Water quality suspended matter detection system and method based on sensing data

Abstract

The application provides a water quality suspended matter detection system and method based on sensing data, and belongs to the technical field of water quality suspended matter detection.The gain of a mixed optical signal is compensated based on backward spontaneous radiation noise power, so that the influence of sensor gain drift and environmental interference on the signal baseline is effectively eliminated, an adaptive variational mode decomposition algorithm is used to decompose the normalized signal into multiple intrinsic mode functions, the signal strongly related to the physical characteristics of the target suspended particulate matter is accurately extracted, the intrinsic mode functions selected through weighting and superposition are used to generate a particulate matter characteristic waveform signal with optimized signal-to-noise ratio, the time-frequency characteristics of the target signal are further highlighted, a dynamic threshold peak detection algorithm is used to accurately separate each independent particulate matter signal event, multi-dimensional feature vectors are extracted from each signal event, and a pre-trained event classifier is used to distinguish the target particulate matter from interference, so that the detection accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water quality suspended matter detection, and in particular to a water quality suspended matter detection system and method based on sensing data. BACKGROUND

[0002] With the substantial increase in water consumption and sewage discharge brought about by industrial and agricultural production, urban development and population growth, a large amount of industrial wastewater and domestic sewage is discharged into water bodies, leading to serious deterioration of water environmental quality, which makes it increasingly important to accurately monitor key indicators of water quality such as suspended matter concentration. At present, the detection of suspended matter in water mainly relies on traditional weight method, optical sensor method, acoustic measurement method and other methods. The weight method, as a benchmark measurement method, is simple to operate and accurate in measurement, but the process is tedious and time-consuming, and only discrete point data can be obtained, which is difficult to reflect the dynamic changes of water quality in a large range of water area. The optical method is easily disturbed by water color, bubbles, dissolved organic matter and instrument drift, and the measurement result is not stable enough, and it can only provide total suspended matter information, and cannot effectively distinguish the size distribution and specific composition of particulate matter. Although acoustic measurement can be directly and quickly measured in a large area, the signal bandwidth is limited by the transducer, and most single-frequency instruments cannot effectively distinguish the cooperative change of suspended matter linearity and concentration. At the same time, the existing sensing technology generally lacks self-calibration capability, which is easy to cause measurement benchmark drift due to optical window pollution, element aging or environmental temperature change, and it is difficult to provide long-term reliable monitoring data. Therefore, there is an urgent need for a new water quality suspended matter detection technology with high reliability, which can run in real time and online, has strong anti-interference ability, can not only accurately measure the total concentration of suspended matter but also analyze the particle size distribution of particulate matter, and has self-calibration function, in order to overcome the inherent defects of traditional methods.

[0003] Therefore, it is necessary to provide a water quality suspended matter detection system and method based on sensing data to solve the above technical problems. SUMMARY

[0004] To solve the above technical problems, the present application provides a water quality suspended matter detection system and method based on sensing data, which has the beneficial effect of not only accurately measuring the total concentration of suspended matter but also analyzing the particle size distribution of particulate matter and having self-calibration function.

[0005] The present application provides a water quality suspended matter detection method based on sensing data, comprising:

[0006] S1: acquiring a mixed light signal containing a main return light signal and a backscattering spontaneous emission noise collected by a sensor; based on the power of the backscattering spontaneous emission noise, generating standardized waveform time series data by dynamic gain compensation for the mixed light signal;

[0007] S2: Perform adaptive variational mode decomposition on the standardized waveform time series data to obtain multiple essential mode functions. Based on the preset mapping table of particle size and mode function center frequency and the correlation coefficient between the essential mode functions and the standardized waveform time series data, select the target essential mode function set.

[0008] S3: Superimpose the essential mode functions in the target essential mode function set to generate particulate matter characteristic waveform signals;

[0009] S4: The dynamic threshold peak detection algorithm is used to identify the characteristic waveform signal of particulate matter, obtain multiple independent signal events, and extract the multi-dimensional feature vector corresponding to each independent signal event;

[0010] S5: Input all multidimensional feature vectors into a pre-trained event classifier to obtain target independent signal events labeled as target suspended particulate matter, and calculate the equivalent particle size and relative mass of the target independent signal events based on a regression model;

[0011] S6: Within a preset independent time window, count the number of all target independent signal events and accumulate the corresponding relative mass. Perform correction and compensation through a counting correction algorithm, and output the particulate matter number concentration and total mass concentration of suspended matter within the preset independent time window.

[0012] Preferably, in step S1, the step of generating standardized waveform timing data includes:

[0013] The mixed optical signal, which includes the main return optical signal and the backscattered spontaneous emission noise, is converted into a voltage signal.

