A floating water pollution detection system based on multi-parameter sensor integration

Abstract

The application belongs to the technical field of sensors, and particularly relates to a floating type water pollution detection system based on multi-parameter sensor integration, which comprises a sensor module, a signal acquisition module, a crosstalk compensation module, a sensor health state evaluation module, a blind source separation processing module, a data fusion and output module, a remote communication and data management module and a self-cleaning maintenance module, the sensor module integrates an electrochemical sensor and an optical sensor; the crosstalk compensation module quantitatively strips electromagnetic crosstalk between sensors by using an independent component analysis algorithm and a recursive least square method; the sensor health state evaluation module realizes online estimation of zero point drift and sensitivity attenuation through an extended Kalman filter; the application effectively solves the technical problem that multi-sensor signal crosstalk and sensor aging are difficult to distinguish and compensate in the prior art, and improves the accuracy and long-term reliability of water quality monitoring data.

Description

Technical Field

[0001] This invention belongs to the field of sensor technology, specifically a floating water pollution detection system based on multi-parameter sensor integration. Background Technology

[0002] Floating water pollution detection systems are core equipment for real-time monitoring of the aquatic environment. They can realize continuous and automated monitoring of various water quality parameters, providing key data support for water quality assessment, pollution early warning and environmental governance. The acquisition accuracy and signal reliability of various types of sensors directly determine the detection performance and data credibility of the overall water quality monitoring system.

[0003] Existing floating water pollution detection systems integrate electrochemical and optical sensors on the same floating monitoring platform, sharing power supply and data acquisition links. They simultaneously complete the signal acquisition and data output of various water quality parameters, enabling routine and unmanned water quality monitoring operations in outdoor water bodies and meeting basic water environment monitoring needs.

[0004] Existing monitoring technologies neglect electromagnetic crosstalk and chemical cross-coupling issues between multiple sensors, cannot eliminate nonlinear systematic measurement errors caused by environmental variables, and have difficulty distinguishing between actual water quality fluctuations, sensor signal crosstalk, and data changes caused by equipment aging. Under long-term outdoor operation conditions, the uncertainty of monitoring data is significantly amplified, which can easily lead to water quality misjudgment or early warning failure, affecting the accuracy of water quality monitoring.

[0005] Therefore, the present invention provides a floating water pollution detection system based on multi-parameter sensor integration. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by the present invention to solve its technical problem is as follows: The floating water pollution detection system based on multi-parameter sensor integration of the present invention includes a sensor module, a signal acquisition module, a crosstalk compensation module, a sensor health status assessment module, a blind source separation and processing module, a data fusion and output module, a remote communication and data management module, and a self-cleaning and maintenance module.

[0008] The sensor module is used to sense water quality parameters. The module includes at least one set of electrochemical sensors and at least one set of optical sensors. The electrochemical sensors include a pH electrode, a dissolved oxygen electrode, and an ammonia nitrogen electrode. The optical sensors include a turbidity sensor and a chemical oxygen demand sensor. Both the electrochemical sensors and the optical sensors are electrically connected to the signal acquisition module.

[0009] The signal acquisition module is electrically connected to the sensor module and is configured to amplify, filter, and perform analog-to-digital conversion on the raw signals output by each sensor. The module has a built-in multi-channel synchronous sampling circuit to ensure the timing consistency of the signals from each sensor.

[0010] The crosstalk compensation module is electrically connected to the signal acquisition module and is configured to separate the mixed signals. The module first uses the independent component analysis algorithm to initially separate the electrochemical sensor signal and the optical sensor signal. Then, it constructs the transfer function matrix between the sensors to characterize the signal coupling relationship, uses the recursive least squares method to estimate the transfer function parameters online, completes the quantitative stripping of the crosstalk signal, and remaps the corrected output signals of each sensor to the corresponding physical parameter space.

[0011] The sensor health status assessment module is communicatively connected to the signal acquisition module and the crosstalk compensation module. It is configured to perform real-time monitoring and degradation diagnosis of sensor performance. This module establishes a time-varying characteristic model for each sensor and extracts multi-dimensional health features such as signal amplitude, noise level, response time, and temperature coefficient. Combined with the extended Kalman filter algorithm, it realizes online estimation of sensor zero drift and sensitivity decay. When the sensor performance index is detected to deviate from the preset threshold range, the module automatically triggers the corresponding calibration or replacement reminder.

[0012] The blind source separation processing module interacts with the crosstalk compensation module and the sensor health status assessment module. It is configured to separate the real water quality change component and the sensor aging component from the compensated sensor output signal. The module models the sensor output signal as a linear mixture of the real water quality signal and the aging noise signal, and uses a kurtosis-based blind source separation algorithm to achieve statistically independent separation of the two. The physical rationality of the separation result is further verified by time-domain recursive analysis and frequency-domain feature comparison.

[0013] The data fusion and output module is communicatively connected to the blind source separation and processing module and is configured to perform weighted fusion processing on the separated real water quality change components. The module dynamically adjusts the weight coefficients of each sensor in the fusion process according to the health status assessment results of each sensor, and uses an adaptive weighted fusion algorithm to generate the final water quality monitoring results and output the results to a display terminal or remote monitoring platform.

[0014] The remote communication and data management module is communicatively connected to the data fusion and output module and is configured to realize the remote transmission of monitoring data and the reception of instructions. The module sends the processed monitoring data to the remote monitoring center through a wireless transmission protocol, and at the same time receives and executes calibration instructions, parameter configuration instructions and data query instructions issued by the monitoring center.

[0015] The self-cleaning maintenance module is communicatively connected to the sensor module and the main control module, and is configured to periodically clean and maintain the sensor probe. The module removes biological deposits and particulate deposits on the sensor surface by combining high-pressure water rinsing and ultrasonic vibration, ensuring the sensor's response sensitivity and measurement accuracy.

[0016] Preferably, the electrochemical sensor and the optical sensor in the sensor module operate using a time-division multiplexing strategy, staggering the working windows of each sensor in time to reduce electromagnetic crosstalk generated when the sensors work simultaneously.

[0017] Preferably, the independent component analysis algorithm used in the crosstalk compensation module is an improved algorithm based on the principle of maximizing information. It improves the separation performance of the algorithm under non-Gaussian signal conditions by introducing a nonlinear activation function, and combines principal component analysis to reduce feature dimensionality and reduce computational complexity.

[0018] Preferably, the sensor health status assessment module adopts a subspace decomposition method based on a multi-signal classification algorithm to project the sensor output signal into a high-dimensional subspace. By calculating the signal subspace projection error, a quantitative diagnosis of sensor performance degradation is achieved. When the projection error exceeds a preset threshold, an aging alarm is automatically triggered.

[0019] Preferably, the blind source separation algorithm used in the blind source separation processing module is a fast fixed-point algorithm, which achieves rapid separation of source signals by iteratively optimizing higher-order statistics, and uses time-domain recursive analysis and frequency-domain power spectral density comparison to ensure the physical rationality of the separation results.

