Water quality probe dynamic correction system and method based on multi-source data fusion

Abstract

The invention provides a water quality probe dynamic correction system and method based on multi-source data fusion, and the method comprises the steps: collecting water quality parameters in real time, evaluating the passivation degree of a probe through an electrochemical impedance spectrum and a reaction time index by a probe state monitoring unit, and collecting, storing and preprocessing multi-source data by a data collection and preprocessing unit. The multi-source data fusion and correction algorithm unit establishes a dynamic correction model based on the probe state and the multi-source data; and the intelligent control and feedback unit controls the cleaning device according to the correction result and provides maintenance suggestions. The real-time dynamic correction of the water quality probe data is realized, and the data accuracy and reliability are obviously improved. Through a probe state evaluation mechanism, the passivation degree of the probe can be monitored in real time, and the failure risk of the probe is warned in advance; probe monitoring data and laboratory test data are effectively fused, the advantages of the probe monitoring data and the laboratory test data are fully utilized, and continuous monitoring and high precision are combined.

Description

Technical Field

[0001] This invention belongs to the field of water quality testing technology, and relates to a dynamic calibration system and method for water quality probes based on multi-source data fusion. Background Technology

[0002] Currently, online water quality monitoring systems mainly rely on underwater probes for continuous real-time detection. However, during long-term operation, underwater probes are susceptible to the effects of high concentrations of sludge, suspended solids, algae, etc., leading to the formation of a passivation layer on the probe surface and causing deviations in sensor measurement data. Existing technologies mainly address the probe passivation problem through the following methods.

[0003] Regular manual cleaning solution: This solution requires technicians to clean and maintain the probes periodically, resulting in a large workload, high labor costs, and inability to guarantee real-time data accuracy. It is only suitable for scenarios with a small number of monitoring points and cannot meet the needs of large-scale monitoring networks.

[0004] Automated mechanical cleaning devices: This approach typically uses brushes, ultrasonic waves, or high-pressure water jets for automatic cleaning. However, it is difficult to completely remove biofilms and chemical deposits, and the cleaning frequency is hard to precisely match the degree of contamination. Furthermore, mechanical cleaning may damage the probe surface, affecting the sensor's lifespan.

[0005] Simple data correction algorithms: This approach relies on linear or nonlinear correction based on fixed parameters, making it unsuitable for complex situations involving dynamic changes in water quality and probe performance degradation. Furthermore, it lacks real-time assessment of the probe's condition, resulting in limited correction accuracy.

[0006] Existing technologies using algorithmic correction mainly suffer from the following problems:

[0007] 1. Relying solely on physical cleaning or simple data calibration cannot fundamentally solve the problem of probe passivation; it only addresses the symptoms, not the root cause.

[0008] 2. Lack of a real-time assessment mechanism for probe status: It is impossible to quantitatively assess the degree of probe passivation and it is difficult to predict the trend of probe performance changes;

[0009] 3. Inability to effectively integrate online monitoring and laboratory data: Failure to fully utilize high-precision laboratory data to correct online data;

[0010] 4. Unable to dynamically adapt to changes in water quality: The correction parameters are fixed and cannot adapt to different water quality conditions and pollution levels;

[0011] 5. Lack of intelligent decision-making mechanism: It is unable to intelligently decide the timing and intensity of cleaning based on the probe status.

[0012] Therefore, developing a dynamic calibration method for water quality probes based on multi-source data fusion is of great practical significance. Summary of the Invention

[0013] The purpose of this invention is to address the problems existing in the prior art. This invention proposes a dynamic correction system and method for water quality probes based on multi-source data fusion. By fusing online probe monitoring data, laboratory test data, historical data, and environmental auxiliary parameters, a dynamic correction model is constructed to achieve real-time and accurate correction of water quality probe data, thereby solving the data deviation problem caused by probe passivation.

