Self-adaptive multispectral correction system and method for water quality probe in high-turbidity sewage
By using a multispectral correction system and method, the problem of decreased measurement accuracy of water quality probes in high-turbidity wastewater has been solved, achieving high-accuracy and long-life water quality monitoring. It is applicable to monitoring multiple types of water quality and multiple parameters, and reduces equipment costs.
Patent Information
- Application Number
- CN202511019348.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-23
- Publication Date
- 2025-11-07
AI Technical Summary
Existing water quality probes suffer from interference in high-turbidity wastewater, and their accuracy decreases with prolonged use, making them unable to cope with changes in water quality conditions and probe aging.
The system employs a multi-wavelength spectral acquisition module, a turbidity feature identification and separation module, an adaptive multivariate correction module, a machine learning-enhanced prediction module, and a probe aging compensation module. Through the adaptive multispectral correction method, it corrects the characteristic wavelength signals under turbidity conditions, calculates the actual concentration of the target parameter, and uses machine learning to optimize the prediction accuracy, performing dynamic fusion and automatic probe diagnosis and compensation.
It improves the measurement accuracy of water quality probes in high-turbidity wastewater, reduces measurement errors, extends the service life of probes, and has universal applicability to multiple water quality types and multi-parameter monitoring scalability, thereby reducing equipment costs.
Smart Images
Figure CN120908124A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of water quality monitoring, and particularly relates to a self-adaptive multi-spectrum correction system and method for a water quality probe in high-turbidity sewage. BACKGROUND
[0002] The water quality probe is mainly used for water quality monitoring. Key parameters in water bodies are monitored in real time by the water quality probe, and the key parameters include the contents of COD and ammonia nitrogen.
[0003] Two main problems are faced by the existing water quality probe in actual application: one is measurement interference caused by different turbidity of sewage, and the other is the decline of measurement accuracy caused by factors such as scaling on the surface of the probe and attenuation of the light source as the use time is prolonged.
[0004] At present, the water quality probe on the market mostly adopts a calibration method with fixed parameters, which cannot cope with the dual challenges of changes in water quality conditions and aging of the probe. Especially in the case of high-turbidity sewage or long-term use, the measurement error often exceeds the acceptable range. SUMMARY
[0005] In view of the above technical deficiencies, the purpose of the present application is to provide a self-adaptive multi-spectrum correction system and method for a water quality probe in high-turbidity sewage, which is used to significantly improve the measurement accuracy of the water quality probe in different turbidity of sewage and solve the problem of accuracy decline caused by the increase of the use time of the probe.
[0006] To solve the above technical problems, the present application adopts the following technical solution: a self-adaptive multi-spectrum correction system for a water quality probe in high-turbidity sewage, comprising: a multi-wavelength spectrum acquisition module for simultaneously measuring spectrum signals of multiple characteristic wavelengths; a turbidity feature recognition and separation module for correcting the characteristic wavelength signals under turbidity conditions according to the measured spectrum signals; a self-adaptive multivariate correction module for calculating the actual concentration of the target parameter based on the characteristic wavelength signals corrected under the turbidity conditions; a machine learning enhanced prediction module for learning historical measurement data based on a machine learning algorithm, generating a predicted concentration, and dynamically fusing the predicted concentration with the result obtained by the self-adaptive multivariate correction module; a self-adaptive parameter updating module for updating the weight of the self-adaptive multivariate correction algorithm based on recent historical data, and feeding back the information to the turbidity feature recognition and separation module and the self-adaptive multivariate correction module; a probe aging compensation module for calculating the attenuation compensation coefficient of the probe at different stages, and feeding back the attenuation compensation coefficient to the self-adaptive multivariate correction module.