[0020] Preferably, the data fusion and output module uses an adaptive weighted fusion algorithm to dynamically adjust the weights based on the health status assessment results of each sensor. When a sensor is determined to be in an aging state, its weight coefficient in the fusion process is automatically reduced. The fused water quality parameters are converted into a comprehensive water quality level output through a water quality index model.

[0021] Preferably, the remote communication and data management module uses a low-power wide area network transmission protocol to realize the remote transmission of monitoring data. The data transmission frequency can be configured according to monitoring needs, and it supports remote firmware upgrades and remote calibration of sensor parameters.

[0022] Preferably, the cleaning cycle and duration of the self-cleaning maintenance module can be adaptively adjusted according to the water quality. When the sensor response sensitivity decreases or the noise level increases, the cleaning program is automatically triggered. During the cleaning process, signal acquisition is automatically shielded to avoid data interference.

[0023] The present invention has the following beneficial effects: 1. By setting up a crosstalk compensation module and a blind source separation processing module, this invention achieves coordinated suppression of crosstalk from multiple sensors and sensor aging components, solving the technical problem of difficulty in distinguishing and compensating for signal crosstalk and sensor aging in traditional technologies. This significantly improves the system accuracy and long-term reliability of water quality monitoring data, providing high-quality data support for water quality assessment and pollution early warning.

[0024] 2. By setting up a sensor health status assessment module, this invention realizes online monitoring and degradation diagnosis of sensor performance. It can identify aging phenomena such as electrode surface contamination, sensitive film performance degradation, and zero-point drift in real time, solving the technical problem that sensor aging cannot be identified online in traditional technologies. It avoids false alarms or missed alarms caused by sensor failure and improves the data reliability of the monitoring system.

[0025] 3. By setting up a self-cleaning maintenance module, this invention realizes automated cleaning and maintenance of the sensor probe, which can regularly remove biological attachments and particulate deposits, maintain the sensor's response sensitivity and measurement accuracy, solve the maintenance problem of floating water pollution detection systems under long-term unattended operation conditions, extend the service life of the sensor, and reduce system operation and maintenance costs.

[0026] 4. By setting up a remote communication and data management module, this invention realizes real-time remote transmission of monitoring data and sensor status information. Maintenance personnel can grasp the water quality and sensor health status without being physically present on site. It supports remote calibration and parameter configuration functions, solves the maintenance difficulties of floating water pollution detection systems deployed in remote water areas, and improves the intelligent management level of the system. Attached Figure Description

[0027] The invention will now be further described with reference to the accompanying drawings.

[0028] Figure 1 This is a structural block diagram of a floating water pollution detection system based on multi-parameter sensor integration according to the present invention; Figure 2 This is a schematic diagram of the process by which the data fusion and output module dynamically adjusts the weighting coefficients based on the health status of the sensors and generates water quality monitoring results in this invention. Detailed Implementation

[0029] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0030] Example 1: like Figure 1 and Figure 2As shown, this embodiment is applied to the application scenario of a floating water pollution detection system for real-time water quality monitoring of lakes, rivers or nearshore waters. The system is deployed on a floating platform on the surface of the water body to be monitored. Through integrated multi-parameter sensors, it realizes continuous automatic monitoring of water quality parameters such as pH value, dissolved oxygen, ammonia nitrogen, turbidity and chemical oxygen demand, and sends the processed monitoring data to a remote monitoring center via a wireless network.

[0031] This application scenario deploys a complete system including a sensor module, a signal acquisition module, a crosstalk compensation module, a sensor health status assessment module, a blind source separation and processing module, a data fusion and output module, a remote communication and data management module, and a self-cleaning and maintenance module. Each module is interconnected through a preset communication interface and data path to form a complete water quality monitoring and signal processing link.

[0032] The sensor module is installed in the underwater detection chamber of the floating platform. The module includes at least one set of electrochemical sensors and at least one set of optical sensors. The electrochemical sensors include a pH electrode, a dissolved oxygen electrode, and an ammonia nitrogen electrode. These electrodes are connected to the corresponding channels of the signal acquisition module through independent shielded cables. The pH electrode uses glass membrane ion selection technology, and its output signal is a potential change in the mV range, with a typical measurement range of 0 to 14 and a sensitivity of approximately 59 mV / pH. The dissolved oxygen electrode uses a polarographic or galvanic cell working principle, forming a weak polarization current between the anode and cathode, and its output signal is a current change in the nA to μA range, with a typical measurement range of 0 to 20 mg / L. The ammonia nitrogen electrode uses gas-sensitive membrane technology to detect the change in gas chamber pressure caused by changes in ammonia nitrogen concentration in the water, and its output signal is a potential difference in the mV range. All these electrochemical sensors are electrically connected to the signal acquisition module. The sensor probe is equipped with a temperature-compensated thermistor to collect water temperature data in real time to correct temperature-related measurement errors. The optical sensors include a turbidity sensor and a chemical oxygen demand (COD) sensor. The turbidity sensor works on the principle of 90-degree scattering or transmission. The light source emits incident light of a specific wavelength, and the detector measures the intensity of the scattered light or the attenuation of the transmitted light at a 90-degree angle to the incident light. The output signal is a photocurrent change at the μA level, with a typical measurement range of 0 to 1000 NTU. The COD sensor works on the principle of ultraviolet absorption or ozone oxidation. Chemical oxygen demand (COD) is estimated by measuring the degree of absorption of ultraviolet light of a specific wavelength by water or the spectral changes of the oxidation reaction process. The output signal is a weak current at the μA level. The optical sensor is also electrically connected to the signal acquisition module. The light source driving circuit and photoelectric detection circuit of each optical sensor are equipped with independent electromagnetic shielding to reduce the influence of external electromagnetic interference on the weak photocurrent signal.

[0033] The signal acquisition module is located in the main control compartment of the floating platform. This module is connected to the sensor module via a waterproof cable. The signal acquisition module has a built-in multi-channel synchronous sampling circuit, which includes three functional sub-modules: a preamplifier module, a programmable filter module, and a multi-channel analog-to-digital converter module. The preamplifier module is equipped with a dedicated low-noise instrumentation amplifier chip for each sensor channel. The amplifier of the electrochemical sensor channel is set with a fixed gain to adapt to the signal range from mV to μA. The amplifier of the optical sensor channel is set with a programmable gain to adapt to the signal amplitude changes under different turbidity or COD concentrations. The input of the preamplifier module is equipped with a low-pass filter network to suppress high-frequency noise and power frequency interference signals. The cutoff frequency of the filter network is set to half of the sampling frequency to meet the requirements of the Nyquist sampling theorem. The programmable filtering module uses digital filtering technology to achieve more accurate filtering characteristics. This module runs a finite impulse response (FIR) filtering algorithm or an infinite impulse response (IRS) filtering algorithm internally. According to the preset filtering parameters, it performs further noise reduction processing on the amplified analog signal. The filtering parameters include filter type selection, passband boundary frequency, stopband boundary frequency, and passband ripple coefficient. These parameters can be configured independently according to the signal characteristics of different sensors. The multi-channel analog-to-digital converter (ADC) module adopts a combination architecture of synchronous sample-and-hold circuit and multi-channel ADC chip to ensure that each sensor channel completes signal sampling at the same time. The resolution of this module is set to 24 bits to meet the precise quantization requirements of small signal changes, and the sampling rate is set to 10 times per second to ensure real-time response to signal changes. The ADC result is sent to the subsequent crosstalk compensation module for signal processing through a serial peripheral interface.