[0014] The voltage signal is sampled and converted into a discrete-time digital waveform sequence;

[0015] The discrete-time digital waveform sequence is processed by a digital bandpass filter to extract the backscattered spontaneous emission noise component.

[0016] The power of the back-emitted spontaneous noise is calculated and compared with the preset calibration curve to obtain the actual gain.

[0017] Based on the preset target gain and actual gain, the gain compensation coefficient is calculated, and the gain compensation coefficient is used to perform gain compensation on the discrete-time digital waveform sequence to generate standardized waveform timing data.

[0018] Preferably, in step S2, the number of decomposition modes and the penalty factor in the adaptive variational mode decomposition are parameter combinations adaptively determined by an optimization algorithm.

[0019] Preferably, in step S2, the parameter combination of the number of decomposed modes and the penalty factor in the adaptive variational mode decomposition takes the sparsity of multiple essential mode functions as the optimization objective.

[0020] Preferably, in step S2, the screening step of the target essential mode function set includes:

[0021] Based on a pre-defined mapping table between particle size and modal function center frequency, a subset of essential modal functions whose center frequencies match the target particle size range are selected from multiple essential modal functions.

[0022] The correlation coefficient between each essential mode function in the essential mode function subset and the standardized waveform time series data is calculated, and a second screening is performed using a preset correlation coefficient threshold to obtain the target essential mode function set.

[0023] Preferably, in step S3, the superposition is linear superposition, and the correlation coefficient between each essential mode function participating in the superposition and the standardized waveform time series data is normalized and used as the weight coefficient of the superposition.

[0024] Preferably, the following steps are included before step S4:

[0025] Calculate the average signal-to-noise ratio of the particulate matter characteristic waveform signal within a preset signal-to-noise ratio time window;

[0026] If the average signal-to-noise ratio is higher than the preset first signal-to-noise ratio threshold, then dynamic threshold peak detection is performed;

[0027] If the average signal-to-noise ratio is lower than the preset first signal-to-noise ratio threshold but higher than the preset second signal-to-noise ratio threshold, then the particulate matter characteristic waveform signal is subjected to noise reduction preprocessing based on wavelet transform before dynamic threshold peak detection is performed.

[0028] If the average signal-to-noise ratio is lower than the preset second signal-to-noise ratio threshold, the data within the preset signal-to-noise ratio time window is determined to be invalid and discarded. At the same time, a sensor cleaning and calibration prompt message is generated, wherein the first signal-to-noise ratio threshold is greater than the second signal-to-noise ratio threshold.

[0029] Preferably, in step S4, the dynamic threshold in the dynamic threshold peak detection algorithm is calculated based on the local statistical characteristics of the particulate matter characteristic waveform signal within a preset sliding time window.

[0030] Preferably, in step S6, the counting correction algorithm is based on the Poisson distribution theory, and its correction factor is calculated based on the average time interval of the target independent signal events.

[0031] This invention provides a water quality suspended solids detection system based on sensor data, comprising:

[0032] The signal standardization module is used to acquire the mixed optical signal, which includes the main return optical signal and the backspontaneous emission noise, collected by the sensor; based on the power of the backspontaneous emission noise, the mixed optical signal is used to generate standardized waveform timing data through dynamic gain compensation.

[0033] The mode decomposition and screening module is used to perform adaptive variational mode decomposition on standardized waveform time series data to obtain multiple essential mode functions. Based on the preset mapping table of particle size and mode function center frequency and the correlation coefficient between essential mode functions and standardized waveform time series data, the target essential mode function set is screened out.

[0034] The signal waveform reconstruction module is used to superimpose the essential mode functions in the target essential mode function set to generate particulate characteristic waveform signals;

[0035] The pulse event detection module is used to identify particulate matter characteristic waveform signals through a dynamic threshold peak detection algorithm, obtain multiple independent signal events, and extract the multi-dimensional feature vector corresponding to each independent signal event.

[0036] The target recognition and inversion module is used to input all multidimensional feature vectors into a pre-trained event classifier to obtain target-independent signal events labeled as target suspended particulate matter, and calculate the equivalent particle size and relative mass of the target-independent signal events based on a regression model.

[0037] The result calculation and output module is used to count the number of all target independent signal events and accumulate the corresponding relative mass within a preset independent time window. It performs correction and compensation through a counting correction algorithm and outputs the particulate number concentration and total mass concentration of suspended matter within the preset independent time window.

