Method for analyzing transport information of multi-element ions in confinement space

Abstract

The invention discloses a multi-ion transport information analysis method in a confinement space, and relates to the field of bionic nanofluidic ionology, and the method comprises the following steps: S1, constructing a nanoscale confinement space, and injecting a multi-ion mixed solution; s2, acquiring an I-V curve, a C-V curve and an I-t curve; s3, extracting features from the curve and screening the features; s4, two machine learning models are selected for ion type classification and concentration proportion prediction, and the screened feature vectors are used for training; and S5, detecting a to-be-detected sample by using the trained model to obtain ion classification and concentration. According to the method, the transmission behavior of a multi-element ion mixing system under the nanoscale is systematically analyzed, the influence of ion physicochemical characteristics, interaction rules and ion types on the transmission process is comprehensively considered, multi-dimensional information is extracted and analyzed by utilizing machine learning, information expressed by each type of ions in the transmission process is explored, and the accuracy of the transmission process is improved. The cooperative transmission mechanism of the multi-element ion system in the confinement environment is clarified.

Description

Technical Field

[0001] This invention relates to the field of biomimetic nanofluidics, and more particularly to a method for analyzing multi-component ion transport information in a confined space. Background Technology

[0002] Multiple ion channels exist on the cell membranes of living organisms, playing a core regulatory role in physiological processes such as cellular material exchange, signal transduction, and energy conversion. To explore the signal transduction mechanisms of biological ion channels, researchers have constructed various artificial biomimetic nanochannels. Current research on biomimetic nanochannels largely focuses on the transport behavior of single ions, laying an important foundation for revealing the basic principles of ion transport. However, signal transduction in biological ion channels often involves the synergistic transport of multiple ions; therefore, understanding the coupling mechanism of multi-ion transport within nanochannels is more crucial. Although some scholars have explored the characteristics of multi-ion transport within channels, the complexity of multi-ion coupling behavior (inter-ion interactions, etc.) makes it difficult to effectively simulate the characteristic signal expression of multi-ion synergistic transport in biological ion channels.

[0003] To deepen our understanding of the mechanisms of multi-ion signal transport and expression in biological systems, this study constructed a confined nanochannel model. By systematically analyzing the transport behavior of multi-ion mixed systems at the nanoscale, and comprehensively considering the influence of ion physicochemical properties, interaction laws, and ion types on the transport process, machine learning was used to extract and analyze multidimensional information, exploring the information expressed by each ion during transport to elucidate the cooperative transport mechanism of multi-ion systems under confined conditions. This study provides a new perspective for understanding complex ion transport processes in organisms and offers important theoretical basis and experimental guidance for subsequent research simulating the function of natural biological nanochannels. Summary of the Invention

[0004] To address the above problems, this invention proposes a method for analyzing multi-ion transport information in a confined space. This method involves constructing a nanoscale confined space and injecting a multi-ion mixed solution with known ion types and concentration ratios. Electrochemical signals of multi-ion transport within the nanoscale confined space are collected, yielding IV, CV, and It curves. Feature vectors are extracted from the IV, CV, and It curves to train a machine learning model. The trained model is then used for ion type classification and concentration ratio prediction. This method enables simultaneous detection of multiple ions without complex sample pretreatment, allowing direct analysis of mixed solutions. Furthermore, it offers fast measurement speeds, making it suitable for online real-time monitoring.

[0005] A method for analyzing the transport information of multiple ions in a confined space, the specific steps of which are as follows:

[0006] S1, construct a nanoscale confined space and inject a multi-electrode ion mixture solution with known ion types and concentration ratios;

[0007] S2, collect the electrochemical signals of multi-ion transport in a multi-ion mixed solution in a nanoscale confined space, and obtain IV curves, CV curves and It curves;

[0008] S3: Extract features from the IV curve, CV curve and It curve and perform feature filtering to obtain the filtered feature vector;

[0009] S4. Select the first machine learning model for ion type classification and the second machine learning model for concentration ratio prediction. Use the selected feature vectors as the dataset to train the two machine learning models to obtain the trained first machine learning model and the trained second machine learning model.

