Water quality on-line monitoring system based on full spectrum analysis

Abstract

The invention relates to the field of water quality monitoring, and discloses a water quality on-line monitoring system based on full spectrum analysis, which comprises a full spectrum water quality analysis reference module, an equipment consistency correction module, a full spectrum water quality analysis derivative module and a quality monitoring module, wherein the reference module collects full-spectrum data through reference equipment to construct a water quality analysis reference model; the correction module uses the ResNet-C neural network to correct the spectral data difference of different devices; the derivative module combines the correction data and the reference model to realize water quality index prediction; the quality monitoring module calculates a quality evaluation coefficient through a multi-dimensional index and triggers early warning; the method is based on the Lambert-Beer theorem and the neural network architecture, solves the problems of high reagent consumption and poor equipment consistency in the traditional monitoring technology, and is suitable for real-time online monitoring of the surface water quality.

Description

Technical Field

[0001] The present invention relates to the field of water quality monitoring, and particularly to an on-line water quality monitoring system based on full-spectrum analysis. Background Art

[0002] On-line monitoring of surface water quality, as an important technical means for water environment management, provides key data support and decision-making basis for pollution prevention and control and water quality improvement. Traditional on-line water quality monitoring is mainly based on chemical analysis methods. Although it strictly follows the laboratory standard detection process and has high data accuracy, there are obvious technical limitations: on the one hand, it requires a large amount of chemical reagents and takes a long time for a single detection; on the other hand, the equipment maintenance frequency is high, not only the operation cost remains high, but also there may be secondary pollution of the water environment caused by the use of reagents.

[0003] In recent years, the full-spectrum on-line water quality monitoring technology based on UV-Vis (ultraviolet-visible) spectroscopy has received extensive attention due to its significant technical advantages. This technology has the characteristics of zero reagent consumption, real-time monitoring, low maintenance requirements, etc., and is developing rapidly in the field of surface water monitoring. However, the existing technology still faces important challenges: due to the limitations of hardware performance, there are significant differences in the spectral scanning results of different devices for the same water sample. The above technical defects lead to the need for each monitoring device to establish an independent water quality analysis model through a large number of surface water samplings and CMA (China Metrology Accreditation) certification tests, which greatly increases the production and application costs of the devices and restricts the large-scale promotion of this technology. Summary of the Invention

[0004] The purpose of the present invention is to provide an on-line water quality monitoring system based on full-spectrum analysis to solve the above technical problems.

[0005] The purpose of the present invention can be achieved by the following technical solutions: An on-line water quality monitoring system based on full-spectrum analysis, comprising: a full-spectrum water quality analysis reference module, a device consistency correction module, and a full-spectrum water quality analysis derivative module; The full-spectrum water quality analysis reference module is used to construct a water quality analysis reference model based on the full-spectrum data of water samples collected by a reference device; The device consistency correction module is used to correct the spectral data differences of different devices to make the spectral data of the device to be corrected match that of the reference device; The full-spectrum water quality analysis derivative module is used to combine the corrected spectral data with the reference model to realize the prediction of water quality indicators.

[0006] As a further technical solution, the full-spectrum water quality analysis reference module includes: Surface water sample collection unit, which is used to collect surface water samples with different pollution levels, including excellent water quality, slightly polluted water quality, moderately polluted water quality and severely polluted water quality. Water quality detection unit, which is used to detect the standard water quality indicators of the collected surface water samples. The standard water quality indicator detection includes chemical oxygen demand, ammonia nitrogen, total phosphorus and total nitrogen. Spectral scanning unit, which is used to scan the surface water sample and pure water through a reference device to obtain the corresponding full-spectrum light intensity data. Absorbance calculation unit, which calculates the full-spectrum absorbance of the surface water sample based on Lambert-Beer's law. Reference model construction unit, which takes the full-spectrum absorbance data of the surface water sample as the model input and the water quality indicator concentration as the model output, and constructs a full-spectrum water quality analysis reference model based on the ResNet (Residual Network)-B neural network architecture.