[0014] A first aspect of the present invention provides a dynamic calibration system for a water quality probe based on multi-source data fusion, comprising:

[0015] A multi-parameter online water quality monitoring unit is used to collect water quality parameters in real time;

[0016] The probe status monitoring unit is used to assess the passivation degree of the probe through electrochemical impedance spectroscopy and reaction time indicators.

[0017] The data acquisition and preprocessing unit is used to acquire, store, and preprocess multi-source data, and to identify and process outliers.

[0018] A multi-source data fusion and correction algorithm unit is used to establish a dynamic correction model based on probe status and multi-source data;

[0019] The intelligent control and feedback unit is used to control the cleaning device and provide maintenance suggestions based on the calibration results.

[0020] Preferably, the water quality parameters include pH, turbidity, dissolved oxygen concentration, conductivity, and ammonia nitrogen concentration.

[0021] A second aspect of the present invention provides a method for applying a dynamic calibration system for water quality probes based on multi-source data fusion, comprising:

[0022] S1. Multi-parameter online water quality monitoring unit collects water quality parameters in real time;

[0023] S2. The probe status monitoring unit assesses the passivation degree of the probe through electrochemical impedance spectroscopy and reaction time indicators;

[0024] S3. Data Acquisition and Preprocessing Unit: Acquires, stores, and preprocesses multi-source data, and performs outlier identification and processing.

[0025] S4. The multi-source data fusion and correction algorithm unit establishes a dynamic correction model based on probe status and multi-source data;

[0026] S5. The intelligent control and feedback unit controls the cleaning device and provides maintenance suggestions based on the calibration results.

[0027] Preferably, in S2, the degree of probe passivation is quantitatively evaluated by defining the Probe Status Index (PSI):

[0028]

[0029] In the formula: R t R0 is the current electrochemical impedance value of the probe surface (unit: Ω); R0 is the standard impedance value of the probe under clean calibration conditions (unit: Ω); R0 max T represents the maximum permissible value for impedance deviation (unit: Ω); t T0 is the current probe response time (in seconds); T0 is the standard response time (in seconds) of the probe in a clean state; T0 is the current probe response time (in seconds); T0 is the current probe response time (in seconds). max The maximum permissible deviation of response time (in seconds); S t This is a dimensionless index representing the signal stability at the current moment.

[0030] The calculation formula is

[0031] Where x i For n consecutive measurements (n=20), It is the arithmetic mean;

[0032] S0: Standard signal stability under probe cleaning conditions; S max : The maximum permissible deviation of signal stability;

[0033] α, β, γ are weighting coefficients that satisfy the constraints α + β + γ = 1 and α, β, γ > 0;

[0034] Accordingly, the passivation degree of the probe is defined according to the following classification:

[0035] PSI∈[0,0.2]: The probe is in good condition and working normally;

[0036] PSI∈(0.2,0.5]: The probe is slightly dulled; enhanced monitoring is recommended.

[0037] PSI∈(0.5,0.8]: The probe is moderately passivated and requires data calibration;

[0038] PSI∈(0.8,1.0]: The probe is severely passivated and needs to be cleaned immediately.

[0039] Preferably, in S4, a multi-source data fusion model based on weighted Kalman filtering is established to achieve optimal fusion of probe data and laboratory data:

[0040] System state equations:

[0041]

[0042] System observation equations:

[0043]

[0044] State prediction process:

[0045]

[0046] Covariance prediction:

[0047]

[0048] Kalman gain calculation:

[0049]

[0050] State estimation update:

[0051]

[0052] Covariance update:

[0053]

[0054] In the formula: Let be the state estimation vector of the true water quality parameters at time k; Let k be the prior state estimate at time k; The state transition matrix describes the dynamic characteristics of the system. To control the input matrix; This is the control vector, which includes environmental impact factors; w k-1 The process noise is assumed to be zero-mean Gaussian white noise, and the covariance matrix is... Let k be the observation vector at time k, including probe data and laboratory data; The observation matrix describes the relationship between the state and the observations; v k For observation noise, the covariance matrix is ​​R. k ;K k This is the Kalman gain matrix; P is the prior error covariance matrix; k Q is the posterior error covariance matrix; k-1 R is the process noise covariance matrix; k To observe the noise covariance matrix.

