A method for intelligent extension of instrument range based on sample dilution optimization

Abstract

This invention relates to the field of instrument range extension technology, specifically to an intelligent range extension and optimization method for instruments based on sample dilution. The method includes: obtaining an effective range threshold using an intelligent sensor; statistically analyzing the relative standard deviation and signal-to-noise ratio of the concentration range within the range to obtain an accuracy distribution curve; performing a trial test on the sample to be tested; if the response value is within the range, direct detection is performed; if it exceeds the range, dilution parameters are matched according to the type of contaminant in the sample, and an intelligent prediction model is activated to recommend a dilution factor based on the initial signal strength and sample prediction; when the prediction confidence is lower than the threshold, a gradient scan is triggered; parallel sample measurements are performed on the diluted solution to calculate the relative standard deviation; for samples where the difference from the segment boundary is less than the threshold, two adjacent dilution factors are executed, and mutual verification is performed through the consistency of concentration in the overlapping area; full-process detection is performed for different dilution factors, a dilution correction model is established, and the correction coefficients of the dilution factors are updated, along with updates to the range boundary and accuracy distribution curve.

Description

Technical Field

[0001] This invention relates to the field of instrument range extension technology, specifically to an intelligent extension and optimization method for instrument range based on sample dilution. Background Technology

[0002] In fields such as chemical analysis, biological detection, and environmental monitoring, various analytical instruments typically have specific measurement ranges. When the concentration or content of the sample to be tested exceeds the linear detection range of the instrument, the sample needs to be diluted to bring it into the measurable range. Traditional sample dilution methods mainly rely on the operator's experience and judgment, selecting a fixed dilution factor based on the estimated concentration of the sample. This approach has several limitations: Firstly, for samples with unknown concentrations or large ranges, it is difficult to accurately predict the appropriate dilution factor, often requiring multiple attempts to obtain effective measurement results, leading to cumbersome operation, long processing time, and large sample consumption. Secondly, using a fixed dilution factor cannot fully utilize the instrument's range. When the diluted sample concentration is too low, it will approach the lower detection limit, significantly reducing measurement accuracy, while a dilution factor that is too small may still exceed the measurement range.

[0003] While some automated dilution devices exist, most can only execute preset dilution programs and lack the ability to dynamically adjust the dilution strategy based on actual measurement results. Furthermore, during multiple dilution processes, errors from each step accumulate and propagate. How to control these accumulated errors within an acceptable range while extending the measurement range remains a problem that current technology has not effectively solved. Particularly for high-concentration samples, selecting the optimal combination of dilution path and dilution factor to maximize the measurement range while ensuring measurement accuracy remains a challenging problem in this field.

[0004] To address this, a method for intelligent expansion and optimization of instrument range based on sample dilution is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent expansion and optimization method for instrument range based on sample dilution. By establishing an accuracy distribution curve and mapping matrix, and combining an intelligent prediction model and gradient scanning mechanism, intelligent expansion of the instrument range is achieved.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent range extension optimization of instruments based on sample dilution includes: Linear regression analysis of the instrument was performed using standard solutions, and the effective range threshold was obtained through intelligent sensors. The relative standard deviation and signal-to-noise ratio of the concentration range within the range were statistically analyzed to obtain the accuracy distribution curve. The extended range was divided into multiple sub-ranges, and a mapping matrix between the concentration range, dilution factor and target measurement range was established. The sample to be tested is tested in its original form. If the response value is within the range, it is directly detected. If it exceeds the range, the dilution parameters are matched according to the type of pollutant in the sample, and the intelligent prediction model is activated to predict and recommend the dilution factor based on the initial signal intensity and sample properties. When the prediction confidence is lower than the set threshold, gradient scanning is triggered to continuously dilute and monitor the response values ​​at each level in real time. When the response value enters the target measurement range and the rate of change is stable, the dilution level is locked. The relative standard deviation of the diluent is calculated by performing parallel sample determinations. For dilutions with a distance difference from the segment boundary less than the first threshold, two adjacent dilution schemes are executed simultaneously, and the consistency of concentration in the overlapping area is mutually verified. The entire process was tested for different dilution factors, a dilution correction model was established, the correction coefficients for the dilution factors were updated, and the range boundary and accuracy distribution curves were updated.

[0007] Preferably, the process of obtaining the accuracy distribution curve specifically includes: selecting multiple concentration points at equal intervals within the effective measurement range threshold; preparing a standard solution for each concentration point and performing multiple parallel measurements; collecting the raw data of the response signal output by the intelligent sensor in the instrument; performing statistical analysis on the multiple measurement data of each concentration point to obtain the average value and standard deviation, and obtaining the relative standard deviation; simultaneously analyzing the signal waveforms collected in each measurement to identify the peak amplitude of the effective signal and the baseline noise amplitude of the measurement gap, and obtaining the signal-to-noise ratio; identifying the concentration range where the relative standard deviation is lower than the first threshold and the signal-to-noise ratio is higher than the second threshold, and marking it as the high-precision measurement area.

[0008] Preferably, the process of establishing the mapping matrix specifically includes: The upper limit of the effective measurement range threshold is set as the first boundary point, and the upper limit of the target extended measurement range is set as the second boundary point, thus obtaining the concentration span between the two. A segmentation strategy is determined based on the order of magnitude of the concentration span, dividing the area from the first boundary point to the second boundary point into sub-intervals. For each sub-interval, the lower limit, upper limit, and concentration boundary value of the high-precision measurement area are obtained, thus determining the dilution factor range required for the concentration range of the sub-interval to fall into the high-precision measurement area after dilution. For two adjacent sub-intervals, the difference in concentration boundary values ​​is calculated, and a concentration overlap region is set at the boundary between the two sub-intervals. The concentration range of the overlap region is simultaneously detected using the recommended dilution factors of the two adjacent sub-intervals.

[0009] Preferably, the intelligent prediction model includes a data preprocessing layer, a feature calculation layer, a model inference layer, and a confidence evaluation layer; The data preprocessing layer receives the pollutant type and initial response signal intensity of the current sample to be tested. It performs one-heat encoding on the pollutant type to convert it into a pollutant vector, and normalizes the initial response signal intensity to map the value to a standard range. The pollutant vector and signal intensity are then combined into an input feature vector. The feature calculation layer calculates the ratio of the initial signal intensity to the effective range threshold from the input feature vector as an over-range feature, queries the dilution factor distribution of similar pollutants as a priori features, and combines the input feature vector, over-range feature, and prior features into an extended feature vector. The model inference layer performs forward calculation on the extended feature vector, outputting the predicted dilution factor and its corresponding probability distribution, and selects the dilution factor with the highest probability as the recommended value. The confidence assessment layer calculates a confidence index based on the predicted probability distribution. When the difference between the highest probability value and the second highest probability value is greater than a preset difference threshold, it is considered high confidence; when the difference is less than the preset difference threshold, it is considered low confidence. The confidence assessment result is then output to the gradient scan trigger module.

[0010] Preferably, the gradient scanning process specifically includes: Multiple adjacent sub-intervals covering the concentration corresponding to the initial test response signal are obtained from the mapping matrix, and the corresponding recommended dilution factors are extracted to form a candidate dilution factor sequence. Each dilution factor in the sequence is executed sequentially on the test sample, and the response signal intensity of the diluted solution is measured after mixing. For each dilution level, its dilution factor and corresponding response signal intensity are recorded, and the ratio of the current level response signal to the previous level response signal is calculated. The ratio of the current level dilution factor to the previous level dilution factor is also calculated to obtain the linearity coefficient. It is determined whether the response signal of the current level falls into the target measurement interval, and at the same time, it is determined whether the linearity coefficient is close to the theoretical value. When the response signal falls into the target measurement interval and the deviation of the linearity coefficient from the theoretical value is less than the preset deviation threshold, the gradient scan is terminated and the current dilution level is locked. If no level that meets the conditions is found after traversing all candidate dilution factors, the dilution level whose response signal is closest to the center value of the target measurement interval is selected as the final selection.