[0007] As a further technical solution, the device consistency correction module includes: Standard solution preparation unit, which is used to prepare m groups of standard solutions with different concentrations. Spectral data collection unit, which is used to scan the standard solution through a reference device and the device to be corrected respectively to obtain the corresponding full-spectrum light intensity data of the standard solution. Data conversion unit, which is used to convert the full-spectrum light intensity data of the standard solution of the device to be corrected from a one-dimensional array with a length of n 2 to a two-dimensional array of n×n. Calibration model construction unit, which takes the two-dimensional array of the full-spectrum light intensity data of the standard solution of the device to be corrected as the model input and the one-dimensional array of the full-spectrum light intensity data of the standard solution of the reference device as the model output, and constructs a full-spectrum device consistency calibration model based on the ResNet-C neural network architecture.

[0008] As a further technical solution, the full-spectrum water quality analysis derivative module includes: Spectral data collection unit, which is used to perform full-spectrum scanning on the surface water to be measured and pure water through the device to be corrected to obtain the corresponding full-spectrum light intensity data. Data correction unit, which is used to convert the full-spectrum light intensity data of the water sample of the device to be corrected into a two-dimensional array and then input it into the device consistency correction model to obtain the corrected full-spectrum light intensity data. Absorbance calculation unit, which calculates the full-spectrum absorbance of the corrected surface water sample based on Lambert-Beer's law. Water quality prediction unit, which is used to input the calculated full-spectrum absorbance data of the surface water sample into the full-spectrum water quality analysis reference model and output the predicted water quality indicator concentration value.

[0009] As a further technical solution, the system further includes: A quality monitoring module for real-time monitoring of the reliability of the model prediction results.

[0010] As a further technical solution, the quality monitoring module includes: A data verification unit for obtaining the detection results of the same sample, comparing them with the predicted water quality index concentration values, and calculating the absolute error and relative error; A model performance evaluation unit for performing performance tests on the full-spectrum water quality analysis benchmark model and the full-spectrum equipment consistency correction model using an independent verification dataset at a set period to obtain a first performance evaluation index F1, a second performance evaluation index F2, and a third performance evaluation index F3; A confidence calculation unit for generating a confidence score C for each predicted water quality index concentration value based on a Bayesian neural network, where the value range of the confidence score is 0-1; Example: Bayesian neural network: calculates the uncertainty of the predicted value through the Monte Carlo dropout method and outputs the confidence score C. For example, C = 0.9 means that the predicted value has a 90% probability of falling within ±10% of the true value; A quality analysis unit for substituting the relative error K, the first performance evaluation index F1, the second performance evaluation index F2, the third performance evaluation index F3, and the confidence score C into the formula: Calculate to obtain the quality evaluation coefficient ; is the relative error threshold; where is the correlation index, ; is the standardized th performance evaluation index, represents the actual value of the th performance evaluation index, , are respectively the lower limit value and the upper limit value of the th performance evaluation index, , are weight coefficients used to adjust the importance of the relative error K and the performance evaluation index combination term, and the value range is preset by the system, is the confidence adjustment coefficient used to amplify or reduce the effect intensity of the confidence score C on the final result, is the power coefficient of the th performance evaluation index used to highlight or weaken the influence of this index on the result; An early warning trigger unit that triggers an early warning when the quality evaluation coefficient does not belong to the preset early warning threshold range , .

[0011] As a further technical solution, the calculation method of the correlation index is as follows: ; represents the th and th performance evaluation indexes and The correlation weight between them ranges from [0, 1], is the th standardized performance evaluation index.

[0012] As a further technical solution, the acquisition methods of the first performance evaluation index F1, the second performance evaluation index F2, and the third performance evaluation index F3 are as follows: Calculated through the formula ; where represents the total number of times, is the concentration value of the true water quality index at the th time, is the predicted concentration value of the water quality index, is the average value of the true water quality index concentration.

[0013] As a further technical solution, the specific working process of the early warning trigger unit is as follows: When the quality evaluation coefficient , trigger the early warning of insufficient model prediction reliability; When the quality evaluation coefficient , trigger the early warning of data anomaly.