[0034] The crosstalk compensation module is connected to the signal acquisition module via a high-speed data bus. This module is configured to separate mixed signals to eliminate coupling interference between electrochemical sensor signals and optical sensor signals. The module runs an independent component analysis algorithm with an improved architecture based on the principle of information maximization. A nonlinear activation function is introduced into the feature extraction network to improve the separation performance of the algorithm under non-Gaussian signal conditions. The nonlinear activation function adopts the form of a modified hyperbolic tangent function or a sign function. Its mathematical expression can be described as y equal to the weighted sum of the activation function applied to the network output. This activation function forms a nonlinear mapping in the higher-order statistical property space of the signal, enabling the algorithm to effectively distinguish between electrochemical signals and optical signals with different probability distribution characteristics. This module also combines principal component analysis to perform feature dimensionality reduction to reduce computational complexity. The specific implementation method is as follows: First, a covariance matrix is ​​constructed for the multi-channel sensor signal. By solving the eigenvalues ​​and eigenvectors of the covariance matrix, several principal components with a cumulative contribution rate reaching a preset percentage are selected as the input dimensions for subsequent independent component analysis. This significantly reduces the amount of computation while preserving the main information of the signal. The statistical independent components are calculated based on the information maximization principle and the nonlinear activation function. These independent components characterize the potential source signal representation of the sensor signal under the statistical independence criterion.

[0035] This module further characterizes the signal coupling relationship by constructing a transfer function matrix between sensors. The construction process of the transfer function matrix is ​​as follows: First, synchronous sampling data of electrochemical sensor signals and optical sensor signals within the same time period are acquired, and two signal sets are constructed, namely electrochemical signal vector and optical signal vector. Each element of the transfer function matrix represents the coupling influence coefficient of the electrochemical signal channel on the optical signal channel. This coupling influence coefficient is initialized and estimated by analyzing the correlation characteristics of the two types of signals in the frequency domain and time domain. After initialization, the online parameter estimation stage is entered. In the online estimation stage, the recursive least squares method is used to update the parameters of the transfer function matrix in real time to adapt to the time-varying characteristics of the signal coupling relationship. The implementation method of the recursive least squares method is as follows: a forgetting factor is set to balance the weight contribution of historical data and real-time data. The value range of the forgetting factor is usually set between 0.95 and 0.99. When the forgetting factor is close to 1, the algorithm assigns a high degree of confidence to historical data; when the forgetting factor is close to 0.95, the algorithm focuses more on tracking rapid changes in the signal. Within each sampling period, the algorithm calculates the prediction error based on the current input signal vector and the parameter estimates from the previous time step. The prediction error and forgetting factor are then substituted into the parameter update formula to iteratively correct each element of the transfer function matrix. The parameter update formula for the recursive least squares method can be described as follows: First, the gain vector is calculated as the product of the covariance matrix and the input vector divided by a normalization term that includes a forgetting factor. Then, the covariance matrix is ​​updated as the difference between the previous covariance matrix and the current input contribution term. Finally, the parameter estimate is updated as the sum of the product of the previous estimate, the gain vector, and the prediction error. Through the above recursive calculation process, the transfer function matrix parameters can continuously track the dynamic evolution of the coupling relationship between sensors, thereby achieving quantitative stripping of crosstalk signals.

[0036] This module remaps the corrected output signals of each sensor to the corresponding physical parameter space. The mapping process is achieved through the inversion of the transfer function model: First, the estimated value of the component caused by crosstalk from the electrochemical sensor in the optical sensor signal is calculated based on the updated transfer function matrix. Then, the estimated value of the crosstalk component is subtracted from the original measurement value of the optical sensor to obtain the pure optical signal. Finally, the pure optical signal is converted into the corresponding turbidity value or chemical oxygen demand value through the calibration curve of each optical sensor. The signal correction process of the electrochemical sensor is similar. The crosstalk effect of the optical sensor on the electrochemical sensor is eliminated by the inverse mapping of the transfer function matrix. The output result of the crosstalk compensation module is sent to the sensor health status assessment module and the blind source separation processing module through the data bus for subsequent analysis and processing.

[0037] The sensor health status assessment module maintains communication with the signal acquisition module and crosstalk compensation module. This module is configured to perform real-time monitoring and degradation diagnosis of sensor performance. The module first establishes a time-varying characteristic model for each sensor. The time-varying characteristic model expresses the sensor output signal as a combination function of the real physical quantity, zero-point offset, sensitivity scaling factor, and measurement noise. The zero-point offset characterizes the degree of drift of the sensor reference potential, and the sensitivity scaling factor characterizes the change in the sensor's response capability to the measured physical quantity. The mathematical expression of this time-varying characteristic model can be described as follows: the sensor output measurement value is equal to the sensitivity scaling factor multiplied by the real value of the measured physical quantity plus the zero-point offset plus the measurement noise term. The ideal value of the sensitivity scaling factor is 1. When the sensor ages, this factor will gradually deviate from 1. The ideal value of the zero-point offset is 0. When the sensor drifts, this offset will gradually accumulate.

[0038] This module characterizes the sensor's performance status by extracting multidimensional health features composed of signal amplitude, noise level, and response time. The signal amplitude feature is extracted as follows: Within a preset time window, statistical analysis is performed on the sensor output signal to calculate the signal mean and standard deviation. The signal mean reflects the current level of the measured physical quantity, and the signal standard deviation reflects the degree of signal fluctuation. The noise level feature is extracted as follows: High-frequency component analysis is performed on the sensor output signal. High-frequency noise components are separated using bandpass filtering, and the power spectral density integral of these high-frequency components is calculated as a quantitative indicator of the noise level. The response time characteristics are extracted as follows: When the sensor receives a step change excitation of a standard concentration or standard physical quantity, the response time is the time required for the sensor output signal to change from an initial value to a certain percentage of a final stable value. When a decrease in sensor response sensitivity or an increase in noise level is detected, a cleaning assessment process is automatically triggered to determine the urgency of cleaning and maintenance of the sensor probe.