[0010] S5. Obtain the feature vector of the multi-ion mixed solution sample to be tested. Input the filtered feature vector of the sample to be tested into the trained first machine learning model for ion classification and recognition. Input it into the trained second machine learning model for concentration ratio prediction to obtain the ion classification and concentration of the sample to be tested.

[0011] Preferably, the nanoscale confined space is a single nanopore in glass.

[0012] Preferably, the features extracted from the I-V curve, CV curve, and It curve include features from the I-V curve, CV curve, It curve, and principal component analysis features. Specifically, the features extracted from the I-V curve include rectification ratio, positive and negative conductance, current statistics, voltage range characteristics, nonlinear characteristics, frequency domain characteristics, and differential conductance characteristics. The features extracted from the CV curve include cyclic volt-ampere area, hysteresis characteristics, peak characteristics, capacitance characteristics, symmetry characteristics, and electrochemical reversibility indices. The features extracted from the It curve include steady-state characteristics, dynamic response characteristics, and correlation characteristics. The principal component analysis features are obtained by capturing the overall morphological information of the I-V and C-V curves using principal component analysis. Preferably, the feature selection is as follows:

[0013] Redundancy removal is achieved by using a Pearson correlation coefficient threshold. The Pearson correlation coefficient of the features is calculated and features with a correlation coefficient greater than the preset threshold are removed to obtain the redundancy-free features.

[0014] Mutual information is used to evaluate and sort the information content of redundancy removal features and ion types in descending order;

[0015] Select the top-ranked features as the filtered feature vector.

[0016] Preferably, the multi-electrode ion mixture solution with known ion types and concentration ratios is an electrolyte solution; the cations in the multi-electrode ion mixture solution include K+.+ Na + Li + Ca 2+ Mg 2+ Fe 2+ Co 2+ Zn 2+ Cu 2+ Mn 2+ Cr 3+ and Fe 3+ One or more of the following; the anions in a multi-ion mixed solution include Cl. - NO3 - SO4 2- SO3 2- NO2 - I - N 3- MoO4 2- Se 2- SeO3 2- SeO4 2- and SO3 2- One or more of the following; the multi-ion mixed solution includes cations and anions of different valence states; the ion concentration of the multi-ion mixed solution is between 0.1 and 100 mM / L.

[0017] Preferably, the first machine learning model is selected through the following steps:

[0018] Select several machine learning models;

[0019] Each machine learning model is trained for ion type classification using the dataset;

[0020] For each machine learning model, five-fold cross-validation was used to optimize hyperparameters, and grid search was used to determine the optimal combination of hyperparameters.

[0021] The best-performing machine learning model is selected as the first machine learning model based on cross-validation accuracy and test accuracy.

[0022] The second machine learning model is selected through the following steps:

[0023] Select several machine learning models;

[0024] Each machine learning model was trained to predict concentration ratios using the dataset.

[0025] For each machine learning model, five-fold cross-validation was used to optimize hyperparameters, and grid search was used to determine the optimal combination of hyperparameters.

[0026] The best-performing machine learning model is selected as the second machine learning model based on regression evaluation metrics.

[0027] Preferably, the machine learning models used for ion type classification include gradient boosting models, K-nearest neighbor models, random forest models, decision tree models, support vector machine models, XGBoost models, and logistic regression models; the machine learning models used for concentration ratio prediction include support vector regression models, random forest regression models, XGBoost regression models, Ridge regression models, linear regression models, and Lasso regression models.

[0028] Preferably, the first machine learning model is a gradient boosting model; and the second machine learning model is a support vector regression model.

[0029] Preferably, before training the two machine learning models using the feature vectors as the dataset, the method further includes: preprocessing the dataset, specifically as follows:

[0030] Baseline correction was performed using the air-tube method.

[0031] Missing value imputation is performed using the median as the filler.

[0032] Standardization involves fitting the mean and standard deviation to the training set in the dataset.

[0033] Outlier removal is performed using robust Z-score and derivative peak rules to remove outliers from the training set in the dataset.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) This invention constructs a nanoscale confined space and injects a multi-ion mixed solution with known ion types and concentration ratios; collects electrochemical signals of multi-ion transport in the nanoscale confined space to obtain IV curves, CV curves and It curves; extracts feature vectors from the IV curves, CV curves and It curves to train a machine learning model, and uses the trained machine learning model for ion type classification and concentration ratio prediction; uses a glass single nanopore as a nanoscale confined space, and only a single nanochannel is needed to realize the simultaneous detection of multiple ions; no complicated sample pretreatment is required, and mixed solutions can be directly analyzed; and the measurement speed is fast, making it suitable for online real-time monitoring.