[0011] Preferably, the dilution correction model includes a standard substance testing layer, a deviation calculation layer, a model fitting layer, and a dynamic update layer; The standard substance testing layer dilutes standard samples using the recommended dilution factors in the mapping matrix to obtain the dilution factor, theoretical concentration after dilution, and actual measured concentration. The deviation calculation layer uses the dilution factor, theoretical concentration after dilution, and actual measured concentration to infer the original concentration, compares the inferred original concentration with the standard value of the standard substance, and calculates the absolute and relative deviations. The model fitting layer fits the deviation-dilution factor relationship curve to obtain the correction equation, and calculates the correction coefficient corresponding to each dilution factor based on the correction equation. The dynamic update layer periodically repeats the standard substance testing and deviation calculation process, performs trend analysis on the newly obtained deviation data and historical deviation data, and triggers the model reconstruction process when the mean deviation value of multiple consecutive periods shows a unidirectional drift and the drift amplitude exceeds a preset threshold, using the latest data to refit the correction equation and update the correction coefficients.

[0012] Preferably, the process of updating the range boundary and accuracy distribution curve specifically includes: when the dynamic update layer triggers the model reconstruction process, the range boundary update is started synchronously to obtain the upper limit of the linear response under the current state. The newly obtained upper limit of the linear response is compared with the original effective range threshold. If the difference exceeds a preset ratio, the effective range threshold is updated. The accuracy distribution test is re-performed on the concentration interval within the range to obtain new relative standard deviation and signal-to-noise ratio data. The updated accuracy distribution curve is plotted to identify the new high-precision measurement area boundary value. Based on the updated effective range threshold and high-precision measurement area boundary value, the recommended dilution factor for each sub-interval is recalculated to determine whether the recommended dilution factor has changed. When the recommended dilution factor changes, the dilution factor field and the target concentration interval field after dilution in the corresponding sub-interval in the mapping matrix are updated to keep the concentration boundary value of the sub-interval unchanged. The updated effective range threshold, accuracy distribution curve parameters, and mapping matrix are synchronously written into the database, and the range-related derived feature calculation rules are updated in the feature calculation layer of the intelligent prediction model.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention obtains the effective measurement range threshold and accuracy distribution curves through linear regression of standard solutions, signal-to-noise ratio, and relative standard deviation analysis, and establishes a mapping matrix between concentration range and dilution factor, forming a data-driven basis for dilution decisions. When the sample response exceeds the range, the intelligent prediction model automatically recommends a dilution factor based on the type of contaminant and the initial signal intensity, and introduces a confidence judgment mechanism. At low confidence levels, gradient scanning is triggered to verify the response stability and linearity step by step, thereby accurately determining the dilution level. This method significantly reduces the number of experiments and human error, allowing samples exceeding the range to directly reach the effective measurement area in one go, improving dilution accuracy and overall detection efficiency.

[0014] 2. This invention utilizes concentration segmentation and interval overlap design to enable parallel detection at dual dilution factors in the boundary region between adjacent intervals. This achieves mutual verification of concentration consistency in the overlapping area, effectively eliminating systematic biases caused by unstable boundary judgments and abrupt dilution changes. Simultaneously, this invention performs full-process standard material verification for each dilution factor, constructs a bias-dilution factor relationship model, fits a dilution correction equation, and continuously corrects the correction coefficient through a dynamic update mechanism, ensuring the data always maintains optimal calibration. This dilution correction and consistency verification system ensures that measurement results under different dilution strategies are mutually traceable and aligned, significantly improving data reliability and cross-interval continuity.

[0015] 3. This invention introduces a dynamic update mechanism. Through periodic standard sample testing and deviation trend monitoring, when a continuous unidirectional drift in deviation data is detected and exceeds a threshold, the dilution correction model reconstruction and range boundary update are automatically triggered. Simultaneously, the accuracy distribution is retested and the recommended dilution factor in the mapping matrix is ​​updated, ensuring the algorithm always matches the instrument's current actual performance. The updated range, accuracy parameters, and mapping rules are synchronously written back to the database and drive real-time adjustments to relevant features in the intelligent prediction model, ensuring the inference results are consistent with the latest accuracy model. This self-learning and adaptive mechanism reduces the accumulation of long-term operating errors, enabling the instrument to maintain stable and accurate scalability throughout its lifespan, significantly reducing manual calibration and maintenance interventions. Attached Figure Description

[0016] Figure 1 A flowchart of an instrument range intelligent expansion optimization method based on sample dilution is provided for this invention; Figure 2 This is a schematic diagram of the intelligent prediction model structure provided by the present invention; Figure 3 This is a schematic diagram of the dilution correction model structure provided by the present invention; Figure 4 A schematic diagram illustrating the intelligent extension of the instrument range provided by this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0018] Please see Figure 1 This invention provides a method for intelligent expansion and optimization of instrument range based on sample dilution, the technical solution of which is as follows: Linear regression analysis of the instrument was performed using standard solutions, and the effective range threshold was obtained through intelligent sensors. The relative standard deviation and signal-to-noise ratio of the concentration range within the range were statistically analyzed to obtain the accuracy distribution curve. In this embodiment, the intelligent sensor is a spectroscopic sensor, electrochemical sensor, or chromatographic detector equipped with digital signal processing capabilities, capable of acquiring complete waveform data of the response signal in real time. The process of obtaining the effective range threshold is as follows: a series of standard solutions with increasing concentrations are prepared, starting from a low concentration close to the detection limit, and gradually increasing the concentration until the instrument response signal shows a nonlinear deviation. Linear regression analysis is performed on the response signals of all concentration points, and the correlation coefficient and residual distribution of the fitted straight line are calculated. When the correlation coefficient is lower than 0.995 or the residual exceeds 5%, it is determined that the concentration point exceeds the linear range. The previous concentration point is determined as the upper limit threshold of the effective range, and the first concentration point with a signal-to-noise ratio higher than 10:1 is determined as the lower limit threshold of the effective range.

[0019] Furthermore, the process of obtaining the accuracy distribution curve specifically includes: selecting multiple concentration points at equal intervals within the effective measurement range threshold; preparing standard solutions for each concentration point and performing multiple parallel measurements; collecting the raw data of the response signals output by the intelligent sensor in the instrument; performing statistical analysis on the multiple measurement data for each concentration point to obtain the average value and standard deviation, and obtaining the relative standard deviation; simultaneously analyzing the signal waveforms collected in each measurement to identify the peak amplitude of the effective signal and the baseline noise amplitude of the measurement gap, and obtaining the signal-to-noise ratio; identifying the concentration range where the relative standard deviation is lower than the first threshold and the signal-to-noise ratio is higher than the second threshold, and marking it as the high-precision measurement area.