[0014] The early warning thresholds , can be dynamically adjusted according to historical monitoring data and business requirements.

[0015] Advantages of the present invention: (1) By collecting surface water samples with different pollution levels, covering excellent, mild, moderate, and severe pollution, and combining the core water quality indexes of chemical oxygen demand and ammonia nitrogen detected by standards, the present invention uses a neural network model to analyze the full-spectrum absorbance data, realizes high-precision prediction of water quality indexes, avoids the defects of long time consumption and large reagent consumption of traditional chemical analysis methods, and greatly improves the monitoring efficiency; (2) The present invention uses a neural network model to calibrate the spectral data differences of different devices, and realizes data matching between the device to be calibrated and the reference device through the spectral scan data of the standard solution. There is no need to establish a water quality analysis model independently for each device, significantly reducing the sampling, detection, and modeling costs during the device production and maintenance processes, and laying a foundation for the large-scale application of the full-spectrum monitoring technology; (3) The quality monitoring module in the present invention comprehensively considers multi-dimensional parameters such as relative error, model performance indicators, prediction confidence, and indicator correlation. It dynamically evaluates the reliability of the model prediction results through a preset algorithm, automatically triggers an early warning mechanism according to the threshold, and monitors the data quality in real time to ensure the accuracy and stability of the monitoring results, providing reliable data support for water environment management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 is a schematic diagram of the system structure of the present invention; Figure 2 is a schematic diagram of the system principle of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Please refer to Figure 1 - Figure 2 As shown, the present invention is an on-line water quality monitoring system based on full-spectrum analysis, including: a full-spectrum water quality analysis reference module, a device consistency correction module, and a full-spectrum water quality analysis derivative module; The full-spectrum water quality analysis reference module is used to construct a water quality analysis reference model based on the full-spectrum data of water samples collected by a reference device; The device consistency correction module is used to correct the spectral data differences of different devices so that the spectral data of the device to be corrected matches that of the reference device; The full-spectrum water quality analysis derivative module is used to combine the corrected spectral data with the reference model to realize the prediction of water quality indicators.

[0020] The full-spectrum water quality analysis reference module includes: A surface water sample collection unit for collecting surface water samples with different pollution levels, including excellent water quality, slightly polluted water quality, moderately polluted water quality, and severely polluted water quality; Sampling point setting: covering surface water monitoring sections (such as river sections, lake vertical lines), and determining the sampling depth according to the HJ495-2009 standard; Sampling frequency: at least 1 time per day, and encrypted to 1 time per hour during the pollution outbreak period; Water sample type: including Class I (excellent), Class II-III (slightly polluted), Class IV (moderately polluted), Class V, and inferior Class V (severely polluted) water quality, covering all categories of the Surface Water Environment Quality Standard (GB3838-2002); A water quality detection unit for detecting water quality indicators of surface water samples collected according to CMA standards; the standard water quality indicator detection includes chemical oxygen demand, ammonia nitrogen, total phosphorus, and total nitrogen; A spectral scanning unit for scanning surface water samples and pure water through a reference device to obtain corresponding full-spectrum light intensity data; An absorbance calculation unit for calculating the full-spectrum absorbance of surface water samples based on the Lambert-Beer theorem; A reference model construction unit for constructing a full-spectrum water quality analysis reference model based on the ResNet-B neural network architecture, with the full-spectrum absorbance data of surface water samples as the model input and the water quality indicator concentration as the model output.

[0021] The device consistency correction module includes: A standard solution preparation unit for preparing m groups of standard solutions with different concentrations; standard substances: Humic acid (simulating organic pollution, concentration range 1 - 50 mg / L); Potassium nitrate (simulating nitrogen pollution, concentration range 0.1 - 10 mg / L); Potassium dihydrogen phosphate (simulating phosphorus pollution, concentration range 0.01 - 1 mg / L); Turbidity standard solution (Formazin polymer, turbidity range 1 - 100 NTU); A spectral data acquisition unit for scanning the standard solutions through a reference device and the device to be corrected respectively to obtain the corresponding full-spectrum light intensity data of the standard solutions; A data conversion unit for converting the full-spectrum light intensity data of the standard solutions of the device to be corrected from a one-dimensional array of length n 2 to a two-dimensional array of n×n; A correction model construction unit for constructing a full-spectrum device consistency correction model based on the ResNet-C neural network architecture, with the two-dimensional array of the full-spectrum light intensity data of the standard solutions of the device to be corrected as the model input and the one-dimensional array of the full-spectrum light intensity data of the standard solutions of the reference device as the model output.

