Seawater quality AI fusion intelligent monitoring system and method based on full spectrum data

Abstract

The invention discloses a seawater quality AI fusion intelligent monitoring system and method based on full spectrum data, and relates to the technical field of seawater quality monitoring. The system comprises a full-spectrum data acquisition module, a water quality initial evaluation and characteristic wave band identification module, an environment adaptive wave band selection module, a weight dynamic distribution and anomaly detection module, a multi-dimensional data fusion and optimization module and an abnormal event response and report module. The system collects original spectral data through a spectral sensor, establishes a basic water quality model after preprocessing, dynamically selects an effective wave band according to real-time environmental parameters, allocates weights, outputs water quality parameters after fusing data of a multi-dimensional sensor, and finally responds to abnormal events and generates a report. The system realizes intelligent and real-time monitoring of seawater quality, and can effectively deal with sudden pollution events and capture water quality fluctuation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of seawater quality monitoring, in particular to an AI fusion intelligent seawater quality monitoring system and method based on full-spectrum data, which realizes real-time monitoring and abnormal response of seawater quality through spectral sensing, multi-dimensional data fusion and self-adaptive algorithm. BACKGROUND

[0002] In the current seawater quality monitoring field, the traditional monitoring method has significant technical bottlenecks. The laboratory sampling analysis based on chemical reagents requires manual high-frequency sampling, which faces the problems of low monitoring frequency and slow response speed, and it is difficult to capture sudden changes in water quality. Although the fixed monitoring equipment such as the national monitoring station has high accuracy, it has high purchase cost and needs to be installed on land, and the deployment flexibility is limited. Although the lightweight equipment such as the micro station has certain real-time performance, it lacks anti-pollution ability, and the monitoring accuracy is easily disturbed in complex seawater environments such as high turbidity and high salinity. In addition, the existing system relies on single spectrum or sensor data, lacks multi-dimensional data fusion mechanism, has limited dynamic recognition ability for seawater pollutants (such as organic matter and suspended particles), and the wave band selection strategy is fixed, which cannot be adjusted adaptively according to real-time environmental parameters, resulting in insufficient monitoring reliability in complex water quality scenarios.

[0003] Based on the above problems, the present application provides an AI fusion intelligent seawater quality monitoring system and method based on full-spectrum data to solve one or more of the above problems. SUMMARY

[0004] The purpose of the present application is to provide an AI fusion intelligent seawater quality monitoring system and method based on full-spectrum data to solve the problems in the prior art.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] The AI fusion intelligent monitoring system for seawater quality based on full-spectrum data comprises a full-spectrum data acquisition module, a water quality preliminary evaluation and characteristic waveband identification module, an environment adaptive waveband selection module, a weight dynamic distribution and anomaly detection module, a multi-dimensional data fusion and optimization module, and an abnormal event response and report module. The full-spectrum data acquisition module acquires original spectral data of seawater bodies in real time through a spectral sensor and records absorbance changes at different wavelengths. The water quality preliminary evaluation and characteristic waveband identification module establishes a basic water quality model based on chemometrics after preprocessing the spectral data and outputs a preliminary evaluation result. The adaptability of each waveband data is identified by comparing the model prediction value with the actual measurement value. The environment adaptive waveband selection module dynamically generates an effective waveband set and establishes a redundant waveband switching mechanism according to real-time water quality environmental parameters through a preset waveband interference evaluation rule. The weight dynamic distribution and anomaly detection module allocates evaluation weights to each waveband through statistical analysis, constructs a weighted fusion model, and sets a threshold rule to identify abnormal data points. The multi-dimensional data fusion and optimization module integrates data of other types of environmental sensors except the spectral sensor, corrects the spectral analysis result through a weighted fusion algorithm, and outputs water quality parameters. The abnormal event response and report module triggers an alarm and generates an analysis report containing pollution type speculation and impact range prediction when detecting water quality abnormalities.

[0007] The full-spectrum data acquisition module comprises a light source and light path transmission unit, a sample cell and transmitted light detection unit, and a data acquisition and transmission unit.

[0008] The light source and light path transmission unit generates light radiation of a specified wavelength range using a tunable laser and a continuous light source, and adjusts light intensity and irradiation mode through an optical control system. Then, the light source is divided into a reference light beam and a measurement light beam using double-beam compensation technology, and the complex light is decomposed into monochromatic light through a transmission grating.

[0009] The sample cell and transmitted light detection unit prevents pollutants from adhering through the hydrophobic and oleophobic coating characteristics on the surface of the sample cell, and removes pollutants relying on an automatic cleaning mechanism. Then, the light intensity change of transmitted light is detected using double-beam compensation technology, the optical signal is converted into an electrical signal through an arrayed photodetector, and the absorbance value at different wavelengths is recorded based on the absorbance formula A=KLC, where A is the absorbance, representing the degree of attenuation of light after passing through the water body; K is the absorption coefficient, determined by the optical properties of the water body material; L is the light path thickness in the sample cell; and C is the concentration of the water quality target parameter.

[0010] The data acquisition and transmission unit digitizes, acquires and processes the electrical signals output by the sample cell and the transmitted light detection unit, integrates spectral data and performs format regularization; supports network transmission and RS485 communication protocol, and transmits the collected absorbance data and other environmental parameters to the subsequent processing module in real time to provide basic data support for water quality preliminary evaluation and characteristic waveband identification.

[0011] The water quality preliminary evaluation and characteristic waveband identification module includes a spectral data preprocessing unit, a basic water quality model establishment unit, a preliminary evaluation result output unit and a waveband adaptability identification unit.

[0012] The spectral data preprocessing unit filters the original spectral data to eliminate noise interference, eliminates the influence of light intensity fluctuations through normalization algorithm, compensates the spectral data in combination with environmental parameters, removes the nonlinear lifting caused by non-dissolved substances, so that the spectrum only reflects the absorption characteristics of dissolved substances, and then converts the original absorbance data A into standard concentration value C under the unified optical path and substance characteristics by calculating the known absorption coefficient K and sample cell optical path L according to the absorbance formula A=KLC, eliminates the difference in optical path and environmental interference, so that the spectral data is comparable, and provides preprocessed data for the establishment of basic water quality model and the identification of characteristic waveband;

[0013] The basic water quality model establishment unit aligns the preprocessed spectral data with the actual water quality measurement data collected synchronously in time stamp and matches the characteristic parameters, fuses the physical relationship between absorbance and substance concentration in optical mechanism and the nonlinear fitting capability of AI algorithm, adopts partial least squares regression algorithm to screen the characteristic waveband of standardized spectral data, establishes the quantitative relationship between absorbance and water quality parameters, and realizes the preliminary prediction of water quality parameters after cross-validation;

[0014] The preliminary evaluation result output unit performs structured arrangement and format standardization processing on the water quality parameter prediction value generated by the basic water quality model establishment unit, forms a preliminary evaluation report containing parameter prediction values, measurement time stamp and spectral data source identification, and outputs the report to the waveband adaptability identification unit through a standardized data interface as the basic data of the comparison between model prediction value and actual measurement value, and supports sending the evaluation results to the subsequent abnormal event response and report module in the form of network transmission and RS485 communication protocol for storage and further processing;

[0015] The waveband adaptability recognition unit receives the water quality parameter prediction value of the preliminary evaluation result output unit and the actual measurement data collected synchronously, compares the prediction value of the basic water quality model with the actual measurement value waveband by waveband, calculates the prediction error corresponding to each waveband spectral data, the prediction error being the mean square error and the absolute error, then analyzes the representation ability of different wavebands to the water quality parameter according to the prediction error, determines the waveband with strong adaptability as a characteristic waveband for the error less than a preset threshold, and determines the waveband with weak adaptability conversely, thereby recognizing a characteristic waveband set having a key role in water quality monitoring and outputting the set to the environmental adaptive waveband selection module to provide a basis for dynamically generating an effective waveband set.