[0039] This module activates the Extended Kalman Filter (EKF) algorithm to perform online estimation of sensor zero-point drift and sensitivity decay when abnormal fluctuations exist in multidimensional health characteristics. The state vector of the EKF contains three state variables: the measured physical quantity, the sensor zero-point offset, and the sensitivity scaling factor. The observation vector is the actual output measurement value of the sensor. The state transition equation describes the evolution of each state variable over time. The measured physical quantity is described by a random walk model to describe its slow change characteristics, the zero-point offset is described by a random walk model to describe its drift accumulation characteristics, the sensitivity scaling factor is described by a random walk model to describe its decay accumulation characteristics, and the observation equation describes the nonlinear relationship between the sensor output measurement value and the state variable. This nonlinear relationship originates from the product operation of the sensitivity scaling factor and the measured physical quantity. The prediction step of the extended Kalman filter first uses the state transition equation to predict the state prior estimate at the current time based on the state estimate at the previous time. At the same time, it updates the covariance matrix of the state estimate to reflect the degree of determinism of the prediction process. The update step of the extended Kalman filter first calculates the observation residual, which is equal to the difference between the actual observation value and the predicted observation value. Then, the Kalman gain matrix is ​​calculated to balance the contribution of state prediction and observation update to the final state estimate, and the state prior estimate is corrected based on the Kalman gain matrix and observation residuals to obtain the state posterior estimate. Through the above iterative calculation process, the extended Kalman filter algorithm can track the time-varying trajectory of zero-point offset and sensitivity scaling factor in real time, thereby realizing online diagnosis of sensor aging. When the sensor performance index is detected to deviate from the preset threshold range, the module automatically triggers the corresponding calibration or replacement reminder and sends the diagnostic results to the data fusion and output module to adjust the subsequent fusion weights.

[0040] The blind source separation and processing module maintains data interaction simultaneously with the crosstalk compensation module and the sensor health status assessment module. This module is configured to separate the real water quality change component and the sensor aging component from the compensated sensor output signal. This module models the sensor output signal as a linear mixture of the real water quality signal and the aging noise signal. The linear mixture model can be expressed as: The sensor output signal vector is equal to a weighted linear combination of the real water quality signal vector and the aging noise signal vector. The weighting coefficients reflect the contribution ratio of each type of signal to the final measurement. This module uses a kurtosis-based blind source separation algorithm to achieve statistically independent separation of the two types of signals. Kurtosis, as a fourth-order statistic, can effectively characterize the degree to which the signal probability distribution deviates from the Gaussian distribution. The real water quality signal and the aging noise signal usually have different kurtosis values ​​and can therefore be distinguished. The fast fixed-point algorithm achieves rapid separation of source signals by iteratively optimizing the kurtosis objective function. The implementation method of this algorithm is as follows: First, initialize each element of the separation matrix to a random value or an initial value set according to the principal component analysis results. Then, in each iteration, calculate the fourth-order cumulant or kurtosis value of the current separation result. Adjust the elements of the separation matrix according to the gradient direction of the kurtosis value to maximize or minimize the absolute value of kurtosis. Continue iterating until the separation matrix converges to a stable state. The convergence criterion of the fast fixed-point algorithm is that the change in the elements of the separation matrix between two adjacent iterations is less than the preset convergence threshold or the number of iterations reaches the preset maximum number of iterations.

[0041] This module further verifies the physical rationality of the separation results through time-domain recursive analysis and frequency-domain feature comparison. The verification method of time-domain recursive analysis is as follows: The recursive correlation coefficients between sampling points at different times are calculated for the real water quality change components obtained after separation. The calculation of the recursive correlation coefficients is based on recursive graph analysis technology, which quantifies the dynamic evolution characteristics of the system by constructing a recursive matrix of the distance between time series sample points. The calculated recursive correlation coefficients are compared with the preset physical constraint range. If the recursive correlation coefficients fall within the physical reasonable range, the separation result is verified; otherwise, a re-separation process is triggered. The verification method of frequency domain feature comparison is as follows: the power spectral density of the real water quality change components and the aging noise components obtained after separation are calculated separately. By comparing the energy distribution characteristics of the two types of signals in the frequency domain, it is determined whether the separation result conforms to prior physical knowledge. For example, real water quality changes usually exhibit low-frequency slow-change characteristics, while sensor aging noise usually exhibits high-frequency or specific frequency band characteristic spectra. The separation results that pass the verification are sent to the data fusion and output module for weighted fusion processing.

[0042] The data fusion and output module is communicatively connected to the blind source separation and processing module. This module is configured to perform weighted fusion processing on the separated real water quality change components. The module dynamically adjusts the weight coefficients of each sensor in the fusion process based on the health status assessment results of each sensor. The strategy for determining the weight coefficients is as follows: The system acquires performance status information for each sensor sent by the sensor health status assessment module, including estimated values ​​for zero-point offset, sensitivity scaling factor, and noise level. Based on this performance status information, it calculates a health score for each sensor. The health score is calculated as follows: Set the normal threshold range for zero-point offset and the normal fluctuation range for sensitivity scaling factor. When the estimated zero-point offset of a sensor falls within the normal threshold range, the zero-point stability score of the sensor is full; otherwise, points are deducted based on the degree of offset. When the estimated value of the sensitivity scaling factor of a sensor falls within the normal fluctuation range, the sensitivity stability score of the sensor is full; otherwise, the score is deducted according to the degree of deviation. The zero-point stability score and the sensitivity stability score are weighted and summed according to a preset ratio to obtain the overall health score of the sensor. The overall health scores of each sensor are normalized to obtain the initial weight coefficient of each sensor. When the health status estimation deviation of a certain sensor exceeds a preset deviation threshold, its weight coefficient in the fusion process is automatically reduced. The preset deviation threshold takes into account the differences in sensor type and measurement accuracy requirements. For electrochemical sensors, the deviation threshold is set when the sensitivity scaling factor deviates from 1 by more than 15% or the zero-point offset exceeds 5% of the range. For optical sensors, the deviation threshold is set when the sensitivity scaling factor deviates from 1 by more than 10% or the noise level exceeds twice the normal level.

[0043] This module uses an adaptive weighted fusion algorithm to generate the final water quality monitoring results. The implementation of the adaptive weighted fusion algorithm is as follows: the real water quality change components obtained from each sensor are multiplied by their corresponding dynamic weight coefficients, and then the weighted components are summed to obtain the estimated water quality parameters after fusion. The calculation process of the adaptive weighted fusion algorithm can be described as follows: the fusion result is equal to the real water quality change component of the first sensor multiplied by the first weight coefficient, plus the real water quality change component of the second sensor multiplied by the second weight coefficient, and so on up to the last sensor. The sum of all weight coefficients is equal to 1 to ensure the unbiasedness of the fusion result. The fused water quality parameters include estimated pH value, estimated dissolved oxygen concentration, estimated ammonia nitrogen concentration, estimated turbidity, and estimated chemical oxygen demand. This module further converts the fused water quality parameters into a comprehensive water quality level output through a water quality index model. The water quality index model calculates the individual water quality index of each parameter based on the deviation of the actual measured value of each water quality parameter from the standard limit. Then, the individual water quality indices are combined into a comprehensive water quality index by weighted average or geometric average, and the water quality level is divided according to the numerical range of the comprehensive water quality index. This module outputs the water quality anomaly level indicator, sensor aging degree indicator and sensor probe cleaning and maintenance priority indicator to the local display terminal of the floating platform through the display interface or to the remote monitoring platform through the communication interface.