[0020] Specifically, in acquiring the accuracy distribution curve, multiple concentration test points are first selected within the determined effective measurement range threshold according to a preset equal-interval principle. For example, a test point is set every 10 mg / L within the 0-100 mg / L range. For each selected concentration point, a standard solution of the corresponding concentration is prepared using standard substances, and no fewer than six parallel repeated measurements are performed. The raw response signal data of each measurement is collected in real time by an intelligent sensor, including the complete time series data of the signal waveform. For each concentration point, the system performs statistical analysis on the dataset obtained from multiple parallel measurements, calculates the arithmetic mean and standard deviation of the measured values ​​at that concentration point, and then obtains the relative standard deviation (RSD) value, which directly reflects the measurement precision at that concentration point. Simultaneously, the signal waveform data acquired in each measurement is analyzed. Signal processing algorithms are used to identify and extract the peak amplitude of the effective analytical signal. At the same time, noise amplitude in the baseline region is acquired during measurement intervals when there is no sample signal. The signal-to-noise ratio (SNR) value for that concentration point is calculated by the ratio of the peak amplitude to the noise amplitude. The data structure of the accuracy distribution curve includes two independent curves: a relative standard deviation (RSD) curve and a signal-to-noise ratio (SNR) curve. The RSD curve uses concentration as the x-axis and the percentage of relative standard deviation (RSD) as the y-axis. A continuous RSD-concentration curve is obtained by cubic spline interpolation of the RSD statistics for each test concentration point. The SNR curve uses concentration as the x-axis and the SNR value as the y-axis. The SNR curve is obtained by calculating the SNR data for each test concentration point using... Three-dimensional spline interpolation is performed to obtain continuous SNR-concentration curves. The data points of these two curves and the interpolation parameters are stored in a database. The first threshold is determined according to the instrument type and detection requirements. For spectroscopic instruments, the first threshold is set to 3% to 5%, and for chromatographic instruments, the first threshold is set to 5% to 10%. The second threshold is determined according to the signal detection principle. For optical detection instruments, the second threshold is set to 10:1 to 20:1, and for electrochemical detection instruments, the second threshold is set to 5:1 to 10:1. Subsequently, the RSD and SNR data of all test concentration points are comprehensively evaluated to identify continuous concentration intervals that simultaneously satisfy RSD below the first threshold and SNR above the second threshold. These intervals are marked as high-precision measurement areas.

[0021] In this embodiment, a precision distribution map within the measurement range was established through multi-point precision characterization, identifying the instrument's optimal operating range, i.e., the high-precision measurement zone, providing a scientific basis for the subsequent determination of the dilution factor. This method considers not only measurement repeatability (characterized by RSD) but also detection capability (characterized by SNR), with these dual indicators ensuring the reliability of the target measurement zone.

[0022] The extended measurement range is divided into multiple sub-ranges, and a mapping matrix between the concentration range, dilution factor and target measurement range is established. Furthermore, the process of establishing the mapping matrix specifically includes: The upper limit of the effective measurement range threshold is set as the first boundary point, and the upper limit of the target extended measurement range is set as the second boundary point, thus obtaining the concentration span between the two. A segmentation strategy is determined based on the order of magnitude of the concentration span, dividing the area from the first boundary point to the second boundary point into sub-intervals. For each sub-interval, the lower limit, upper limit, and concentration boundary value of the high-precision measurement area are obtained, thus determining the dilution factor range required for the concentration range of the sub-interval to fall into the high-precision measurement area after dilution. For two adjacent sub-intervals, the difference in concentration boundary values ​​is calculated, and a concentration overlap region is set at the boundary between the two sub-intervals. The concentration range of the overlap region is simultaneously detected using the recommended dilution factors of the two adjacent sub-intervals.

[0023] Specifically, in the process of establishing the mapping matrix, the upper limit concentration value of the original effective range of the instrument is obtained as the first dividing point, such as 100 mg / L. At the same time, the upper limit concentration value of the target range to be expanded is obtained as the second dividing point, such as 10000 mg / L. The concentration span is calculated by dividing the two by 100. Based on the order-of-magnitude characteristics of the concentration span, a scientific segmentation strategy is determined. For example, for an expansion requirement spanning two orders of magnitude, the concentration range between the first dividing point and the second dividing point is divided into several sub-intervals using the logarithmic division principle. For example, 100-316 mg / L is the first sub-interval, 316-1000 mg / L is the second sub-interval, 1000-3160 mg / L is the third sub-interval, and 3160-10000 mg / L is the fourth sub-interval. The specific implementation of the segmentation strategy is as follows: calculate the ratio of the concentration range to the upper limit of the effective range. When the ratio is less than 10, divide the range into two to three sub-intervals. When the ratio is between 10 and 100, divide the range into three to five sub-intervals. When the ratio is greater than 100, divide the range into five to eight sub-intervals. The width of each sub-interval is evenly distributed according to the logarithmic scale to ensure that each sub-interval covers the same order of magnitude. For each sub-interval, the lower and upper concentration limits, as well as the concentration boundary values ​​of the previously determined high-precision measurement zone (e.g., 20–80 mg / L), are extracted. The dilution factor required for the sample in the sub-interval to fall exactly into the high-precision measurement zone after dilution is determined by calculating the concentration ratio. For example, the first sub-interval (100–316 mg / L) requires a dilution of 1.25–4 times, and the second sub-interval (316–1000 mg / L) requires a dilution of 4–12.5 times. The dilution factor is calculated by dividing the upper concentration limit of the sub-interval by the upper concentration limit of the high-precision measurement zone and rounding up to the standard dilution factor series as the recommended dilution factor for the sub-interval. The standard dilution factor series includes 2x, 5x, 10x, 20x, 50x, 100x, etc., to ensure the repeatability of the dilution operation and the accuracy of the pipette. For each sub-interval, the system calculates its target concentration range after dilution, that is, divides the lower and upper limits of the sub-interval concentration by the recommended dilution factor, respectively, to verify whether the diluted concentration falls completely within the high-precision measurement range. If it does not, the dilution factor is adjusted until the requirements are met. To ensure the continuity and verifiability of measurements between adjacent sub-intervals, boundary analysis is performed on adjacent sub-intervals to calculate the numerical difference between their concentration boundary values. A concentration overlap region of a certain width is artificially set at the boundary between two sub-intervals, for example, in the range of 300–330 mg / L. Samples within this overlap region will be simultaneously detected in parallel using the recommended dilution factors of the first and second sub-intervals. The results obtained from the two dilution schemes are cross-compare and verified.The width of the overlapping region is set to within ±10% of the boundary value of adjacent sub-intervals to ensure that samples near the boundary can be covered by both dilution schemes simultaneously. When measuring samples within the overlapping region, the system automatically performs parallel dilution and measurement at two dilution factors, compares the original concentrations calculated from the two measurement results, and considers both dilution schemes to be effective when the relative deviation is less than 10%, taking the average value as the final result; when the relative deviation is greater than 10%, the scheme result whose diluted concentration is closer to the center value of the high-precision measurement area is selected.

[0024] The data structure of the mapping matrix adopts a relational data table format and includes the following fields: sub-interval number, lower limit of sub-interval concentration, upper limit of sub-interval concentration, recommended dilution factor, lower limit of target concentration after dilution, upper limit of target concentration after dilution, preceding sub-interval number, subsequent sub-interval number, and overlapping area flag.

[0025] In this embodiment, by establishing a mapping matrix between concentration ranges and dilution factors, structured management and rapid query response of the extended measurement range are achieved. The segmentation strategy is adaptively adjusted according to the concentration span, which not only ensures the rationality of the dilution factor within each sub-range but also avoids the increase in system complexity caused by excessive subdivision. The overlapping area design at the boundary of adjacent sub-ranges provides a dual detection scheme for samples near the boundary. The results obtained by the two dilution factors are mutually verified, which effectively eliminates the measurement error that may be caused by improper selection of segment boundaries and significantly improves the reliability and continuity of measurement results throughout the entire extended measurement range.