[0016] The environmental adaptive waveband selection module comprises an interference evaluation rule execution unit, an effective waveband dynamic generation unit and a redundant waveband switching mechanism unit.

[0017] The interference evaluation rule execution unit obtains real-time spectral data and environmental parameters from the full-spectrum data collection module based on a preset waveband interference evaluation rule, analyzes the interference degree of each waveband to the water body substance by a chemometrics algorithm, and the analysis mode is specifically that the prediction error of each waveband under the same environmental parameter in the historical data is compared with a preset threshold to determine the adaptability of the waveband data, the waveband with an error less than the threshold is determined to have strong adaptability, and conversely, the waveband is determined to have weak adaptability.

[0018] The effective waveband dynamic generation unit screens the waveband with a prediction error less than a threshold from the full-spectrum range according to the interference evaluation result, and dynamically combines the waveband into an effective waveband set; this process fuses the optical mechanism and the nonlinear fitting ability of the AI algorithm, so that the waveband in the set can reflect the spectral characteristics of the current water quality parameter, and provides input data for the subsequent weighted fusion model.

[0019] The redundant waveband switching mechanism unit establishes a waveband replacement mechanism based on real-time interference evaluation: when the error of a waveband in the effective waveband set exceeds the threshold, the system automatically selects a standby waveband with the same type of spectral characteristics from a preset redundant waveband library for replacement, the redundant waveband library contains standby wavebands in the ultraviolet and visible regions, which are classified according to spectral feature similarity to maintain the continuity of spectral monitoring; the switching logic relies on the double-beam compensation technology and the absorbance formula A=KLC.

[0020] The weight dynamic allocation and anomaly detection module comprises a weight statistical allocation unit and an anomaly threshold detection unit.

[0021] The weight statistical distribution unit obtains the spectral data of the effective wave band set from the environment adaptive wave band selection module, and constructs a wave band weight matrix. Specifically, first, the spectral data is standardized to eliminate the dimensional differences of different wave bands; then, principal component analysis (PCA) is used to reduce the dimension of the standardized data, the number of principal components m is determined according to the Kaiser criterion, and the covariance matrix is calculated and eigenvalue decomposition is performed to obtain the principal components and the corresponding wave band load coefficients, wherein the load coefficient is a standardized eigenvector element, representing the linear correlation degree of the wave band and the principal component; then, the sum of the squares of the load coefficients of each wave band in all principal components is taken as the variance contribution ratio of the wave band, and the ratio of this value to the sum of the squares of all wave bands reflects the explanation weight of the wave band to the total variance of the spectral data; then, through normalization processing, the variance contribution ratio of each wave band is divided by the sum of the variance contribution ratios of all wave bands, so that the sum of the weight coefficients is 1, and finally the weight coefficients of each wave band are obtained and a wave band weight matrix is constructed, which is used for weighted calculation of the wave band data in the weighted fusion model;

[0022] The abnormal threshold detection unit performs the following operations on the weighted and fused spectral data: first, a sliding window is used to smooth the data to reduce the influence of random noise; then, the mean and standard deviation of the data in the window are calculated, and the mean ± 3 times the standard deviation is taken as the abnormal threshold range; when the data of a wave band exceeds this range, it is determined as an abnormal data point, and the time and wave band position of the abnormality are recorded; at the same time, the system counts the number of consecutive abnormal data points, and when the proportion of abnormal points exceeds a preset percentage of the total data, the abnormal early warning mechanism is triggered, and the standby wave band is started through the redundant wave band switching mechanism.

[0023] The multi-dimensional data fusion and optimization module includes a data integration unit, a weighted fusion calculation unit and an outlier suppression unit.

[0024] The data integration unit integrates the sensor data of conductivity, dissolved oxygen, pH, temperature, turbidity, chlorophyll a, water level and flow rate, and aligns the time stamp with the spectral data of the full-spectrum data acquisition module; the specific process is as follows: first, standardize the format of each sensor data to unify the data unit; then, through the RS485 communication protocol and network transmission protocol of the data acquisition and transmission unit, integrate the sensor data and the spectral data into a structured data set, so that the time of different types of data is consistent, and provide unified input for subsequent fusion calculation;

[0025] The weighted fusion calculation unit uses a weighted fusion algorithm with adjustable weight coefficients, and the calculation formula is as follows:

[0026]

[0027] wherein, The estimated value of the water quality parameter after fusion, y 光谱 The water quality parameter value obtained by spectral analysis, y 传感器 The water quality parameter value measured by the auxiliary sensor, w 光谱 And w 传感器 The weight coefficients of the spectral data and the sensor data, respectively, the weight coefficients are determined based on the information amount of the spectral data and the reliability test of the sensor data; in specific implementation, first, the weight is adjusted dynamically according to the water quality scene, and then the spectral data and the sensor data are weighted and summed through matrix operation, and the fused water quality parameter is output;

[0028] The abnormal value suppression unit suppresses data anomalies through validity verification rules and a sliding window average method; specifically: first, set the sensor data validity verification rules to eliminate abnormal measurement values exceeding the threshold; then, the effective data is smoothed by using the sliding window average method, and the calculation formula is as follows:

[0029]

[0030] Among them, The smoothed value at the i-th moment, y j The original data at the j-th moment in the window; when the sensor data exceeds the mean ± 3 times the standard deviation for 3 consecutive windows, the adaptive adjustment strategy is triggered, and the weight of other sensor data is temporarily increased, and the weight of the sensor is distributed to other sensors, which refers to all effective sensors in the data integration unit except the abnormal sensor, specifically including: conductivity sensor, dissolved oxygen sensor, pH sensor, temperature sensor, turbidity sensor, chlorophyll a sensor, water level sensor and flow rate sensor;

[0031] The method of weight distribution is as follows: the initial weight wi of each sensor is determined based on the information amount of the spectral data and the reliability test of the sensor; when the sensor A appears data anomaly of exceeding the mean ± 3 times the standard deviation for 3 consecutive windows, its weight w A Is distributed according to the current reliability score proportion of the remaining sensors, and the calculation formula is as follows:

[0032]

[0033] Among them, w i The newly added weight of sensor i, r i The real-time reliability score of sensor i, r j Indicates the real-time reliability score of each sensor without anomaly, the range is 0-100, which is calculated based on the stability of historical data and the validity check result of current data, and n is the number of non-anomalous sensors; in the reliability score, the stability of historical data is determined by the coefficient of variation CV of the sensor data in the past 24 hours, and the calculation formula of CV is as follows:

[0034]

[0035] Wherein, the standard deviation and the mean value are statistical values of the sensor monitoring data in the past 24 hours, the smaller the CV value, the higher the score, the current effectiveness is based on whether the data exceeds the threshold value, the score is 100 if it does not exceed, and it is deducted by 20 points for each time it exceeds until it is reduced to 0, and finally the abnormal sensor weight is redistributed according to the reliability score of each sensor.

[0036] The abnormal event response and reporting module comprises an abnormality detection unit, an alarm triggering unit and a report generation and output unit.