[0055] Preferably, adaptive filtering is achieved by dynamically adjusting the probe passivation state and measuring the noise covariance according to the following formula:

[0056] R k =R0·(1+λ·PSI) 2 );

[0057] Where: R0 is the reference measurement noise covariance matrix under normal probe conditions; λ is the amplification coefficient, which controls the degree of influence of passivation on measurement uncertainty, and its value range is: λ∈[2,10].

[0058] Preferably, the weighted fusion formula for probe data and laboratory data is:

[0059]

[0060] in: These are the estimated optimal water quality parameters after fusion; For probe measurement data; Laboratory test data (processed with time interpolation); ω k This is a dynamic weighting coefficient, with a value range of [0,1].

[0061] Dynamic weighting coefficient ω k Determined by both PSI and data availability:

[0062]

[0063] Where: PSI is the probe passivation state index; n is the sensitivity parameter, which controls the steepness of the weight change, and a value of n=2 is recommended; ξ k For laboratory data availability factors;

[0064]

[0065] Where: t k t represents the current time; lab The timestamp for the most recent laboratory data; σ t This is the time decay parameter.

[0066] Preferably, the method also includes the step of building an LSTM (Long Short-Term Memory) network model using laboratory data and historical data, and dynamically updating the correction parameters.

[0067] Preferably, the basic equations of the LSTM are constructed as follows:

[0068] Forgotten Gate:

[0069] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0070] Input Gate:

[0071] i t =σ(W i ·[ht-1 ,x t ]+b i );

[0072] Candidate values:

[0073]

[0074] Cell status update:

[0075]

[0076] Output gate:

[0077] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0078] Hidden state:

[0079] h t =o t *tanh(C t );

[0080] In the formula: x t The input vector contains water quality parameters, environmental factors, and PSI features; h t Let C be the hidden state vector at time t; t W is the cell state vector at time t; f W i W C W o This is the weight matrix;

[0081] b f ,b i ,b C ,b o Here, σ is the bias vector; σ is the Sigmoid activation function. * indicates element-wise multiplication (Hadamard product);

[0082] Output layer computation:

[0083] o t =W ho h t +b o ;

[0084] Among them o t This is the predicted correction parameter vector.

[0085] Preferably, combining prediction accuracy and model complexity, a comprehensive loss function is used to calculate the comprehensive loss value:

[0086]

[0087] in: The true value measured in the laboratory; θ represents the predicted value from the LSTM model. j These are the trainable parameters of the model; λ is the L2 regularization coefficient, and a value of 10 is recommended. -4 ~10 -6 N is the number of training samples.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] This invention enables real-time dynamic correction of water quality probe data, significantly improving data accuracy and reliability. Specifically, through a probe status assessment mechanism, this invention can monitor the probe passivation level in real time and provide early warning of probe failure risks; it effectively integrates probe monitoring data and laboratory test data, fully utilizing the advantages of both to achieve a combination of continuous monitoring and high precision; and it employs an adaptive learning mechanism that continuously optimizes the correction model as data accumulates, improving the long-term operational stability of the system.

[0090] This invention can effectively reduce the frequency of manual maintenance, lower maintenance costs, and extend the lifespan of the probe.

[0091] This invention is applicable to various water quality parameters and different water environments, and has a wide range of application scenarios and promotional value. Attached Figure Description

[0092] Figure 1 This is a system architecture diagram of the present invention.

[0093] Figure 2 This is a structural diagram of the probe status monitoring unit.

[0094] Figure 3 This is a flowchart of the data fusion and correction algorithm.

[0095] Figure 4 This is a comparison chart of test results for a system implementation example. Detailed Implementation

[0096] To enable those skilled in the art to better understand the technical solutions of the present invention, the preferred embodiments of the present invention are described below in conjunction with specific examples. However, it should be understood that the accompanying drawings are for illustrative purposes only and should not be construed as limiting the present patent. To better illustrate the embodiments, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable for those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings. The positional relationships described in the drawings are for illustrative purposes only and should not be construed as limiting the present patent.

