimensional convolutional neural network. The detection method includes: Step 10): Electrode preparation and sample configuration; the working electrode is a self-made ITO electrode. ITO conductive glass is cut into strips of 0.4 cm × 3.5 cm, and insulating tape is used to cover part of the glass surface to form the ITO electrode. One end, with a 0.4 cm × 0.4 cm area exposed, serves as the working end, while the other end is partially exposed for connection to the wire. A platinum wire electrode is used as the counter electrode, with non-aqueous Ag / Ag... + The electrode serves as a reference electrode; The mixed sample was prepared by dissolving five antioxidants of different concentrations—butylated hydroxyanisole (BHA), tert-butylhydroquinone (TBHQ), butylated hydroxytoluene (BHT), propyl gallate (PG), and dihydroguaiacoate (NDGA)—in an acetonitrile solution containing 0.1 mol / L TEAP.

[0022] Sample preparation followed the pretreatment method based on the NY / T 1602-2008 standard. Accurately weigh 1 g of edible oil sample and place it in a 4 mL centrifuge tube. Add 1.6 mL of methanol, vortex for 3 min, let stand for 2 min, centrifuge at 3000 r / min for 5 min, and transfer the supernatant to a 10 mL volumetric flask. Extract the residue three times with 1.6 mL of methanol each time, combining the supernatants in a 10 mL volumetric flask, and dilute to volume with methanol. After evaporating the solution to dryness using a rotary evaporator, dilute to 10 mL with disodium hydrogen phosphate-citric acid buffer solution.

[0023] Step 20): Electrochemical data acquisition; using SWV technology, scanning range 0~2 V, amplification 4 mV, amplitude 25 mV, frequency 15 Hz; each sample was scanned 10 times, and the average value was taken as the raw data.

[0024] Step 30): Data preprocessing; smoothing using Savitzky-Golay filtering; baseline correction using polynomial fitting to remove background; data normalization to the [0,1] interval.

[0025] Step 40): CNN model construction and training; detailed experimental steps are as follows: (1) First, import the necessary Python packages: numpy, pandas, matplotlib, scipy, keras, tensorflow, sklearn.

[0026] (2) Then, the data file is read from the directory, the file order is randomly shuffled, and some processing is performed on the data. Specifically, based on the value of the first column in the file, the data is filtered and selected, only data within a certain range is selected. Then, the selected data is added to the DataFrame object X, and its corresponding label is added to the list target. The final result is a dataset X and a label set target, which are used for subsequent machine learning and deep learning tasks.

[0027] (3) Change the marker to a floating-point number marker to facilitate subsequent data processing and analysis.

[0028] (4) Data preprocessing: A function called SG was used to smooth each data point. A function called preprocess was used to perform baseline trimming on each data point. A function called enlarge was used to augment each data point. The purpose of these data processing steps is likely to improve data quality, reduce noise, and facilitate subsequent analysis.

[0029] (5) Split the data for the PLS model: The training_test_split function was used to split the dataset into a training set and a test set, with the training set accounting for 85% of the total dataset and the test set accounting for 15%.

[0030] (6) Establishing the PLS Regression Model: The PLSRegression class was used to create the PLS regression model. Then, the GridSearchCV function was used to fine-tune the model parameters. The training set data was fitted using the `fit` method, and the prediction accuracy of the test set was calculated and printed using the `score` method. Next, the `predict` method was used to predict the test set, and the root mean square error (RMSE) was calculated using the `mean_squared_error` function. Finally, a for loop was used to output the first 38 prediction results and their corresponding true values, making it easy to view the accuracy and error of the prediction results.

[0031] (7) Split the dataset for the neural network: Divide the original data into training, validation, and test sets. Use the Stratified Shuffle Split function to split X_new and target into training-validation and test sets in a ratio of 85%:15%. On the training-validation set, use the Stratified Shuffle Split function again to split the training-validation set into training and validation sets in a ratio of 85%:15%. Use the np.expand_dims function to convert the features of each sample from a two-dimensional array into a three-dimensional array.