[0044] The remote communication and data management module is connected to the data fusion and output module. This module is configured to realize the remote transmission of monitoring data and the reception of instructions. The module uses a low-power wide area network (LPWAN) transmission protocol to realize the remote transmission of monitoring data. The LPWAN protocol includes at least one of LoRa protocol, NB-IoT protocol or Sigfox protocol. This protocol has the advantages of wide coverage, low power consumption and stable connection, making it suitable for monitoring application scenarios in remote waters. The data transmission frequency can be configured according to monitoring needs. Under normal monitoring conditions, the data transmission frequency is set to send a complete data packet once every 15 minutes. In the event of an abnormal water quality alarm, the data transmission frequency is automatically increased to once per minute to meet emergency response requirements. The complete data packet includes the original measurement values ​​of each sensor, the measurement values ​​after crosstalk compensation, the health status assessment results, the blind source separation results, the estimated water quality parameters after fusion, and the comprehensive water quality level label. This module also receives and executes calibration instructions, parameter configuration instructions, and data query instructions issued by the monitoring center. The calibration instructions include the sensor zero-point calibration parameters and sensitivity calibration coefficients. After receiving the calibration instructions, this module updates the calibration parameters of the sensor module and performs in-situ calibration operations. The parameter configuration instructions include parameter adjustments for the crosstalk compensation algorithm, parameter modifications for the filter, and alarm threshold settings updates. After receiving the parameter configuration instructions, the module modifies the corresponding configuration parameters and saves them to local storage. The data query instructions include historical data query requests for a specified time period. After receiving the data query instructions, the module retrieves historical data that meets the conditions from the local data storage and transmits it back to the monitoring center via a remote communication link. The module also supports remote firmware upgrade functionality. The monitoring center can send firmware upgrade packages to the module's storage and perform firmware burning at an appropriate time to achieve remote updates of system functions.

[0045] The self-cleaning maintenance module communicates with the sensor module and the main control module. This module is configured to periodically clean and maintain the sensor probe to remove biological deposits and particulate matter. The module includes two cleaning execution modules: a high-pressure water flushing submodule and an ultrasonic vibration submodule. The high-pressure water flushing submodule is located on the periphery of the sensor probe. This submodule includes a miniature high-pressure water pump, a nozzle array, and a pipeline control valve. The high-pressure water flow generated by the miniature high-pressure water pump acts on the surface of the sensor probe through the nozzle array in a multi-angle spray manner. The nozzle array is designed to cover the main exposed surface of the sensor probe to ensure thorough cleaning. The pipeline control valve controls the on / off state and spray timing of the high-pressure water flow according to the cleaning control signal. The ultrasonic vibration submodule includes an ultrasonic transducer and an ultrasonic driver. The ultrasonic transducer is attached and fixed to the mounting base of the sensor probe. The ultrasonic driver generates a high-frequency AC signal to drive the ultrasonic transducer to generate ultrasonic vibration. The ultrasonic vibration acts on the surface of the sensor probe and its sensitive membrane layer through mechanical transmission. The cavitation effect of the ultrasonic waves can effectively peel off biofilm, algae and mineral deposits attached to the probe surface. The cleaning cycle and duration of this module can be adaptively adjusted according to the water quality. The cleaning control module analyzes the noise level change trend and response sensitivity change trend sent by the sensor health status assessment module in real time. When the sensor response sensitivity drops below a preset threshold or the noise level rises above a preset threshold, the cleaning assessment process is automatically triggered. The cleaning assessment process calculates the cleaning urgency score based on the degree of sensitivity drop and noise rise, and divides the score into multiple urgency levels. When performing cleaning operations, this module automatically sends a signal shielding command to the signal acquisition module to pause the signal acquisition function, so as to avoid physical disturbances generated during the cleaning process from interfering with normal measurement data. After cleaning and maintenance are completed, this module automatically sends a cleaning completion signal to the signal acquisition module to restore the signal acquisition function, and at the same time updates the local maintenance record and reports the cleaning and maintenance execution status to the remote monitoring platform.

[0046] The workflow of this system and the collaborative operation of its modules are as follows: After system startup, the sensor module enters the initialization state. Each electrochemical and optical sensor completes power-on self-test and preheating processes. The preheating time for electrochemical sensors is typically set to 5 to 10 minutes to ensure that the ion-selective membrane reaches a stable operating state, while the preheating time for optical sensors is typically set to 2 to 3 minutes to ensure that the light source and photodetector reach stable operating characteristics. After the sensor module initialization is complete, it enters the data acquisition mode. Each sensor continuously outputs raw analog signals reflecting water quality parameters according to a preset sampling frequency. The signal acquisition module continuously monitors the signal output of each sensor channel. When the sensor signal exceeds the minimum signal threshold, the current sampling process is initiated. The sampling process first captures the analog signals of each sensor channel at the same trigger time through a multi-channel synchronous sample-and-hold circuit. This trigger signal is generated by a timer inside the signal acquisition module to ensure the timing consistency of sampling for each channel. After synchronous sample-and-hold, the analog signals of each channel are amplified by a preamplifier to adapt to the input range of the analog-to-digital converter. The gain of the preamplifier is pre-configured according to the signal characteristics of each channel. The amplified analog signals are digitally filtered by a process-controlled filter to remove mixed power frequency interference and environmental noise. The filtered digital signals of each channel are temporarily stored in the buffer of the signal acquisition module for subsequent processing. The signal acquisition module encapsulates the digital signals in the buffer into sampling data frames at fixed time intervals and sends them to the crosstalk compensation module for crosstalk separation processing.

[0047] After receiving the sampled data frame, the crosstalk compensation module initiates the crosstalk compensation processing flow. This flow first performs mean removal and pre-whitening preprocessing on the signals of each channel in the data frame to improve the convergence performance of the subsequent independent component analysis algorithm. The mean removal process is achieved by calculating the arithmetic mean of each channel signal within a preset sample window and then subtracting this mean from the original signal. The pre-whitening process transforms the original signal vector into a whitened vector with the covariance matrix as the identity matrix through principal component analysis to eliminate the second-order correlation between signal channels. After the preprocessing is completed, the independent component analysis stage begins. In this stage, a multi-layer feature extraction network is constructed, and nonlinear activation functions are set between network layers to achieve the extraction of higher-order statistics. The selection of nonlinear activation functions takes into account the balance requirements of computational complexity and separation performance. The independent component analysis algorithm searches in the feature space for the optimal separation matrix under the condition of minimizing mutual information or maximizing entropy. During the algorithm iteration process, the elements of the separation matrix are continuously updated until the objective function converges to a stable state.

[0048] After the independent component analysis algorithm converges, it obtains a sequence of statistical independent components. These independent components need to be mapped back to the original sensor physical space through the transfer function matrix to achieve quantitative removal of crosstalk. The elements of the transfer function matrix are estimated online by recursive least squares method to adapt to the time-varying characteristics of the coupling relationship between sensors. The forgetting factor of recursive least squares method is adaptively adjusted according to the speed of signal coupling change. When a sudden change in coupling strength is detected, the forgetting factor is appropriately reduced to speed up the tracking speed. When the coupling strength tends to stabilize, the forgetting factor is appropriately increased to improve the estimation accuracy. The recursive least squares method predicts the crosstalk contribution of each channel based on the current input signal vector in each iteration; Then, the prediction error is calculated and the elements of the transfer function matrix are corrected accordingly. After the transfer function matrix is ​​updated, the crosstalk estimate of each sensor channel is calculated and the crosstalk estimate is subtracted from the original measurement signal to obtain the corrected sensor output signal. The corrected sensor output signal is converted into the corresponding physical parameter value through the inverse calibration curve mapping and then sent to the sensor health status assessment module and the blind source separation processing module.