[0036] (4) This invention expands the existing ion-based computing system by mimicking the parallel computing process of information carried by multiple ions in the neuronal system of the brain through multi-ion transport and mathematical analysis. Attached Figure Description

[0037] The present invention will now be described in further detail with reference to the accompanying drawings;

[0038] Figure 1 This is a flowchart of a method for analyzing multi-component ion transport information in a confined space according to an embodiment of the present invention;

[0039] Figure 2 This is a confusion matrix heatmap of the method for analyzing multi-component ion transport information in a confined space according to an embodiment of the present invention.

[0040] Figure 3 This is the feature scatter matrix of the method for analyzing multi-element ion transport information in a confined space according to an embodiment of the present invention;

[0041] Figure 4 This is the It feature correlation matrix of the method for analyzing multi-element ion transport information in a confined space according to an embodiment of the present invention;

[0042] Figure 5 This is a feature importance ranking diagram of the method for analyzing multi-element ion transport information in a confined space according to an embodiment of the present invention;

[0043] Figure 6 This is a scatter plot of regression prediction for the method of analyzing multi-element ion transport information in a confined space according to an embodiment of the present invention.

[0044] Figure 7 This is a comparison chart of It curves for the method of analyzing multi-element ion transport information in a confined space according to an embodiment of the present invention;

[0045] Figure 8 This is a bar chart comparing the model classification performance of the multi-element ion transport information analysis method in confined space according to an embodiment of the present invention.

[0046] Figure 9 This is a bar chart comparing the model concentration prediction performance of the multi-element ion transport information analysis method in a confined space according to an embodiment of the present invention. Detailed Implementation

[0047] The present invention will be further described below through specific embodiments.

[0048] like Figure 1 As shown, the method for analyzing the transport information of multiple ions in a confined space comprises the following steps:

[0049] S1, construct a nanoscale confined space and inject a multi-element ion mixture solution with known ion types and concentration ratios.

[0050] This embodiment utilizes a nanoconfined channel device to measure the current-voltage, capacitance-voltage, and current-time (IV / CV / IT) characteristics of mixed ionic solutions. The nanoconfined channel is constructed from a single glass nanopore with a tip radius of 67 nm. Ag / AgCl electrodes are positioned at both ends to apply voltage and measure ionic current. During the experiment, a voltage is applied across the electrodes.

[0051] The ion concentration of the multi-component ion solution is between 0.1 and 100 mM / L; the cations in the multi-component ion mixed solution include K+. + Na + Li + Ca 2+ Mg 2+ Fe 2+ Co 2+ Zn 2+ Cu 2+ Mn 2+ Cr 3+ and Fe 3+ One or more of the following; the anions in a multi-ion mixed solution include Cl. - NO3 - SO4 2- SO3 2- NO2 - I - N 3- MoO4 2- Se 2- SeO3 2- SeO4 2- and SO3 2- One or more of the following; multi-valent ionic solutions include ions of different valence states, commonly including but not limited to monovalent, divalent, trivalent, and tetravalent cations and anions.

[0052] In this embodiment, the multi-ion mixed solution sample is a mixture of FeCl3 and KCl, NaCl and LiCl solutions in different volume ratios with the same concentration. The concentration is 1 mM and the volume ratios are 1:0, 2:8, 3:7, 4:6, 1:1, 6:4, 7:3, 8:2 and 0:1, respectively.

[0053] S2 collects the electrochemical signals of multi-ion transport in a nanoscale confined space of a multi-ion mixed solution, and obtains IV curves, CV curves and It curves.

[0054] During the experiment, the baseline current of the empty tube (containing only solvent) was first measured at 0V and 1V for subsequent baseline correction. In all IV / CV experiments, the voltage scan range was set to -1.0V to +1.0V, and the ion current response within the channel was recorded using a linear voltage scan method. After each sample was injected into the nanochannel, it was allowed to equilibrate for a short time before IV / CV scanning data acquisition began. Using this apparatus and procedure, IV / CV curve data of the nanoconfined channel under different ion combinations and concentrations were obtained. In the IT experiments, a fixed voltage of +1V was applied, and the current change over time was recorded.