[0037] The abnormality detection unit obtains the weighted fused water quality parameter data from the weight dynamic allocation and abnormality detection module, and performs abnormality identification, specifically: first, according to the preset abnormal threshold of each parameter; Then calculate the real-time change rate of the parameter using a sliding window, when a parameter exceeds the threshold for 3 consecutive windows, it is determined as an abnormal event;

[0038] The alarm triggering unit triggers an alarm when the abnormality detection unit confirms an abnormality by the following process: first, determine the alarm level according to the type of abnormal parameter, the alarm level is divided into warning and emergency; Then, through network transmission and RS485 communication protocol, the alarm information is transmitted to the superior environmental protection detection center, the client and the DCS system; At the same time, start the on-site sound and light alarm device, the alarm duration is preset, until manual confirmation;

[0039] The report generation and output unit generates an analysis report based on the abnormal event data, specifically: first, by comparing the spectral characteristics of the abnormal parameters with the built-in benzene, toluene and phenol absorption spectrum library, combined with the algorithm model of the AIOP platform to infer the pollution type, and determine that the organic matter pollution and inorganic ion exceed the standard; Then, use the water flow velocity collected by the flow sensor and the water level parameter obtained by the water level sensor to calculate the pollution diffusion radius through the fluid dynamics model, the calculation formula is as follows:

[0040]

[0041] Wherein, R is the diffusion radius, v is the flow rate, t is the duration of the abnormality, C is the measured concentration, and C0 is the background concentration; Finally, the report content containing the pollution type, the influence range and the treatment suggestion is displayed in the form of COD real-time change graph and pollution diffusion thermodynamic graph on the ecological environment IEP platform large screen and mobile terminal APP, and the output format follows the established specification of visual large screen and Web management terminal.

[0042] The seawater quality AI fusion intelligent monitoring method based on full-spectrum data comprises the following steps:

[0043] S1. Real-time acquisition of raw spectral data of seawater, and collection of environmental parameters such as conductivity, dissolved oxygen, pH, temperature, turbidity, chlorophyll a, water level, and flow velocity to provide basic data for subsequent water quality assessment.

[0044] S2. The raw spectral data is filtered and normalized, and the influence of non-dissolved substances is compensated by environmental parameters. The data is then converted into standard concentration values. A basic water quality model is established based on chemometrics methods and AI algorithms. The predicted values ​​are compared with the actual measured values, and characteristic bands with strong adaptability are identified.

[0045] S3. Based on real-time water quality environmental parameters and according to the preset band interference assessment rules, bands with small prediction errors are selected from the full spectrum range and dynamically combined into an effective band set. When the error of a certain band in the effective band exceeds the threshold, a spare band of the same type is selected from the redundant band library to replace it, so as to ensure the continuity of spectral monitoring.

[0046] S4. Standardize the spectral data of the effective band set, use principal component analysis to calculate the variance contribution ratio of each band, construct a band weight matrix after normalization, use it for weighted fusion model, and perform sliding window smoothing on the weighted fused spectral data. Use the mean ± 3 times the standard deviation as the threshold to identify outlier data points.

[0047] S5. Integrate various environmental sensor data, align them with spectral data using timestamps and standardize their formats. Use a weighted fusion algorithm with adjustable weight coefficients to correct the spectral analysis results and output water quality parameters. At the same time, suppress data anomalies through validity verification rules and sliding window averaging.

[0048] S6. Obtain data from the weighted and fused water quality parameters, determine whether an abnormal event has occurred based on the preset abnormal threshold and real-time change rate, and if an abnormality is confirmed, determine the alarm level according to the abnormality type and transmit alarm information. At the same time, compare spectral characteristics with the built-in spectral library to infer the pollution type, calculate the pollution diffusion radius using parameters such as flow velocity and water level, and generate an analysis report containing the pollution type, the scope of impact and treatment suggestions, which is displayed in the form of charts on the designated platform.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] 1. Reagent-free monitoring and efficient maintenance: Spectroscopic measurement avoids secondary pollution caused by chemical reagents. Combined with a special coating and automatic cleaning function, it reduces equipment maintenance costs and enables long-term stable deployment.

[0051] 2. Dynamic band optimization and anti-interference: Through the environmental adaptive band selection module, effective bands are dynamically selected based on real-time parameters such as turbidity and salinity. Combined with the redundant band switching mechanism, the monitoring stability in highly polluted and highly fluctuating seawater environments is improved.

[0052] 3. Multi-dimensional data fusion and accuracy improvement: Integrate spectral data with parameters from multiple sensors such as conductivity and dissolved oxygen, correct spectral analysis results through weighted fusion algorithms, and combine AI model optimization to improve the measurement accuracy of complex water quality parameters (such as COD and total nitrogen);

[0053] 4. Real-time early warning and intelligent response: Through the abnormal event response module, based on dynamic thresholds and pollution diffusion models, it can quickly detect, alarm and predict the impact range of water quality anomalies, providing decision support for pollution emergency response;

[0054] 5. Intelligent and Adaptive Capabilities: The built-in AIOP algorithm platform supports local / remote model iteration and upgrades, and combined with principal component analysis and dynamic weight allocation, it enhances the system's adaptive learning capabilities to new pollutants or environmental changes. Attached Figure Description

[0055] Fig. 1 This is a system organization diagram of the seawater quality AI fusion intelligent monitoring system based on full-spectrum data of the present invention;

[0056] Fig. 2 This is a system workflow diagram of the seawater quality AI fusion intelligent monitoring system based on full-spectrum data of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Example: Figs. 1-2 As shown, the present invention provides a technical solution.

[0059] The AI fusion intelligent monitoring system for seawater quality based on full-spectrum data comprises a full-spectrum data acquisition module, a water quality preliminary evaluation and characteristic waveband identification module, an environment adaptive waveband selection module, a weight dynamic distribution and anomaly detection module, a multi-dimensional data fusion and optimization module, and an abnormal event response and report module. The full-spectrum data acquisition module acquires original spectral data of seawater bodies in real time through a spectral sensor and records absorbance changes at different wavelengths. The water quality preliminary evaluation and characteristic waveband identification module establishes a basic water quality model based on chemometrics after preprocessing the spectral data and outputs a preliminary evaluation result. The adaptability of each waveband data is identified by comparing the model prediction value with the actual measurement value. The environment adaptive waveband selection module dynamically generates an effective waveband set and establishes a redundant waveband switching mechanism according to real-time water quality environmental parameters through a preset waveband interference evaluation rule. The weight dynamic distribution and anomaly detection module allocates evaluation weights to each waveband through statistical analysis, constructs a weighted fusion model, and sets a threshold rule to identify abnormal data points. The multi-dimensional data fusion and optimization module integrates data of other types of environmental sensors except the spectral sensor, corrects the spectral analysis result through a weighted fusion algorithm, and outputs water quality parameters. The abnormal event response and report module triggers an alarm and generates an analysis report containing pollution type speculation and impact range prediction when detecting water quality abnormalities.

[0060] The full-spectrum data acquisition module comprises a light source and light path transmission unit, a sample cell and transmitted light detection unit, and a data acquisition and transmission unit.

[0061] The light source and light path transmission unit generates light radiation of a specified wavelength range using a tunable laser and a continuous light source, and adjusts light intensity and irradiation mode through an optical control system. Then, the light source is divided into a reference beam and a measurement beam using double-beam compensation technology, and the complex light is decomposed into monochromatic light through a transmission grating.