[0022] The full-spectrum water quality analysis derivative module includes: A spectral data acquisition unit for performing full-spectrum scanning on the surface water to be measured and pure water through the device to be corrected to obtain the corresponding full-spectrum light intensity data; A data correction unit for converting the full-spectrum light intensity data of the water samples of the device to be corrected into a two-dimensional array and then inputting it into the device consistency correction model to obtain the corrected full-spectrum light intensity data; An absorbance calculation unit for calculating the full-spectrum absorbance of the corrected surface water samples based on the Lambert-Beer theorem; A water quality prediction unit for inputting the calculated full-spectrum absorbance data of surface water samples into a full-spectrum water quality analysis benchmark model and outputting the predicted water quality index concentration values.

[0023] In this embodiment, the full-spectrum water quality analysis benchmark module: for the full-spectrum benchmark device, using a large amount of surface water sample data with different water quality conditions, based on the ResNet-B neural network architecture, constructs a full-spectrum water quality analysis benchmark model; because it focuses on the construction of the full-spectrum water quality analysis model of the benchmark device, it avoids the high cost of constructing a full-spectrum water quality analysis model for each device separately; by adopting the ResNet-B neural network architecture, that is, moving the downsampling of the residual branch of the standard ResNet neural network architecture to the subsequent 3×3 convolution, a large amount of loss of the full-spectrum information of the water sample is avoided, thus improving the accuracy of water quality prediction.

[0024] The device consistency correction module: using the full-spectrum light intensity data of the standard solution on the benchmark device and the device to be corrected, based on the ResNet-C neural network architecture, constructs a full-spectrum device consistency correction model; aiming at the light intensity data of the water sample rather than the absorbance data, it avoids the spectral scanning of pure water and the absorbance calculation of the water sample, and the operation is simpler; using the standard solution to replace the surface water sample avoids the labor cost of outdoor sampling; by adopting the ResNet-C neural network architecture, that is, replacing the 7×7 convolution kernel in the stem_block of the standard ResNet neural network architecture with three 3×3 convolution kernels, the number of parameters and the amount of calculation are significantly reduced, and it is beneficial to the feature extraction of the full-spectrum of the water sample; through the application of deep learning algorithms, full-spectrum device consistency correction is realized, breaking through the hardware technology limitations of the full spectrum.

[0025] The full-spectrum water quality analysis derivative module: connecting in series the full-spectrum device consistency correction model and the full-spectrum water quality analysis benchmark model to realize full-spectrum surface water quality analysis based on the device to be corrected; through the "correction + benchmark" mode, the full-spectrum water quality analysis benchmark model can be easily adapted to each device; it reduces the product R & D cost of the full-spectrum water quality online monitoring device, reduces the product R & D complexity, improves the device calibration efficiency, and thus reduces the industrial application cost of the full-spectrum water quality online monitoring device.

[0026] The system further includes: A quality monitoring module for monitoring the reliability of the model prediction results in real time.

[0027] The quality monitoring module includes: A data verification unit for obtaining the test results of the same sample, comparing them with the predicted water quality index concentration values, and calculating the absolute error and relative error; for example: Comparison frequency: At least 5 samples are selected every week and sent to a third-party CMA laboratory for testing, covering different pollution levels; Error calculation: Absolute Error (AE) = |Laboratory value - Predicted value|; Relative Error (K) = |Laboratory value - Predicted value| / Laboratory value × 100%, when the laboratory value is 0, K = 0.

