Ion concentration analysis method and system

Abstract

The invention relates to the technical field of data processing, and discloses an ion concentration analysis method and system. The method comprises the following steps: carrying out spectral data acquisition on a plurality of monitoring points in a hydroxide ion bin to obtain an original spectral data set of each monitoring point; performing spectral amplitude, shape and information difference measurement on the original spectral data set to obtain comprehensive spectral characteristic data of each monitoring point; inputting the comprehensive spectral feature data of each monitoring point into a preset partial least square regression model for ion concentration prediction to obtain ion concentration prediction data of each monitoring point; and inputting the ion concentration prediction data of each monitoring point into a preset integrated cost-sensitive deep neural network model to perform ion concentration anomaly detection to obtain a target ion concentration anomaly detection result, thereby improving the ion concentration detection accuracy of the hydroxide ion bin.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an ion concentration analysis method and system. Background Art

[0002] Accurate ion concentration monitoring can greatly improve the operational performance and safety of hydroxide ion chambers. However, traditional ion concentration monitoring methods often rely on offline chemical analysis techniques, which are not only time-consuming but also often cannot meet the needs of rapid response because they cannot provide real-time data.

[0003] Although some existing online monitoring methods such as conductivity measurement and sensors have been widely used, these methods can usually only provide limited information and cannot capture more complex ion changes, such as the interaction of multiple ions and concentration dynamics. In other words, the accuracy of existing technologies is low. Summary of the invention

[0004] The present application provides an ion concentration analysis method and system for improving the ion concentration detection accuracy of a hydroxide ion chamber.

[0005] In a first aspect, the present application provides an ion concentration analysis method, the ion concentration analysis method comprising: Collecting spectral data from multiple monitoring points in the hydrogen and oxygen ion chamber to obtain an original spectral data set for each monitoring point; The spectral amplitude, shape and information difference measurements are respectively performed on the original spectral data set to obtain comprehensive spectral feature data of each monitoring point; The comprehensive spectral characteristic data of each monitoring point is input into a preset partial least squares regression model to predict the ion concentration, and the predicted ion concentration data of each monitoring point is obtained; The predicted ion concentration data of each monitoring point is input into the preset integrated cost-sensitive deep neural network model for ion concentration anomaly detection to obtain the target ion concentration anomaly detection result.

[0006] In a second aspect, the present application provides an ion concentration analysis system, the ion concentration analysis system comprising: The acquisition module is used to collect spectral data from multiple monitoring points in the hydroxide ion chamber to obtain an original spectral data set for each monitoring point; A measurement module is used to measure the spectral amplitude, shape and information difference of the original spectral data set to obtain comprehensive spectral feature data of each monitoring point; A prediction module is used to input the comprehensive spectral feature data of each monitoring point into a preset partial least squares regression model to predict the ion concentration, thereby obtaining the predicted ion concentration data of each monitoring point; The monitoring module is used to input the ion concentration prediction data of each monitoring point into a preset integrated cost-sensitive deep neural network model to perform ion concentration anomaly detection and obtain the target ion concentration anomaly detection result.

[0007] The third aspect of the present application provides an ion concentration analysis device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the ion concentration analysis device performs the above-mentioned ion concentration analysis method.

[0008] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned ion concentration analysis method.

[0009] In the technical solution provided by the present application, by deploying a multi-channel spectrometer in the hydroxyl ion chamber to collect spectral data in real time, the change of ion concentration can be continuously monitored to ensure the timeliness and continuity of the data. Through fine spectral amplitude, shape and information difference measurement, the slight differences between the monitoring points can be accurately identified and analyzed, which enhances the accuracy of the data and the meticulousness of the analysis. The partial least squares regression model is used to predict the ion concentration of the comprehensive spectral feature data, and the factors highly correlated with the ion concentration are effectively extracted from the complex spectral information. The data structure is simplified by dimensionality reduction technology, while retaining the most critical information for ion concentration prediction. Not only the accuracy of the prediction is improved, but also the prediction model can be dynamically adjusted based on historical and real-time data to adapt to possible environmental changes and equipment aging. By inputting the prediction data into the integrated cost-sensitive deep neural network model for ion concentration anomaly detection, the ability to identify potential risks is enhanced, thereby improving the accuracy of ion concentration detection in the hydroxyl ion chamber. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 This is a schematic diagram of an embodiment of the ion concentration analysis method in the embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of an ion concentration analysis system in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application embodiment provides an ion concentration analysis method and system. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the ion concentration analysis method in the embodiment of the present application includes: Step S101, collecting spectral data from multiple monitoring points in the hydroxide ion chamber to obtain an original spectral data set for each monitoring point; It is understandable that the execution subject of the present application may be an ion concentration analysis system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0014] Specifically, multiple monitoring points are pre-set in the hydroxide ion chamber, and the setting of the points is based on the expected monitoring needs and ion distribution characteristics. In order to ensure the accuracy and coverage of data collection, each monitoring point is equipped with a multi-channel spectrometer, which can operate according to the set wavelength range and acquisition frequency. When implementing data collection, the multi-channel spectrometer performs spectral scanning at each monitoring point according to predetermined parameters to obtain a first spectral data set. In order to improve data quality and ensure the accuracy of subsequent analysis, the first spectral data set is cleaned, including the removal of outliers and the interpolation of missing values, to obtain a second spectral data set. The second spectral data set is marked with data labels, which are based on factors such as the geographical location of the monitoring point and the acquisition time, so that each set of data can be accurately traced and analyzed. The labeled spectral data set is effectively divided into data sets, and the data is allocated to the original spectral data set corresponding to each monitoring point, so that the data of each monitoring point can be analyzed independently of other points, thereby providing more accurate input during model training and prediction, and ensuring the efficiency and accuracy of the entire ion concentration analysis method.