[0049] After receiving the corrected sensor output signal, the sensor health status assessment module initiates the sensor health status monitoring process. This process first acquires sensor signal samples within a preset time window to construct a signal segment for feature extraction. After the signal segment is constructed, the signal amplitude feature, noise level feature, and response time feature of the signal segment are extracted to form a multidimensional health feature vector. The signal amplitude feature extraction is achieved by calculating the arithmetic mean and standard deviation of the sample points within the signal segment. The noise level feature extraction is achieved by calculating the power of high-frequency components after separating them by bandpass filtering. The response time feature extraction is achieved by analyzing the rise time of the signal step response curve. The multidimensional health feature vector is compared with a preset feature threshold to detect whether there are abnormal fluctuations. When abnormal fluctuations are detected, the extended Kalman filter algorithm is activated to estimate the sensor zero-point drift and sensitivity decay online.

[0050] When initializing the extended Kalman filter algorithm, the initial values ​​of the state vector and the state estimation covariance matrix are set. The initial values ​​of the state vector include the initial estimates of the measured physical quantity, the initial estimates of the zero-point offset, and the initial estimates of the sensitivity scaling factor. The initial estimate of the measured physical quantity is set to the current measurement value of the sensor. The initial estimate of the zero-point offset is set to 0. The initial estimate of the sensitivity scaling factor is set to 1. The parameters of the state transition model and the observation model are pre-configured according to the sensor type and historical operating data to ensure that the model can accurately describe the aging evolution law of the sensor. When the extended Kalman filter algorithm enters the prediction step, it uses the state transition equation to calculate the state prior estimate at the current time based on the state posterior estimate at the previous time step, and updates the state estimation covariance matrix based on the process noise covariance matrix. When the Extended Kalman Filter (EKF) algorithm enters the update step, it first calculates the predicted observation value based on the observation model, then calculates the observation residual, which is the difference between the actual observation value and the predicted observation value. Based on the observation residual and the observation noise covariance matrix, it calculates the Kalman gain matrix. Finally, it corrects the state prior estimate based on the Kalman gain matrix and the observation residual to obtain the state posterior estimate. The EKF algorithm continuously iterates and updates the state estimate, thereby tracking the time-varying trajectory of the zero-point offset and the sensitivity scaling factor in real time. When the zero-point offset or sensitivity scaling factor is detected to be outside the preset normal range, the sensor is determined to be in an aging state and an aging alarm is triggered. At the same time, the aging state information is sent to the data fusion and output module to adjust the weight coefficient of the sensor in the fusion process.

[0051] After receiving the crosstalk-compensated sensor output signal and the aging diagnosis information sent by the sensor health status assessment module, the blind source separation processing module starts the blind source separation process. This process models the sensor output signal as a linear mixture of real water quality signal and aging noise signal, and then uses a kurtosis-based blind source separation algorithm to separate the two types of signal sources. The blind source separation algorithm first initializes the separation matrix. The dimension of the separation matrix is ​​the number of sensor channels multiplied by the number of sensor channels. The initial elements are set to random values ​​or according to the eigenvectors of the signal covariance matrix. After the separation matrix is ​​initialized, it enters the iterative optimization stage. The fast fixed-point algorithm first calculates the current separation result, i.e., the product of the separation matrix and the input signal vector, in each iteration. Then, the kurtosis value of the separation result is calculated, which is the fourth-order cumulant divided by the square of the second-order moment. The sign of the kurtosis value reflects the direction of deviation of the signal distribution from the Gaussian distribution. Positive kurtosis indicates that the signal distribution is sharper than the Gaussian distribution, and negative kurtosis indicates that the signal distribution is flatter than the Gaussian distribution. Real water quality signals usually exhibit positive kurtosis characteristics, while aging noise signals usually exhibit negative kurtosis characteristics. Therefore, the two types of signal sources can be distinguished according to the kurtosis sign. The fast fixed-point algorithm adjusts the elements of the separation matrix according to the gradient direction of the kurtosis value so that the absolute value of the kurtosis of each component of the separation result gradually increases, thereby achieving statistical independent separation of the signal. The iterative process continues until the change of the separation matrix elements is less than the convergence threshold or the number of iterations reaches the preset upper limit.

[0052] After the blind source separation algorithm converges, the separated signal components are obtained. This module needs to verify the physical rationality of the separation results to ensure that the separation effect meets the expected physical constraints. The verification method of time-domain recursive analysis is to calculate the recursive correlation coefficient between the sampling points of the separated signal at different times. The calculation of the recursive correlation coefficient first constructs a recursive graph matrix. The elements of this matrix represent whether the distance between two sampling points in the time series is less than the preset recursive threshold. Then, based on the recursive graph matrix, the recursive quantitative analysis indicators, including recursion rate, determinism, and recursive entropy, are calculated. These recursive quantitative indicators are combined to form the recursive correlation coefficient. The recursive correlation coefficient reflects the time evolution structure characteristics of the separated signal. Real water quality signals usually exhibit high recursiveness and determinism due to the constraints of physical processes, while aging noise signals usually exhibit low recursiveness and determinism due to the influence of random factors. The recursive correlation coefficient of the separated signal is compared with the preset physical constraint range. If it exceeds the constraint range, the separation result is determined to be unreasonable and a re-separation process is triggered. The verification method of frequency domain feature comparison is to calculate the power spectral density of the real water quality component and the aging noise component after separation, and then compare the energy distribution difference of the two types of signals in the frequency domain. The real water quality signal usually has a higher energy ratio in the low frequency band, and the aging noise signal usually has a significant energy peak in the high frequency band or a specific frequency band. If the frequency domain characteristics of the separation result are consistent with the expected physical characteristics, the verification is passed. The separation result after verification is sent to the data fusion and output module for weighted fusion processing.

