Liquid analysis instrument calibration error correction method based on multi-parameter linkage

Abstract

The present application relates to the technical field of analytical instrument error compensation, and discloses a liquid analytical instrument calibration error correction method based on multi-parameter linkage, comprising the following steps: firstly, synchronously collecting original measurement signals of a liquid analytical instrument at a current time, environmental auxiliary parameters reflecting liquid cross interference, and instrument hardware state parameters reflecting aging of underlying hardware of the liquid analytical instrument; then, extracting time sequence statistical features from the original measurement signals, and splicing the time sequence statistical features, the environmental auxiliary parameters, and the instrument hardware state parameters. By synchronously collecting the original measurement signals, the environmental auxiliary parameters, and the instrument hardware state parameters of the liquid analytical instrument, the comprehensiveness and accuracy of multi-parameter linkage state representation are improved from the source, and through a model self-adaptive updating mechanism, the instrument can realize long-term dynamic error self-adaptive compensation in the whole life cycle by relying on only a small amount of field calibration data.

Description

Technical Field

[0001] This invention relates to the field of analytical instrument error compensation technology, specifically to a method for correcting calibration errors in liquid analytical instruments based on multi-parameter linkage. Background Technology

[0002] Liquid analysis instruments (such as pH meters, conductivity meters, and multi-parameter water quality spectrometers) are widely used in environmental monitoring, industrial process control, and laboratory biochemical analysis. In actual measurement, the measured liquid system is often a complex physicochemical mixture, and its measurement accuracy is easily affected by the cross-interference of various matrix effects such as liquid temperature, pressure, turbidity, flow rate, and even the concentration of coexisting ions. Therefore, high-precision calibration and error correction of liquid analysis instruments are the core technical links to ensure the reliability of monitoring data.

[0003] Most existing error correction methods for liquid analysis instruments rely on static compensation using single or dual parameters. For example, they use built-in temperature sensors and traditional physicochemical empirical formulas (such as the Nernst equation) for basic linear temperature conversion. However, real industrial or natural water environments contain highly complex nonlinear multi-parameter coupled interferences. Relying solely on basic mechanism models cannot completely cover and eliminate the nonlinear residuals caused by these cross-interferences. To address this issue, some existing technologies have begun to introduce purely data-driven machine learning algorithms (such as artificial neural networks) to attempt to establish a black-box mapping model between multidimensional environmental parameters and measurement errors. Although such pure black-box models can achieve good fitting results under standard laboratory conditions, they lack physical interpretability. Once the actual operating conditions exceed the distribution range of the factory training dataset, the model is prone to divergence or outputs incorrect calibration values ​​that violate physical common sense, resulting in poor generalization ability and robustness under complex and variable operating conditions.

[0004] Meanwhile, during long-term continuous operation, the sensing probes of liquid analyzers inevitably undergo irreversible physical aging phenomena such as electrode wear, surface scaling, impedance changes, and internal light source attenuation. This time-varying drift at the hardware level can cause a fundamental shift in the instrument's original response curve. Existing multi-parameter calibration models are often fixed at the factory, and their input characteristics are limited to external environmental interference with the liquid, completely stripping away the real-time perception of the instrument's underlying hardware status. As the instrument's service life increases, the fixed calibration model will gradually become ineffective. In order to maintain measurement accuracy, users need to frequently return the instrument to the factory or use a large number of standard solutions of different concentrations for global recalibration on-site. This not only greatly increases the instrument's operation and maintenance costs and workload, but also fails to truly achieve long-term, dynamic error adaptive compensation throughout the sensor's entire life cycle. Therefore, this invention designs a liquid analyzer calibration error correction method based on multi-parameter linkage to address the problems mentioned above. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for correcting calibration errors in liquid analyzers based on multi-parameter linkage. This method solves the problems of existing liquid analyzers struggling to handle multi-parameter nonlinear cross-interference correction in complex fluid matrices, as well as the failure of solidified calibration models due to physical aging of sensors over long-term operation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for correcting calibration errors in liquid analysis instruments based on multi-parameter linkage, comprising the following steps:

[0007] S1. First, synchronously acquire the original measurement signal of the liquid analyzer at the current moment, the environmental auxiliary parameters reflecting liquid cross-interference, and the instrument hardware status parameters reflecting the aging of the underlying hardware of the liquid analyzer.