[0015] Step S102, respectively measuring the spectral amplitude, shape and information difference of the original spectral data set to obtain comprehensive spectral feature data of each monitoring point; Specifically, the original spectral data set is curve fitted to obtain the spectral curve of each monitoring point. The curve shows the basic graphics of the spectral data and also reflects the spectral response of the monitoring point under a specific environment. The spectral amplitude is compared and analyzed to obtain the spectral amplitude difference between different monitoring points. The comparison result is obtained by comparing the peak height and overall energy distribution of the spectral curves of each point. The difference in spectral amplitude is directly related to the change in ion concentration. By comparing the spectral shapes, the differences in peak shape symmetry, width and sharpness of the spectral curves of each monitoring point are analyzed. Shape comparison helps to understand the types of substances and state changes that may exist between different points. The spectral information measurement is performed on the spectral curve of each monitoring point, and the information richness of the spectral data is quantified by calculating the information entropy or other statistical measurement values ​​of each spectral curve, and the spectral information measurement differences between multiple monitoring points are obtained. The information measurement differences between different points help determine which monitoring points may be affected by external interference or internal parameter changes. According to the spectral amplitude difference comparison results, spectral shape difference comparison results and spectral information measurement differences, the spectral feature bands of the spectral curve of each monitoring point are extracted and fused. By selecting the most representative and distinctive bands in the spectral data of each monitoring point, and then effectively fusing these bands, comprehensive spectral feature data for each point can be formed.

[0016] According to the results of the spectral amplitude difference comparison, the spectral curve of each monitoring point is analyzed, and the band reflecting the most significant change in spectral intensity is extracted to obtain the first spectral characteristic band. Identify those bands with large amplitude changes, which can often indicate changes in ion concentration or differences in chemical composition. Perform a second round of feature extraction based on the results of the spectral shape difference comparison. Analyze the characteristic differences in shape of each spectral curve, such as peak width, symmetry, and sharpness, and extract bands with significant shape differences. These bands are defined as the second spectral characteristic bands. The analysis of spectral shape helps identify changes caused by physical or chemical processes, which may be caused by specific ions in the monitoring environment. According to the difference in spectral information measurement, characteristic bands with high information content are extracted from each spectral curve to obtain the third spectral characteristic band. Use information entropy or other statistical methods to evaluate the information contribution of each band, and select bands with large information content as key features. This metric reflects the amount of information carried in the spectral data, which helps to mine and utilize those bands that may be most critical for ion concentration prediction. The first, second and third spectral characteristic bands are screened for similarity to identify the bands that appear repeatedly in different feature extraction steps, ensuring that the target spectral characteristic bands finally selected are representative and consistent from multiple angles. By comparing the overlap and similarity of each characteristic band, the target spectral characteristic band that best represents the characteristics of each monitoring point is selected. The characteristic weights of the target spectral characteristic bands of each spectral curve are calculated, and the weight data reflects the importance and contribution of each characteristic band in the overall analysis. The weights may be determined based on the effectiveness of the bands in the ion concentration prediction model, or their ability to distinguish different chemical components. According to the calculated characteristic weight data, the target spectral characteristic bands of each spectral curve are weighted and combined to form comprehensive spectral characteristic data for each monitoring point. The weighted combination retains the unique information of each characteristic band and enhances the accuracy and robustness of the model in predicting ion concentrations.

[0017] Step S103, inputting the comprehensive spectral characteristic data of each monitoring point into a preset partial least squares regression model to predict the ion concentration, and obtaining the predicted ion concentration data of each monitoring point; Specifically, the comprehensive spectral feature data of each monitoring point is transformed into a standard normal variable, and the original spectral data is converted into standardized spectral feature data to ensure that the data has a uniform proportion and distribution before further analysis, thereby effectively avoiding the deviation of the model during the training process and improving the accuracy of the prediction. The standardized spectral feature data of each monitoring point is input into a preset partial least squares regression model. In this model, a spectral feature data matrix that can reveal the main change trend of the data is constructed by performing matrix conversion processing on the input data. The conversion is achieved by the partial least squares regression method, the core of which is to find the most important component, that is, the latent variable of the model, by maximizing the covariance between the dependent variable and the independent variable. The scores of the spectral feature data matrix in multiple latent variable spaces are calculated, and the score matrix of each monitoring point is constructed accordingly. The score matrix reflects the main change characteristics and patterns of the spectral data of each monitoring point, and provides key explanatory variables for the prediction of ion concentration. Through the score matrix of these latent variables, the main factors affecting the change of ion concentration are effectively captured. By obtaining the regression coefficient of the partial least squares regression model and combining the score matrix obtained above, the ion concentration prediction data of each monitoring point is calculated. The regression coefficient provides a specific conversion method from the latent variable score to the predicted ion concentration, ensuring the accuracy and reliability of the prediction. Through a comprehensive data processing and model application process, the ion concentration of each monitoring point is predicted, and the complex relationship between spectral data and ion concentration is obtained.