[0026] The sample to be tested is tested in its original form. If the response value is within the range, it is directly detected. If it exceeds the range, the dilution parameters are matched according to the type of pollutant in the sample, and the intelligent prediction model is activated to predict and recommend the dilution factor based on the initial signal intensity and sample properties. When the prediction confidence is lower than the set threshold, gradient scanning is triggered to continuously dilute and monitor the response values ​​at each level in real time. When the response value enters the target measurement range and the rate of change is stable, the dilution level is locked. Furthermore, the intelligent prediction model includes a data preprocessing layer, a feature calculation layer, a model inference layer, and a confidence evaluation layer, referencing... Figure 2 ; The data preprocessing layer receives the pollutant type and initial response signal intensity of the current sample to be tested. It performs one-heat encoding on the pollutant type to convert it into a pollutant vector, and normalizes the initial response signal intensity to map the value to a standard range. The pollutant vector and signal intensity are then combined into an input feature vector. The feature calculation layer calculates the ratio of the initial signal intensity to the effective range threshold from the input feature vector as an over-range feature, queries the dilution factor distribution of similar pollutants as a priori features, and combines the input feature vector, over-range feature, and prior features into an extended feature vector. The model inference layer performs forward calculation on the extended feature vector, outputting the predicted dilution factor and its corresponding probability distribution, and selects the dilution factor with the highest probability as the recommended value. The confidence assessment layer calculates a confidence index based on the predicted probability distribution. When the difference between the highest probability value and the second highest probability value is greater than a preset difference threshold, it is considered high confidence; when the difference is less than the preset difference threshold, it is considered low confidence. The confidence assessment result is then output to the gradient scan trigger module.

[0027] Specifically, the intelligent prediction model employs a four-layer architecture to intelligently recommend the optimal dilution factor. The data preprocessing layer, serving as the model's input interface, receives two key types of information: first, the pollutant type of the current sample (e.g., classification identifiers for heavy metals, organic matter, and inorganic ions); and second, the initial measured response signal intensity value. This layer first performs one-heat encoding on the discrete classification variable of pollutant type, converting it into a computer-recognizable pollutant vector form. For example, "heavy metals" is encoded as [1, 0, 0, 0], and "organic matter" is encoded as [0, 1, 0, 0]. Simultaneously, it performs one-heat encoding on the initial measured response signal intensity value. The continuous numerical variable of intensity is normalized using a max-min normalization method to map the original value to a standard range of 0-1, eliminating the influence of dimensions. The processed pollutant vector and normalized signal intensity are then combined to form a unified input feature vector, which is passed to the next layer. The feature calculation layer receives the input feature vector and performs feature engineering. First, it calculates the ratio of the initial measured signal intensity to the instrument's effective range threshold. This ratio directly reflects the degree to which the sample exceeds its range, serving as an important derived feature. Simultaneously, it queries the historical database to extract the statistical characteristics of the dilution factor distribution from previous detections of similar pollutants, including the most... Commonly used dilution factors, their mean and variance, etc., are used as prior knowledge features. The input feature vector, overrange feature, and prior features are concatenated to form an extended feature vector. The model inference layer is the core computational unit for prediction. It uses a trained neural network or other machine learning model to perform forward propagation calculations on the extended feature vector. The model output layer provides the predicted probability distribution for all possible dilution factors, such as a 5% probability for 2x dilution, a 65% probability for 5x dilution, and a 28% probability for 10x dilution. The dilution factor with the highest probability (5x in this example) is selected as the recommended value output. Confidence assessment... The layer performs a reliability assessment on the probability distribution of the model output, extracts the maximum probability value and the second maximum probability value, and calculates the difference between the two as a confidence quantification index. When the difference is greater than a preset difference threshold (e.g., 0.3), it indicates that the model has a clear preference for the recommendation result, and is judged as a high-confidence output, which can be directly diluted using the recommended value. When the difference is less than the threshold, it indicates that the probabilities of multiple dilution factors are close, the model judgment is uncertain, and is judged as a low-confidence output (the threshold is set to 0.2 in this embodiment). At this time, the confidence judgment result is sent to the gradient scan trigger module to start the gradient scan process to obtain a more reliable dilution scheme.

[0028] The training process of the neural network includes the following steps: During initial deployment, at least one hundred standard samples of different concentrations are collected, covering the entire range of the extended measurement range. Each standard sample is actually diluted according to the dilution factor recommended by the mapping matrix and measured. The pollutant type label and the initial measured response signal intensity are recorded as input features, and the effective dilution factor used is used as the training label. The label represents the dilution factor category using one-hot encoding. The collected dataset is randomly divided into a training set and a validation set in an 8:2 ratio. The training set is used for model parameter learning, and the validation set is used to evaluate the model's generalization performance. Cross-entropy is used as the loss function, which measures the difference between the probability distribution predicted by the model and the true label distribution; the smaller the difference, the more accurate the model prediction. The Adam optimization algorithm is used for parameter updates. This algorithm combines momentum and adaptive learning rate adjustment, enabling rapid convergence to the optimal solution. The initial learning rate is set to 0.001, the training batch size is set to thirty-two samples, and the total number of training rounds is set to one hundred. After each training round, the model performance is evaluated on the validation set, and the validation set loss value is recorded. When the validation set loss no longer decreases for ten consecutive rounds, the model is considered to have learned sufficiently, and training is stopped early to prevent overfitting. After training is complete, the model parameters are saved to memory for subsequent inference.

[0029] In this embodiment, the intelligent prediction model integrates multi-dimensional information such as sample attributes, signal strength, and historical experience to achieve intelligent recommendation of dilution factors, significantly reducing the number of trial and error iterations in traditional methods and improving detection efficiency. The data preprocessing layer ensures the standardization and comparability of input data, the feature calculation layer enhances the model's predictive ability through derived features and prior knowledge, and the model inference layer provides probabilistic prediction results rather than a single deterministic value. The confidence assessment layer, by quantifying the reliability of the model's predictions, achieves an adaptive decision-making mechanism: at high confidence levels, the recommended value is directly used to quickly complete the dilution; at low confidence levels, gradient scanning is triggered for precise verification. This intelligent triage mechanism maximizes detection efficiency while ensuring accuracy, achieving an optimal balance between accuracy and efficiency.

[0030] When the sample to be tested contains multiple contaminants, the intelligent prediction model also includes a multi-contaminant synergistic analysis layer, specifically: The sample contains a list of multiple pollutant components and the initial response signal intensity of each pollutant. The recommended dilution factor for each pollutant is queried separately to form a candidate set of pollutant-dilution factors. The consistency of the recommended dilution factors for different pollutants in the candidate set is analyzed, and the coefficient of variation of all recommended dilution factors is calculated. When the coefficient of variation is lower than the preset consistency threshold, the median of the candidate dilution factors is selected as the unified dilution scheme; when the coefficient of variation is higher than the preset consistency threshold, it is determined to be a dilution requirement conflict, and the dominant pollutant with the highest response signal intensity and the minor pollutant with a response signal intensity significantly lower than the dominant pollutant are identified. For the dominant pollutant, the recommended dilution factor is used for the first dilution measurement. At the same time, the dilution of the minor pollutant is assessed to see if it will be lower than the detection limit. If the minor pollutant is lower than the detection limit after dilution, the original sample is used for a second independent dilution measurement using the recommended dilution factor of the minor pollutant. This forms a graded dilution strategy to obtain the accurate concentration of each pollutant.