[0062] The sample cell and transmitted light detection unit prevents pollutants from adhering through the hydrophobic and oleophobic coating characteristics on the surface of the sample cell, and removes pollutants relying on an automatic cleaning mechanism. Then, the light intensity change of transmitted light is detected using double-beam compensation technology, the optical signal is converted into an electrical signal through an arrayed photodetector, and the absorbance value at different wavelengths is recorded based on the absorbance formula A=KLC, where A is the absorbance, representing the degree of attenuation of light after passing through the water body; K is the absorption coefficient, determined by the optical properties of the water body material; L is the light path thickness in the sample cell; and C is the concentration of the water quality target parameter.

[0063] The data acquisition and transmission unit digitizes, acquires and processes the electrical signals output by the sample cell and the transmitted light detection unit, integrates spectral data and performs format regularization; supports network transmission and RS485 communication protocol, and transmits the collected absorbance data and other environmental parameters to the subsequent processing module in real time to provide basic data support for water quality preliminary evaluation and characteristic waveband identification.

[0064] The water quality preliminary evaluation and characteristic waveband identification module includes a spectral data preprocessing unit, a basic water quality model establishment unit, a preliminary evaluation result output unit and a waveband adaptability identification unit.

[0065] The spectral data preprocessing unit filters the original spectral data to eliminate noise interference, eliminates the influence of light intensity fluctuations through normalization algorithm, compensates the spectral data in combination with environmental parameters, removes the nonlinear lifting caused by non-dissolved substances, so that the spectrum only reflects the absorption characteristics of dissolved substances, and then converts the original absorbance data A into standard concentration value C under the unified optical path and substance characteristics by calculating the known absorption coefficient K and sample cell optical path L according to the absorbance formula A=KLC, eliminates the difference in optical path and environmental interference, so that the spectral data is comparable, and provides preprocessed data for the establishment of basic water quality model and the identification of characteristic waveband;

[0066] The basic water quality model establishment unit aligns the preprocessed spectral data with the actual water quality measurement data collected synchronously in time stamp and matches the characteristic parameters, fuses the physical relationship between absorbance and substance concentration in optical mechanism and the nonlinear fitting capability of AI algorithm, adopts partial least squares regression algorithm to screen the characteristic waveband of standardized spectral data, establishes the quantitative relationship between absorbance and water quality parameters, and realizes the preliminary prediction of water quality parameters after cross-validation;

[0067] The preliminary evaluation result output unit performs structured arrangement and format standardization processing on the water quality parameter prediction value generated by the basic water quality model establishment unit, forms a preliminary evaluation report containing parameter prediction values, measurement time stamp and spectral data source identification, and outputs the report to the waveband adaptability identification unit through a standardized data interface as the basic data of the comparison between model prediction value and actual measurement value, and supports sending the evaluation results to the subsequent abnormal event response and report module in the form of network transmission and RS485 communication protocol for storage and further processing;

[0068] The waveband adaptability recognition unit receives the water quality parameter prediction value of the preliminary evaluation result output unit and the actual measurement data collected synchronously, compares the prediction value of the basic water quality model with the actual measurement value waveband by waveband, calculates the prediction error corresponding to each waveband spectral data, the prediction error being the mean square error and the absolute error, then analyzes the representation ability of different wavebands to the water quality parameter according to the prediction error, determines the waveband with strong adaptability as a characteristic waveband for the error less than a preset threshold, and determines the waveband with weak adaptability conversely, thereby recognizing a characteristic waveband set having a key role in water quality monitoring and outputting the set to the environmental adaptive waveband selection module to provide a basis for dynamically generating an effective waveband set.

[0069] The environmental adaptive waveband selection module comprises an interference evaluation rule execution unit, an effective waveband dynamic generation unit and a redundant waveband switching mechanism unit.

[0070] The interference evaluation rule execution unit obtains real-time spectral data and environmental parameters from the full-spectrum data collection module based on a preset waveband interference evaluation rule, analyzes the interference degree of each waveband to the water body substance by a chemometrics algorithm, and the analysis mode is specifically that the prediction error of each waveband under the same environmental parameter in the historical data is compared with a preset threshold to determine the adaptability of the waveband data, the waveband with an error less than the threshold is determined to have strong adaptability, and conversely, the waveband is determined to have weak adaptability.

[0071] The effective waveband dynamic generation unit screens the waveband with a prediction error less than a threshold from the full-spectrum range according to the interference evaluation result, and dynamically combines the waveband into an effective waveband set; this process fuses the optical mechanism and the nonlinear fitting ability of the AI algorithm, so that the waveband in the set can reflect the spectral characteristics of the current water quality parameter, and provides input data for the subsequent weighted fusion model.

[0072] The redundant waveband switching mechanism unit establishes a waveband replacement mechanism based on real-time interference evaluation: when the error of a waveband in the effective waveband set exceeds the threshold, the system automatically selects a standby waveband with the same type of spectral characteristics from a preset redundant waveband library for replacement, the redundant waveband library contains standby wavebands in the ultraviolet and visible regions, which are classified according to spectral feature similarity to maintain the continuity of spectral monitoring; the switching logic relies on the double-beam compensation technology and the absorbance formula A=KLC.

[0073] The weight dynamic allocation and anomaly detection module comprises a weight statistical allocation unit and an anomaly threshold detection unit.

[0074] The weighted statistical allocation unit obtains spectral data of the effective band set from the environmental adaptive band selection module and constructs a band weight matrix. Specifically: First, the spectral data is standardized to eliminate dimensional differences between different bands; then, principal component analysis (PCA) is used to reduce the dimensionality of the standardized data. According to the Kaiser criterion, principal components with eigenvalues ​​≥1 are selected, and the number of principal components m is determined. Then, by calculating the covariance matrix and performing eigenvalue decomposition, the loading coefficients of each principal component and its corresponding band are obtained. The loading coefficients are standardized eigenvector elements, representing the degree of linear correlation between the band and the principal component. Next, the sum of the squares of the loading coefficients of each band in all principal components is taken as the variance contribution ratio of that band. This value accounts for the proportion of the sum of the squares of all bands, reflecting the explanatory weight of the band on the total variance of the spectral data. Then, through normalization, the variance contribution ratio of each band is divided by the sum of the variance contribution ratios of all bands, so that the sum of the weight coefficients is 1. Finally, the weight coefficients of each band are obtained, and a band weight matrix is ​​constructed. This matrix is ​​used to perform weighted calculations on the data of each band in the weighted fusion model.

[0075] The anomaly threshold detection unit performs the following operations on the weighted fused spectral data: First, it uses a sliding window to smooth the data and reduce the impact of random noise; then, it calculates the mean and standard deviation of the data within the window, and uses the mean ± 3 times the standard deviation as the anomaly threshold range; when a certain band of data exceeds this range, it is determined to be an abnormal data point, and the time and band location of the anomaly are recorded; at the same time, the system counts consecutive abnormal data points, and when the proportion of abnormal points exceeds a preset percentage of the total data, it triggers an anomaly warning mechanism and starts a backup band through a redundant band switching mechanism.

[0076] The multi-dimensional data fusion and optimization module includes a data integration unit, a weighted fusion calculation unit, and an outlier suppression unit;

[0077] The data integration unit integrates sensor data such as conductivity, dissolved oxygen, pH, temperature, turbidity, chlorophyll a, water level, and flow rate, and aligns the timestamps with the spectral data from the full-spectrum data acquisition module. The specific process is as follows: First, the data from each sensor is standardized in format to unify the data units; then, through the RS485 communication protocol and network transmission protocol of the data acquisition and transmission unit, the sensor data and spectral data are integrated into a structured dataset, ensuring that the timestamps of different types of data are consistent, providing a unified input for subsequent fusion calculations.