[0055] S3 extracts features from the IV curve, CV curve, and It curve, and performs feature filtering to obtain the feature vector.

[0056] In this embodiment, a total of 20 IV curve features, 17 CV curve features, and 7 It curve features were extracted. Among them, the IV curve features specifically include several aspects:

[0057] Rectification ratio: This reflects the asymmetric ion transport characteristics of nanochannels;

[0058] Positive and negative conductance: characterizing ion transport capabilities under positive and negative bias voltages, respectively;

[0059] Current statistical characteristics: including mean, standard deviation, skewness, kurtosis, etc., which describe the characteristics of current distribution;

[0060] Voltage range characteristics: current response in the low-voltage region (|V|≤0.3V), medium-voltage region (0.3V<|V|≤0.7V), and high-voltage region (|V|>0.7V);

[0061] Nonlinear characteristics: quadratic fitting coefficient, residual standard deviation, etc., characterize the degree of nonlinearity of the IV curve;

[0062] Frequency domain characteristics: maximum amplitude and average frequency after FFT transformation;

[0063] Differential conductance characteristics: mean, standard deviation, and maximum value of dI / dV;

[0064] The selected IV curve features total 20, specifically: rectification ratio: |I(+1V)| / |I(-1V)|; forward conductance: G+=I(+1V) / 1V; backward conductance: G-=|I(-1V)| / 1V; mean current: I(V); standard deviation current: iv_current_std; skewness current: iv_current_skewness; kurtosis current: iv_current_kurtosis; and low-voltage average current: iv_lowV_current_mean (|V| ≤ 0.3V). Average current in the medium-voltage section: iv_midV_current_mean (0.3 < |V| ≤ 0.7 V). Average current in the high-voltage section: iv_highV_current_mean (|V| > 0.7 V). Mean differential conductance: dIdV_mean (central difference). Standard deviation of differential conductance: dIdV_std. Maximum differential conductance: dIdV_max. Origin slope: dI / dV = 0. Quadratic coefficients: b of the fit I(V) = aV + bV^2 + c. Standard deviation of quadratic residuals: iv_residual_std. Power exponent: n of |I| = k|V|^n. FFT dominant peak frequency fft_dom_freq (the dominant angular frequency after FFT of I(V)). FFT dominant peak amplitude fft_dom_amp. Current area iv_area_total: \displaystyle\int |I(V)|\,dV.

[0065] The characteristics of a CV curve specifically include several aspects:

[0066] Cyclic voltammetry area: The integrated area of ​​the positive and negative scans, reflecting the total amount of charge transport;

[0067] Hysteresis characteristics: The closed area enclosed by the forward and reverse scan curves and the hysteresis factor characterize electrochemical reversibility;

[0068] Peak characteristics: current values, potential values, and peak potential separation of oxidation and reduction peaks;

[0069] Capacitor characteristics: capacitive current and equivalent capacitance in the low-voltage region;

[0070] Symmetry characteristics: the symmetry index of positive and negative currents;

[0071] Electrochemical reversibility index: reversibility parameters calculated based on peak potential separation degree;

[0072] A total of 17 CV curve features were selected, specifically: `cv_area_positive` (forward scan area, V>0 interval, dV); `cv_mean_current_positive` (absolute value of average current during forward scan); `cv_area_negative` (forward scan area, V<0 interval, dV); `cv_mean_current_negative` (absolute value of average current during reverse scan); `cv_area_total` (total scan area, dV); `cv_hysteresis_area` (hysteresis area, dV, |I_f(V)-I_r(V)|); `cv_hysteresis_factor` (hysteresis factor = hysteresis area / total scan area); `cv_oxidation_peak_current` (oxidation peak current, maximum value during forward scan). `cv_oxidation_peak_potential`: Oxidation peak potential. `cv_reduction_peak_current`: Reduction peak current (backscan minimum). `cv_reduction_peak_potential`: Reduction peak potential. `cv_peak_current_ratio`: Peak current ratio. `cv_peak_potential_separation`: Peak potential separation. `cv_capacitive_current`: Mean "capacitive current" in the low-voltage region. `cv_capacitance`: Equivalent capacitance. `cv_current_symmetry`: Current symmetry. `cv_reversibility_index`: Reversibility index.