[0008] S2. Subsequently, time-series statistical features are extracted from the original measurement signal, and the time-series statistical features, the environmental auxiliary parameters, and the instrument hardware status parameters are concatenated to construct a fusion feature vector for characterizing the multi-parameter linkage state.

[0009] S3. At this time, based on the preset physicochemical mechanism model, according to the original measurement signal and the environmental auxiliary parameters, the basic mechanism correction value for eliminating the explicit linear error is calculated, the fused feature vector is input into the pre-trained data-driven model, and the nonlinear residual prediction value caused by the cross-coupling of multi-dimensional parameters is output.

[0010] S4. Finally, the basic mechanism correction value and the nonlinear residual prediction value are fused and reconstructed to obtain and output the final calibration measurement result of the liquid analysis instrument.

[0011] Preferably, the original measurement signal includes the sensor's uncorrected output value; the environmental auxiliary parameters include at least one of the temperature, pressure, and turbidity of the liquid being measured; and the instrument hardware status parameters include at least one of the sensor impedance, light source emission current, and reference voltage of the liquid analysis instrument.

[0012] Preferably, the process of extracting time-series statistical features from the original measurement signal includes:

[0013] Create a sliding window with a set time duration;

[0014] Calculate the mean, variance, and rate of change of all the original measurement signals within the sliding window, and use the mean, variance, and rate of change as the time series statistical features.

[0015] Preferably, the physicochemical mechanism model is a pre-set theoretical compensation function;

[0016] The process of calculating the basic mechanism correction value for eliminating explicit linearity error based on the original measurement signal and the environmental auxiliary parameters includes:

[0017] Substituting the original measurement signal and the environmental auxiliary parameters into the theoretical compensation function, the basic mechanism correction value calculated and output by the theoretical compensation function is obtained.

[0018] Preferably, the pre-trained data-driven model is obtained through an offline pre-training phase; in the offline pre-training phase:

[0019] Obtain the true concentration baseline value of the target liquid;

[0020] The deviation between the true concentration benchmark value and the basic mechanism correction value is defined as the target nonlinear residual for the pre-trained data-driven model to learn.

[0021] Preferably, the process of fusing and reconstructing the basic mechanism correction value and the nonlinear residual prediction value to obtain and output the final calibration measurement result of the liquid analysis instrument includes:

[0022] The basic mechanism correction value and the nonlinear residual prediction value are summed to obtain the initial reconstruction measurement value;

[0023] The initial reconstructed measurement values ​​are filtered using a dynamic smoothing filtering algorithm to eliminate transient distortions, resulting in the final calibration measurement result, which is then output and displayed.

[0024] Preferably, the dynamic smoothing filtering algorithm is an exponentially weighted moving average filtering algorithm;

[0025] The process of filtering the initial reconstructed measurement values ​​using a dynamic smoothing filtering algorithm includes:

[0026] The final calibration measurement result at the current moment is obtained by weighted summation of the initial reconstructed measurement value at the current moment, the final calibration measurement result at the previous moment, and a preset smoothing factor.

[0027] Preferably, after outputting the final calibration measurement results of the liquid analyzer, the process further includes an adaptive update of the pre-trained data-driven model based on incremental learning, wherein the adaptive update process includes:

[0028] At the calibration time, obtain the reference nominal value of the standard solution;

[0029] Based on the original measurement signal and the environmental auxiliary parameters at the calibration time, calculate the corresponding basic mechanism correction value;

[0030] The difference between the reference nominal value and the corresponding calculated basic mechanism correction value is taken as the actual calibration residual.

[0031] Preferably, the adaptive update process further includes:

[0032] Obtain the nonlinear residual prediction value output by the pre-trained data-driven model at the calibration time;

[0033] Calculate the absolute error between the actual calibration residual and the predicted nonlinear residual;

[0034] Determine whether the absolute error is greater than a preset trigger threshold. If so, trigger the parameter fine-tuning procedure for the pre-trained data-driven model.