[0097] like Figure 1 and Figure 2 As shown, the dynamic correction system for water quality probes based on multi-source data fusion proposed in this invention mainly consists of the following main modules:

[0098]

[0099]

[0100] This invention proposes a method for applying the aforementioned dynamic calibration system for water quality probes based on multi-source data fusion, such as... Figure 3 As shown, it includes:

[0101] S1. Multi-parameter online water quality monitoring unit collects water quality parameters in real time;

[0102] S2. The probe status monitoring unit assesses the passivation degree of the probe through electrochemical impedance spectroscopy and reaction time indicators;

[0103] S3. Data Acquisition and Preprocessing Unit: Acquires, stores, and preprocesses multi-source data, and performs outlier identification and processing.

[0104] S4. The multi-source data fusion and correction algorithm unit establishes a dynamic correction model based on probe status and multi-source data;

[0105] S5. The intelligent control and feedback unit controls the cleaning device and provides maintenance suggestions based on the calibration results.

[0106] The core algorithm of this invention includes the following parts:

[0107] 1. Probe passivation status assessment

[0108] Define the Probe Status Index (PSI) to quantitatively assess the current state of the probe:

[0109]

[0110] Detailed parameter definition:

[0111] R t : Current electrochemical impedance value of the probe surface (unit: Ω);

[0112] o Measurement method: Measured at a specific frequency using an electrochemical impedance spectroscopy instrument;

[0113] o Physical meaning: Reflects the degree of contaminant accumulation on the probe surface;

[0114] R0: Standard impedance value of the probe under clean and calibrated conditions (unit: Ω);

[0115] o Acquisition method: Reference value obtained after probe factory calibration or on-site cleaning;

[0116] R max : Maximum permissible value of impedance deviation (unit: Ω);

[0117] Setting principle: Determined according to probe type and accuracy requirements, usually 20%-50% of R0;

[0118] T t : Current probe response time (unit: seconds);

[0119] o Measurement method: The time required to switch from the standard solution to the water sample to be tested and reach a stable reading of 90%;

[0120] o Physical meaning: Reflects the change in mass transfer resistance on the probe surface;

[0121] T0: Standard response time (in seconds) under clean probe conditions;

[0122] Reference values: pH probes typically last 10-30 seconds, DO probes last 30-60 seconds. max : Maximum permissible deviation of response time (unit: seconds);

[0123] o Set a standard: generally not exceeding 200% of the standard response time;

[0124] S t Current signal stability index (dimensionless);

[0125] o Calculation formula:

[0126] o in which x i For n consecutive measurements (n=20), It is the arithmetic mean;

[0127] Physical meaning: Reflects the degree of fluctuation in the measured signal;

[0128] S0: Standard signal stability under probe cleaning conditions;

[0129] o Reference value: A stability index measured in a standard solution;

[0130] S max : The maximum permissible deviation of signal stability;

[0131] o Setting basis: Determined according to the accuracy level of the sensor.

[0132] Weighting coefficients:

[0133] α,β,γ: Weighting coefficients, satisfying the constraints α+β+γ=1 and α,β,γ>0.

[0134] Recommended weighting coefficients for different probe types:

[0135] Probe type α (Impedance weight) β (response time weight) γ (stability weight) pH probe 0.4 0.3 0.3 Dissolved oxygen probe 0.3 0.4 0.3 Turbidity probe 0.2 0.3 0.5 conductivity probe 0.5 0.2 0.3 ammonia nitrogen probe 0.4 0.4 0.2

[0136] PSI index grading standards:

[0137] PSI∈[0,0.2]: The probe is in good condition and working normally.

[0138] PSI∈(0.2,0.5]: The probe is slightly dulled; enhanced monitoring is recommended.

[0139] PSI∈(0.5,0.8]: The probe is moderately passivated and requires data correction.