[0018] Step S104: input the predicted ion concentration data of each monitoring point into a preset integrated cost-sensitive deep neural network model to perform ion concentration anomaly detection to obtain a target ion concentration anomaly detection result.

[0019] Specifically, the in-warehouse location information of each monitoring point is obtained. The abnormality of ion concentration may be closely related to the specific location of the monitoring point. After obtaining the location information, it is encoded and converted into an in-warehouse location code. The purpose of encoding is to digitize the location information so that the neural network model can process this information. The ion concentration change characteristics of the ion concentration prediction data of each monitoring point are calculated to determine the ion concentration change value of each monitoring point at different time points. These change values ​​can reveal potential trends or mutations, which are crucial for identifying anomalies. The ion concentration change values ​​are encoded to obtain the ion concentration change code so that they can be effectively processed by the deep neural network model. The in-warehouse location code and ion concentration change code of each monitoring point are input into the preset integrated cost-sensitive deep neural network model. The model includes an input layer composed of multiple base classifier models, which are designed based on cost-sensitive algorithms and can optimize the processing of different cost-sensitive problems, such as the trade-off between false positives and false negatives. After receiving the input data, each base classifier processes the data and gives the prediction result at the output layer. Through the processing of the integrated model, the position code in the bin and the ion concentration change code of each monitoring point are used to detect ion concentration anomalies. The model evaluates all input data and identifies data patterns that may indicate ion concentration anomalies based on the deep learning capabilities of the neural network.

[0020] Process and analyze the bin position code and multiple ion concentration change codes of each monitoring point. In the input layer, the bin position code and multiple ion concentration change codes of each monitoring point are vector-mapped to generate the ion concentration input vector of each monitoring point. Convert all important monitoring information into a format that the model can effectively process. Input each ion concentration input vector into multiple base classifier models configured in the model. After receiving the input vector, each base classifier independently calculates the abnormal probability of the ion concentration corresponding to the vector, through the complex calculation process inside the model, including weight adjustment, activation function processing and other standard operations of multi-layer neural networks. Each classifier responds to specific features and patterns in the data set to enhance the sensitivity and accuracy of the model to abnormal conditions. In order to integrate the outputs of multiple base classifiers and improve the reliability of decision-making, a hard voting mechanism is used to comprehensively evaluate the abnormal probabilities of each ion concentration. In the hard voting mechanism, each base classifier is regarded as one vote for the judgment of whether there is an abnormality, and the final decision on whether each monitoring point is abnormal is based on the majority vote principle. The voting mechanism simplifies the decision-making process, while taking advantage of the advantages of ensemble learning and reducing the possibility of misjudgment of a single model. The initial ion concentration anomaly detection results of all monitoring points are summarized to form the target ion concentration anomaly detection results of the entire hydroxyl ion warehouse. The result summary provides a warehouse-level anomaly monitoring view and can provide a comprehensive analysis and explanation of the abnormal status of different areas or monitoring points.

[0021] In the embodiment of the present application, by deploying a multi-channel spectrometer in the hydroxide ion warehouse to collect spectral data in real time, the change of ion concentration can be continuously monitored to ensure the timeliness and continuity of the data. Through fine spectral amplitude, shape and information difference measurement, the slight differences between each monitoring point can be accurately identified and analyzed, which enhances the accuracy of the data and the meticulousness of the analysis. The partial least squares regression model is used to predict the ion concentration of the comprehensive spectral feature data, and the factors highly correlated with the ion concentration are effectively extracted from the complex spectral information. The data structure is simplified by dimensionality reduction technology, while retaining the most critical information for ion concentration prediction. Not only the accuracy of the prediction is improved, but also the prediction model can be dynamically adjusted based on historical and real-time data to adapt to possible environmental changes and equipment aging. By inputting the prediction data into the integrated cost-sensitive deep neural network model for ion concentration anomaly detection, the ability to identify potential risks is enhanced, thereby improving the accuracy of ion concentration detection in the hydroxide ion warehouse.

[0022] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Pre-set multiple monitoring points in the hydroxyl ion chamber and define the wavelength range and acquisition frequency of the multi-channel spectrometer; (2) using a multi-channel spectrometer to collect spectral data from multiple monitoring points according to the wavelength range and the collection frequency, thereby obtaining a first spectral data set; (3) removing outliers and interpolating missing values ​​from the first spectral data set to obtain a second spectral data set; (4) labeling the second spectral data set according to the multiple monitoring points to obtain a labeled spectral data set; (5) Divide the labeled spectral dataset to obtain the original spectral dataset of each monitoring point.