[0035] Preferably, the parameter fine-tuning procedure includes:

[0036] Based on the difference between the actual calibration residual and the predicted value of the nonlinear residual, a prediction loss function is constructed;

[0037] The gradient descent algorithm is used to iteratively fine-tune and update the model parameters of the pre-trained data-driven model using the prediction loss function;

[0038] After the parameters are updated, the original model parameters in the pre-trained data-driven model are replaced with the updated model parameters, and the process of synchronously acquiring the original measurement signals, environmental auxiliary parameters, and instrument hardware status parameters continues to be executed in a loop.

[0039] This invention provides a method for correcting calibration errors in liquid analysis instruments based on multi-parameter linkage. It has the following beneficial effects:

[0040] 1. This invention overcomes the limitations of traditional calibration methods that rely solely on external environmental parameters for compensation by simultaneously acquiring the original measurement signals, environmental auxiliary parameters, and instrument hardware status parameters of the liquid analysis instrument, thereby improving the comprehensiveness and accuracy of multi-parameter linkage status characterization from the source.

[0041] 2. This invention extracts time-series statistical features such as mean, variance, and rate of change from the original measurement signal and constructs a fusion feature vector, so that the data input to the correction model fully contains the dynamic time-series information of fluid changes, which significantly enhances the instrument's anti-interference capability under dynamic fluid conditions.

[0042] 3. This invention constructs a gray-box model architecture driven by both mechanism and data by calculating the basic mechanism correction value based on the physicochemical mechanism model and fusing and reconstructing the nonlinear residual prediction value output by the data-driven model. This architecture not only utilizes the mechanism model to eliminate explicit linear errors and anchor a safe physical benchmark, avoiding the risk of output divergence of the pure black-box model under extreme conditions, but also leverages the advantages of the data-driven model in handling multi-parameter nonlinear cross-coupling interference, effectively improving the generalization ability and robustness of the calibration model.

[0043] 4. This invention utilizes a dynamic smoothing filtering algorithm to filter the initial reconstructed measurement values ​​after fusion, effectively suppressing transient signal distortion caused by microbubble interference, sudden changes in flow velocity, or high-frequency random electromagnetic noise in the fluid pipeline. While accurately preserving the long-term trend of the measured liquid concentration, it ensures the stability and reliability of the instrument's final output value.

[0044] 5. This invention introduces an incremental learning-based model adaptive update mechanism to calculate the error between the actual calibration residual and the predicted residual at regular calibration times, and uses the gradient descent algorithm to iteratively fine-tune the parameters of the data-driven model, enabling the instrument to achieve long-term dynamic error adaptive compensation throughout its entire life cycle with only a small amount of on-site calibration data. Attached Figure Description

[0045] Figure 1 This is one of the schematic diagrams of the method flow of the present invention;

[0046] Figure 2 This is a second schematic diagram of the method flow of the present invention;

[0047] Figure 3 This is the third schematic diagram of the method flow of the present invention;

[0048] Figure 4 This is the fourth schematic diagram of the method flow of the present invention;

[0049] Figure 5 This is the fifth schematic diagram of the method flow of the present invention;

[0050] Figure 6 This is the sixth schematic diagram of the method flow of the present invention. Detailed Implementation

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see the appendix Figure 1 -Appendix Figure 6 This invention provides a method for correcting calibration errors in liquid analysis instruments based on multi-parameter linkage, comprising the following steps:

[0053] S1. First, synchronously acquire the original measurement signal of the liquid analyzer at the current moment, the environmental auxiliary parameters reflecting liquid cross-interference, and the instrument hardware status parameters reflecting the aging of the underlying hardware of the liquid analyzer. The original measurement signal includes the uncorrected output value of the sensor; the environmental auxiliary parameters include at least one of the temperature, pressure, and turbidity of the liquid being measured; the instrument hardware status parameters include at least one of the sensor impedance, light source emission current, and reference voltage of the liquid analyzer.