[0028] The model performance evaluation unit performs performance tests on the full-spectrum water quality analysis benchmark model and the full-spectrum equipment consistency correction model using an independent validation dataset at a set period, obtaining the first performance evaluation index F1, the second performance evaluation index F2, and the third performance evaluation index F3; The confidence calculation unit generates a confidence score C for each predicted water quality index concentration value based on a Bayesian neural network, and the value range of the confidence score is 0 - 1; The quality analysis unit substitutes the relative error K, the first performance evaluation index F1, the second performance evaluation index F2, the third performance evaluation index F3, and the confidence score C into the formula: Calculate to obtain the quality evaluation coefficient ; Among them, is the correlation index, ; is the th performance evaluation index after standardization, represents the actual value of the th performance evaluation index, , are respectively the lower limit value and the upper limit value of the th performance evaluation index, , are weight coefficients used to adjust the importance of the relative error K and the performance evaluation index combination term, and the value range is preset by the system, is the confidence adjustment coefficient used to amplify or reduce the effect intensity of the confidence score C on the final result, is the th power coefficient of the performance evaluation index, used to highlight or weaken the influence of this index on the result, is the relative error threshold; , can be determined through historical data training, such as grid search and genetic algorithms, to avoid subjectivity in manual setting. For example, in a scenario where errors are given priority attention, can be set to increase the weight of the relative error. Index sensitivity adjustment: The power coefficient allows "penalizing" or "rewarding" a single index; for example, if F2 is crucial for water quality safety, , which significantly increases the quality assessment coefficient when F2 is large , forced triggering of warning; dynamic adjustment of confidence: The parameter amplifies or reduces the effect of confidence C; when When C=0.5, the low confidence results will be significantly reduced. Even if the error is small, it may trigger an early warning, which is suitable for high-risk monitoring scenarios, such as drinking water sources.

[0029] Early warning trigger unit, when the quality assessment coefficient Does not fall within the preset warning threshold range [ , ], trigger an early warning; The preset warning threshold range is determined by the historical quantile method, specifically: Data preparation; Collect at least 3 months of historical data on quality assessment coefficients (G), exclude data from abnormal periods such as equipment failures and manual intervention, and retain valid data from normal operation, recorded as set S.

[0030] Calculate the quantile threshold: The first threshold ( ): Take the 5% quantile (P5) of S, that is, only 5% of the normal data is less than this value; the second threshold ( ): Take the 95% quantile (P95) of S, that is, only 5% of the normal data are greater than this value.

[0031] In this embodiment, the data verification unit, the model performance evaluation unit, and the confidence calculation unit are used to realize the whole chain quality monitoring from data collection, model training to prediction results, avoid the one-sidedness of single indicator evaluation, and realize the whole process reliability monitoring; at the same time, the quality evaluation coefficient is linked with the preset threshold to ensure that when the model performance degrades or the data is abnormal, such as when the prediction error exceeds the limit or the model is overfitted, the warning is triggered in time to buy troubleshooting time for the operation and maintenance personnel, avoid misleading decision-making with erroneous data, and realize the timeliness of the warning mechanism; the confidence score C is used to convert the uncertainty of the prediction result into a quantifiable indicator in the range of 0-1, which is convenient for users to intuitively judge the reliability of the data, improve the transparency and credibility of the monitoring system, and realize the quantification of data credibility; in the formula , integrating relative error K, performance evaluation indicators F1, F2, F3, confidence C and correlation indicators , which not only reflects the degree of deviation between the prediction result and the true value, but also considers the overall performance of the model, the prediction credibility and the logical relationship between the indicators, avoids misjudgment in a single dimension, and realizes a comprehensive evaluation in multiple dimensions; through the weight coefficient , and power coefficient , according to the monitoring requirements of different water quality indicators, such as priority control pollutants, the importance of each factor can be dynamically adjusted to enhance the adaptability of the system to complex monitoring scenarios; standardized processing can eliminate the dimensional differences of different indicators and ensure the comparability of each parameter in the formula; the confidence level C amplifies or reduces the influence of the comprehensive error through exponential operation, which conforms to the logic that the results with high confidence level should be more credible.