[0140] PSI∈(0.8,1.0]: The probe is severely passivated and needs to be cleaned immediately.

[0141] 2. Multi-source data fusion model

[0142] Establish a multi-source data fusion model based on weighted Kalman filtering to achieve optimal fusion of probe data and laboratory data:

[0143] System state equations:

[0144]

[0145] System observation equations:

[0146]

[0147] State prediction process:

[0148]

[0149] Covariance prediction:

[0150]

[0151] Kalman gain calculation:

[0152]

[0153] State estimation update:

[0154]

[0155] Covariance update:

[0156]

[0157] Parameter matrix definition:

[0158] The true water quality parameter state estimation vector at time k;

[0159] Example:

[0160] The prior state estimate at time k;

[0161] The state transition matrix describes the dynamic characteristics of the system.

[0162] For water quality parameters, they are usually set as identity matrix I or diagonally dominant matrix;

[0163] Control input matrix;

[0164] Control vectors, which include environmental impact factors;

[0165] o Example: u k-1 =[Temperature,Flow_rate,Rainfall] T ;

[0166] w k-1 Process noise, assumed to be zero-mean Gaussian white noise, with a covariance matrix of Q. k-1 ;

[0167] The observation vector at time k, including probe data and laboratory data;

[0168] o Components:

[0169] The observation matrix describes the relationship between the state and the observations;

[0170] v k Observation noise, with a covariance matrix of R k ;

[0171] K k Kalman gain matrix;

[0172] Prior error covariance matrix;

[0173] P k : Posterior error covariance matrix;

[0174] Q k-1 Process noise covariance matrix;

[0175] R k : Observation noise covariance matrix.

[0176] 3. Dynamic correction algorithm

[0177] Adaptive adjustment of observation noise covariance:

[0178] Adaptive filtering is achieved by dynamically adjusting the measurement noise covariance based on the probe passivation state.

[0179] R k =R0·(1+λ·PSI) 2 );

[0180] Where: R0: the reference measurement noise covariance matrix under normal probe conditions;

[0181] λ: Amplification factor, which controls the degree of influence of passivation on measurement uncertainty;

[0182] Recommended value range for o: λ∈[2,10];

[0183] Physical meaning: The larger λ is, the more significant the effect of PSI on the filter.

[0184] Multi-source data weighted fusion strategy:

[0185] The weighted fusion formula for probe data and laboratory data is as follows:

[0186]

[0187] in:

[0188] Estimated optimal water quality parameters after fusion;

[0189] Probe measurement data;

[0190] Laboratory test data (processed with time interpolation);

[0191] ω k : Dynamic weighting coefficient, with a value range of [0,1].

[0192] Dynamic weight calculation:

[0193] weight ω k Determined by both PSI and data availability:

[0194]

[0195] in:

[0196] PSI: Probe passivation index;

[0197] n: Sensitivity parameter, which controls the steepness of the weight change; a value of n=2 is recommended.

[0198] ξ k Laboratory data availability factor;

[0199]

[0200] in:

[0201] t k : The current moment;

[0202] t lab : Timestamp of the most recent lab data;

[0203] σ t Time decay parameter, recommended value is 12-24 hours;

[0204] 4. Adaptive learning mechanism

[0205] An LSTM (Long Short-Term Memory) network model is built using laboratory and historical data, and the calibration parameters are dynamically updated.

[0206] LSTM fundamental equations:

[0207] Forgotten Gate:

[0208] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0209] Input Gate:

[0210] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0211] Candidate values:

[0212]

[0213] Cell status update:

[0214]

[0215] Output gate:

[0216] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0217] Hidden state:

[0218] ht =o t *tanh(C t ).