[0073] The specific characteristics of the It curve include:

[0074] Steady-state characteristics: steady-state current value, steady-state standard deviation, and noise level;

[0075] Dynamic response characteristics: current change slope, initial-to-final current ratio, time constant;

[0076] Correlation characteristics: Autocorrelation time after detrending reflects the temporal correlation of ion transport;

[0077] In addition, five principal component analysis (PCA) features were extracted to capture the overall morphological information of the I–V or C–V curves, denoted as IV_PC1, IV_PC2, CV_PC1, CV_PC2, and CV_PC3, respectively; at the same time, four meta-information features (temperature, aperture size, batch number, and scan type) were retained.

[0078] In the feature selection stage, redundancy is first removed using the Pearson correlation threshold. The highly correlated features were then evaluated and ranked using mutual information, which is defined as follows:

[0079]

[0080] Based on mutual information scores, the 30 most discriminative features are retained as feature vectors. In this embodiment, the top 10 features include: it_mean (It, average current), it_correlation_time (correlation time), it_cv (coefficient of variation), it_std (standard deviation), it_noise_std (noise standard deviation), iv_forward_conductance (forward conductance), it_evolution_ratio (current evolution ratio), it_slope (slope), it_evolution_ratio_div_it_mean (ratio of evolution ratio to mean), and IV_PC2 (second principal component). The remaining 20 input features are as follows (grouped by source):

[0081] I–V: rectification_ratio, iv_backward_conductance, dIdV_max, dIdV_mean, iv_lowV_current_mean (|V| ≤ 0.3V), iv_midV_current_mean (0.3 < |V| ≤ 0.7V), iv_highV_current_mean (|V| > 0.7V), iv_quadratic_coeff (quadratic fitting coefficient, b in I(V) = aV + bV^2 + c), iv_residual_std (quadratic fitting residual standard deviation).

[0082] C–V: cv_area_total (cyclic voltaic area), hysteresis_area (hysteresis area), redox_peak_current_ox (oxidation peak current), redox_peak_current_red (reduction peak current), delta_peak_potential (peak potential separation, Delta E_p), lowV_capacitance (low-voltage region capacitance), capacity_equivalent (equivalent capacitance), symmetry_index (symmetry index, measuring the relative symmetry between |I(+V)| and |I(-V)|), reversibility_index (reversibility index, 1-A_{\mathrm{hys}} / A_{\mathrm{CV}}).

[0083] I–t: it_steady_current (steady-state current, mean of the last 20% of the time window), it_time_constant (time constant, obtained by fitting I(t)=I_{\infty}+(I_0-I_{\infty})e^{-t / \tau}).

[0084] S4. Select the first machine learning model for ion type classification and the second machine learning model for concentration ratio prediction. Use the feature vector as the dataset to train the two machine learning models to obtain the trained first machine learning model and the trained second machine learning model.

[0085] Dataset partitioning and preprocessing pipeline. Stratified sampling was used to partition the samples into training and test sets at an 80:20 ratio, ensuring a consistent proportion of each ion type. The preprocessing pipeline included baseline correction, missing value imputation, and normalization.

[0086] Baseline correction (empty pipe method): I–V / C–V use 0V baseline, I–t use +1V baseline.

[0087]

[0088]

[0089] Missing value imputation: Imputation is performed based on the number of bits in the training set (median filling).

[0090] Standardization (z-score / StandardScaler): Fits the mean and standard deviation using only the training set and applies them to the training / test set, avoiding information leakage. It is represented as:

[0091]

[0092] Outlier rules (training set): Robust Z-score and derivative peak removal rules are used together.

[0093] ;

[0094] And peak detection at the curve level ( or Exceed The test set is only truncated by quantiles, without any removal.

[0095] S5. Obtain the feature vector of the multi-ion mixed solution sample to be tested. Input the feature vector of the sample to be tested into the first machine learning model trained for ion classification and recognition. Input it into the second machine learning model trained for concentration ratio prediction to obtain the ion classification and concentration of the sample to be tested.