[0031] A new collaborative analysis layer for multi-pollutant samples has been added to the intelligent prediction model. This layer works in conjunction with the existing data preprocessing, feature calculation, model inference, and confidence assessment layers. When multiple pollutants are detected in a sample, this layer automatically activates, receives spectral or multi-channel data, and generates a pollutant list containing name, category, and initial measurement information. It queries or predicts recommended dilution factors for each pollutant, forms a candidate set, and calculates the mean, standard deviation, and coefficient of variation to assess dilution consistency. When the coefficient of variation is below the threshold (0.2–0.4), it indicates that the dilution requirements are similar, and the median is selected as the uniform dilution factor for a single dilution to achieve efficient synchronous detection. If the coefficient of variation is above the threshold, it indicates that the dilution requirements differ significantly, and the system activates a tiered dilution strategy. First, the dominant and secondary pollutants are determined based on the initial signal intensity. The first dilution measurement is performed using the recommended factor for the dominant pollutant, and the measurability of the secondary pollutants at this factor is assessed. If they are still measurable, they are measured together; otherwise, a second independent dilution is initiated, and these secondary pollutants are measured separately according to their recommended factors, or a compromise factor is used for batch measurement when the dilution factors are close. Ultimately, the system integrates the results of multiple rounds of measurement to form a complete concentration dataset, and records the dilution factor, measurement conditions and quality control parameters of each pollutant, balancing detection efficiency and data quality, and realizing intelligent optimization of multi-pollutant detection.

[0032] In this embodiment, the multi-pollutant collaborative analysis layer, through intelligent evaluation and adaptive strategy selection, solves the long-standing dilemma of dilution scheme selection in multi-component sample detection. For samples with consistent dilution requirements, the adoption of a unified scheme ensures the measurement quality of all pollutants while maximizing detection efficiency, achieving the ideal state of simultaneous determination of multiple components with a single dilution. For samples with conflicting dilution requirements, the innovation of the tiered dilution strategy lies in its intelligent identification of dominant and secondary pollutants, and its targeted selection of the optimal dilution factor for pollutants at different concentration levels. This ensures that each pollutant is detected within its optimal measurement range, avoiding measurement inaccuracies or data loss caused by any pollutant due to an inappropriate dilution factor.

[0033] Furthermore, the gradient scanning process specifically includes: Multiple adjacent sub-intervals covering the concentration corresponding to the initial test response signal are obtained from the mapping matrix, and the corresponding recommended dilution factors are extracted to form a candidate dilution factor sequence. Each dilution factor in the sequence is executed sequentially on the test sample, and the response signal intensity of the diluted solution is measured after mixing. For each dilution level, its dilution factor and corresponding response signal intensity are recorded, and the ratio of the current level response signal to the previous level response signal is calculated. The ratio of the current level dilution factor to the previous level dilution factor is also calculated to obtain the linearity coefficient. It is determined whether the response signal of the current level falls into the target measurement interval, and at the same time, it is determined whether the linearity coefficient is close to the theoretical value. When the response signal falls into the target measurement interval and the deviation of the linearity coefficient from the theoretical value is less than the preset deviation threshold, the gradient scan is terminated and the current dilution level is locked. If no level that meets the conditions is found after traversing all candidate dilution factors, the dilution level whose response signal is closest to the center value of the target measurement interval is selected as the final selection.

[0034] Specifically, the gradient scanning process serves as a supplementary verification mechanism for the intelligent prediction model when it has low confidence. It employs a multi-level dilution verification strategy. Upon triggering the gradient scan, it queries and extracts multiple adjacent sub-intervals from the established mapping matrix that cover the concentration value corresponding to the initial test response signal. For example, if the estimated concentration corresponding to the initial test signal is 500 mg / L, it extracts the second sub-interval (316–1000 mg / L) containing this concentration, along with its adjacent first sub-interval (100–316 mg / L) and third sub-interval (1000–3160 mg / L). Recommendations are then extracted from these sub-intervals. The dilution factors are used to form a candidate dilution factor sequence arranged from smallest to largest. Then, the sample to be tested is subjected to each dilution factor in the sequence in turn. After each dilution, the sample is thoroughly mixed and the intensity of the response signal of the diluted solution is measured and recorded immediately. For each dilution level, a data recording table is established, which includes the dilution factor of that level and the corresponding measured response signal intensity value. Based on the data of two adjacent levels, the intensity ratio of the response signal of the current level to that of the previous level is calculated. At the same time, the fold ratio of the current level dilution factor to that of the previous level is also calculated. The linearity coefficient is obtained by dividing the intensity ratio by the fold ratio. For each dilution level, a dual judgment is performed: first, it is judged whether the response signal intensity of the current level has fallen into the previously determined target measurement interval (i.e., the high-precision measurement area); second, it is judged whether the calculated linearity coefficient is close to the theoretical value of 1.0. When the response signal falls into the target measurement interval and the deviation of the linearity coefficient from the theoretical value is less than the preset deviation threshold (e.g., 0.15), it means that a dilution level that is both in the optimal measurement interval and maintains good dilution linearity has been found. At this time, the gradient scan is immediately terminated and the current dilution level is locked as the final solution. If an ideal level that meets both conditions is not found after traversing all candidate dilution factor sequences, the alternative strategy is activated. Among all tested dilution levels, the dilution level whose response signal is closest to the center value of the target measurement interval is selected as the final choice.

[0035] In this embodiment, the gradient scanning mechanism provides a systematic verification and optimization scheme for low-confidence prediction. By performing multiple consecutive dilutions and monitoring the response values ​​in real time, the optimal dilution factor can be accurately found. Compared with traditional single-point verification or blind trial and error, this scheme establishes a complete sequence of dilution levels and tests each level sequentially, ensuring the comprehensiveness of the search. By introducing a linearity coefficient as a verification indicator, it is required not only that the response signal falls within the target range, but also that the dilution process maintains a good linear relationship. This eliminates possible matrix interference, adsorption loss, or uneven dilution anomalies, significantly improving the reliability of the finally selected dilution factor.

[0036] The gradient scan process also includes a dilution process quality control mechanism, specifically: Record the timestamp and ambient temperature of the dilution operation when performing each dilution level; perform a blank background test on the dilution before measuring the response signal of the dilution and calculate the background signal intensity; for each dilution level, perform a second repeated measurement after a preset time interval after completing the first measurement and calculate the relative deviation between the two measurements. When the background signal strength exceeds the preset background threshold, cross-contamination or container contamination is determined, the container cleaning procedure is triggered, and the dilution level is re-executed; when the relative deviation between two measurements exceeds the preset repeatability threshold, the diluent is determined to be unstable or unevenly mixed, the mixing time is extended, and a third measurement is performed. Monotonicity tests are performed on the response signal sequences of all dilution levels, and the theoretical agreement between the ratio of response signals of adjacent levels and the ratio of dilution factors is calculated. When the response signal of a certain dilution level is found to deviate from the monotonically decreasing trend or the theoretical agreement is abnormal, the level is marked as an abnormal data point, the data point is removed, and the measurement of that dilution level is re-executed.

[0037] A quality control mechanism for the dilution process is embedded in the gradient scanning architecture, running throughout the entire scanning process. Real-time monitoring and verification of operations and measurements at each dilution level are conducted. Timestamps and ambient temperatures are recorded during each dilution level execution, providing a basis for subsequent traceability and anomaly analysis. After obtaining the diluted solution, a blank background test is performed first. If the background signal exceeds the threshold, cross-contamination or container contamination is determined, and automatic cleaning, rinsing, and repeated verification are initiated. Only after passing the verification can formal measurements be performed. Repeat measurements are performed at several-minute intervals after the initial sample measurement. The relative deviation is calculated and compared with the repeatability threshold (3%–10%). If the deviation exceeds... If the mixing time is limited, a third measurement should be performed, and the average value should be taken or the result marked as a stability issue. After completing all dilution level measurements, the system performs a monotonicity test. Based on the principle that signal strength should be inversely proportional to the dilution factor, the decreasing trend of the response signal sequence and the theoretical compliance coefficient (ideal value is 1.0) are analyzed. If a significant deviation is found at a certain level (e.g., >1.2 or <0.8), it is marked as an abnormal data point and removed for retesting. Abnormalities may be caused by pipetting errors, matrix effects, contamination, or instrument malfunctions. The retested data should be retested to ensure compliance with the pattern. If the abnormality persists, the cause and corrective measures should be recorded.