[0054] Specifically, the measurement error of a liquid system is not only related to transient environmental parameters, but also affected by fluid dynamic fluctuations and the underlying hardware status of the instrument. At the discrete current sampling time k, the system performs synchronous acquisition of multi-source data:

[0055] First, acquire the raw measurement signal. This signal is the sensor’s raw physical output value without any correction (e.g., weak electrode voltage or raw absorbance without temperature compensation).

[0056] Obtain environmental auxiliary parameters, defined as vectors. This parameter includes at least one of the temperature, pressure, and turbidity of the measured liquid, and is used to characterize the cross-interference of the external environment on the measurement. Simultaneously, it acquires the instrument hardware status parameters, defined as a vector. The parameter includes at least one of the sensor impedance, light source emission current, and reference voltage of the liquid analyzer. The principle of introducing hardware state parameters is that it can separate the error caused by the instrument's own hardware drift from the interference of the external environment from the source.

[0057] S2. Subsequently, time-series statistical features are extracted from the original measurement signals, and the time-series statistical features, environmental auxiliary parameters, and instrument hardware status parameters are concatenated to construct a fusion feature vector to characterize the multi-parameter linkage state. The process of extracting time-series statistical features from the original measurement signals includes: establishing a sliding window with a set time length; calculating the mean, variance, and rate of change of all original measurement signals within the sliding window, and using the mean, variance, and rate of change as time-series statistical features.

[0058] Specifically, to eliminate sensor response hysteresis caused by the dynamic flow of fluid in the pipeline, this invention does not directly use transient data, but instead extracts time-series statistical features. The specific process is as follows: Establish a time-series statistical feature... sliding window ; Calculate all raw measurement signals within the sliding window mean ,variance and rate of change The calculation formulas are as follows:

[0059] ;

[0060] ;

[0061] ;

[0062] Finally, the mean ,variance rate of change With the current time-ahead environment auxiliary parameter vector and instrument hardware status parameter vector Concatenate the features to construct a fused feature vector. :

[0063] ;

[0064] This fused feature vector comprehensively characterizes the multi-parameter linkage state of the current measurement system.

[0065] S3. At this point, based on the preset physicochemical mechanism model, the basic mechanism correction value for eliminating explicit linear errors is calculated according to the original measurement signal and environmental auxiliary parameters. The fused feature vector is input into the pre-trained data-driven model, and the predicted value of the nonlinear residual caused by the cross-coupling of multi-dimensional parameters is output. The physicochemical mechanism model is a preset theoretical compensation function. The process of calculating the basic mechanism correction value for eliminating explicit linear errors according to the original measurement signal and environmental auxiliary parameters includes: substituting the original measurement signal and environmental auxiliary parameters into the theoretical compensation function to obtain the basic mechanism correction value calculated by the theoretical compensation function. The pre-trained data-driven model is obtained through offline pre-training. In the offline pre-training stage: the true concentration benchmark value of the target liquid is obtained; the deviation between the true concentration benchmark value and the basic mechanism correction value is defined as the target nonlinear residual for the pre-trained data-driven model to learn.

[0066] Specifically, because purely data-driven models are prone to output divergence when encountering abnormal operating conditions that exceed the training set distribution, this invention introduces a physicochemical mechanism model as a theoretical anchoring benchmark. The physicochemical mechanism model is a pre-defined theoretical compensation function. (For example, the classic Nernst equation or empirical temperature conversion formula), based on the original measurement signal and environmental auxiliary parameters Substituting this into the theoretical compensation function, we obtain the basic mechanism correction value. :

[0067] ;

[0068] Taking the most common linear compensation for temperature and pressure as an example, its function can be expanded as follows:

[0069] ;

[0070] in, and These are the current temperature and pressure values ​​extracted from environmental auxiliary parameters, respectively. and The set standard reference constant; and As the corresponding mechanism compensation coefficient constant, the implementation of this step effectively eliminates the explicit, low-dimensional linear error in the system, providing a physically meaningful safety benchmark for subsequent processing;