[0219] Parameter description:

[0220] x t : Input vector, containing features such as water quality parameters, environmental factors, and PSI;

[0221] o-dimensionality: typically 8-15 dimensions;

[0222] o composition: [pH,DO,Turbidity,EC,Temperature,PSI,Δt,...];

[0223] h t The hidden state vector at time t;

[0224] o-dimensionality: 64-128 dimensions are recommended;

[0225] C t : The cell state vector at time t;

[0226] W f W i W C W o Weight matrix;

[0227] o-dimensionality: (n h +n x )×n h , where n h Let n be the dimension of the hidden layer. x For input dimensions;

[0228] b f ,b i ,b C ,b o Bias vector;

[0229] σ(·): Sigmoid activation function

[0230] *: Element-wise multiplication (Hadamard product).

[0231] Output layer computation:

[0232] o t =W ho h t +b o ;

[0233] Among them o t This is the predicted correction parameter vector.

[0234] Loss Function Design:

[0235] A comprehensive loss function is adopted, combining prediction accuracy and model complexity:

[0236]

[0237] in:

[0238] The true value measured in the laboratory;

[0239] LSTM model predictions;

[0240] θ j : Trainable parameters of the model;

[0241] λ: L2 regularization coefficient, a value of 10 is recommended. -4 ~10 -6 ;

[0242] N: Number of training samples.

[0243] Model training strategy:

[0244] Optimization algorithm: Adam optimizer;

[0245] Learning rate: Initial learning rate 0.001, using exponential decay;

[0246] Batch size: 32-64 samples;

[0247] Training period: Determined based on the amount of data, typically 100-500 epochs;

[0248] Early stopping mechanism: Training is stopped when the validation set loss shows no improvement for 10 consecutive epochs.

[0249] like Figure 4 As shown in the practical application case, the method of the present invention is used for dynamic correction. The filled black line part (CCD) is too low, and the fused red line is corrected.

[0250] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic calibration system for water quality probes based on multi-source data fusion, characterized in that, include: A multi-parameter online water quality monitoring unit is used to collect water quality parameters in real time; The probe status monitoring unit is used to assess the passivation degree of the probe through electrochemical impedance spectroscopy and reaction time indicators. The data acquisition and preprocessing unit is used to acquire, store, and preprocess multi-source data, and to identify and process outliers. A multi-source data fusion and correction algorithm unit is used to establish a dynamic correction model based on probe status and multi-source data; The intelligent control and feedback unit is used to control the cleaning device and provide maintenance suggestions based on the calibration results.

2. The dynamic calibration system for a water quality probe based on multi-source data fusion according to claim 1, characterized in that, The water quality parameters include pH, turbidity, dissolved oxygen concentration, conductivity, and ammonia nitrogen concentration.

3. A method for applying to the dynamic calibration system of a water quality probe based on multi-source data fusion as described in claim 1, characterized in that, include: S1. Multi-parameter online water quality monitoring unit collects water quality parameters in real time; S2. The probe status monitoring unit assesses the passivation degree of the probe through electrochemical impedance spectroscopy and reaction time indicators; S3. Data Acquisition and Preprocessing Unit: Acquires, stores, and preprocesses multi-source data, and performs outlier identification and processing. S4. The multi-source data fusion and correction algorithm unit establishes a dynamic correction model based on probe status and multi-source data; S5. The intelligent control and feedback unit controls the cleaning device and provides maintenance suggestions based on the calibration results.

4. The method according to claim 3, characterized in that: In S2, the degree of probe passivation is quantitatively evaluated by defining the Probe Status Index (PSI): In the formula: R t R0 is the current electrochemical impedance value of the probe surface (unit: Ω); R0 is the standard impedance value of the probe under clean calibration conditions (unit: Ω); R0 max The maximum permissible value for impedance deviation (unit: Ω); T t T0 is the current probe response time (in seconds); T0 is the standard response time (in seconds) of the probe in a clean state. T max The maximum permissible deviation in response time (in seconds); S t This is a dimensionless index representing the signal stability at the current moment. The calculation formula is Where x i For n consecutive measurements (n=20), It is the arithmetic mean; S0: Standard signal stability under probe cleaning conditions; S max : The maximum permissible deviation of signal stability; α, β, γ are weighting coefficients that satisfy the constraints α + β + γ = 1 and α, β, γ > 0; Accordingly, the passivation degree of the probe is defined according to the following classification: PSI∈[0,0.2]: The probe is in good condition and working normally; PSI∈(0.2,0.5]: The probe is slightly dulled; enhanced monitoring is recommended. PSI∈(0.5,0.8]: The probe is moderately passivated and requires data calibration; PSI∈(0.8,1.0]: The probe is severely passivated and needs to be cleaned immediately.