[0038] The automated implementation of the quality control mechanism avoids over-reliance on operator experience and judgment. Even less experienced operators can complete high-quality measurements under the guidance and monitoring of the system, improving the operability of the method and the consistency of the results. For applications requiring large-scale sample testing or unattended automated testing, the embedded quality control mechanism provides reliable quality assurance, enabling the automated system to autonomously judge data quality, identify problems, and take corrective measures, achieving true intelligent quality management. This mechanism realizes automated quality assurance and anomaly self-correction from single operation to the entire process, ensuring that the data used to determine the optimal dilution level is authentic, reliable, traceable, and scientifically consistent.

[0039] The relative standard deviation of the diluent is calculated by performing parallel sample determinations. For dilutions with a distance difference from the segment boundary less than the first threshold, two adjacent dilution schemes are executed simultaneously, and the consistency of concentration in the overlapping area is mutually verified. The entire process was tested for different dilution factors, a dilution correction model was established, the correction coefficients for the dilution factors were updated, and the range boundary and accuracy distribution curves were updated.

[0040] Furthermore, the dilution correction model includes a standard substance testing layer, a deviation calculation layer, a model fitting layer, and a dynamic update layer, as referenced. Figure 3 ; The standard substance testing layer dilutes standard samples using the recommended dilution factors in the mapping matrix to obtain the dilution factor, theoretical concentration after dilution, and actual measured concentration. The deviation calculation layer uses the dilution factor, theoretical concentration after dilution, and actual measured concentration to infer the original concentration, compares the inferred original concentration with the standard value of the standard substance, and calculates the absolute and relative deviations. The model fitting layer fits the deviation-dilution factor relationship curve to obtain the correction equation, and calculates the correction coefficient corresponding to each dilution factor based on the correction equation. The dynamic update layer periodically repeats the standard substance testing and deviation calculation process, performs trend analysis on the newly obtained deviation data and historical deviation data, and triggers the model reconstruction process when the mean deviation value of multiple consecutive periods shows a unidirectional drift and the drift amplitude exceeds a preset threshold, using the latest data to refit the correction equation and update the correction coefficients.

[0041] Specifically, the standard substance testing layer serves as the data acquisition module for model building. It selects certified standard substances with accurately known concentrations as the test objects and performs actual dilution operations according to various dilution factors (such as 2x, 5x, 10x, 20x, etc.) recommended in different sub-intervals of the mapping matrix. The dilution operations are strictly performed according to the actual sample testing procedure. Instrument measurements are taken for each dilution factor to obtain the actual measured concentration value. Three sets of key data are recorded: dilution factor, theoretical concentration after dilution (standard value of the standard substance divided by the dilution factor), and actual measured concentration. The deviation calculation layer performs error analysis on the test data. The actual measured concentration at each dilution factor is multiplied by that dilution factor to obtain the back-calculated original concentration value. This back-calculated original concentration is then compared with the standard value (true value) of the standard substance to calculate the absolute deviation and relative deviation. The relative deviation comprehensively reflects the cumulative effect of various systematic errors in the dilution process, such as pipetting error, uneven mixing, container adsorption, and volatilization loss. The model fitting layer collects the relative deviation datasets corresponding to different dilution factors. With the dilution factor as the independent variable and the relative deviation as the dependent variable, the deviation-dilution factor relationship curve is fitted using the least squares method or other numerical fitting methods. The correction equation is obtained through fitting. Based on the fitted correction equation, for each dilution factor used in the mapping matrix, the expected deviation value under that factor is calculated by substituting it into the equation, and then the correction coefficient is obtained (correction coefficient = 1 / (1 + relative deviation)). This correction coefficient is used to multiply the diluted measured concentration by the dilution factor and then by the correction coefficient to obtain the corrected original concentration. The dynamic update layer implements continuous optimization and maintenance of the model. According to the preset time period (such as monthly or quarterly) or at key nodes such as instrument maintenance and reagent replacement, the standard material testing and deviation calculation process is repeatedly executed to obtain the deviation data of the new period. The newly obtained deviation data is compared with the historical deviation data of multiple periods to perform time series trend analysis, calculate the mean deviation of multiple consecutive periods, and determine whether the mean has a unidirectional continuous drift (continuous increase or continuous decrease). When the unidirectional drift of the mean deviation is detected to exceed the preset threshold (such as a drift of more than 20% relative to the initial value), it indicates that the system characteristics have changed significantly and the original correction model is no longer applicable. At this time, the model reconstruction process is automatically triggered, the outdated historical data is discarded, and the latest data of the most recent periods are used to re-execute the fitting process to generate new correction equations and correction coefficients, thus completing the model update and iteration.

[0042] In this embodiment, the dilution correction model establishes a mathematical model of the dilution process error through systematic standard substance testing, achieving quantitative compensation for systematic deviations and significantly improving the accuracy of back-calculating the original concentration after dilution. Compared with traditional methods that assume an ideal, error-free dilution process, this approach acknowledges and quantifies various unavoidable error sources in actual dilution operations, and uses correction coefficients for mathematical correction, making the measurement results closer to the true value.

[0043] Furthermore, the process of updating the range boundary and accuracy distribution curve specifically includes: when the dynamic update layer triggers the model reconstruction process, the range boundary update is initiated synchronously to re-obtain the upper limit of the linear response under the current state. The newly obtained upper limit of the linear response is compared with the original effective range threshold. If the difference exceeds a preset ratio, the effective range threshold is updated. The accuracy distribution test is re-performed on the concentration interval within the range to obtain new relative standard deviation and signal-to-noise ratio data. The updated accuracy distribution curve is plotted, and the new high-precision measurement area boundary value is identified. Based on the updated effective range threshold and high-precision measurement area boundary value, the recommended dilution factor for each sub-interval is recalculated, and it is determined whether the recommended dilution factor has changed. When the recommended dilution factor changes, the dilution factor field and the target concentration interval field after dilution in the corresponding sub-interval in the mapping matrix are updated, while keeping the concentration boundary value of the sub-interval unchanged. The updated effective range threshold, accuracy distribution curve parameters, and mapping matrix are synchronously written into the database, and the range-related derived feature calculation rules are updated in the feature calculation layer of the intelligent prediction model.