[0032] The calculation method of the correlation index is as follows: ; represents the th and the th performance evaluation indicators and The correlation weight between them ranges from [0, 1]. is the th performance evaluation indicator after standardization.

[0033] In this embodiment, a method for obtaining the correlation index is provided, that is , through the correlation weight and the standardized index difference , the degree of association between different performance indicators is quantified; when the indicators are highly correlated (such as and ≈ ), the value approaches 0 to avoid repeated evaluation; when the indicators are strongly independent, such as or and have significant differences, the value increases, indicating that multiple indicators need to be considered comprehensively; the correlation weight can be set by industry experts according to water quality monitoring experience. For example, the correlation between F1 and F3 is usually higher than that between F2 and F3, making the mathematical model more in line with the actual business needs and avoiding the blindness of pure data-driven; taking the average of the summation results, that is, dividing by 3, can control the value in the interval of [0, 2] (because ∈[0, 1], the maximum absolute difference is 1, and the total difference of 3 pairs of indicators is at most 6, and the average is 2), preventing extreme values from overly interfering with the quality evaluation coefficient.

[0034] The acquisition methods of the first performance evaluation indicator F1, the second performance evaluation indicator F2, and the third performance evaluation indicator F3 are as follows: Calculated by the formula ; Among them, represents the total number of times, is the th true water quality indicator concentration value, is the predicted concentration value of the water quality index, is the average value of the actual water quality index concentration.

[0035] In this embodiment, a method for obtaining the first performance evaluation index F1, the second performance evaluation index F2, and the third performance evaluation index F3 is provided. By calculating the linear weighted mean error, which is sensitive to small errors and suitable for reflecting the overall deviation degree of the predicted value; amplifying the weight of extreme errors through squaring operations to highlight the robustness of the model to outliers; evaluating the model interpretability to reflect the overall fitting degree between the predicted value and the actual value ( = 1 represents a perfect fit).

[0036] The specific working process of the warning trigger unit is as follows: When the quality evaluation coefficient is satisfied, a warning of insufficient reliability of the model prediction is triggered; When the quality evaluation coefficient is satisfied, a warning of data anomaly is triggered.

[0037] The warning thresholds 、 can be dynamically adjusted according to historical monitoring data and business requirements.

[0038] It should be noted that the calculation formulas and each parameter participating in the operation in the present invention are all pre - processed by dimensionless processing, and the process of dimensionless processing is well - known in the industry and will not be described herein.

[0039] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the present invention application should still fall within the scope covered by the patent of the present invention.

Claims

1. An on-line water quality monitoring system based on full-spectrum analysis, characterized in that, Including: A full-spectrum water quality analysis benchmark module, an equipment consistency correction module, and a full-spectrum water quality analysis derivative module; The full-spectrum water quality analysis benchmark module is used to construct a water quality analysis benchmark model based on the full-spectrum data of water samples collected by a benchmark device; The equipment consistency correction module is used to correct the spectral data differences of different devices so that the spectral data of the device to be corrected matches that of the benchmark device; The full-spectrum water quality analysis derivative module is used to combine the corrected spectral data with the benchmark model to realize the prediction of water quality indicators.

2. The on-line water quality monitoring system based on full-spectrum analysis according to claim 1, characterized in that, The full-spectrum water quality analysis benchmark module includes: A surface water sample collection unit for collecting surface water samples with different pollution levels; A water quality detection unit for detecting standard water quality indicators of the collected surface water samples; the standard water quality indicator detection includes chemical oxygen demand, ammonia nitrogen, total phosphorus, and total nitrogen; A spectral scanning unit for scanning surface water samples and pure water through a benchmark device to obtain corresponding full-spectrum light intensity data; An absorbance calculation unit for calculating the full-spectrum absorbance of surface water samples based on the Lambert-Beer theorem; A benchmark model construction unit for constructing a full-spectrum water quality analysis benchmark model based on the ResNet-B neural network architecture with the full-spectrum absorbance data of surface water samples as the model input and the water quality indicator concentration as the model output.