[0032] (8) Defining and Training a Convolutional Neural Network (CNN) Model: First, the TensorFlow library was imported to create the neural network and define related operations. The Keras layer and metric modules were also imported to build and train the neural network. A sequential model was defined, and several convolutional layers, pooling layers, and fully connected layers were added. This model uses a series of convolutional and pooling layers, as well as several fully connected layers, to extract features from the input data and perform classification or regression prediction. Specifically, this model contains three convolutional layers, using 16, 64, and 16 convolutional kernels respectively, and ReLU activation and max-pooling layers are added sequentially. A fully connected layer is added after the convolutional layers to convert the feature vectors output by the convolutional layers into a scalar output value. The model was compiled, and the loss function, optimizer, and evaluation metrics were set. Mean squared error (MSE) was used as the loss function, the Adam optimizer was used for optimization, and a custom R2 score and the Keras-provided root mean squared error (RMSE) were used as evaluation metrics. The model is trained using the `fit()` function. Here, the number of training epochs is set to 300, the batch size to 8, and a validation set is used to validate the model after each training epoch. The training process returns a `history` object that records the training loss and changes in the evaluation metrics, which can be used for subsequent visualization and analysis.

[0033] Finally, the accuracy of the model in concentration prediction was evaluated using different metrics. The root mean square error (RMSE) was used to represent the error between the detected concentration and the standard concentration. This parameter provides information about the model's fit to the data, as shown in Equation 1: (1); In the formula, It is the concentration of each sample i detected by the model. is the standard value of the response variable (concentration), and n is the number of samples in the set. The relative percentage error (RE%) between the standard and detected concentrations is calculated using Equation 2 to evaluate the model's predictive ability.

[0034] (2); In the formula, c m It is the average of all standard concentrations in the response set (training and validation).

[0035] Step 50): Actual sample testing For unknown edible oil extracts, perform pretreatment, SWV scanning and signal preprocessing according to steps 10)-30). Then, input the preprocessed voltammetric curve into the final CNN model trained in step 40) to directly obtain the predicted concentrations of five antioxidants in the sample.

[0036] Example 2: The electrochemical behavior of the antioxidant prepared in Example 1 on the electrode was investigated. The specific operation is as follows: SWV scanning was performed on single antioxidant solutions. Figure 2 As shown, the oxidation potentials of TBHQ, BHA, BHT, PG, and NDGA are 0.93, 1.03, 1.53, 1.30, and 1.07 V, respectively. The oxidation peaks of NDGA and BHA completely overlap, while the other three partially overlap. This indicates that in mixed samples, the current intensity at the potential corresponding to each antioxidant is determined not only by the concentration of a single antioxidant but also by other antioxidants. Therefore, univariate calibration methods cannot be used for the detection of mixed samples containing multiple antioxidants.

[0037] Example 3: The linear range of the antioxidant prepared in Example 1 was investigated. The specific procedures are as follows: The response signal of each antioxidant on the electrode was acquired using IT technology. For example... Figure 3 As shown in AE, the response signals of TBHQ, PG, NDGA, BHT and BHA decrease stepwise. Since the applied potential is greater than the oxidation potential of each antioxidant, an oxidation current is generated with each drop, and finally, as the concentration increases, saturation is reached. Figure 3 FJ represents the calibration curves for TBHQ, PG, NDGA, BHT, and BHA. The linear range was determined based on the calibration curve for each antioxidant, and the detection limit was calculated using S / N=3. The results are shown in Table 1.

[0038] Table 1. Linear calibration results for single antioxidants

[0039] Example 4 examines the evaluation of the results of PLS ​​and deep learning in Example 1. The specific operation is as follows: The data was trained using the model, and the results are shown in Tables 2 and 3. Figure 4 This is a comparison of the prediction results of PLS ​​and CNN models. In the figure, A~E are the results of the PLS models for BHA, BHT, NDGA, PG, and TBHQ, and F~J are the results of the corresponding neural networks. It can be clearly seen that the neural network models are generally better than the PLS models. The PLS models perform better for easily distinguishable components (such as BHT) but worse for easily confused components. Furthermore, due to the limited number of training iterations, deep learning performs very well in predicting training set data, but poorly in predicting test set data.