[0023] Specifically, multiple monitoring points are pre-set in the hydrogen and oxygen ion chamber. The selection of the points should be based on the uniformity and representativeness of the ion distribution in the chamber to ensure that the collected data can fully reflect the ion concentration state in the entire chamber. Define the wavelength range and acquisition frequency of the multi-channel spectrometer. The setting of these parameters should take into account the characteristics of the monitoring target and the specific needs of the monitoring environment. For example, the wavelength range should cover the characteristic spectral lines absorbed or emitted by key ions, and the acquisition frequency should be high enough to capture the rapid changes in ion concentration. Use the multi-channel spectrometer to collect spectral data for the monitoring points according to the set wavelength range and acquisition frequency. Collect the spectral data of each monitoring point to form a first spectral data set. Perform data cleaning on the first spectral data set, including removing outliers and interpolating missing values. Outlier removal is to exclude those data points that are non-representative or may be caused by instrument errors through statistical analysis methods, such as setting thresholds or using standard deviation judgments. Missing value interpolation is to fill in the gaps in the data set through various techniques such as linear interpolation, spline interpolation or more complex machine learning methods, such as decision tree interpolation, to ensure the integrity and coherence of the data and obtain the second spectral data set. The second spectral data set is labeled with data labels. By labeling the monitoring point information for each data point, the necessary background information is provided for subsequent data analysis. The addition of labels makes it easier to track the source of each data point and also helps in grouping or classification in subsequent analysis. The labeled spectral data set is divided into data sets, and the data is separated by monitoring points to form an independent original spectral data set for each monitoring point. By identifying the labels and assigning the data to the corresponding sets, each data set represents the spectral information of a specific monitoring point. In this way, it is ensured that the data of each monitoring point is considered independently when performing data analysis and model building, thereby improving the accuracy and pertinence of the analysis.

[0024] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Perform curve fitting on the original spectral data set to obtain the spectral curve of each monitoring point; (2) Compare the spectral amplitude of the spectral curve of each monitoring point to obtain the spectral amplitude difference comparison results between multiple monitoring points; (3) Compare the spectral shapes of the spectral curves of each monitoring point to obtain the spectral shape difference comparison results between multiple monitoring points; (4) Perform spectral information measurement on the spectral curve of each monitoring point to obtain the spectral information measurement difference between multiple monitoring points; (5) Based on the comparison results of spectral amplitude difference, spectral shape difference and spectral information metric difference, the spectral characteristic bands of the spectral curve of each monitoring point are extracted and fused to obtain the comprehensive spectral characteristic data of each monitoring point.

[0025] Specifically, curve fitting is performed on the original spectral data set to obtain the spectral curve of each monitoring point. Discrete data points are converted into a continuous spectral curve through mathematical models such as polynomial fitting, Gaussian fitting or Lorentz fitting. Curve fitting makes the data smoother and easier to analyze, and can more accurately depict the main characteristics and trends of the spectrum. The spectral amplitude of the spectral curve of each monitoring point is compared. The peak height and overall intensity of the curve are quantitatively analyzed. By comparing the spectral amplitudes of different monitoring points, the spatial differences in ion concentrations in the monitoring area can be effectively revealed. The spectral shapes are compared to observe the morphological characteristics such as width, symmetry and sharpness of the spectral curve of each monitoring point. Differences in spectral shapes often reflect different chemical components or differences in the behavior of the same component under different environmental conditions. Differential analysis of spectral information metrics is performed. The amount and quality of information contained in each spectral curve is measured by calculating information entropy or other statistical indicators. This metric reflects the efficiency and differences in the expression of chemical information by the spectral data of each monitoring point, which helps to identify which monitoring point data may contain more chemical or physical information. According to the results of spectral amplitude difference comparison, spectral shape difference comparison and spectral information metric difference, the spectral characteristic bands of the spectral curve of each monitoring point are extracted and fused. Identify those spectral regions that are significant in amplitude, shape or information metric, and extract the data of these regions for analysis and fusion. Feature extraction is achieved through signal processing techniques such as wavelet transform, Fourier transform or feature selection methods based on machine learning. Fusion of these characteristic bands requires consideration of their statistical importance and contribution to the prediction model, and uses algorithms such as principal component analysis or linear discriminant analysis to optimize the feature combination, ultimately forming comprehensive spectral characteristic data for each monitoring point.