[0007] The application discloses a self-adaptive multi-spectrum correction method for a water quality probe in high-turbidity sewage, adopts a self-adaptive multi-spectrum correction system for the water quality probe in the high-turbidity sewage, and comprises the following steps, S1, spectrum collection: based on a multi-wavelength spectrum collection probe, a specific wavelength combination is selected in an ultraviolet-visible spectrum range, and spectrum signals of multiple characteristic wavelengths are measured at the same time; S2, processing of original spectrum signals, correction of characteristic wavelength signals under turbidity conditions; S3, calculation of attenuation compensation coefficients of probes in different stages; S4, calculation of actual concentrations of target parameters based on the characteristic wavelength signals under the corrected turbidity conditions and in combination with the attenuation compensation coefficients; S5, machine learning model based on gradient boosting decision tree, learning of historical measurement data, generation of predicted concentrations, and dynamic fusion of the predicted concentrations and the actual concentrations of the target parameters.
[0008] In S1, the specific wavelength combination comprises a main wavelength group and a reference wavelength group, the main wavelength group is used for measuring characteristic wavelengths of target parameters, and the reference wavelength group is used for turbidity feature identification and device state monitoring; the target parameters comprise COD and ammonia nitrogen.
[0009] The main wavelength group comprises: for organic matter characteristic absorption; for aromatic compound absorption; for ammonia nitrogen compound absorption; for color reaction absorption; The reference wavelength group comprises: for low-absorption-zone reference wavelengths; for near-infrared reference wavelengths; for turbidity characteristic wavelengths.
[0010] In S2, a turbidity feature identification and separation algorithm is adopted to correct the characteristic wavelength signals under turbidity conditions, and the turbidity feature identification and separation algorithm is as follows: ; In the formula, denotes a spectrum signal of a corrected characteristic wavelength ; denotes a spectrum signal of a measured characteristic wavelength ; a spectral signal representing a turbidity characteristic wavelength; represents an adaptive correction coefficient, determined by the following formula: ; to is historical calibration data, and t is the probe usage time.
[0011] In S3, the attenuation compensation coefficient of the probe at different stages is calculated by using a probe performance attenuation model, and the probe performance attenuation model is: ; In the formula: represents the probe performance attenuation coefficient at time t; represents an initial rapid attenuation parameter; represents a long-term slow attenuation parameter; represents an attenuation rate constant; The formula for calculating the attenuation compensation coefficient of the probe is: .
[0012] In S4, the actual concentration of the target parameter is calculated by using an adaptive multivariate correction algorithm, and the adaptive multivariate correction algorithm is: ; In the formula: is the concentration of the target parameter; : is a weight coefficient related to turbidity B; is a time-dependent bias term; where the weight coefficient is determined by a turbidity piecewise function: ; In the formula: represents the linear coefficient of the first segment; represents the linear coefficient of the second segment; represents the linear coefficient of the third segment; represents the quadratic coefficient of the extremely high turbidity interval; represents the intercept of the first segment; represents the intercept of the second segment; represents the intercept of the third segment, represents the intercept of the extremely high turbidity interval; a coefficient of a linear term representing an extremely high turbidity interval; represents a turbidity value of a current water sample; represents a turbidity threshold value of the first segment; represents a turbidity threshold value of the second segment; represents a turbidity threshold value of the third segment.
[0013] In S5, the machine learning model based on gradient boosting decision tree is: ; In the formula: represents a target parameter concentration of a final output; represents a concentration calculated by a basic algorithm; represents a concentration predicted by a machine learning model; represents a dynamic fusion coefficient, ranging from 0 to 1; The input feature vector X of the machine learning model is defined as: ; In the formula: T represents water temperature; and pH represents a pH value; represents a specific wavelength ratio, ; wherein, represents a main wavelength group for organic matter characteristic absorption; represents a main wavelength group for aromatic compound absorption.
[0014] After S5, an adaptive parameter updating mechanism is further included, the adaptive parameter updating mechanism performs periodic self-checking and parameter updating, the adaptive parameter updating mechanism updates adaptive multivariate correction algorithm weights based on recent historical data, and a calculation formula of a weight update amount is: ; In the formula: represents a weight update amount; : represents a learning rate; : represents an error gradient based on recent historical data .