[0078] The weighted fusion calculation unit adopts a weighted fusion algorithm with adjustable weight coefficients, and the calculation formula is as follows:

[0079]

[0080] in, The estimated values ​​of the merged water quality parameters, y 光谱 The water quality parameter values ​​obtained from spectral analysis, y 传感器 To assist the sensor in measuring water parameters, w 光谱 and w 传感器 The weighting coefficients are for spectral data and sensor data, respectively. The weighting coefficients are determined based on the information content of spectral data and the reliability test of sensor data. In the specific implementation, the weights are first dynamically adjusted according to the water quality scenario, and then the spectral data and sensor data are weighted and summed through matrix operations to output the fused water quality parameters.

[0081] The outlier suppression unit suppresses data anomalies through validity verification rules and a sliding window averaging method. Specifically: first, sensor data validity verification rules are set to remove outlier measurements that exceed the threshold; then, a sliding window averaging method is used to smooth the valid data, and the calculation formula is as follows:

[0082]

[0083] in, Let y be the smoothed value at time i. j This represents the raw data at time j within the window. When a sensor's data exceeds the mean ± 3 times the standard deviation for three consecutive windows, an adaptive adjustment strategy is triggered, temporarily increasing the weight of other sensor data and allocating the weight of this sensor to other sensors. Other sensors refer to all valid sensors in the data integration unit except for abnormal sensors, specifically including: conductivity sensor, dissolved oxygen sensor, pH sensor, temperature sensor, turbidity sensor, chlorophyll a sensor, water level sensor, and flow rate sensor.

[0084] The weighting method is as follows: the initial weight wi of each sensor is determined based on the spectral data information content and sensor reliability testing; when sensor A exhibits an anomaly where its data exceeds the mean ± 3 times the standard deviation for three consecutive windows, its weight wi is adjusted. A The remaining sensors are allocated according to their current reliability scores, calculated as follows:

[0085]

[0086] Among them, w i The added weights for sensor i, r i For the real-time reliability score of sensor i, r j The real-time reliability score for each sensor that is not abnormal ranges from 0 to 100. It is calculated based on the stability of historical data and the validity of current data, where n is the number of sensors that are not abnormal. In the reliability score, the stability of historical data is determined by the coefficient of variation (CV) of the sensor's data over the past 24 hours. The formula for calculating CV is as follows:

[0087]

[0088] Among them, the standard deviation and mean refer to the statistical values ​​of the sensor's monitoring data in the past 24 hours. The smaller the CV value, the higher the score. The current validity is based on whether the data exceeds the threshold. If it does not exceed the threshold, the score is 100. For each time it exceeds the threshold, 20 points are deducted until it drops to 0. Finally, the weight of abnormal sensors is redistributed based on the reliability scores of each sensor.

[0089] The abnormal event response and reporting module includes an abnormality detection unit, an alarm triggering unit, and a report generation and output unit;

[0090] The anomaly detection unit obtains the weighted and fused water quality parameter data from the weighted dynamic allocation and anomaly detection module, and performs anomaly identification. Specifically, it first uses the preset anomaly threshold for each parameter; then it uses a sliding window to calculate the real-time change rate of the parameter. When a parameter exceeds the threshold for three consecutive windows, it is determined to be an abnormal event.

[0091] When the anomaly detection unit confirms an anomaly, the alarm triggering unit triggers an alarm through the following process: First, the alarm level is determined according to the type of abnormal parameter, and the alarm level is divided into two levels: warning and emergency; then, the alarm information is transmitted to the superior environmental monitoring center, client and DCS system through network transmission and RS485 communication protocol; at the same time, the on-site audible and visual alarm device is activated, and the alarm duration is preset until manual confirmation.

[0092] The report generation and output unit generates an analysis report based on abnormal event data. Specifically: First, by comparing the spectral characteristics of abnormal parameters with the built-in absorption spectral libraries of benzene, toluene, and phenol, and combining the algorithm model of the AIOP platform, the pollution type is inferred, and it is determined to be organic pollution and excessive inorganic ions. Then, using the water flow velocity collected by the flow velocity sensor and the water level parameters obtained by the water level sensor, the pollution diffusion radius is calculated through a fluid dynamics model. The calculation formula is as follows:

[0093]

[0094] Where R is the diffusion radius, v is the flow velocity, t is the duration of the anomaly, C is the measured concentration, and C0 is the background concentration; finally, the report content, including the pollution type, the scope of impact, and treatment recommendations, will be displayed on the ecological environment IEP platform's large screen and mobile APP in the form of a real-time COD change map and a pollution diffusion heat map, and the output format will follow the established specifications of the visualization large screen and the web management terminal.

[0095] The AI-based intelligent monitoring method for seawater quality based on full-spectrum data includes the following steps:

[0096] S1. Real-time acquisition of raw spectral data of seawater, and collection of environmental parameters such as conductivity, dissolved oxygen, pH, temperature, turbidity, chlorophyll a, water level, and flow velocity to provide basic data for subsequent water quality assessment.

[0097] S2. The raw spectral data is filtered and normalized, and the influence of non-dissolved substances is compensated by environmental parameters. The data is then converted into standard concentration values. A basic water quality model is established based on chemometrics methods and AI algorithms. The predicted values ​​are compared with the actual measured values, and characteristic bands with strong adaptability are identified.

[0098] S3. Based on real-time water quality environmental parameters and according to the preset band interference assessment rules, bands with small prediction errors are selected from the full spectrum range and dynamically combined into an effective band set. When the error of a certain band in the effective band exceeds the threshold, a spare band of the same type is selected from the redundant band library to replace it, so as to ensure the continuity of spectral monitoring.

[0099] S4. Standardize the spectral data of the effective band set, use principal component analysis to calculate the variance contribution ratio of each band, construct a band weight matrix after normalization, use it for weighted fusion model, and perform sliding window smoothing on the weighted fused spectral data. Use the mean ± 3 times the standard deviation as the threshold to identify outlier data points.

[0100] S5. Integrate various environmental sensor data, align them with spectral data using timestamps and standardize their formats. Use a weighted fusion algorithm with adjustable weight coefficients to correct the spectral analysis results and output water quality parameters. At the same time, suppress data anomalies through validity verification rules and sliding window averaging.

[0101] S6. Obtain data from the weighted and fused water quality parameters, determine whether an abnormal event has occurred based on the preset abnormal threshold and real-time change rate, and if an abnormality is confirmed, determine the alarm level according to the abnormality type and transmit alarm information. At the same time, compare spectral characteristics with the built-in spectral library to infer the pollution type, calculate the pollution diffusion radius using parameters such as flow velocity and water level, and generate an analysis report containing the pollution type, the scope of impact and treatment suggestions, which is displayed in the form of charts on the designated platform.

[0102] Suppose that the nearshore waters of a coastal city are polluted with organic matter due to industrial wastewater discharge. The local environmental protection bureau has activated a seawater quality AI-integrated intelligent monitoring system based on full-spectrum data to monitor the seawater quality in the area in real time. The monitored seawater temperature ranges from 20-25℃, and the salinity is 30-35‰. The initial background water quality parameters are: CODcr 10mg / L, total nitrogen 1.2mg / L, and turbidity 5NTU. The system needs to quickly identify anomalies before the pollutants spread and generate an analysis report on the pollution type and its impact range.