[0026] In a specific embodiment, the execution step extracts and fuses the spectral characteristic bands of the spectral curve of each monitoring point according to the spectral amplitude difference comparison result, the spectral shape difference comparison result and the spectral information metric difference, and the process of obtaining the comprehensive spectral characteristic data of each monitoring point can specifically include the following steps: (1) Extracting the spectral characteristic bands of the spectral curve of each monitoring point according to the spectral amplitude difference comparison results to obtain a plurality of first spectral characteristic bands of each spectral curve; (2) extracting the spectral characteristic bands of the spectral curve of each monitoring point according to the spectral shape difference comparison results, and obtaining a plurality of second spectral characteristic bands of each spectral curve; (3) extracting the spectral characteristic bands of the spectral curve of each monitoring point according to the spectral information measurement difference, and obtaining multiple third spectral characteristic bands of each spectral curve; (4) performing similarity screening on the multiple first spectral characteristic bands, the multiple second spectral characteristic bands, and the multiple third spectral characteristic bands of each spectral curve to obtain multiple target spectral characteristic bands of each spectral curve; (5) Calculating the weights of multiple target spectral characteristic bands of each spectral curve to obtain characteristic weight data of each target spectral characteristic band; (6) Based on the characteristic weight data, multiple target spectral characteristic bands of each spectral curve are weighted and combined to obtain the comprehensive spectral characteristic data of each monitoring point.

[0027] Specifically, according to the results of the spectral amplitude difference comparison, the spectral characteristic bands of the spectral curve of each monitoring point are extracted respectively to identify those areas with significant amplitude changes, which represent the presence and concentration changes of specific ions or chemical substances. The characteristic bands are determined by analyzing the peak value and overall energy distribution in the spectral curve to ensure that the extracted first spectral characteristic bands can accurately reflect the changes in concentration. According to the results of the spectral shape difference comparison, the spectral curve of each monitoring point is subjected to a second round of spectral characteristic band extraction to obtain multiple second spectral characteristic bands of each spectral curve. This stage focuses on the morphological characteristics of the spectral curve, such as the width, symmetry and sharpness of the peak, which can reveal the molecular structure or interaction characteristics of the substance. The spectral shape is analyzed to obtain important information related to chemical properties. According to the spectral information metric difference, the spectral curve of each monitoring point is subjected to a third round of characteristic band extraction. The information content in the spectral data is evaluated using statistical methods or information theory. Methods such as calculating information entropy can help identify which bands contain the most information and obtain the third spectral characteristic band. The first, second and third spectral characteristic bands are screened for similarity. By comparing the similarities of these bands, characteristic bands that can represent the common characteristics or significant differences of each monitoring point are screened out. The screening is based on algorithms such as cluster analysis or principal component analysis to ensure that the target spectral characteristic bands finally selected are highly representative and discriminative. The target spectral characteristic bands are weighted and the weights are determined according to the importance of each band in characterizing chemical properties and concentration changes. The weight calculation can use machine learning algorithms such as feature importance scores in support vector machines or random forests, which can effectively quantify the contribution of each band. The target spectral characteristic bands of each spectral curve are weighted and combined according to the feature weight data. The combination process takes into account the individual influence of each band and the interaction between them to form comprehensive spectral characteristic data for each monitoring point.

[0028] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Perform standard normal variable transformation on the comprehensive spectral characteristic data of each monitoring point to obtain the standardized spectral characteristic data of each monitoring point; (2) Inputting the standardized spectral feature data of each monitoring point into a preset partial least squares regression model, performing data matrix conversion through the partial least squares regression model, and obtaining the spectral feature data matrix of each monitoring point; (3) Calculate the scores of the spectral feature data matrix in multiple latent variable spaces and construct a score matrix for each monitoring point; (4) Obtain the regression coefficient of the partial least squares regression model and calculate the predicted ion concentration data for each monitoring point based on the score matrix and regression coefficient.

[0029] Specifically, the comprehensive spectral feature data of each monitoring point is transformed into a standard normal variable so that each feature data conforms to the standard normal distribution. Standardization processing is helpful for subsequent statistical analysis and machine learning models. The optimization of algorithm performance depends on the standardized form of the input data, which can reduce the influence between features of different dimensions and ensure that the influence of each feature in the model is balanced. The standardized spectral feature data of each monitoring point is input into the preset partial least squares regression model. Partial least squares regression is a statistical method for processing data with multicollinearity and is suitable for spectral data analysis. The partial least squares model converts high-dimensional spectral features into a smaller number of latent variables through data matrix transformation. These latent variables are designed to capture the maximum information explaining the relationship between spectral data and target ion concentration. The scores of the spectral feature data matrix in multiple latent variable spaces are calculated. These scores reflect the performance of each monitoring point in each latent variable space in the partial least squares model, which is basically the projection of the original spectral data in the new latent variable space. The score matrix of each monitoring point shows the behavior and trend of the data in the model. The regression coefficient of the partial least squares regression model is obtained. These coefficients are obtained during the model training process and describe the specific contribution of each potential variable to the predicted ion concentration. Using the regression coefficients, the predicted ion concentration data for each monitoring point is calculated through the score matrix. The scores in the score matrix are multiplied by the corresponding regression coefficients to obtain the predicted ion concentration value.