[0044] Specifically, the process of updating the range boundary and accuracy distribution curve is closely linked to the dynamic update layer of the dilution correction model, achieving coordinated optimization of the entire system parameters. When the dynamic update layer detects that the deviation drift exceeds the threshold and triggers the model reconstruction process, it determines that the basic characteristics of the instrument have changed, and simultaneously starts the range boundary update program. The instrument response characteristics are tested using a newly prepared standard solution, and the linearity of the response signal is monitored in real time by gradually increasing the concentration. The concentration point at which the response signal begins to deviate nonlinearly (e.g., the degree of deviation from linearity exceeds 5%) is identified and determined as the upper limit of the linear response under the current state, i.e., the new upper limit of the range. The newly measured upper limit of the linear response is compared with the original effective range threshold, and the relative percentage difference between the two is calculated. If the difference exceeds the preset percentage threshold (e.g., 10%), it is determined that the range has changed significantly, and the effective range threshold needs to be updated to the newly measured value. The update time and reason are recorded. If the difference is within the threshold range, the original effective range threshold remains unchanged. After determining the new measurement range boundaries, a comprehensive accuracy distribution test was conducted on the concentration intervals within the entire range. Test concentration points were selected at equal intervals within the new range, and multiple parallel measurements were performed at each point to collect response signal data. New relative standard deviation (RSD) and signal-to-noise ratio (SNR) datasets were statistically calculated, and an updated accuracy distribution curve was plotted based on the new data. The updated accuracy distribution curve was analyzed to re-identify concentration intervals that simultaneously meet both RSD and SNR threshold requirements, determining the new high-precision measurement area boundary value. This boundary value may have shifted or changed in width relative to the original boundary. Subsequently, based on the updated effective range... The system recalculates the recommended dilution factor for each sub-interval using the threshold (first dividing point) and the new high-precision measurement zone boundary value, combined with the unchanged target extended range upper limit (second dividing point). It then compares the newly calculated value with the original stored value in the mapping matrix, checking for changes in the recommended dilution factor for each sub-interval. If a change is detected, the system updates the key fields of the corresponding sub-interval in the mapping matrix, including updating the dilution factor field to the new calculated value and updating the target concentration range field after dilution to the new high-precision measurement zone range. However, it maintains the sub-interval's own concentration boundary values ​​(lower and upper limits) to ensure consistent coverage of the extended range. The system synchronously writes all updated key parameters, including the new effective range threshold value, characteristic parameters of the accuracy distribution curve (such as the boundary coordinates of the high-precision measurement zone, RSD and SNR values ​​of each concentration point, etc.), and the updated complete mapping matrix data, into the backend database via database transaction operations, ensuring data consistency and traceability.Simultaneously, the system accesses the feature calculation layer configuration of the intelligent prediction model and updates the derived feature calculation rules related to the measurement range. For example, the calculation of the over-range feature requires the use of a new effective range threshold, and the calculation of the center value of the target measurement area requires the use of a new high-precision measurement area boundary. This ensures that the intelligent prediction model uses the latest system parameters for feature engineering and inference calculations, and maintains parameter consistency among all modules of the entire method.

[0045] In this embodiment, the dynamic update mechanism of the range boundary and accuracy distribution curves enables the entire system to adaptively evolve, allowing the method to proactively track and adapt to long-term changes in instrument characteristics. During long-term use, factors such as sensor aging, light source attenuation, and changes in detector sensitivity can cause drift in the range boundary and accuracy distribution. Traditional methods with fixed parameters will gradually become inaccurate due to system drift. Compared to traditional methods that require periodic manual recalibration, this solution's automated updates significantly reduce maintenance workload, improve the system's intelligence and practical value, and are particularly suitable for online monitoring scenarios requiring long-term continuous operation, ensuring the continuous reliability of the extended range function.

[0046] This invention provides an intelligent range extension and optimization method for instrument measurement based on sample dilution. Through the collaboration of intelligent sensing and machine learning models, it achieves dynamic range extension and accuracy optimization of the instrument. The method uses standard solutions for linear regression analysis to determine the effective range threshold, combines relative standard deviation and signal-to-noise ratio to generate an accuracy distribution curve, and establishes a mapping matrix between concentration range, dilution factor, and target measurement range. When the sample response exceeds the range, the system automatically recommends a dilution factor based on the pollutant type and initial measurement signal using an intelligent prediction model. If the prediction confidence is insufficient, a gradient scanning mechanism is activated to monitor response changes in real time and lock in the optimal dilution level. A dilution correction model is used to perform deviation analysis and dynamic correction on the detection results of standard samples, achieving continuous optimization of the dilution factor. The system can also automatically adjust the range boundary and accuracy distribution curve based on model update results to ensure measurement linearity and stability at different dilution factors. See details below. Figure 4 Overall, this method enables intelligent adaptive expansion of the instrument's measurement range, significantly improving the detection accuracy and stability of high-concentration samples, and is suitable for various high-precision analytical scenarios such as environmental monitoring and biochemical detection.

[0047] Example 2: This invention provides an intelligent range extension optimization method for instruments based on sample dilution. This method achieves intelligent range extension by establishing a precision distribution curve and mapping matrix, combined with an intelligent prediction model and gradient scanning mechanism. Building upon the basic technical solution, it also includes a matrix effect identification and adaptive compensation module, further improving the accuracy, adaptability, and reliability of the method in the detection of complex real-world samples. The newly added technical solutions are described in detail below with reference to specific embodiments.

[0048] The matrix effect recognition and adaptive compensation module specifically includes: After preliminary concentration determination of the actual sample using a standard curve, a pure standard solution with a close concentration is prepared, or a standard substance of known concentration is added to the sample using the standard addition method. The matrix factor is obtained by measuring the response signal of the pure standard solution and the response signal of the actual sample, or the response increment before and after the standard addition. When the matrix factor deviates from 1.0 by more than the preset matrix threshold, a significant matrix effect is determined. A matrix type feature library is established. Matrix feature vectors are extracted based on sample source information, pollutant type, and response signal characteristic parameters. Samples are then classified into predefined matrix types using a clustering algorithm. For each matrix type, a dedicated calibration curve is established using standard spiked samples of the corresponding matrix to obtain the response slope and intercept for that matrix type. When diluting and testing actual samples, the corresponding dedicated calibration curve is selected according to the identified matrix type for concentration calculation. For samples with unidentified matrix types, the standard addition method is used, in which a standard substance of known concentration is added to the diluted sample for secondary measurement, and a temporary calibration relationship is established by the difference in response values ​​before and after spiking.

[0049] In this embodiment, the determination of lead is used as an example to illustrate the matrix effect identification and compensation process. First, an original sample of industrial wastewater was tested. The initial response signal indicated a lead concentration of approximately 500 mg / L, far exceeding the instrument's effective range limit of 100 mg / L. Based on the intelligent prediction model's recommendation of a five-fold dilution scheme, the sample was diluted to the target measurement range before measurement, yielding a diluted concentration of 95 mg / L. The original sample concentration was calculated to be 475 mg / L. To verify the influence of the matrix effect, a pure lead standard solution with a concentration of 95 mg / L was prepared and measured under the same conditions. The obtained response signal was significantly higher than that of the diluted wastewater solution. The matrix factor was calculated as the wastewater dilution response value divided by the pure standard solution response value, yielding 0.85. This value significantly deviates from the theoretical value of 1.0. If the sample concentration exceeds the preset matrix threshold of 0.1, the system determines that there is a significant matrix inhibition effect. Subsequently, the system extracts the sample's characteristic information, including the sample source being electroplating wastewater, the pollutant type being heavy metal, and the waveform characteristic parameters of the response signal, forming a matrix feature vector. This vector is then categorized into the established electroplating wastewater matrix type using a clustering algorithm. For this matrix type, a pre-established calibration curve specific to the electroplating wastewater matrix is ​​invoked. This curve is established using standard spiked samples containing similar matrix components, and its response slope and intercept are significantly different from the calibration curve of the pure standard solution. The sample concentration is recalculated using the specific calibration curve, yielding a corrected lead concentration of 558 mg / L. This result is highly consistent with the concentration value obtained independently through the standard addition method, demonstrating the effectiveness of matrix compensation.

[0050] By applying a matrix effect identification and adaptive compensation module, the interference of complex matrices in real samples on measurement results was successfully resolved. This embodiment quantitatively calculates the matrix factor to clearly identify the existence and intensity of matrix effects, avoiding the misinterpretation of matrix interference as a true characteristic of sample concentration. The application of a matrix type feature library and intelligent classification algorithm can quickly identify the matrix category of a sample and automatically select the corresponding dedicated calibration curve, eliminating the need for operator experience-based judgment and improving the automation level and consistency of results. Especially for samples with unknown matrix types, the standard addition method, as a universal compensation scheme, establishes a calibration relationship by adding a spike to the sample's own matrix, fundamentally eliminating the influence of matrix effects and ensuring the accuracy and reliability of measurement results.