[0038] Compared with related technologies, the water quality suspended solids detection system and method based on sensor data provided by the present invention have the following beneficial effects:

[0039] This invention utilizes dynamic gain compensation based on back-emission noise power to generate standardized waveform time-series data from hybrid optical signals. This effectively eliminates the influence of sensor gain drift and environmental interference on the signal baseline, providing a stable and reliable signal foundation for subsequent processing. An adaptive variational mode decomposition algorithm is employed to decompose the standardized signal into multiple essential mode functions (EMFs). A dual screening process is then performed, combining the mapping relationship between particle size and the center frequency of the EMFs, as well as the correlation coefficients between each EMF and the original signal. This accurately extracts signal components strongly correlated with the physical properties of the target suspended particles, increasing the proportion of useful signals. Weighted superposition of the screened EMFs generates a signal-to-noise ratio (SNR) optimized particle characteristic waveform signal, further highlighting the time-frequency characteristics of the target signal. Finally, a dynamic threshold peak detection algorithm is employed, and intelligent preprocessing based on real-time SNR assessment is introduced. The system employs a processing mechanism that adaptively selects between direct detection and noise reduction followed by detection or discarding invalid data, effectively avoiding errors caused by low-quality data. It accurately separates each independent particulate matter signal event, extracts multi-dimensional feature vectors from each event, and uses a pre-trained event classifier to distinguish target particles from interfering substances, significantly improving target identification accuracy. Based on a regression model, it calculates the equivalent particle size and relative mass of each target particle, achieving refined measurement of the physical properties of suspended matter. Finally, it counts the number of particles and accumulates the mass within a preset independent time window, using a Poisson distribution-based counting correction algorithm to compensate for missed count errors caused by signal overlap. Ultimately, it synchronously outputs the particle number concentration and total mass concentration, thus achieving stable, reliable, and comprehensive online monitoring of suspended matter concentration and particle size distribution in water bodies under complex hydrological conditions. Attached Figure Description

[0040] Figure 1 This is a flowchart of a water quality suspended solids detection method based on sensor data according to the present invention;

[0041] Figure 2 This is a modular structure diagram of a water quality suspended solids detection system based on sensor data according to the present invention. Detailed Implementation

[0042] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the drawings, not all structures. Moreover, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0043] It should also be noted that, for ease of description, the accompanying drawings show only the parts relevant to the invention and not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the drawings. The process may correspond to a method, function, procedure, subroutine, subprogram, etc.

[0044] Example 1

[0045] A method for detecting suspended solids in water based on sensor data, in the specific implementation process, such as... Figure 1 As shown, a flowchart of a water quality suspended solids detection method based on sensor data according to the present invention is illustrated, including:

[0046] Step S1: Acquire the mixed optical signal containing the main return optical signal and the backspinning spontaneous emission noise collected by the sensor; based on the power of the backspinning spontaneous emission noise, generate standardized waveform timing data for the mixed optical signal through dynamic gain compensation.

[0047] Specifically, in step S1, the steps for generating standardized waveform timing data include:

[0048] The mixed optical signal, which includes the main return optical signal and the backscattered spontaneous emission noise, is converted into a voltage signal.

[0049] The voltage signal is sampled and converted into a discrete-time digital waveform sequence;

[0050] The discrete-time digital waveform sequence is processed by a digital bandpass filter to extract the backscattered spontaneous emission noise component.

[0051] The power of the back-emitted spontaneous noise is calculated and compared with the preset calibration curve to obtain the actual gain.

[0052] Based on the preset target gain and actual gain, the gain compensation coefficient is calculated, and the gain compensation coefficient is used to perform gain compensation on the discrete-time digital waveform sequence to generate standardized waveform timing data.