[0030] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Obtain the in-warehouse location information of each monitoring point, and encode the in-warehouse location information to obtain the in-warehouse location code of each monitoring point; (2) Calculating the ion concentration change characteristics of the ion concentration prediction data of each monitoring point to obtain multiple ion concentration change values ​​of each monitoring point; (3) Encoding multiple ion concentration change values ​​to obtain multiple ion concentration change codes for each monitoring point; (4) Inputting the position code of each monitoring point in the warehouse and the multiple ion concentration change codes into a preset integrated cost-sensitive deep neural network model, wherein the preset integrated cost-sensitive deep neural network model includes an input layer, multiple base classifier models composed of cost-sensitive deep neural networks, and an output layer; (5) By integrating the cost-sensitive deep neural network model, the warehouse position encoding and multiple ion concentration change encodings of each monitoring point are used to detect ion concentration anomalies, and the target ion concentration anomaly detection results are obtained.

[0031] Specifically, the in-warehouse location information of each monitoring point is obtained, and the information is effectively encoded. The location information encoding is to convert the physical location into a data format that the model can understand and process, usually using numerical encoding or one-hot encoding methods. The ion concentration change characteristics of the ion concentration prediction data of each monitoring point are calculated. Important statistical features such as change amplitude, trend and periodicity are extracted from the time series data. These change values ​​can be obtained by calculating the ion concentration difference at consecutive time points or using time series analysis technology. The ion concentration change characteristics reflect the dynamic changes in the ion concentration of each monitoring point in different time periods. The obtained ion concentration change values ​​are encoded and converted into a format suitable for processing by the machine learning model. The ion concentration change encoding can use methods similar to position encoding, such as numericalization or one-hot encoding, to ensure the uniqueness and identifiability of each change value. Through this encoding, the model can more accurately identify and learn the patterns and laws of ion concentration changes. After the encoding is completed, the in-warehouse location encoding and ion concentration change encoding of each monitoring point are input into a preset integrated cost-sensitive deep neural network model. The model includes an input layer, a hidden layer composed of multiple base classifier models, and an output layer. Each base classifier is a cost-sensitive deep neural network that optimizes for different types of errors, such as false positives and false negatives. Cost sensitivity is achieved by adjusting the loss function during network training or by adjusting the weight distribution of the data set, so that the model focuses more on the more costly types of errors. After the input data is processed through these layers, the integrated cost-sensitive deep neural network model determines the final anomaly detection result by combining the outputs of each base classifier using techniques such as voting or weighted averaging. This integrated approach improves the robustness and accuracy of the model, combines the judgments of multiple models, and reduces the risks that may arise from relying on a single model. The model outputs the target ion concentration anomaly detection results for each monitoring point. These results are based on a comprehensive analysis of the specific location of the monitoring point and the changes in ion concentration, and can effectively indicate which monitoring points may have ion concentration anomalies.

[0032] In a specific embodiment, the execution step performs ion concentration anomaly detection on the in-bin position code and multiple ion concentration change codes of each monitoring point by integrating the cost-sensitive deep neural network model, and the process of obtaining the target ion concentration anomaly detection result can specifically include the following steps: (1) In the input layer, the bin position code of each monitoring point and multiple ion concentration change codes are vector-mapped to generate the ion concentration input vector of each monitoring point; (2) Inputting the ion concentration input vector of each monitoring point into multiple base classifier models, and calculating the ion concentration abnormality probability of the ion concentration input vector in the base classifier models; (3) A hard voting mechanism is used to conduct a comprehensive evaluation of the probability of abnormal ion concentration and obtain the initial ion concentration abnormality detection results for each monitoring point; (4) The initial ion concentration anomaly detection results of each monitoring point are summarized to obtain the target ion concentration anomaly detection results of the hydroxide ion bin.

[0033] Specifically, the position code in the bin and the ion concentration change code of each monitoring point are vector-mapped. The classification or sequence data is converted into a numerical vector that can be effectively processed by the machine learning algorithm. Usually, this conversion can be achieved by one-hot encoding to ensure the uniqueness and mutual exclusivity of each category, or by converting all features to the same measure through a standardization method, thereby avoiding the deviation caused by the difference in feature dimensions during the model training process. After the encoding is completed, the ion concentration input vector of each monitoring point contains all the necessary information, which maps the characteristics of the position and ion concentration change. The ion concentration input vector is input into multiple base classifier models. Each base classifier model is an independent deep neural network, which is used to identify and predict abnormal ion concentrations. After receiving the input vector, each network will analyze the current data point according to the pattern learned during the training process and calculate the abnormal probability of ion concentration. These calculations are based on the weights and biases of the network, which are optimized during the network training through methods such as back propagation and gradient descent to ensure that the model can accurately identify abnormal patterns. In order to improve the accuracy and reliability of decision-making, a hard voting mechanism is used to comprehensively evaluate the abnormal probability of ion concentration of all base classifiers. The hard voting mechanism means that the output of each classifier (abnormal or non-abnormal) will be regarded as one vote, and the final classification decision will be determined by majority voting. By combining the independent decisions of multiple models to improve the overall prediction performance, especially when dealing with environmental data with uncertainty and variability, it can significantly reduce the possible misjudgment of a single model. The initial ion concentration anomaly detection results of each monitoring point are summarized to obtain the target ion concentration anomaly detection results of the entire hydroxyl ion warehouse. The anomaly detection results of each monitoring point are analyzed and the information is integrated to ensure that the safety status of the entire environment can be comprehensively evaluated. The comprehensive evaluation not only provides detailed information about potential problems at specific monitoring points, but also reveals possible systemic risks or abnormal patterns in the entire warehouse.