[0051] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent expansion and optimization of instrument range based on sample dilution, characterized in that, include: Linear regression analysis of the instrument was performed using standard solutions, and the effective range threshold was obtained through intelligent sensors. The process involves statistically analyzing the relative standard deviation and signal-to-noise ratio (SNR) within the concentration range to obtain an accuracy distribution curve. The extended measurement range is divided into multiple sub-ranges, and a mapping matrix is ​​established between the concentration range, dilution factor, and target measurement range. Specifically, obtaining the accuracy distribution curve includes: selecting multiple concentration points at equal intervals within the effective measurement range threshold; preparing standard solutions for each concentration point and performing multiple parallel measurements; collecting the raw response signal data output by the intelligent sensor in the instrument; statistically analyzing the multiple measurement data for each concentration point to obtain the average value and standard deviation, thus obtaining the relative standard deviation; simultaneously analyzing the signal waveforms acquired in each measurement to identify the peak amplitude of the effective signal and the baseline noise amplitude of the measurement gap, thus obtaining the SNR; and identifying concentration ranges with a relative standard deviation below a first threshold and an SNR above a second threshold, marking them as high-precision measurement areas. The sample to be tested is tested in its original state. If the response value is within the range, it is directly detected. If it exceeds the range, the dilution parameters are matched according to the type of contaminant in the sample, and the intelligent prediction model is activated to predict and recommend a dilution factor based on the initial signal intensity and sample properties. When the prediction confidence is lower than a set threshold, a gradient scan is triggered, and the sample is continuously diluted while the response values ​​at each level are monitored in real time. When the response value enters the target measurement range and the rate of change is stable, the dilution level is locked. The intelligent prediction model includes a data preprocessing layer, a feature calculation layer, a model inference layer, and a confidence assessment layer. The data preprocessing layer receives the pollutant type and initial response signal intensity of the current sample to be tested. It performs one-heat encoding on the pollutant type to convert it into a pollutant vector, and normalizes the initial response signal intensity to map the value to a standard range. The pollutant vector and signal intensity are then combined into an input feature vector. The feature calculation layer calculates the ratio of the initial signal intensity to the effective range threshold from the input feature vector as an over-range feature, queries the dilution factor distribution of similar pollutants as a priori feature, and combines the input feature vector, the over-range feature, and the priori feature into an extended feature vector. The model inference layer performs forward calculation on the extended feature vector, outputting the predicted dilution factor and its corresponding probability distribution, and selects the dilution factor with the highest probability as the recommended value. The confidence assessment layer calculates a confidence index based on the predicted probability distribution. When the difference between the highest probability value and the second highest probability value is greater than a preset difference threshold, it is determined to be high confidence; when the difference is less than the preset difference threshold, it is determined to be low confidence. The confidence assessment result is then output to the gradient scan trigger module. The gradient scanning process specifically includes: Multiple adjacent sub-intervals covering the concentration corresponding to the initial test response signal are obtained from the mapping matrix, and the corresponding recommended dilution factors are extracted to form a candidate dilution factor sequence. Each dilution factor in the sequence is executed sequentially on the test sample, and the response signal intensity of the diluted solution is measured after mixing. For each dilution level, the dilution factor and the corresponding response signal intensity are recorded, and the ratio of the current level response signal to the previous level response signal is calculated. The ratio of the current level dilution factor to the previous level dilution factor is also calculated to obtain the linearity coefficient. It is determined whether the response signal of the current level falls into the target measurement interval, and at the same time, it is determined whether the linearity coefficient is the theoretical value. When the response signal falls into the target measurement interval and the deviation of the linearity coefficient from the theoretical value is less than the preset deviation threshold, the gradient scan is terminated and the current dilution level is locked. If no level that meets the conditions is found after traversing all candidate dilution factors, the dilution level whose response signal is closest to the center value of the target measurement interval is selected as the final selection. The relative standard deviation of the diluent is calculated by performing parallel sample determinations. For dilutions with a distance difference from the segment boundary less than the first threshold, the two adjacent dilution factor schemes are executed, and the consistency of concentration in the overlapping area is mutually verified. Full-process testing was conducted for different dilution ratios, a dilution correction model was established, the correction coefficients for the dilution ratios were updated, and the range boundaries and accuracy distribution curves were updated. The dilution correction model includes a standard substance testing layer, a deviation calculation layer, a model fitting layer, and a dynamic update layer. The standard substance testing layer dilutes standard samples using the recommended dilution factors in the mapping matrix to obtain the dilution factor, theoretical concentration after dilution, and actual measured concentration. The deviation calculation layer uses the dilution factor, theoretical concentration after dilution, and actual measured concentration to infer the original concentration, compares the inferred original concentration with the standard value of the standard substance, and calculates the absolute and relative deviations. The model fitting layer fits the deviation-dilution factor relationship curve to obtain the correction equation, and calculates the correction coefficient corresponding to each dilution factor based on the correction equation. The dynamic update layer periodically repeats the standard substance testing and deviation calculation process, performs trend analysis on the newly obtained deviation data and historical deviation data, and triggers the model reconstruction process when the mean deviation value of multiple consecutive periods shows a unidirectional drift and the drift amplitude exceeds a preset threshold, using the latest data to refit the correction equation and update the correction coefficients.

2. The intelligent expansion and optimization method for instrument range based on sample dilution according to claim 1, characterized in that: The process of establishing the mapping matrix specifically includes: The upper limit of the effective measurement range threshold is set as the first boundary point, and the upper limit of the target extended measurement range is set as the second boundary point, thus obtaining the concentration span between the two. A segmentation strategy is determined based on the order of magnitude of the concentration span, dividing the area from the first boundary point to the second boundary point into sub-intervals. For each sub-interval, the lower limit, upper limit, and concentration boundary value of the high-precision measurement area are obtained, thus determining the dilution factor range required for the concentration range of the sub-interval to fall into the high-precision measurement area after dilution. For two adjacent sub-intervals, the difference in concentration boundary values ​​is calculated, and a concentration overlap region is set at the boundary between the two sub-intervals. The concentration range of the overlap region is simultaneously detected using the recommended dilution factors of the two adjacent sub-intervals.

3. The intelligent expansion and optimization method for instrument range based on sample dilution according to claim 1, characterized in that: The process of updating the measurement range boundary and accuracy distribution curve specifically includes: when the dynamic update layer triggers the model reconstruction process, the measurement range boundary update is started synchronously to obtain the upper limit of the linear response under the current state. The newly obtained upper limit of the linear response is compared with the original effective measurement range threshold. If the difference exceeds the preset ratio, the effective measurement range threshold is updated. The accuracy distribution test is re-performed on the concentration interval within the measurement range to obtain new relative standard deviation and signal-to-noise ratio data. The updated accuracy distribution curve is plotted, and the new high-precision measurement area boundary value is identified. Based on the updated effective measurement range threshold and high-precision measurement area boundary value, the recommended dilution factor for each sub-interval is recalculated, and it is determined whether the recommended dilution factor has changed. When the recommended dilution factor changes, the dilution factor field and the target concentration interval field after dilution in the corresponding sub-interval in the mapping matrix are updated to keep the concentration boundary value of the sub-interval unchanged. The updated effective measurement range threshold, accuracy distribution curve parameters, and mapping matrix are synchronously written into the database, and the range-related derived feature calculation rules are updated in the feature calculation layer of the intelligent prediction model.

Images