[0096] Multi-model performance comparison and optimization. This embodiment evaluates the performance of seven machine learning algorithms on ion classification and concentration prediction tasks. For each algorithm, five-fold cross-validation is used for hyperparameter optimization, and grid search is used to determine the optimal parameter combination. It should be noted that the search method for the optimal parameter combination can be other than grid search, such as random search, Bayesian optimization, successive halving / overbandwidth, and early stopping strategies, etc., and the specific settings can be set as needed. This embodiment does not impose any restrictions.

[0097] In this embodiment, the results of the ion type classification task are as follows:

[0098] Gradient boosting (optimal): Test accuracy 92.48%, cross-validation accuracy 91.77% ± 0.50%;

[0099] K-Nearest Neighbors: Test accuracy 90.42%, cross-validation accuracy 89.45% ± 0.72%;

[0100] Random Forest: Test accuracy 90.11%, cross-validation accuracy 90.89% ± 0.88%;

[0101] Decision tree: Test accuracy 86.51%, cross-validation accuracy 84.70% ± 2.96%;

[0102] Support Vector Machine: Test accuracy 81.87%, cross-validation accuracy 81.35% ± 1.61%;

[0103] XGBoost: Test accuracy 80.33%, cross-validation accuracy 81.66% ± 1.21%;

[0104] Logistic Regression: Test accuracy 57.98%, Cross-validation accuracy 59.41% ± 1.88%.

[0105] In this embodiment, the results of the concentration ratio prediction task are as follows:

[0106] Support Vector Regression (Optimal): MAE = 0.0654, R² = 0.9398, MAPE = 14.91%;

[0107] Random forest regression: MAE=0.0701, R²=0.9040, MAPE=17.81%;

[0108] XGBoost regression: MAE=0.0772, R²=0.8947, MAPE=17.77%;

[0109] Ridge regression: MAE=0.0925, R²=0.8356, MAPE=20.88%;

[0110] Linear regression: MAE=0.0926, R²=0.8354, MAPE=20.86%;

[0111] Lasso regression: MAE=0.1172, R²=0.7421, MAPE=25.06%.

[0112] Model robustness and generalization ability analysis. Learning curve analysis revealed that the gradient boosting and random forest models showed little difference in performance between the training and validation sets, indicating that the models possess good generalization ability.

[0113] Figure 1 Confusion matrix analysis showed that the recognition accuracy of the three ion types was relatively balanced, with F1 scores of 0.93, 0.91, and 0.92 for Fe+K, Fe+Na, and Fe+Li, respectively. Residual analysis indicated that the concentration prediction error followed an approximately normal distribution with no obvious systematic bias. Figure 2 The confusion matrix heatmap shows the classification results of the gradient boosting model on the test set. The dark areas on the diagonal represent the number of correctly classified samples, and the off-diagonal elements represent misclassifications. The abundance of dark areas on the diagonal indicates a balanced classification across the three classes (Fe+K, Fe+Na, Fe+Li); the off-diagonal distributions are all <10%, indicating no systematic confusion. This demonstrates that the classifier generalizes well. Figure 3Six features with high contribution to the classification task were selected from the feature scatter matrix: it_mean (It, average current, nA), it_correlation_time (autocorrelation time, s), it_cv (coefficient of variation, dimensionless), it_std (standard deviation, nA), it_noise_std (noise standard deviation, nA), and iv_forward_conductance (forward conductance, nS). In the subplot of the i-th row and j-th column of the scatter matrix, the horizontal axis represents the j-th feature, and the vertical axis represents the i-th feature; the diagonal lines provide the one-dimensional probability density of each feature. Different colors represent different ion types (Fe+K, Fe+Na, Fe+Li). This reveals the coupling relationship between It dynamics and forward conductance, while it_correlation_time has low correlation with other features, providing an independent information dimension. Figure 4 The It feature correlation matrix reveals the correlation between time-domain features, where steady-state current is strongly positively correlated with the mean conductance, while the time constant is less correlated with other features, indicating that it provides an independent information dimension. Figure 5 The feature importance ranking plot shows the relative importance of the top 20 most important features in the gradient boosting model. it_mean, it_correlation_time, and it_cv rank in the top three, confirming the key role of the It dynamic feature in ion recognition. Figure 6 The regression prediction scatter plot shows the comparison between the predicted values ​​and the true values ​​of the support vector regression model. The data points are closely distributed near the ideal prediction line (red dashed line). R²=0.940, MAE=0.065, indicating that the model has excellent concentration prediction performance. Figure 7 The It curve comparison chart shows the average It response curves of different ion types under a 1V bias voltage. The Fe+K system exhibits the highest steady-state current (approximately 58nA), followed by Fe+Na (approximately 39nA), and Fe+Li has the lowest (approximately 35nA), reflecting the differences in the mobility of different ions. Figure 7 The accuracy of seven classification algorithms is compared, with gradient boosting leading with an accuracy of 92.5%. Figure 8 The comparison of R² values ​​for six regression algorithms is shown, with support vector regression showing the best performance with an R² value of 0.940.