[0040] Table 2 Evaluation of Training Set Results

[0041] Table 3 Evaluation of Test Set Results

[0042] For easier viewing, a calibration curve was created using predicted and actual concentrations. It can be seen that the neural network model's predicted concentrations are more concentrated and closer to the actual concentrations. However... Figure 5 It shows how the fit and loss function change with the number of training iterations, indicating that the neural network model is running successfully and also helps us to monitor the model's training progress.

[0043] It should be noted that the core of this invention is not the detection of specific antioxidants, but rather the provision of a universal "electrochemical + 1D-CNN" framework. To demonstrate its universality, the target analyte can be replaced with other substances to be detected, for example: Mixtures of heavy metal ions (such as Pb) 2+ Cd 2+ Cu 2+ The signal can be acquired using differential pulse voltammetry; or multiple antibiotics or neurotransmitters; or multiple cancer biomarkers (such as CEA, AFP, PSA).

[0044] By simply collecting mixed sample SWV data of the corresponding target object and reconstructing and training the CNN model according to the procedure in Example 1, a new dedicated quantitative detection method can be established. This demonstrates that the core of the scope of protection of this invention lies in the innovation of the method architecture, rather than the specific application object.

[0045] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Modifications, equivalent substitutions, and improvements made by those skilled in the art within the spirit and principles of the present invention should all be included within the scope of protection of the present invention.

Claims

1. A method for simultaneous quantitative detection of multiple electrochemical components based on a one-dimensional convolutional neural network, characterized in that, The process includes the following steps: S1: The square wave voltammetry method is used to scan the mixed sample containing multiple target analytes and acquire a one-dimensional mixed current-potential signal; S2: Preprocess the hybrid current-potential signal; S3: Construct a one-dimensional convolutional neural network regression model, using the preprocessed signal as input and the concentration vector of each target analyte as output label, and train the one-dimensional convolutional neural network regression model; S4: Input the signal obtained after processing the mixed sample to be tested through steps S1 and S2 into the trained one-dimensional convolutional neural network regression model, and the output is the predicted concentration of each target analyte.

2. The method for simultaneous quantitative detection of multiple electrochemical components based on a one-dimensional convolutional neural network according to claim 1, characterized in that, The preprocessing includes at least baseline correction, filtering and smoothing, and normalization; the filtering and smoothing uses Savitzky-Golay filtering; the baseline correction uses a polynomial fitting method; and the normalization scales the signal amplitude to the [0, 1] interval.

3. The method for simultaneous quantitative detection of multiple electrochemical components based on a one-dimensional convolutional neural network according to claim 2, characterized in that, The preprocessing also includes signal enhancement and noise filtering steps to improve the quality of the model input data.

4. The method for simultaneous quantitative detection of multiple electrochemical components based on a one-dimensional convolutional neural network according to claim 1, characterized in that, The one-dimensional convolutional neural network regression model includes at least an input layer, a one-dimensional convolutional layer, an activation function layer, a pooling layer, and a fully connected output layer connected in sequence.

5. The method for simultaneous quantitative detection of multiple electrochemical components based on a one-dimensional convolutional neural network according to claim 4, characterized in that, The one-dimensional convolutional neural network regression model also includes a regularization layer to prevent overfitting.

6. The method for simultaneous quantitative detection of multiple electrochemical components based on a one-dimensional convolutional neural network according to claim 1, characterized in that, In step S3, the model training uses the Adam optimizer for optimization and the mean squared error (MSE) is used as the loss function.

7. A simultaneous quantitative detection system for multiple electrochemical components based on a one-dimensional convolutional neural network, used to implement the method described in any one of claims 1-6, characterized in that, include: An electrochemical detection module is used to acquire square wave voltammetric signals from mixed samples; The signal preprocessing module is used to perform baseline correction, filtering, and normalization on the acquired signals; The feature extraction and modeling module has a built-in pre-trained one-dimensional convolutional neural network regression model, which is used to receive the pre-processed signal, extract features, and output the concentration prediction results of each component. The results output module is used to output the concentration of each component and the corresponding evaluation index.

8. The electrochemical multi-component simultaneous quantitative detection system based on a one-dimensional convolutional neural network according to claim 7, characterized in that, The electrochemical detection module employs a three-electrode system: an ITO electrode as the working electrode, a platinum wire electrode as the counter electrode, and a non-aqueous Ag / Ag reference electrode. + electrode.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

Images