[0071] In real-world liquid systems, complex nonlinear cross-interferences arise from the coexistence of multiple ions, turbidity coupling, and hardware aging. These interferences cannot be fully explained by the fundamental formulas. Therefore, this invention employs a pre-trained data-driven model. (Such as neural network models or support vector regression models) implicitly map this nonlinear feature. In the offline pre-training stage: first, obtain the true concentration baseline value of the target liquid. The actual concentration benchmark value will then be used. Correction value of basic mechanism The deviation between them is defined as the target nonlinear residual learned by the pre-trained data-driven model. :

[0072] ;

[0073] During the online operation phase: the fused feature vector constructed above will be... Input to a parameter set In the pre-trained data-driven model, the output is the nonlinear residual prediction value caused by the cross-coupling of multidimensional parameters at the current time step. :

[0074] .

[0075] S4. Finally, the basic mechanism correction value and the nonlinear residual prediction value are fused and reconstructed to obtain and output the final calibration measurement result of the liquid analyzer. The process of fusing and reconstructing the basic mechanism correction value and the nonlinear residual prediction value to obtain and output the final calibration measurement result of the liquid analyzer includes: summing the basic mechanism correction value and the nonlinear residual prediction value to obtain the initial reconstructed measurement value; filtering the initial reconstructed measurement value using a dynamic smoothing filter algorithm to eliminate transient distortion and obtain the final calibration measurement result, which is then output and displayed. The dynamic smoothing filter algorithm is an exponential weighted moving average filter algorithm. The process of filtering the initial reconstructed measurement value using the dynamic smoothing filter algorithm includes: weighting and summing the initial reconstructed measurement value at the current time, the final calibration measurement result at the previous time, and a preset smoothing factor to obtain the final calibration measurement result at the current time. After outputting the final calibration measurement result of the liquid analyzer, the process also includes an adaptive update of the pre-trained data-driven model based on incremental learning. The adaptive update process includes... At the calibration time, the reference nominal value of the standard solution is obtained; based on the original measurement signal and environmental auxiliary parameters at the calibration time, the corresponding basic mechanism correction value is calculated; the difference between the reference nominal value and the corresponding calculated basic mechanism correction value is used as the actual calibration residual. The adaptive update process also includes: obtaining the nonlinear residual prediction value output by the pre-trained data-driven model at the calibration time; calculating the absolute error between the actual calibration residual and the nonlinear residual prediction value; determining whether the absolute error is greater than a preset trigger threshold. If so, the parameter fine-tuning procedure for the pre-trained data-driven model is triggered. The parameter fine-tuning procedure includes: constructing a prediction loss function based on the difference between the actual calibration residual and the nonlinear residual prediction value; using the gradient descent algorithm, iteratively fine-tuning and updating the model parameters of the pre-trained data-driven model using the prediction loss function; after the parameter update is completed, replacing the original model parameters in the pre-trained data-driven model with the updated model parameters, and continuing to loop the process of synchronously acquiring the original measurement signal, environmental auxiliary parameters, and instrument hardware status parameters.

[0076] Specifically, after obtaining the theoretical baseline and nonlinear residuals, both need to be reconstructed to recover the true liquid concentration information. First, the basic mechanism correction value... With nonlinear residual prediction values The values ​​are summed to obtain the initial reconstructed measurements. :

[0077] ;

[0078] Because fluid sampling is often accompanied by disturbances such as microbubbles or localized concentration inhomogeneities, directly outputting the initial reconstructed measurement value may cause the instrument display to jump. Therefore, this embodiment uses the exponentially weighted moving average (EMA) filtering algorithm of the dynamic smoothing filtering algorithm to smooth the initial reconstructed measurement value. The final calibration measurement result at the previous moment and the preset smoothing factor constant Perform a weighted summation to calculate the final calibration measurement result at the current time. :

[0079] ;

[0080] In the formula, This filtering process, while preserving the long-term trend of fluid concentration changes, filters out transient high-frequency peaks, and then... Output display;