[0114] This embodiment can simultaneously identify ion species and predict concentration ratios within a single test cycle (total duration approximately 2-3 minutes). The following technical specifications are achieved:

[0115] The accuracy rate of ion identification reaches 92.48%, and it can accurately distinguish between three mixed systems: Fe+K, Fe+Na, and Fe+Li, meeting the qualitative analysis needs of complex electrolyte solutions.

[0116] The average absolute error of the concentration ratio prediction is 0.0654 (approximately 6.5% of the true concentration value), with R²=0.9398, which meets the quantitative analysis requirements for applications such as environmental monitoring and industrial process control.

[0117] Feature importance analysis: It curve features (such as it_mean and it_correlation_time) play a crucial role in ion recognition, indicating that the time-domain dynamic response carries rich ion-specific information. This finding provides theoretical guidance for designing more efficient ion detection protocols.

[0118] Against the backdrop of biomimetic nanofluidic ion science research, this embodiment acquires data on the transport behavior of multiple ions within a confined space, and then analyzes and processes this information to transform it into a dynamic database. The design goal is to efficiently and accurately identify the types, valence states, and proportions of ions in a multi-ion solution, thereby achieving high-efficiency information transmission and processing. This provides a new technical means for realizing ion-based parallel computing. A method system for realizing ion-based parallel computing through reverse mathematical analysis of multi-ion transport in a confined space has been developed. Intelligent design of nanochannels allows for intelligent control of the process and behavior of simultaneous transport of multiple ions in a confined space, acquiring a large number of ion signals. The information carried by the ions is analyzed and extracted using mathematical inverse analysis, enabling parallel processing of large amounts of information and improving the computational efficiency of ion-based computing. The method of confined space multi-ion transport and mathematical inverse analysis proposed in this embodiment analyzes the large amount of information carried by multi-ion transport, thereby improving the parallel computing power of ion-based computing. In the neuronal system of the brain, rapid and efficient information processing and transmission are the result of the interaction and coordination of information carried by multiple ions. By mimicking this parallel computing process through multi-ion transport and mathematical analysis, the existing ion-based computing system is broadened.

[0119] Compared with traditional ion-selective electrode arrays or mass spectrometry analysis methods, this scheme has the following advantages: (1) it can achieve simultaneous detection of multiple ions with only a single nanochannel; (2) it does not require complex sample pretreatment and can directly analyze mixed solutions; (3) it has a fast measurement speed and is suitable for online real-time monitoring; (4) the device structure for implementing the method in this embodiment is simple and easy to miniaturize and integrate. This technology has broad application prospects in water quality monitoring, biomedical diagnosis, chemical process control and other fields.

[0120] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A method for analyzing multi-component ion transport information in a confined space, characterized in that, Includes the following steps: S1, construct a nanoscale confined space and inject a multi-electrode ion mixture solution with known ion types and concentration ratios; S2, collect the electrochemical signals of multi-ion transport in a multi-ion mixed solution in a nanoscale confined space, and obtain IV curves, CV curves and It curves; S3: Extract features from the IV curve, CV curve and It curve and perform feature filtering to obtain the filtered feature vector; S4. Select the first machine learning model for ion type classification and the second machine learning model for concentration ratio prediction. Use the selected feature vectors as the dataset to train the two machine learning models to obtain the trained first machine learning model and the trained second machine learning model. S5. Obtain the feature vector of the multi-ion mixed solution sample to be tested. Input the filtered feature vector of the sample to be tested into the trained first machine learning model for ion classification and recognition. Input it into the trained second machine learning model for concentration ratio prediction to obtain the ion classification and concentration of the sample to be tested.