[0015] The present application has the following beneficial effects: The application measures the spectral signals of multiple characteristic wavelengths simultaneously through a multi-wavelength spectral acquisition module, corrects the characteristic wavelength signals under turbidity conditions through a turbidity feature recognition and separation module, effectively separates turbidity interference and target parameter signals, calculates the actual concentration of the target parameter through an adaptive multivariate correction module, uses historical data to continuously optimize prediction accuracy through a machine learning enhanced prediction module, dynamically fuses the predicted concentration and the result obtained by the adaptive multivariate correction module to improve the accuracy of the measured concentration, realizes automatic diagnosis and compensation of the probe through a probe aging compensation module, reduces measurement errors, and prolongs the effective service life of the probe.
[0016] Meanwhile, the application has multi-water quality type universality, function segmentation according to turbidity, and different thresholds and corresponding coefficients can automatically adapt to different scenes such as chemical wastewater, municipal sewage, industrial circulating water, and surface runoff, solving the limitation of traditional probes that “single parameter adaptation to specific scenes”; the application also has multi-parameter monitoring expandability, the main wavelength group can be flexibly expanded to other water quality parameters (such as total phosphorus and total nitrogen), and only the corresponding characteristic wavelength (such as λ=700 nm for total phosphorus molybdenum blue method color development) needs to be added, without the need to reconstruct the hardware architecture, thereby reducing the equipment cost of multi-parameter monitoring. The application has strong dynamic correction ability and self-maintenance ability, can perform turbidity correction and aging compensation in real time, and is combined with machine learning, and has strong compatibility. BRIEF DESCRIPTION OF DRAWINGS
[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.
[0018] Figure 1 The figure is a structural block diagram of the system of the present application.
[0019] Figure 2 The figure is a flow block diagram of the method of the present application. DETAILED DESCRIPTION
[0020] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0021] Embodiment 1: Referring to Figure 1 As shown in the figure, a high-turbidity sewage water quality probe adaptive multi-spectral correction system comprises: a multi-wavelength spectrum acquisition module for simultaneously measuring spectrum signals of multiple characteristic wavelengths.
[0022] a turbidity feature recognition and separation module for correcting the characteristic wavelength signals under turbidity conditions according to the measured spectrum signals.
[0023] an adaptive multivariate correction module for calculating the actual concentration of the target parameter based on the corrected characteristic wavelength signals under turbidity conditions; a machine learning enhanced prediction module for learning historical measurement data based on a machine learning algorithm, generating a predicted concentration, and dynamically fusing the predicted concentration with the result obtained by the adaptive multivariate correction module; an adaptive parameter update module for updating the adaptive multivariate correction algorithm weight based on recent historical data, and feeding back information to the turbidity feature recognition and separation module and the adaptive multivariate correction module; a probe aging compensation module for calculating attenuation compensation coefficients of probes at different stages, and feeding back the attenuation compensation coefficients to the adaptive multivariate correction module.
[0024] The present application simultaneously measures spectrum signals of multiple characteristic wavelengths through the multi-wavelength spectrum acquisition module, corrects the characteristic wavelength signals under turbidity conditions through the turbidity feature recognition and separation module, effectively separates turbidity interference and target parameter signals, calculates the actual concentration of the target parameter through the adaptive multivariate correction module, continuously optimizes the prediction accuracy using historical data through the machine learning enhanced prediction module, dynamically fuses the predicted concentration with the result obtained by the adaptive multivariate correction module, improves the accuracy of the measured concentration, realizes automatic diagnosis and compensation of the probe through the probe aging compensation module, reduces measurement errors, and prolongs the effective service life of the probe.
[0025] In the present embodiment, the multi-wavelength spectrum acquisition module adopts an ultraviolet-visible spectrum acquisition module, which includes 7 wavelengths of LED light sources and corresponding photodetectors: Main wavelength group: 254 nm, 280 nm, 340 nm, 420 nm, for COD / ammonia nitrogen measurement; Reference wavelength group: 550 nm, 700 nm, 810 nm, for turbidity recognition and equipment monitoring; Integrated temperature sensor, water temperature T and pH electrode for obtaining environmental parameters.
[0026] The multi-wavelength spectrum acquisition module synchronously acquires spectrum signals of 7 wavelengths per second, denoted as [S_measured ( ),...,S_measured ( ). And the original signal is median filtered to remove impulse noise. Synchronize to get water temperature T, unit: ℃, pH value, current time t, unit: day, counted from the start of the probe.