[0103] I. Full-spectrum data acquisition and environmental parameter collection

[0104] The system acquires raw seawater spectral data in real time at a frequency of 10 minutes per acquisition via the spectral sensor of the full-spectrum data acquisition module. The light source and optical path transmission unit utilizes a tunable laser (wavelength range 190-1100nm) and a continuous light source. Dual-beam compensation technology splits the light source into a reference beam and a measurement beam, which are then decomposed into monochromatic light by a transmission grating. A hydrophobic and oleophobic coating on the sample cell surface effectively prevents contaminant adhesion, and an automatic cleaning mechanism is activated hourly. The data acquisition and transmission unit digitizes the electrical signals and simultaneously collects environmental parameters such as conductivity (25mS / cm), dissolved oxygen (6.5mg / L), pH (8.1), temperature (23℃), turbidity (5NTU), chlorophyll a (10μg / L), water level (2.5m), and flow velocity (0.3m / s), transmitting these parameters to subsequent modules via the RS485 protocol.

[0105] II. Spectral Data Preprocessing and Basic Water Quality Model Establishment

[0106] The water quality preliminary assessment and characteristic band identification module filters and normalizes the raw spectral data, and compensates for the influence of non-dissolved substances by incorporating parameters such as temperature and turbidity. It converts the raw absorbance into standard concentration values ​​using the absorbance formula A = KLC (assuming the sample cell optical path length L = 1 cm, and the absorption coefficient K is determined by the water's material properties). The basic water quality model building unit uses a partial least squares regression (PLS) algorithm to align the preprocessed spectral data with synchronously acquired actual measurement data (e.g., laboratory CODcr measurement of 10.5 mg / L, total nitrogen of 1.3 mg / L) with timestamps, constructing a quantitative relationship between absorbance and water quality parameters. Cross-validation shows that the model's prediction mean square error for CODcr is 0.3 mg / L, meeting the preliminary assessment requirements.

[0107] III. Characteristic Band Identification and Dynamic Band Selection

[0108] The band-adaptive identification unit compares the model's predicted values ​​with the actual measured values, calculating the mean square error and absolute error for each band. A preset error threshold of 0.05 (dimensionless) is used; bands with errors less than this threshold are identified as characteristic bands, such as 254nm in the ultraviolet region (corresponding to organic absorption), 450nm in the visible region (corresponding to turbidity), and 750nm in the near-infrared region (corresponding to chlorophyll a). The environmental adaptive band selection module analyzes the band interference level using a chemometric algorithm based on real-time parameters such as turbidity (5 NTU) and temperature (23℃), dynamically generating an effective band set {254nm, 450nm, 750nm}. When the error in a certain band (e.g., 450nm) exceeds the threshold (0.05) due to increased suspended particulate matter, the system selects a 480nm band with similar characteristics from the redundant band library to replace it, maintaining monitoring continuity.

[0109] IV. Dynamic Weight Allocation and Abnormal Data Detection

[0110] After standardizing the spectral data of the effective band set, the weighted statistical allocation unit uses Principal Component Analysis (PCA) for dimensionality reduction. Based on the Kaiser criterion (eigenvalue ≥ 1), the number of principal components is determined to be m = 2, and the sum of squares of the loading coefficients for each band in the principal components is calculated. Assuming the sum of squares of the loading coefficients for 254nm, 450nm, and 750nm are 0.6, 0.3, and 0.1 respectively, the normalized weight coefficients are [0.6, 0.3, 0.1], and a weight matrix is ​​constructed for the weighted fusion model. The anomaly threshold detection unit smooths the fused data using a sliding window (window size N = 5), and calculates the mean ± 3 times the standard deviation as the anomaly threshold. If the 750nm band data exceeds the threshold for three consecutive windows, a redundant band switching mechanism is triggered, replacing it with the backup 700nm band.

[0111] V. Multi-dimensional data fusion and outlier suppression

[0112] The data integration unit aligns sensor data such as conductivity and dissolved oxygen with the timestamps of spectral data, unifying the format to mg / L or dimensionless. The weighted fusion calculation unit uses a formula... Where the weight of the spectral data w 光谱 =0.7, sensor data weight w_sensor =0.3. Assuming the spectral analysis CODcr is 12 mg / L, the sensor measurement is 11.5 mg / L, and the estimated value after fusion is 11.85 mg / L. The outlier suppression unit sets the turbidity sensor effectiveness threshold to 50 NTU. When the turbidity data (e.g., 60 NTU) exceeds the mean ± 3 times the standard deviation for three consecutive windows at a certain moment, an adaptive adjustment strategy is triggered, allocating the sensor weight (originally 0.05) to sensors such as conductivity and dissolved oxygen according to the reliability score ratio of other sensors.

[0113] VI. Abnormal Event Response and Report Generation

[0114] The anomaly detection unit detected that the CODcr fusion value exceeded the preset threshold (10 mg / L, Class III water quality standard) for three consecutive windows (11.85 mg / L, 12.2 mg / L, and 12.5 mg / L), classifying it as an abnormal event. The alarm triggering unit determined the alarm level to be "emergency" based on the abnormal parameter type (CODcr exceeding the standard), transmitted the alarm information to the environmental monitoring center via GPRS network, and simultaneously activated the on-site audible and visual alarms. The report generation and output unit compared the abnormal spectral characteristics with the built-in benzene and toluene absorption spectral library, inferring that the pollution type was organic matter (benzene series) pollution. Using the flow velocity (0.3 m / s) and water level (2.5 m) parameters, the pollution diffusion radius was calculated using the formula for R. Assuming the anomaly duration t = 2 hours, the measured concentration C = 12.5 mg / L, and the background concentration C0 = 10 mg / L, the calculated diffusion radius R = 2700 m. The final report is presented on the Ecological and Environmental Protection IEP platform in the form of a real-time COD change map and a pollution diffusion heat map, including pollution type, impact range and emergency response recommendations.

[0115] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A seawater quality AI fusion intelligent monitoring system based on full-spectrum data, characterized in that: The system includes a full-spectrum data acquisition module, a water quality preliminary assessment and characteristic band identification module, an environmentally adaptive band selection module, a weighted dynamic allocation and anomaly detection module, a multi-dimensional data fusion and optimization module, and an anomaly event response and reporting module. The full-spectrum data acquisition module collects raw spectral data of seawater in real time using a spectral sensor, recording absorbance changes at different wavelengths. The water quality preliminary assessment and characteristic band identification module preprocesses the spectral data, establishes a basic water quality model based on chemometrics, and outputs preliminary assessment results. It then identifies the adaptability of each band by comparing the model's predicted values ​​with actual measured values. The environmentally adaptive band selection module dynamically generates an effective band set based on real-time water quality environmental parameters and preset band interference assessment rules, and establishes a redundant band switching mechanism. The weighted dynamic allocation and anomaly detection module assigns assessment weights to each band through statistical analysis, constructs a weighted fusion model, and then sets threshold rules to identify abnormal data points. The multi-dimensional data fusion and optimization module integrates environmental sensor data from sources other than spectral sensors, corrects the spectral analysis results using a weighted fusion algorithm, and outputs water quality parameters. The abnormal event response and reporting module triggers an alarm and generates an analysis report when it detects an abnormal water quality, including pollution type prediction and impact range prediction.