2. The method for analyzing multi-component ion transport information in a confined space according to claim 1, characterized in that, The nanoscale confined space is a single nanopore in glass.

3. The method for analyzing multi-component ion transport information in a confined space according to claim 1, characterized in that, Features extracted from the I-V, CV, and It curves include features from the I-V, CV, and It curves, as well as features from principal component analysis. Specifically, features extracted from the I-V curve include rectification ratio, positive and negative conductance, current statistics, voltage range characteristics, nonlinear characteristics, frequency domain characteristics, and differential conductance characteristics. Features extracted from the CV curve include cyclic volt-ampere area, hysteresis characteristics, peak value characteristics, capacitance characteristics, symmetry characteristics, and electrochemical reversibility indices. Features extracted from the It curve include steady-state characteristics, dynamic response characteristics, and correlation characteristics. The principal component analysis features are obtained by capturing the overall morphological information of the I-V and C-V curves using principal component analysis.

4. The method for analyzing multi-component ion transport information in a confined space according to claim 1, characterized in that, The feature selection is as follows: Redundancy removal is achieved by using a Pearson correlation coefficient threshold. The Pearson correlation coefficient of the features is calculated and features with a correlation coefficient greater than the preset threshold are removed to obtain the redundancy-free features. Mutual information is used to evaluate and sort the information content of redundancy removal features and ion types in descending order; Select the top-ranked features as the filtered feature vector.

5. The method for analyzing multi-component ion transport information in a confined space according to claim 1, characterized in that, The known ion-mixed solution with known ion types and concentration ratios is an electrolyte solution; the cations in the multi-ion-mixed solution include K+. + Na + Li + Ca 2+ Mg 2+ Fe 2+ Co 2+ Zn 2+ Cu 2+ Mn 2+ Cr 3+ and Fe 3+ One or more of the following; the anions in a multi-ion mixed solution include Cl. - NO3 - SO4 2- SO3 2- NO2 - I - N 3- MoO4 2- Se 2- SeO3 2- SeO4 2- and SO3 2- One or more of the following; the multi-ion mixed solution includes cations and anions of different valence states; the ion concentration of the multi-ion mixed solution is between 0.1 and 100 mM / L.

6. The method for analyzing multi-component ion transport information in a confined space according to claim 1, characterized in that, The first machine learning model was selected through the following steps: Select several machine learning models; Each machine learning model is trained for ion type classification using the dataset; For each machine learning model, five-fold cross-validation was used to optimize hyperparameters, and grid search was used to determine the optimal combination of hyperparameters. The best-performing machine learning model is selected as the first machine learning model based on cross-validation accuracy and test accuracy. The second machine learning model is selected through the following steps: Select several machine learning models; Each machine learning model was trained to predict concentration ratios using the dataset. For each machine learning model, five-fold cross-validation was used to optimize hyperparameters, and grid search was used to determine the optimal combination of hyperparameters. The best-performing machine learning model is selected as the second machine learning model based on regression evaluation metrics.

7. The method for analyzing multi-component ion transport information in a confined space according to claim 6, characterized in that, Machine learning models used for ion type classification include gradient boosting, K-nearest neighbors, random forest, decision tree, support vector machine, XGBoost, and logistic regression; machine learning models used for concentration ratio prediction include support vector regression, random forest regression, XGBoost regression, Ridge regression, linear regression, and Lasso regression.

8. The method for analyzing multi-component ion transport information in a confined space according to claim 7, characterized in that, The first machine learning model is a gradient boosting model; the second machine learning model is a support vector regression model.

9. The method for analyzing multi-component ion transport information in a confined space according to claim 1, characterized in that, Before using the feature vectors as the dataset to train the two machine learning models, the process also includes: preprocessing the dataset, as follows: Baseline correction was performed using the air-tube method. Missing value imputation is performed using the median as the filler. Standardization involves fitting the mean and standard deviation to the training set in the dataset. Outlier removal is performed using robust Z-score and derivative peak rules to remove outliers from the training set in the dataset.

Images