[0053] After receiving the separation results from the blind source separation processing module, the data fusion and output module initiates the adaptive weighted fusion process. This process first calculates the comprehensive health score of each sensor based on the performance status information of each sensor sent by the sensor health status assessment module. The formula for calculating the comprehensive health score is as follows: First, determine whether the zero-point offset exceeds the preset threshold range. If it does, calculate the deduction value based on the degree of offset; otherwise, get full marks. Then, determine whether the sensitivity scaling factor exceeds the preset threshold range. If it does, calculate the deduction value based on the degree of deviation; otherwise, get full marks. Finally, weight the zero-point stability score and the sensitivity stability score according to a weight ratio of 6:4 to obtain the comprehensive health score. Normalize the comprehensive health scores of each sensor to obtain the initial weight coefficient of each sensor. When the health status estimation deviation of a certain sensor exceeds the preset deviation threshold, its weight coefficient in the fusion process is automatically reduced. The specific reduction amount is calculated based on the degree of deviation. After obtaining the dynamic weighting coefficients of each sensor, the actual water quality change components obtained from each sensor are multiplied by their corresponding dynamic weighting coefficients. Then, the weighted components are summed to obtain the estimated values ​​of the fused water quality parameters. The fused water quality parameters are then converted into a comprehensive water quality index model using a water quality index model. The formula for calculating the individual water quality index of each water quality parameter is as follows: When the measured value of a water quality parameter is better than or equal to the water quality standard limit, the single water quality index is equal to 60 plus 40 multiplied by the ratio of the measured value to the standard limit. When the measured value of a water quality parameter is worse than the water quality standard limit, the single water quality index is equal to 60 minus 60 multiplied by the percentage of the measured value that exceeds the standard limit. The individual water quality indices are summed with equal weights to obtain the comprehensive water quality index. Finally, the water quality is classified according to the range of values ​​of the comprehensive water quality index. A comprehensive water quality index of 85 or higher is classified as Class I water quality. A comprehensive water quality index greater than or equal to 70 and less than 85 is classified as Class II water quality. A comprehensive water quality index greater than or equal to 55 and less than 70 is classified as Class III water quality. A comprehensive water quality index greater than or equal to 40 and less than 55 is classified as Class IV water quality. When the comprehensive water quality index is less than 40, it is judged as Class V water quality. This module encapsulates the water quality anomaly level indicator, sensor aging level indicator, and sensor probe cleaning and maintenance priority indicator into an output data packet and sends it to the local display terminal for display. At the same time, the data packet is sent to the remote communication and data management module to realize remote transmission.

[0054] After receiving the output data packet sent by the data fusion and output module, the remote communication and data management module initiates the remote communication process. This process first encapsulates the data packet according to the communication protocol requirements. The encapsulation content includes the data packet header, sensor identifier, timestamp, data of each measurement channel, diagnostic results, and check code field. Then, the encapsulated data packet is sent to the remote monitoring center through the low-power wide area network wireless module. The remote communication module continuously listens for downlink commands from the remote monitoring center. When a calibration command is received, the module parses the command content and extracts the calibration parameters. Then, the calibration parameters are sent to the signal acquisition module to update the sensor calibration configuration and perform in-situ calibration. When a parameter configuration command is received, the module parses the command content and extracts the configuration parameters to be updated. Save the configuration parameters to the local storage and notify the relevant modules to load the new configuration. When a data query command is received, retrieve historical data that meets the conditions from the local data storage according to the time range specified in the command. Historical data is encapsulated and sent to the remote monitoring center. The remote communication and data management module supports remote firmware upgrade functionality when a firmware upgrade command is received. Verify the integrity and validity of the firmware package, then temporarily store the firmware package in local storage and perform the firmware flashing operation during system restart or idle periods to complete the remote update of system functions.

[0055] The self-cleaning maintenance module starts the cleaning process at set intervals according to the preset cleaning cycle, while simultaneously receiving sensor performance status information sent by the sensor health status assessment module in real time to achieve adaptive adjustment of the cleaning cycle. During the cleaning process, a signal shielding command is sent to the signal acquisition module to pause the signal acquisition function to avoid interference with normal measurement data during the cleaning process. After the signal shielding is completed, the high-pressure water flushing submodule starts. The high-pressure water generated by the micro high-pressure water pump is sprayed through the nozzle array at multiple angles onto the surface of the sensor probe. The high-pressure water flushing duration is set to 15 to 30 seconds, and the flushing water pressure is set to 0.3 to 0.5 MPa to ensure that the surface deposits are effectively removed without damaging the sensor probe. After high-pressure water rinsing, the ultrasonic vibration submodule is activated. The high-frequency ultrasonic vibration generated by the ultrasonic transducer is transmitted to the sensor probe surface. The duration of ultrasonic vibration is set to 10 to 20 seconds, and the ultrasonic frequency is set to 40 to 80 kHz to generate an effective cavitation effect to remove stubborn deposits. After the cleaning process is completed, there is a 3 to 5-second delay to allow residual moisture to dissipate. A cleaning completion signal is sent to the signal acquisition module to restore the signal acquisition function. At the same time, the local maintenance record is updated, and the execution status and effect evaluation results of this cleaning and maintenance are reported to the remote monitoring platform. The self-cleaning maintenance module dynamically adjusts the subsequent cleaning cycle based on the sensor performance status information. When the sensor response sensitivity continues to decrease or the noise level continues to increase, the cleaning cycle is automatically shortened to increase the cleaning frequency. When the sensor performance status remains good, the cleaning cycle is automatically extended to reduce unnecessary cleaning operations and mechanical wear on the sensor.

[0056] Example 2: This embodiment is applied to emergency monitoring scenarios where water quality is suddenly polluted. It has the same basic architecture as Embodiment 1, but the collaborative working mode of each module has been adapted to meet the high real-time requirements of emergency response.

[0057] In terms of the sensor module's operating mode, an emergency monitoring scenario employs a time-division multiplexing strategy to stagger the operating windows of the electrochemical and optical sensors in terms of timing, thereby reducing electromagnetic crosstalk generated during simultaneous operation of the sensors. The specific implementation of the time-division multiplexing strategy is as follows: The sampling period is divided into multiple time slots, and each time slot is assigned to a specific sensor channel. For example, the first time slot is assigned to the pH electrode for signal acquisition. The second time slot is allocated to the dissolved oxygen electrode for signal acquisition; The third time slot is allocated to the ammonia nitrogen electrode for signal acquisition; The fourth time slot is allocated to the turbidity sensor for signal acquisition; The fifth time slot is allocated to the chemical oxygen demand sensor for signal acquisition. The duration of the time slot is preset according to the response characteristics of each sensor. The duration of the time slot for the electrochemical sensor is set to 100 milliseconds to ensure that the ion-selective membrane reaches a sufficient response. The duration of the time slot for the optical sensor is set to 200 milliseconds to ensure that the output of the light source and photodetector is stable. Each sensor enters a low-power standby state during non-assigned time slots to reduce overall system energy consumption. The switching between time slots is achieved through timing control logic within the signal acquisition module. The timing control logic precisely controls the enable signal and sampling trigger signal of each sensor channel according to the preset time slot allocation table to ensure the timing accuracy of signal acquisition. The time-division multiplexing strategy effectively reduces the probability of concurrent operation of electrochemical and optical sensors at the same time, thereby reducing the chance of electromagnetic interference from the mV-level electric field generated by the electrochemical sensor to the μA-level current signal of the optical sensor, significantly improving the accuracy and reliability of water quality data in emergency monitoring scenarios.

[0058] In terms of the crosstalk compensation module, the algorithm for emergency monitoring scenarios has been optimized for real-time performance to meet the rapid processing requirements under high sampling rate conditions. The independent component analysis algorithm adopts a simplified network structure to reduce the amount of computation. Specific measures include reducing the number of hidden layers, reducing the number of neurons per layer, and simplifying the computational complexity of nonlinear activation functions. The online estimation of the parameters of the transfer function matrix adopts an adaptive forgetting factor strategy. When a drastic change in water quality is detected, the forgetting factor is appropriately reduced to speed up parameter tracking. When the water quality is detected to be stable, the forgetting factor is appropriately increased to improve estimation accuracy and stability. The calculation process of the recursive least squares method adopts an incremental update method to avoid repeatedly calculating the correlation matrix of historical data, thereby significantly reducing the computational complexity and memory usage of each iteration.