2. The seawater quality AI fusion intelligent monitoring system based on full-spectrum data according to claim 1, characterized in that: The full-spectrum data acquisition module includes a light source and optical path transmission unit, a sample cell and transmitted light detection unit, and a data acquisition and transmission unit. The light source and optical path transmission unit uses a tunable laser and a continuous light source to generate light radiation within a specified wavelength range, and adjusts the light intensity and illumination mode through an optical control system; then, the light source is divided into a reference beam and a measurement beam using dual-beam compensation technology, and polychromatic light is decomposed into monochromatic light through a transmission grating. The sample cell and the transmitted light detection unit prevent contaminant adhesion through the hydrophobic and oleophobic coating on the sample cell surface and remove contaminants using an automatic cleaning mechanism. Then, dual-beam compensation technology is used to detect changes in the intensity of the transmitted light. An array photodetector converts the optical signal into an electrical signal, and the absorbance values ​​at different wavelengths are recorded based on the absorbance formula A = KLC, where A is the absorbance, characterizing the degree of light attenuation after passing through the water; K is the absorption coefficient, determined by the optical properties of the water substance; L is the thickness of the light path within the sample cell; and C is the concentration of the target water quality parameter. The data acquisition and transmission unit digitally acquires and processes the electrical signals output by the sample cell and the transmitted light detection unit, integrates the spectral data and formats it; it supports network transmission and RS485 communication protocol, and transmits the acquired absorbance data and other environmental parameters to the subsequent processing module in real time, providing basic data support for water quality preliminary assessment and characteristic band identification.

3. The seawater quality AI fusion intelligent monitoring system based on full-spectrum data according to claim 1, characterized in that: The water quality preliminary assessment and characteristic band identification module includes a spectral data preprocessing unit, a basic water quality model establishment unit, a preliminary assessment result output unit, and a band adaptation identification unit. The spectral data preprocessing unit filters the raw spectral data to eliminate noise interference, eliminates the influence of light intensity fluctuations through a normalization algorithm, compensates for the spectral data in combination with environmental parameters, removes nonlinear uplift caused by non-dissolved substances, and ensures that the spectrum only reflects the absorption characteristics of dissolved substances. Then, based on the absorbance formula A = KLC, the raw absorbance data A is calculated using the known absorption coefficient K and the optical path length L of the sample cell, and converted into a standard concentration value C under uniform optical path length and material characteristics. This eliminates optical path differences and environmental interference, making the spectral data comparable and providing preprocessed data for the establishment of basic water quality models and the identification of characteristic bands. The basic water quality model building unit is based on chemometrics. It aligns the preprocessed spectral data with the synchronously collected actual water quality measurement data by timestamp and matches the characteristic parameters. By integrating the physical relationship between absorbance and substance concentration in the optical mechanism and the nonlinear fitting capability of the AI ​​algorithm, the partial least squares regression algorithm is used to screen the characteristic bands of the standardized spectral data to establish a quantitative relationship between absorbance and water quality parameters. After cross-validation, the preliminary prediction of water quality parameters is achieved. The preliminary assessment result output unit organizes and standardizes the predicted water quality parameters generated by the basic water quality model building unit, forming a preliminary assessment report containing the predicted values ​​of each parameter, measurement timestamps, and spectral data source identifiers. The report is then output to the band adaptability identification unit through a standardized data interface as the basic data for comparing the model predictions with the actual measurements. The unit also supports sending the assessment results to the subsequent abnormal event response and reporting module for storage and further processing via network transmission and RS485 communication protocol. The band adaptability identification unit receives the water quality parameter prediction values ​​from the preliminary evaluation result output unit and the synchronously collected actual measurement data. It compares the predicted values ​​of the basic water quality model with the actual measurement values ​​band by band, and calculates the prediction error corresponding to the spectral data of each band. The prediction error is the mean square error and the absolute error. Then, it analyzes the characterization ability of different bands for water quality parameters based on the prediction error. For bands with errors less than a preset threshold, they are identified as characteristic bands with strong adaptability, and vice versa. This identifies a set of characteristic bands that play a key role in water quality monitoring, and outputs this set to the environmental adaptive band selection module to provide a basis for dynamically generating an effective band set.

4. The seawater quality AI fusion intelligent monitoring system based on full-spectrum data according to claim 1, characterized in that: The environmental adaptive band selection module includes an interference assessment rule execution unit, an effective band dynamic generation unit, and a redundant band switching mechanism unit. The interference assessment rule execution unit, based on the preset band interference assessment rules, obtains real-time spectral data and environmental parameters from the full-spectrum data acquisition module, and analyzes the degree of interference of each band with water substances through chemometric algorithms. The specific analysis method is to compare the prediction error of each band under the same environmental parameters in historical data with the preset threshold to determine the adaptability of the band data. Bands with errors less than the threshold are considered to have strong adaptability, and vice versa. The effective band dynamic generation unit selects bands with prediction errors less than a threshold from the full spectrum range based on the interference assessment results, and dynamically combines them into an effective band set. This process integrates the nonlinear fitting capabilities of optical mechanisms and AI algorithms, enabling the bands in the set to reflect the spectral characteristics of the current water quality parameters and providing input data for the subsequent weighted fusion model. The redundant band switching mechanism unit establishes a band replacement mechanism based on real-time interference assessment: when the error of a certain band in the effective band set exceeds the threshold, the system automatically selects a spare band with the same spectral characteristics from the preset redundant band library for replacement. The redundant band library includes spare bands in the ultraviolet and visible regions, classified according to the similarity of spectral characteristics to maintain the continuity of spectral monitoring; the switching logic relies on dual-beam compensation technology and absorbance formula A=KLC.

5. The seawater quality AI fusion intelligent monitoring system based on full-spectrum data according to claim 1, characterized in that: The weight dynamic allocation and anomaly detection module includes a weight statistical allocation unit and an anomaly threshold detection unit; The weighted statistical allocation unit obtains spectral data of the effective band set from the environmental adaptive band selection module and constructs a band weight matrix. Specifically: First, the spectral data is standardized to eliminate dimensional differences between different bands; then, principal component analysis (PCA) is used to reduce the dimensionality of the standardized data. According to the Kaiser criterion, principal components with eigenvalues ​​≥1 are selected, and the number of principal components m is determined. Then, by calculating the covariance matrix and performing eigenvalue decomposition, the loading coefficients of each principal component and its corresponding band are obtained. The loading coefficients are standardized eigenvector elements, representing the degree of linear correlation between the band and the principal component. Next, the sum of squares of the loading coefficients of each band in all principal components is taken as the variance contribution ratio of that band. The proportion of this value to the sum of squares of all bands reflects the explanatory weight of the band on the total variance of the spectral data. Then, through normalization, the variance contribution ratio of each band is divided by the sum of the variance contribution ratios of all bands, so that the sum of the weight coefficients is 1. Finally, the weight coefficients of each band are obtained and a band weight matrix is ​​constructed. This matrix is ​​used to perform weighted calculations on the data of each band in the weighted fusion model. The anomaly threshold detection unit performs the following operations on the weighted fused spectral data: First, it uses a sliding window to smooth the data and reduce the impact of random noise; then, it calculates the mean and standard deviation of the data within the window, and uses the mean ± 3 times the standard deviation as the anomaly threshold range; when a certain band of data exceeds this range, it is determined to be an abnormal data point, and the time and band location of the anomaly are recorded; at the same time, the system counts consecutive abnormal data points, and when the proportion of abnormal points exceeds a preset percentage of the total data, it triggers an anomaly warning mechanism and starts a backup band through a redundant band switching mechanism.