[0081] As instruments are deployed over long periods, sensors inevitably experience physical wear and irreversible aging (instantaneous drift), causing the fixed model parameters to gradually become invalid. This invention designs an adaptive update mechanism based on incremental learning, which updates the parameters during on-site maintenance or periodic calibration. Obtain the reference standard value of the standard solution. At this point, based on the original measurement signal and environmental auxiliary parameters at the calibration time, the corresponding basic mechanism correction value is calculated using the mechanism model. , to the benchmark standard value Correction value based on the corresponding calculation mechanism The difference between them is used as the actual calibration residual. :

[0082] ;

[0083] Simultaneously, obtain the nonlinear residual prediction values ​​output by the data-driven model during prediction training. Calculate the actual calibration residual With nonlinear residual prediction values absolute error between The system determines whether the absolute error is greater than a preset trigger threshold. If the value is not greater than 0.5%, it indicates that the current model has not yet experienced severe drift, and the existing model can continue to be used. If the value is greater than 0.5%, it indicates that the model has deviated from the actual hardware state, and the system triggers a parameter fine-tuning procedure for the data-driven model trained for prediction. In the parameter fine-tuning procedure, a prediction loss function is constructed based on the difference between the actual calibration residual and the predicted value of the nonlinear residual. :

[0084] ;

[0085] The gradient descent algorithm is used to drive the model parameter set of the prediction training data using the prediction loss function. Perform iterative micro-updates:

[0086] ;

[0087] In the formula, This represents the number of iteration steps. The incremental learning rate is a constant. This is the partial derivative operator of the loss function with respect to the model parameter set. After the parameter update is complete, the updated model parameters are used. Replace the original model parameters in the data-driven model used for prediction training It continues to perform daily online monitoring, synchronously acquiring raw measurement signals, environmental auxiliary parameters, and instrument hardware status parameters. Through this incremental learning mechanism, long-term dynamic error compensation is achieved throughout the instrument's lifecycle.

[0088] In summary, this invention provides a calibration error correction method for liquid analysis instruments based on multi-parameter linkage. By simultaneously acquiring the original measurement signals, environmental auxiliary parameters, and instrument hardware status parameters of the liquid analysis instrument, it overcomes the limitations of traditional calibration methods that rely solely on external environmental parameters for compensation. It effectively separates the cross-interference from complex external matrices from the physical aging characteristics of the instrument's underlying hardware (such as sensor impedance changes and light source attenuation), fundamentally improving the comprehensiveness and accuracy of the multi-parameter linkage status characterization. By extracting time-series statistical features such as mean, variance, and rate of change from the original measurement signals and constructing a fused feature vector, it overcomes the vulnerability of traditional single-point transient sampling to interference. The limitations imposed by fluid dynamic fluctuations, localized concentration inhomogeneities, and sensor response hysteresis mean that the data input to the correction model fully incorporates the dynamic temporal information of fluid changes. This significantly enhances the instrument's anti-interference capability under dynamic fluid conditions. Furthermore, by introducing an incremental learning-based adaptive update mechanism, the error between the actual calibration residual and the predicted residual is calculated at regular calibration times. The gradient descent algorithm is then used to iteratively fine-tune the parameters of the data-driven model. This solves the problem of time-varying drift caused by physical wear and tear during long-term sensor operation, which leads to the failure of the original fixed model. As a result, the instrument can achieve long-term dynamic error adaptive compensation throughout its entire lifecycle using only a small amount of field calibration data.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for correcting calibration errors in liquid analysis instruments based on multi-parameter linkage, characterized in that, Includes the following steps: S1. First, synchronously acquire the original measurement signal of the liquid analyzer at the current moment, the environmental auxiliary parameters reflecting liquid cross-interference, and the instrument hardware status parameters reflecting the aging of the underlying hardware of the liquid analyzer. S2. Subsequently, time-series statistical features are extracted from the original measurement signal, and the time-series statistical features, the environmental auxiliary parameters, and the instrument hardware status parameters are concatenated to construct a fusion feature vector for characterizing the multi-parameter linkage state. S3. At this time, based on the preset physicochemical mechanism model, according to the original measurement signal and the environmental auxiliary parameters, the basic mechanism correction value for eliminating the explicit linear error is calculated, the fused feature vector is input into the pre-trained data-driven model, and the nonlinear residual prediction value caused by the cross-coupling of multi-dimensional parameters is output. S4. Finally, the basic mechanism correction value and the nonlinear residual prediction value are fused and reconstructed to obtain and output the final calibration measurement result of the liquid analysis instrument.

2. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 1, characterized in that, The original measurement signal includes the sensor's uncorrected output value; the environmental auxiliary parameters include at least one of the temperature, pressure, and turbidity of the liquid being measured; the instrument hardware status parameters include at least one of the sensor impedance, light source emission current, and reference voltage of the liquid analysis instrument.

3. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 1, characterized in that, The process of extracting time-series statistical features from the original measurement signal includes: Create a sliding window with a set time duration; Calculate the mean, variance, and rate of change of all the original measurement signals within the sliding window, and use the mean, variance, and rate of change as the time series statistical features.

4. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 1, characterized in that, The physicochemical mechanism model is a pre-set theoretical compensation function; The process of calculating the basic mechanism correction value for eliminating explicit linearity error based on the original measurement signal and the environmental auxiliary parameters includes: Substituting the original measurement signal and the environmental auxiliary parameters into the theoretical compensation function, the basic mechanism correction value calculated and output by the theoretical compensation function is obtained.

5. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 1, characterized in that, The pre-trained data-driven model is obtained through an offline pre-training phase; in the offline pre-training phase: Obtain the true concentration baseline value of the target liquid; The deviation between the true concentration benchmark value and the basic mechanism correction value is defined as the target nonlinear residual for the pre-trained data-driven model to learn.

6. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 1, characterized in that, The process of fusing and reconstructing the basic mechanism correction value with the nonlinear residual prediction value to obtain and output the final calibration measurement result of the liquid analysis instrument includes: The basic mechanism correction value and the nonlinear residual prediction value are summed to obtain the initial reconstruction measurement value; The initial reconstructed measurement values ​​are filtered using a dynamic smoothing filtering algorithm to eliminate transient distortions, resulting in the final calibration measurement result, which is then output and displayed.

7. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 6, characterized in that, The dynamic smoothing filtering algorithm is an exponentially weighted moving average filtering algorithm. The process of filtering the initial reconstructed measurement values ​​using a dynamic smoothing filtering algorithm includes: The final calibration measurement result at the current moment is obtained by weighted summation of the initial reconstructed measurement value at the current moment, the final calibration measurement result at the previous moment, and a preset smoothing factor.

8. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 1, characterized in that, After outputting the final calibration measurement results of the liquid analyzer, the process further includes an adaptive update of the pre-trained data-driven model based on incremental learning. This adaptive update process includes: At the calibration time, obtain the reference nominal value of the standard solution; Based on the original measurement signal and the environmental auxiliary parameters at the calibration time, calculate the corresponding basic mechanism correction value; The difference between the reference nominal value and the corresponding calculated basic mechanism correction value is taken as the actual calibration residual.

9. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 8, characterized in that, The adaptive update process also includes: Obtain the nonlinear residual prediction value output by the pre-trained data-driven model at the calibration time; Calculate the absolute error between the actual calibration residual and the predicted nonlinear residual; Determine whether the absolute error is greater than a preset trigger threshold. If so, trigger the parameter fine-tuning procedure for the pre-trained data-driven model.

10. The method for correcting calibration errors of liquid analysis instruments based on multi-parameter linkage according to claim 9, characterized in that, The parameter fine-tuning procedure includes: Based on the difference between the actual calibration residual and the predicted value of the nonlinear residual, a prediction loss function is constructed; The gradient descent algorithm is used to iteratively fine-tune and update the model parameters of the pre-trained data-driven model using the prediction loss function; After the parameters are updated, the original model parameters in the pre-trained data-driven model are replaced with the updated model parameters, and the process of synchronously acquiring the original measurement signals, environmental auxiliary parameters, and instrument hardware status parameters continues to be executed in a loop.

Images