[0027] Embodiment 2: Referring to Figure 2 A high turbidity sewage water quality probe adaptive multi-spectral correction method adopts the high turbidity sewage water quality probe adaptive multi-spectral correction system in Embodiment 1. The adaptive multi-spectral correction method includes the following steps: S1, spectrum acquisition: based on a multi-wavelength spectrum acquisition probe, measure the spectrum signals of multiple characteristic wavelengths by selecting a specific wavelength combination in the ultraviolet-visible spectrum range; S2, process the original spectrum signal to correct the characteristic wavelength signal under turbidity conditions; S3, calculate the attenuation compensation coefficient of the probe at different stages; S4, based on the characteristic wavelength signal corrected under turbidity conditions, combine the attenuation compensation coefficient to calculate the actual concentration of the target parameter; S5, based on the gradient boosting decision tree machine learning model, learn the historical measurement data to generate the predicted concentration, and dynamically fuse the predicted concentration with the actual concentration of the target parameter.
[0028] In S1, the specific wavelength combination includes a main wavelength group and a reference wavelength group. The main wavelength group is used to measure the characteristic wavelength of the target parameter, and the reference wavelength group is used for turbidity feature identification and device state monitoring. The target parameter includes COD and ammonia nitrogen.
[0029] The main wavelength group includes: for organic matter characteristic absorption; for aromatic compound absorption; for ammonia nitrogen compound absorption; for color reaction absorption; The reference wavelength group includes: for low absorption area reference wavelength; for near-infrared reference wavelength; for turbidity characteristic wavelength.
[0030] In S2, the step of processing the original spectrum signal to correct the characteristic wavelength signal under turbidity conditions is: extract the reference wavelength signal: T ( = S_measured (810 nm), T ( = S_measured (700 nm).
[0031] By historical calibration data (S_calibrated (t), t ~ ) and current time t, the correction coefficients [a, b, g] are fitted by least squares method: Objective function minimization:∑(S_calibrated (t) - (S_measured (t) - a T (t) - b T (t) - g))²; In the formula: S_calibrated (t) is the spectrum signal of the known standard solution. Correct the main wavelength signal: For each main wavelength , calculate the corrected signal:
[0032] ; In the formula: S_corrected (t, λ) represents the spectrum signal of the corrected characteristic wavelength ; S_measured (t, λ) represents the measured spectrum signal of the characteristic wavelength ; S_turbidity (t, λ) represents the spectrum signal of the turbidity characteristic wavelength; a (t) represents the adaptive correction coefficient, which is determined by the following formula: ; S_calibrated (t) is the historical calibration data, and t is the probe use time. ; to is the historical calibration data, and t is the probe use time.
[0033] In S3, the attenuation compensation coefficient of the probe at different stages is calculated by adopting the probe performance attenuation model, and the probe performance attenuation model is: ; In the formula: a (t) represents the probe performance attenuation coefficient at time t; a0 represents the initial rapid attenuation parameter; a1 represents the long-term slow attenuation parameter; This represents the decay rate constant.
[0034] For example: if The calculation shows that: ; .
[0035] The formula for calculating the attenuation compensation coefficient of the probe is: .
[0036] In S4, an adaptive multivariate correction algorithm is used to calculate the actual concentration of the target parameter. The adaptive multivariate correction algorithm is as follows: ; In the formula: The concentration of the target parameter; : This is the weighting coefficient related to turbidity B; This is a time-dependent bias term.
[0037] Among them, the weighting coefficient Determined by the piecewise turbidity function: ; In the formula: Represents the linear coefficients of the first segment; Indicates the linear coefficients of the second segment; Represents the linear coefficients of the third segment; The coefficient of the quadratic term represents the range of extremely high turbidity. This represents the intercept of the first segment; This represents the intercept of the second segment; This represents the intercept of the third segment. The intercept representing the extremely high turbidity range; The coefficient of the first term represents the extremely high turbidity range; This indicates the turbidity value of the current water sample; This indicates the turbidity threshold for the first segment; This indicates the turbidity threshold for the second segment; This indicates the turbidity threshold of the third segment.