[0034] The above describes the ion concentration analysis method in the embodiment of the present application. The following describes the ion concentration analysis system in the embodiment of the present application. Figure 2 In the embodiment of the present application, an ion concentration analysis system includes: The acquisition module 201 is used to collect spectral data from multiple monitoring points in the hydroxide ion chamber to obtain an original spectral data set for each monitoring point; The measurement module 202 is used to measure the spectral amplitude, shape and information difference of the original spectral data set to obtain the comprehensive spectral feature data of each monitoring point; Prediction module 203, used for inputting comprehensive spectral characteristic data of each monitoring point into a preset partial least squares regression model to predict ion concentration, and obtaining ion concentration prediction data of each monitoring point; The monitoring module 204 is used to input the ion concentration prediction data of each monitoring point into a preset integrated cost-sensitive deep neural network model to perform ion concentration anomaly detection to obtain a target ion concentration anomaly detection result.

[0035] Through the collaboration of the above components, by deploying a multi-channel spectrometer in the hydroxyl ion chamber to collect spectral data in real time, the changes in ion concentration can be continuously monitored to ensure the timeliness and continuity of the data. Through the fine spectral amplitude, shape and information difference measurement, the slight differences between the monitoring points can be accurately identified and analyzed, which enhances the accuracy of the data and the meticulousness of the analysis. The partial least squares regression model is used to predict the ion concentration of the comprehensive spectral feature data, effectively extracting factors highly correlated with the ion concentration from the complex spectral information. The data structure is simplified by dimensionality reduction technology, while retaining the most critical information for ion concentration prediction. Not only does it improve the accuracy of the prediction, but it can also dynamically adjust the prediction model based on historical and real-time data to adapt to possible environmental changes and equipment aging. By inputting the prediction data into the integrated cost-sensitive deep neural network model for ion concentration anomaly detection, the ability to identify potential risks is enhanced, thereby improving the accuracy of ion concentration detection in the hydroxyl ion chamber.

[0036] The present application also provides an ion concentration analysis device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the ion concentration analysis method in the above-mentioned embodiments.

[0037] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are executed on a computer, the computer executes the steps of the ion concentration analysis method.

[0038] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0039] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0040] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for analyzing ion concentration, characterized in that: The ion concentration analysis method comprises: Collecting spectral data from multiple monitoring points in the hydrogen and oxygen ion chamber to obtain an original spectral data set for each monitoring point; The spectral amplitude, shape and information difference measurements are respectively performed on the original spectral data set to obtain comprehensive spectral feature data of each monitoring point; The comprehensive spectral characteristic data of each monitoring point is input into a preset partial least squares regression model to predict the ion concentration, and the predicted ion concentration data of each monitoring point is obtained; The predicted ion concentration data of each monitoring point is input into the preset integrated cost-sensitive deep neural network model for ion concentration anomaly detection to obtain the target ion concentration anomaly detection result.

2. The ion concentration analysis method according to claim 1, characterized in that: The spectral data collection for multiple monitoring points in the hydroxide ion chamber to obtain the original spectral data set for each monitoring point includes: Pre-set multiple monitoring points in the hydroxide ion chamber, and define the wavelength range and acquisition frequency of the multi-channel spectrometer; Using a multi-channel spectrometer, collecting spectral data from the plurality of monitoring points according to the wavelength range and the collection frequency to obtain a first spectral data set; performing outlier removal and missing value interpolation on the first spectral data set to obtain a second spectral data set; Marking the second spectral data set with data labels according to the multiple monitoring points to obtain a labeled spectral data set; The labeled spectral data set is divided into data sets to obtain an original spectral data set for each monitoring point.

3. The ion concentration analysis method according to claim 1, characterized in that: The spectral amplitude, shape and information difference measurement are respectively performed on the original spectral data set to obtain comprehensive spectral feature data of each monitoring point, including: Performing curve fitting on the original spectral data set to obtain a spectral curve of each monitoring point; Comparing the spectrum amplitude of the spectrum curve of each monitoring point to obtain the spectrum amplitude difference comparison result between the multiple monitoring points; Comparing the spectral shapes of the spectral curves of each monitoring point to obtain the spectral shape difference comparison results between the multiple monitoring points; Performing spectral information measurement on the spectral curve of each monitoring point to obtain the spectral information measurement difference between the multiple monitoring points; According to the spectral amplitude difference comparison result, the spectral shape difference comparison result and the spectral information metric difference, the spectral characteristic bands of the spectral curve of each monitoring point are extracted and fused respectively to obtain the comprehensive spectral characteristic data of each monitoring point.