5. The method according to claim 4, characterized in that: In S4, a multi-source data fusion model based on weighted Kalman filtering is established to achieve optimal fusion of probe data and laboratory data: System state equations: System observation equations: State prediction process: Covariance prediction: Kalman gain calculation: State estimation update: Covariance update: In the formula: Let be the state estimation vector of the true water quality parameters at time k; Let k be the prior state estimate at time k; The state transition matrix describes the dynamic characteristics of the system. To control the input matrix; This is the control vector, which includes environmental impact factors; w k-1 The process noise is assumed to be zero-mean Gaussian white noise, with a covariance matrix of Q. k-1 ; Let k be the observation vector at time k, including probe data and laboratory data; The observation matrix describes the relationship between the state and the observations; v k For observation noise, the covariance matrix is ​​R. k ;K k This is the Kalman gain matrix; P is the prior error covariance matrix; k Q is the posterior error covariance matrix; k-1 R is the process noise covariance matrix; k To observe the noise covariance matrix.

6. The method according to claim 5, characterized in that, Based on the dynamic adjustment of the probe passivation state, adaptive filtering is achieved by measuring the noise covariance according to the following formula: R k =R0·(1+λ·PSI 2 ); in: R0 is the reference measurement noise covariance matrix under normal probe conditions; λ is the amplification factor, controlling the passivation pair. The degree of influence of measurement uncertainty, with a value range of λ∈[2,10].

7. The method according to claim 5, characterized in that, The weighted fusion formula for probe data and laboratory data is: in: These are the estimated optimal water quality parameters after fusion; For probe measurement data; For laboratory purposes Verification data (processed with time interpolation); ω k This is a dynamic weighting coefficient, with a value range of [0,1]. Dynamic weighting coefficient ω k Determined by both PSI and data availability: Where: PSI is the probe passivation state index; n is the sensitivity parameter, which controls the steepness of the weight change, and a value of n=2 is recommended; ξ k For laboratory data availability factors; Where: t k t represents the current time; lab The timestamp for the most recent laboratory data; σ t This is the time decay parameter.

8. The method according to claim 5, characterized in that, It also includes the steps of building an LSTM (Long Short-Term Memory) network model using laboratory and historical data, and dynamically updating the correction parameters.

9. The method according to claim 8, characterized in that, The basic equations of LSTM are constructed as follows: Forgotten Gate: f t =σ(W f ·[h t-1 ,x t ]+b f ); Input Gate: i t =σ(W i ·[h t-1 ,x t ]+b i ); Candidate values: Cell status update: Output gate: the t =σ(W o ·[h t-1 ,x t ]+b o ); Hidden state: h t =o t *fishy(C) t ); In the formula: x t The input vector contains water quality parameters, environmental factors, and PSI features; h t Let C be the hidden state vector at time t; t W is the cell state vector at time t; f W i W C W o b is the weight matrix; f ,b i ,b C ,b o Here, σ is the bias vector; σ is the Sigmoid activation function. * indicates element-wise multiplication (Hadamard product); Output layer computation: o t =W ho h t +b o ; Among them o t This is the predicted correction parameter vector.

10. The method according to claim 8, characterized in that, Combining prediction accuracy and model complexity, a comprehensive loss function is used to calculate the comprehensive loss value: in: The true value measured in the laboratory; θ represents the predicted value from the LSTM model. j These are the trainable parameters of the model; λ is the L2 regularization coefficient, and a value of 10 is recommended. -4 ~10 -6 N is the number of training samples.

Images