[0038] When using this method, the first step is to determine the turbidity segment and calculate the current turbidity B. This can be obtained through a reference wavelength ratio or an independent turbidity meter. Here, we assume B = 200 NTU, which falls within the range of... ≤B< Interval.
[0039] The weight coefficient formula is selected again, and In the interval, the second segment formula is used: ; The preset =0.01, =-0.5, corresponding to the weight of =254 nm, then .
[0040] The base concentration is calculated again by an adaptive multivariate correction algorithm: ; Assume, a bias term that increases linearly over time; Then:
[0041] In S5, the machine learning model based on gradient boosting decision trees is: ; In the formula: represents the target parameter concentration of the final output; represents the concentration calculated by the base algorithm; represents the concentration predicted by the machine learning model; represents a dynamic fusion coefficient, ranging from 0 to 1; The input feature vector X of the machine learning model is defined as: ; In the formula: T represents the water temperature; pH represents the pH value; represents the specific wavelength ratio, ; wherein, represents , the main wavelength group for organic matter characteristic absorption; represents , the main wavelength group for aromatic compound absorption.
[0042] The input feature quantity is constructed:
[0043] Assuming , X contains 11 feature values.
[0044] The machine learning model output C_ML =58.5 mg / L, and the dynamic fusion coefficient η=0.7 (adjusted in real time by error): =0.7×57.25 + (1-0.7)×58.5 = 57.625 mg / L.
[0045] Following S5, an adaptive parameter update mechanism is also included. This mechanism performs periodic self-checks and parameter updates. The adaptive parameter update mechanism updates the weights of the adaptive multivariate correction algorithm based on the most recent historical data. The formula for calculating the weight update amount is: ; In the formula: Indicates the amount of weight update; : Indicates the learning rate; : Indicates based on recent historical data The error gradient.
[0046] In this embodiment, periodic self-testing is performed once daily, calibrated using a standard solution, such as COD=60 mg / L, to obtain the measured values. Error from the true value:
[0047] Update the weight coefficients using gradient descent. : ; For example, for corresponding ,like ,but After the update .
[0048] Finally, output and store the results: Output final concentration (e.g., 57.625 mg / L) to the terminal device.
[0049] Store current data ( Spectral signals, environmental parameters, time t, etc., are uploaded to the historical database. This is used for subsequent parameter updates.
[0050] The above description is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.
Claims
1. A high turbidity sewage water quality probe adaptive multi-spectral correction system, characterized in that, The system comprises: a multi-wavelength spectrum acquisition module for simultaneously measuring spectrum signals of multiple characteristic wavelengths; a turbidity feature recognition and separation module for correcting the characteristic wavelength signals under turbidity conditions according to the measured spectrum signals; an adaptive multivariate correction module for calculating the actual concentration of a target parameter based on the corrected characteristic wavelength signals under turbidity conditions; a machine learning enhanced prediction module for learning historical measurement data based on a machine learning algorithm, generating a predicted concentration, and dynamically fusing the predicted concentration with the result obtained by the adaptive multivariate correction module; an adaptive parameter updating module for updating the adaptive multivariate correction algorithm weight based on recent historical data and feeding back the information to the turbidity feature recognition and separation module and the adaptive multivariate correction module; a probe aging compensation module for calculating attenuation compensation coefficients of probes at different stages and feeding back the attenuation compensation coefficients to the adaptive multivariate correction module.
2. A high turbidity sewage water quality probe self-adaptive multi-spectral correction method, characterized in that, The system comprises: S1, spectrum acquisition: based on a multi-wavelength spectrum acquisition probe, simultaneously measuring spectrum signals of multiple characteristic wavelengths by selecting a specific wavelength combination in the ultraviolet-visible spectrum range; S2, processing the original spectrum signals to correct the characteristic wavelength signals under turbidity conditions; S3, calculating attenuation compensation coefficients of probes at different stages; S4, calculating the actual concentration of a target parameter based on the corrected characteristic wavelength signals under turbidity conditions and combining the attenuation compensation coefficients; S5, learning historical measurement data based on a gradient boosting decision tree machine learning model, generating a predicted concentration, and dynamically fusing the predicted concentration with the actual concentration of the target parameter.