[0059] In terms of data fusion and output module processing strategy, emergency monitoring scenarios adopt a multi-level alarm mechanism to achieve rapid response to water quality anomalies of varying severity; The multi-level alarm mechanism has three alarm levels: yellow alert, orange alert, and red alert. The yellow alert condition is that a single water quality parameter exceeds the normal monitoring range, but has not yet reached the hazard threshold. At this time, the system automatically increases the data collection frequency and notifies the operation and maintenance personnel to pay attention. An orange alert is triggered when a single water quality parameter exceeds the hazard threshold or when multiple water quality parameters are abnormal at the same time. In this case, the system will automatically trigger the audible and visual alarm device and send SMS or telephone notifications to multiple preset contacts. A red alert is triggered when water quality parameters undergo drastic changes or the comprehensive water quality index drops to a dangerous level. At this time, the system automatically initiates emergency response procedures, including saving on-site data, notifying relevant departments, and activating backup water quality monitoring equipment. The multi-level alarm mechanism ensures timely detection and rapid response to water quality anomalies in emergency monitoring scenarios, effectively safeguarding the safety of the aquatic environment.

[0060] In terms of transmission strategy for remote communication and data management modules, an adaptive data transmission frequency strategy is adopted for emergency monitoring scenarios to balance the real-time performance of data transmission with the system power consumption requirements. Under normal monitoring conditions, the data transmission frequency is set to send a complete data packet once every 5 minutes to reduce the power consumption of the wireless communication module. Under alarm trigger conditions, the data transmission frequency is automatically increased to send once every 30 seconds to meet the real-time requirements of emergency response. Under major abnormal conditions, the data transmission frequency is further increased to a real-time push mode, that is, data packets are sent immediately after each sampling period to ensure that the monitoring center can keep abreast of the dynamic changes in water quality. The adaptive data transmission frequency strategy effectively controls the average power consumption level of the system while ensuring the real-time performance of emergency response, and extends the continuous operation time of the floating platform in remote waters.

[0061] In terms of the triggering strategy for the self-cleaning maintenance module, the emergency monitoring scenario adopts a hybrid triggering mechanism that combines event triggering and periodic triggering. The periodic triggering mechanism performs cleaning operations at set intervals according to a preset cleaning cycle of 72 hours, i.e., a complete cleaning and maintenance process is performed every three days. The event triggering mechanism dynamically adjusts the cleaning timing based on the real-time monitoring results of the sensor performance status. When a sudden drop in sensor response sensitivity is detected that exceeds the normal fluctuation range, it is determined that a bioattachment event has occurred, and the cleaning process is immediately triggered to quickly restore sensor performance. The hybrid triggering mechanism ensures that the sensor always maintains a good working condition in the emergency monitoring scenario, providing a guarantee for the accuracy of water quality monitoring data.

[0062] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A floating water pollution detection system based on multi-parameter sensor integration, characterized in that, The system includes a processing link, which is configured as follows: Acquire the raw mixed signal output by at least one electrochemical sensor and at least one optical sensor in the water environment to be detected, and trigger crosstalk compensation when there is a coupling correlation between the electrochemical signal and the optical signal in the raw mixed signal; Based on the principle of maximizing information and nonlinear activation function, the statistical independent components between signals are calculated, and the transfer function matrix between sensors is constructed based on the statistical independent components and the dimensionality reduction results of principal component analysis. The parameters of the transfer function matrix are estimated online by recursive least squares method to quantitatively remove the coupling signal, and the corrected output signals of each sensor are remapped to the corresponding physical parameter space. Obtain the time-series data of the corrected sensor output signal within a preset time window, and establish the time-varying characteristic model of each sensor; Based on the time-varying characteristic model, a multidimensional health feature consisting of signal amplitude, noise level and response time is extracted. When there are abnormal fluctuations in the multidimensional health feature, the extended Kalman filter algorithm is activated to perform online estimation of sensor zero drift and sensitivity attenuation. Based on the online estimation results, the actual water quality change component and the sensor aging component are separated from the compensated sensor output signal. The physical rationality of the separation results is verified by time-domain recursive analysis and frequency-domain feature comparison. Based on the estimation bias of the sensor health characteristics, the weight coefficient of each sensor in the fusion process is dynamically determined, and the weighted fusion of the real water quality change components is performed according to the weight coefficient. Based on the weighted fusion results, output water quality anomaly level indicators, sensor aging degree indicators, and sensor probe cleaning and maintenance priority indicators.

2. The floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link, based on the principle of maximizing information and the calculation of statistically independent components between signals using a nonlinear activation function, includes: setting a gradient inversion layer between the feature extraction network and the domain discriminator, which forces the feature extraction network to confuse the feature distributions of the source and target domains during backpropagation to improve separation performance.

3. The floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link constructs the inter-sensor transfer function matrix by: using the covariance matrix of the sensor output signals to characterize the coupling strength between signals, and establishing a bilinear transfer function model based on the coupling strength.

4. The floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link performs online estimation of the transfer function matrix parameters using the recursive least squares method, including: setting a forgetting factor to balance the weight contributions of historical data and real-time data, and iteratively updating the parameter estimates based on the forgetting factor and the current input signal.

5. A floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link performs online estimation based on multidimensional health features and extended Kalman filtering, including: introducing sensor zero-point offset and sensitivity scaling factor into the state vector, and dynamically correcting the estimated value of the state vector according to the observation residual of the sensor output signal.

6. The floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link separates the real water quality change component and the sensor aging component from the compensated sensor output signal by: modeling the sensor output signal as a linear mixture of the real water quality signal and the aging noise signal, and achieving statistical independence separation of the two by optimizing higher-order statistics.

7. A floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link verifies the physical rationality of the separation result through time-domain recursive analysis and frequency-domain feature comparison, including: calculating the recursive correlation coefficient of the separated signal at different times, and comparing the recursive correlation coefficient with the preset physical constraint range.

8. A floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link performs weighted fusion of the real water quality change components according to the weight coefficients, including: when the health status estimation deviation of a certain sensor exceeds a preset deviation threshold, its weight coefficient in the fusion process is automatically reduced, and the fused water quality parameters are converted into a comprehensive water quality level output through a water quality index model.

9. A floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, The processing link outputs a sensor probe cleaning and maintenance priority indicator based on the weighted fusion result, including: automatically triggering a cleaning assessment process when a decrease in sensor response sensitivity or an increase in noise level is detected, and determining the urgency of sensor probe cleaning and maintenance based on the result of the cleaning assessment process.

10. A floating water pollution detection system based on multi-parameter sensor integration according to claim 1, characterized in that, When the processing link triggers crosstalk compensation, it includes: using a time-division multiplexing strategy to stagger the working windows of the electrochemical sensor and the optical sensor in time to reduce signal coupling correlation.

Images