6. The seawater quality AI fusion intelligent monitoring system based on full-spectrum data according to claim 1, characterized in that: The multi-dimensional data fusion and optimization module includes a data integration unit, a weighted fusion calculation unit, and an outlier suppression unit; The data integration unit integrates sensor data such as conductivity, dissolved oxygen, pH, temperature, turbidity, chlorophyll a, water level, and flow rate, and aligns the timestamps with the spectral data from the full-spectrum data acquisition module. The specific process is as follows: First, the data from each sensor is standardized in format to unify the data units; then, through the RS485 communication protocol and network transmission protocol of the data acquisition and transmission unit, the sensor data and spectral data are integrated into a structured dataset, ensuring that the timestamps of different types of data are consistent, providing a unified input for subsequent fusion calculations. The weighted fusion calculation unit adopts a weighted fusion algorithm with adjustable weight coefficients, and the calculation formula is as follows: in, The estimated values ​​of the merged water quality parameters, y 光谱 The water quality parameter values ​​obtained from spectral analysis, y 传感器 To assist the sensor in measuring water parameters, w 光谱 and w 传感器 The weighting coefficients are for spectral data and sensor data, respectively. The weighting coefficients are determined based on the information content of spectral data and the reliability test of sensor data. In the specific implementation, the weights are first dynamically adjusted according to the water quality scenario, and then the spectral data and sensor data are weighted and summed through matrix operations to output the fused water quality parameters. The outlier suppression unit suppresses data anomalies through validity verification rules and a sliding window averaging method. Specifically: first, sensor data validity verification rules are set to remove outlier measurements that exceed the threshold; then, a sliding window averaging method is used to smooth the valid data, and the calculation formula is as follows: in, Let y be the smoothed value at time i. j This represents the raw data at time j within the window. When a sensor's data exceeds the mean ± 3 times the standard deviation for three consecutive windows, an adaptive adjustment strategy is triggered, temporarily increasing the weight of other sensor data and allocating the weight of this sensor to other sensors. Other sensors refer to all valid sensors in the data integration unit except for abnormal sensors, specifically including: conductivity sensor, dissolved oxygen sensor, pH sensor, temperature sensor, turbidity sensor, chlorophyll a sensor, water level sensor, and flow rate sensor. The weighting method is as follows: the initial weight wi of each sensor is determined based on the spectral data information content and sensor reliability testing; when sensor A exhibits an anomaly where its data exceeds the mean ± 3 times the standard deviation for three consecutive windows, its weight wi is adjusted. A The remaining sensors are allocated according to their current reliability scores, calculated as follows: Among them, w i The added weights for sensor i, r i For the real-time reliability score of sensor i, r j The real-time reliability score for each sensor that is not abnormal ranges from 0 to 100. It is calculated based on the stability of historical data and the validity of current data, where n is the number of sensors that are not abnormal. In the reliability score, the stability of historical data is determined by the coefficient of variation (CV) of the sensor's data over the past 24 hours. The formula for calculating CV is as follows: Among them, the standard deviation and mean refer to the statistical values ​​of the sensor's monitoring data in the past 24 hours. The smaller the CV value, the higher the score. The current validity is based on whether the data exceeds the threshold. If it does not exceed the threshold, the score is 100. For each time it exceeds the threshold, 20 points are deducted until it drops to 0. Finally, the weight of abnormal sensors is redistributed based on the reliability scores of each sensor.

7. The seawater quality AI fusion intelligent monitoring system based on full-spectrum data according to claim 1, characterized in that: The abnormal event response and reporting module includes an abnormality detection unit, an alarm triggering unit, and a report generation and output unit; The anomaly detection unit obtains the weighted and fused water quality parameter data from the weighted dynamic allocation and anomaly detection module, and performs anomaly identification. Specifically, it first uses the preset anomaly threshold for each parameter; then it uses a sliding window to calculate the real-time change rate of the parameter. When a parameter exceeds the threshold for three consecutive windows, it is determined to be an abnormal event. When the anomaly detection unit confirms an anomaly, the alarm triggering unit triggers an alarm through the following process: First, the alarm level is determined according to the type of abnormal parameter, and the alarm level is divided into two levels: warning and emergency; then, the alarm information is transmitted to the superior environmental monitoring center, client and DCS system through network transmission and RS485 communication protocol; at the same time, the on-site audible and visual alarm device is activated, and the alarm duration is preset until manual confirmation. The report generation and output unit generates an analysis report based on abnormal event data. Specifically: First, by comparing the spectral characteristics of abnormal parameters with the built-in absorption spectral libraries of benzene, toluene, and phenol, and combining the algorithm model of the AIOP platform, the pollution type is inferred, and it is determined to be organic pollution and excessive inorganic ions. Then, using the water flow velocity collected by the flow velocity sensor and the water level parameters obtained by the water level sensor, the pollution diffusion radius is calculated through a fluid dynamics model. The calculation formula is as follows: Where R is the diffusion radius, v is the flow velocity, t is the duration of the anomaly, C is the measured concentration, and C0 is the background concentration; finally, the report content, including the pollution type, the scope of impact, and treatment recommendations, will be displayed on the ecological environment IEP platform's large screen and mobile APP in the form of a real-time COD change map and a pollution diffusion heat map, and the output format will follow the established specifications of the visualization large screen and the web management terminal.

8. A seawater quality AI fusion intelligent monitoring method based on full-spectrum data, applied to the seawater quality AI fusion intelligent monitoring system based on full-spectrum data as described in any one of claims 1-7, characterized in that: Includes the following steps: S1. Real-time acquisition of raw spectral data of seawater, and collection of environmental parameters such as conductivity, dissolved oxygen, pH, temperature, turbidity, chlorophyll a, water level, and flow velocity to provide basic data for subsequent water quality assessment. S2. The raw spectral data is filtered and normalized, and the influence of non-dissolved substances is compensated by environmental parameters. The data is then converted into standard concentration values. A basic water quality model is established based on chemometrics methods and AI algorithms. The predicted values ​​are compared with the actual measured values, and characteristic bands with strong adaptability are identified. S3. Based on real-time water quality environmental parameters and according to the preset band interference assessment rules, bands with small prediction errors are selected from the full spectrum range and dynamically combined into an effective band set. When the error of a certain band in the effective band exceeds the threshold, a spare band of the same type is selected from the redundant band library to replace it, so as to ensure the continuity of spectral monitoring. S4. Standardize the spectral data of the effective band set, use principal component analysis to calculate the variance contribution ratio of each band, construct a band weight matrix after normalization, use it for weighted fusion model, and perform sliding window smoothing on the weighted fused spectral data. Use the mean ± 3 times the standard deviation as the threshold to identify outlier data points. S5. Integrate various environmental sensor data, align them with spectral data using timestamps and standardize their formats. Use a weighted fusion algorithm with adjustable weight coefficients to correct the spectral analysis results and output water quality parameters. At the same time, suppress data anomalies through validity verification rules and sliding window averaging. S6. Obtain data from the weighted and fused water quality parameters, determine whether an abnormal event has occurred based on the preset abnormal threshold and real-time change rate, and if an abnormality is confirmed, determine the alarm level according to the abnormality type and transmit alarm information. At the same time, compare spectral characteristics with the built-in spectral library to infer the pollution type, calculate the pollution diffusion radius using parameters such as flow velocity and water level, and generate an analysis report containing the pollution type, the scope of impact and treatment suggestions, which is displayed in the form of charts on the designated platform.

Images