3. The fan pitch compensation damping control method for ultra-low frequency oscillation suppression according to claim 2, characterized in that, In S1, the specific wavelength combination includes a main wavelength group and a reference wavelength group, the main wavelength group is used to measure the characteristic wavelengths of the target parameter, and the reference wavelength group is used for turbidity feature recognition and device state monitoring; the target parameter includes COD and ammonia nitrogen.
4. The fan pitch compensation damping control method for ultra-low frequency oscillation suppression according to claim 3, characterized in that, The main wavelength group comprises: for organic matter specific absorption; , for aromatic compound absorption; for ammonia nitrogen compound absorption; , the color reaction is absorbed.
5. The method for fan pitch compensation damping control of ultra-low frequency oscillation suppression according to claim 3, characterized in that, The reference wavelength group comprises: , for low absorption region reference wavelength; , for near infrared reference wavelengths; , for turbidity characteristic wavelength.
6. The fan pitch compensation damping control method for ultra-low frequency oscillation suppression according to claim 2, characterized in that, In S2, a turbidity feature recognition and separation algorithm is used to correct the characteristic wavelength signals under turbidity conditions, and the turbidity feature recognition and separation algorithm is: ； In S3, an probe performance attenuation model is used to calculate the attenuation compensation coefficients of probes at different stages, and the probe performance attenuation model is: a spectral signal representing the corrected characteristic wavelength of the light representing the measured characteristic wavelength of the spectral signal; a spectral signal indicative of a turbidity characteristic wavelength; denotes the adaptive correction factor, determined by the following equation: ； to is the historical calibration data, t is the probe usage time.
7. The draft compensation damping control method for ultra-low frequency oscillation suppression according to claim 2, wherein, In S4, an adaptive multivariate correction algorithm is used to calculate the actual concentration of a target parameter, and the adaptive multivariate correction algorithm is: ； In S5, the gradient boosting decision tree machine learning model is: denotes the probe performance decay coefficient at time t; denotes the initial fast decay parameter; denotes a long-term slow decay parameter; represents the decay rate constant; In S5, the gradient boosting decision tree machine learning model is: 。 8. The method for fan pitch compensation damping control of ultra-low frequency oscillation suppression according to claim 2, characterized in that, The input feature vector X of the machine learning model is defined as: ； In S5, the gradient boosting decision tree machine learning model is: concentration of a target parameter; : weight coefficient related to turbidity B; is a time-dependent bias term; where the weight coefficient determined by the turbidity segmentation function: ； The adaptive parameter updating mechanism is periodically self-checked and updated, and the adaptive parameter updating mechanism updates the adaptive multivariate correction algorithm weight based on recent historical data, and the calculation formula of the weight update amount is: linear coefficient representing the first segment; linear coefficient representing the second segment; linear coefficient representing the third segment; quadratic coefficient representing the very high turbidity interval; intercept representing the first segment; intercept representing the second segment; intercept representing the third segment, intercept representing the very high turbidity interval; linear coefficient representing the very high turbidity interval; turbidity value of the current water sample; turbidity threshold of the first segment; turbidity threshold of the second segment; turbidity threshold of the third segment.
9. The method for fan pitch compensation damping control of ultra-low frequency oscillation suppression according to claim 2, characterized in that, In S5, the gradient boosting decision tree machine learning model is: ； represents the target parameter concentration of the final output; Concentration representing the base algorithm calculation; represents a concentration predicted by the machine learning model; denotes the dynamic fusion coefficient, ranging from [0, 1]; ； T represents water temperature; pH represents pH value; represents a specific wavelength ratio, ; wherein, represents , a main wavelength group for organic matter characteristic absorption; represents , a main wavelength group for aromatic compound absorption.
10. The draft compensation damping control method for ultra-low frequency oscillation suppression according to claim 2, characterized in that, ； denotes the weight update quantity; : denotes the learning rate; : indicates the error gradient based on recent historical data .