3. The on-line water quality monitoring system based on full-spectrum analysis according to claim 1 or 2, characterized in that, The equipment consistency correction module includes: A standard solution preparation unit for preparing m groups of standard solutions with different concentrations; A spectral data collection unit for scanning the standard solutions through a benchmark device and a device to be corrected respectively to obtain the corresponding full-spectrum light intensity data of the standard solutions; A data conversion unit for converting the full-spectrum light intensity data of the standard solution of the device to be calibrated from a one-dimensional array with a length of n 2 into a two-dimensional array of n×n; A correction model construction unit for constructing a full-spectrum equipment consistency correction model based on the ResNet-C neural network architecture with the two-dimensional array of the full-spectrum light intensity data of the standard solutions of the device to be corrected as the model input and the one-dimensional array of the full-spectrum light intensity data of the standard solutions of the benchmark device as the model output.

4. The on-line water quality monitoring system based on full-spectrum analysis according to claim 3, characterized in that The full-spectrum water quality analysis derivative module includes: A spectral data collection unit for performing full-spectrum scanning on the surface water to be measured and pure water through the device to be corrected to obtain the corresponding full-spectrum light intensity data; A data correction unit for converting the full-spectrum light intensity data of the water samples of the device to be corrected into a two-dimensional array and then inputting it into the equipment consistency correction model to obtain the corrected full-spectrum light intensity data; An absorbance calculation unit for calculating the full-spectrum absorbance of the corrected surface water samples based on the Lambert-Beer theorem; A water quality prediction unit for inputting the calculated full-spectrum absorbance data of surface water samples into the full-spectrum water quality analysis benchmark model and outputting the predicted water quality indicator concentration value.

5. The online water quality monitoring system based on full-spectrum analysis according to claim 4, characterized in that, The system further includes: A quality monitoring module for monitoring the reliability of the model prediction results in real time.

6. The on-line water quality monitoring system based on full-spectrum analysis according to claim 5, characterized in that, The quality monitoring module includes: A data verification unit for obtaining the detection results of the same sample, comparing them with the predicted water quality indicator concentration values, and calculating the absolute error and relative error; The model performance evaluation unit performs performance tests on the full-spectrum water quality analysis benchmark model and the full-spectrum device consistency correction model using an independent verification dataset at a set period, and obtains the first performance evaluation index F1, the second performance evaluation index F2, and the third performance evaluation index F3; The confidence calculation unit generates a confidence score C for each predicted water quality index concentration value based on a Bayesian neural network, and the value range of the confidence score is 0-1; The quality analysis unit substitutes the relative error K, the first performance evaluation index F1, the second performance evaluation index F2, the third performance evaluation index F3, and the confidence score C into the formula: to calculate the quality evaluation coefficient ; wherein, is a correlation index, ; is the th performance evaluation index after standardization, represents the actual value of the th performance evaluation index, and are respectively the lower limit value and the upper limit value of the th performance evaluation index, and are weight coefficients, is a confidence adjustment coefficient, is the th power coefficient of the performance evaluation index, is a relative error threshold; Early warning trigger unit, when the quality assessment coefficient does not belong to the preset early warning threshold range , , it triggers an early warning.

7. The on-line water quality monitoring system based on full-spectrum analysis according to claim 6, characterized in that, The calculation method of the correlation index is as follows: ; represents the th and th performance evaluation indicators and the correlation weight between them, and the value range is [0, 1], is the th performance evaluation indicator after standardization.

8. The on-line water quality monitoring system based on full-spectrum analysis according to claim 6, characterized in that The acquisition methods of the first performance evaluation index F1, the second performance evaluation index F2, and the third performance evaluation index F3 are as follows: Calculated by the formula ; Among them, represents the total number of times, is the true water quality index concentration value for the th time, is the predicted water quality index concentration value, is the average value of the true water quality index concentration values.

9. The on-line water quality monitoring system based on full-spectrum analysis according to claim 6, characterized in that, The specific working process of the early warning trigger unit is as follows: When the quality assessment coefficient is reached, a warning of insufficient reliability of model prediction is triggered; When the quality assessment coefficient is reached, a data anomaly warning is triggered.

Images