4. The ion concentration analysis method according to claim 3, characterized in that: According to the spectral amplitude difference comparison result, the spectral shape difference comparison result and the spectral information metric difference, the spectral characteristic bands of the spectral curve of each monitoring point are extracted and fused respectively to obtain the comprehensive spectral characteristic data of each monitoring point, including: Extracting the spectral characteristic bands of the spectral curve of each monitoring point according to the spectral amplitude difference comparison result, and obtaining a plurality of first spectral characteristic bands of each spectral curve; Extracting the spectral characteristic bands of the spectral curve of each monitoring point according to the spectral shape difference comparison result, and obtaining a plurality of second spectral characteristic bands of each spectral curve; Extracting spectral characteristic bands of the spectral curve of each monitoring point according to the spectral information measurement difference, and obtaining a plurality of third spectral characteristic bands of each spectral curve; Performing similarity screening on a plurality of first spectral characteristic bands, a plurality of second spectral characteristic bands, and a plurality of third spectral characteristic bands of each spectral curve to obtain a plurality of target spectral characteristic bands of each spectral curve; Performing weight calculation on multiple target spectral characteristic bands of each spectral curve to obtain characteristic weight data of each target spectral characteristic band; The multiple target spectral characteristic bands of each spectral curve are subjected to band weighted combination according to the characteristic weight data to obtain comprehensive spectral characteristic data of each monitoring point.

5. The ion concentration analysis method according to claim 1, characterized in that: The comprehensive spectral feature data of each monitoring point is input into a preset partial least squares regression model to predict the ion concentration, and the ion concentration prediction data of each monitoring point is obtained, including: Perform standard normal variable transformation on the comprehensive spectral characteristic data of each monitoring point to obtain the standardized spectral characteristic data of each monitoring point; The standardized spectral feature data of each monitoring point is respectively input into a preset partial least squares regression model, and a data matrix is ​​converted through the partial least squares regression model to obtain a spectral feature data matrix of each monitoring point; Calculating the scores of the spectral feature data matrix in multiple latent variable spaces, and constructing a score matrix for each monitoring point; The regression coefficient of the partial least squares regression model is obtained, and the ion concentration prediction data of each monitoring point is calculated according to the score matrix and the regression coefficient.

6. The ion concentration analysis method according to claim 1, characterized in that: The ion concentration prediction data of each monitoring point is input into a preset integrated cost-sensitive deep neural network model to perform ion concentration anomaly detection to obtain a target ion concentration anomaly detection result, including: Obtaining the in-warehouse position information of each monitoring point, and encoding the in-warehouse position information to obtain the in-warehouse position code of each monitoring point; Calculate the ion concentration change characteristics of the ion concentration prediction data of each monitoring point respectively to obtain multiple ion concentration change values ​​of each monitoring point; Encoding the multiple ion concentration change values ​​to obtain multiple ion concentration change codes for each monitoring point; Inputting the in-bin position code of each monitoring point and multiple ion concentration change codes into a preset integrated cost-sensitive deep neural network model, wherein the preset integrated cost-sensitive deep neural network model includes an input layer, multiple base classifier models composed of cost-sensitive deep neural networks, and an output layer; The integrated cost-sensitive deep neural network model is used to perform ion concentration anomaly detection on the in-bin position code and multiple ion concentration change codes of each monitoring point to obtain the target ion concentration anomaly detection result.

7. The ion concentration analysis method according to claim 6, characterized in that: The integrated cost-sensitive deep neural network model is used to perform ion concentration anomaly detection on the in-bin position code and multiple ion concentration change codes of each monitoring point to obtain a target ion concentration anomaly detection result, including: In the input layer, vector mapping is performed on the in-bin position code of each monitoring point and the multiple ion concentration change codes to generate an ion concentration input vector of each monitoring point; Inputting the ion concentration input vector of each monitoring point into the multiple base classifier models respectively, and calculating the ion concentration abnormality probability of the ion concentration input vector in the base classifier models respectively; A hard voting mechanism is used to conduct a comprehensive abnormal evaluation on the abnormal probability of the ion concentration, and obtain the initial ion concentration abnormal detection result of each monitoring point; The initial ion concentration anomaly detection results of each monitoring point are summarized to obtain the target ion concentration anomaly detection results of the hydroxide ion bin.

8. An ion concentration analysis system, characterized in that: The ion concentration analysis system comprises: The acquisition module is used to collect spectral data from multiple monitoring points in the hydroxide ion chamber to obtain an original spectral data set for each monitoring point; A measurement module is used to measure the spectral amplitude, shape and information difference of the original spectral data set to obtain comprehensive spectral feature data of each monitoring point; A prediction module is used to input the comprehensive spectral feature data of each monitoring point into a preset partial least squares regression model to predict the ion concentration, thereby obtaining the predicted ion concentration data of each monitoring point; The monitoring module is used to input the ion concentration prediction data of each monitoring point into a preset integrated cost-sensitive deep neural network model to perform ion concentration anomaly detection and obtain the target ion concentration anomaly detection result.

9. An ion concentration analysis device, characterized in that: The ion concentration analysis device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the ion concentration analysis device to perform the ion concentration analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the ion concentration analysis method according to any one of claims 1 to 7 is implemented.

Images