[0053] In the specific implementation process, firstly, a photodetector converts the mixed optical signal, which includes the main return optical signal and backscattered spontaneous emission noise, collected by the sensor into a continuous analog voltage signal. This analog voltage signal is then sampled by a high-precision analog-to-digital converter at a sampling rate at least twice the highest frequency of the signal, converting it into a discrete-time digital waveform sequence to maintain signal integrity and avoid aliasing. Subsequently, a preset digital bandpass filter is used to filter this discrete-time digital waveform sequence. The center frequency and bandwidth of this filter are precisely set according to the known spectral characteristics of the backscattered spontaneous emission noise, thereby effectively separating and extracting the pure backscattered spontaneous emission noise component from the mixed signal. For example, the backscattered spontaneous emission noise component... The known spectral characteristics of spontaneously radiated noise specifically refer to the power spectral density curve of the back-radiated noise obtained by an optical spectrum analyzer under standard laboratory conditions before the sensor leaves the factory. This curve exhibits stable, broad-spectrum characteristics without sharp peaks. Its key quantitative characteristics include: the center wavelength is located around 1550 nm, the typical 3 dB bandwidth is about 15 nm, that is, the wavelength range corresponding to the power spectral density dropping by half from the peak is from 1542.5 nm to 1557.5 nm, the entire noise spectrum has a significant energy distribution in the range of 1525 nm to 1575 nm, and its power spectral density shape is approximately a Gaussian distribution with a flat top, and the roll-off slope on both sides of the center wavelength is about 0.5 dB per nanometer. Based on these specific known spectral characteristics, the center frequency of the digital bandpass filter was set to the optical frequency corresponding to 1550 nm. Its passband range was precisely set to a 3 dB bandwidth covering the entire 15 nm, appropriately extended to 20 nm to ensure the capture of the main energy of noise, while ensuring a sufficiently steep attenuation rate at the passband edges to effectively suppress the main signal components. Next, the root mean square (RMS) power value of the extracted backscattered noise component was calculated within a preset power calculation time window. This real-time measured RMS power value was then compared with a preset calibration curve established under precise laboratory conditions, showing the correspondence between noise power and the actual system gain. This allowed for the accurate deduction of the current system's actual gain value. Finally... The system compares its current actual gain with the preset ideal target gain, and then calculates the real-time gain compensation coefficient through a proportional-integral controller. This coefficient is then used as a multiplication factor on the original discrete-time digital waveform sequence to adjust its amplitude in real time. Finally, it outputs a standardized waveform timing data with stable amplitude that is independent of the sensor's operating state and environmental interference. In the process of calculating the real-time gain compensation coefficient, the proportional-integral controller processes the input gain difference in parallel through two paths. For example, in the proportional path, the difference between the actual gain measured at the current moment and the preset target gain is multiplied by a pre-set proportional coefficient, directly generating an instantaneous adjustment amount that is proportional to the current difference.Simultaneously, in the integral path, the controller continuously accumulates the gain difference of all consecutive sampling moments over a past period. This accumulated sum is multiplied by a pre-set integral time constant to generate an adjustment amount used to eliminate long-term steady-state errors. Subsequently, the proportional-integral controller algebraically adds the instantaneous adjustment amount generated by the proportional path to the accumulated adjustment amount generated by the integral path, and the sum is directly output as the real-time gain compensation coefficient for the current calculation cycle. The entire calculation process is repeated in each sampling cycle, thereby dynamically tracking and compensating for gain fluctuations caused by changes in sensor state or environment, providing a reliable and consistent benchmark for all subsequent signal processing stages.

[0054] Step S2: Perform adaptive variational mode decomposition on the standardized waveform time series data to obtain multiple essential mode functions. Based on the preset mapping table of particle size and mode function center frequency and the correlation coefficient between the essential mode functions and the standardized waveform time series data, select the target essential mode function set.

[0055] Specifically, in step S2, the number of decomposition modes and the penalty factor in the adaptive variational mode decomposition are parameter combinations adaptively determined by the optimization algorithm.

[0056] Specifically, in step S2, the parameter combination of the number of decomposed modes and the penalty factor in the adaptive variational mode decomposition takes the sparsity of multiple essential mode functions as the optimization objective.

[0057] Specifically, in step S2, the screening steps for the target essential modal function set include:

[0058] Based on a pre-defined mapping table between particle size and modal function center frequency, a subset of essential modal functions whose center frequencies match the target particle size range are selected from multiple essential modal functions.

[0059] The correlation coefficient between each essential mode function in the essential mode function subset and the standardized waveform time series data is calculated, and a second screening is performed using a preset correlation coefficient threshold to obtain the target essential mode function set.

[0060] In the specific implementation process, an adaptive variational mode decomposition algorithm is first applied to the standardized waveform time series data. The core parameters of this algorithm are the number of decomposition modes and the penalty factor, which are adaptively determined through optimization. For example, the sparrow search algorithm is used for parameter optimization. The initial sparrow population size is 50 sparrows. The position vectors of the sparrows are encoded as the parameter combination of the number of decomposition modes and the penalty factor. Each sparrow represents a set of candidate parameter combination solutions. The standardized waveform time series data is then subjected to variational mode decomposition using this set of parameter combinations to obtain a set of essential mode functions. Subsequently, the sum of the envelope entropies of all essential mode functions is calculated as the fitness value of this set of parameters. The lower the envelope entropy, the better the sparsity of the decomposed mode components. Using this as the optimization objective, the sparrow algorithm iteratively searches within a set range of 50 iterations using the position update rules of discoverers, followers, and watchdogs to find the optimal number of decomposition modes and penalty factor parameters that minimize the sum of envelope entropies. By combining these modal components, the sparsest possible decomposition effect is achieved. After parameter optimization, the original signal is subjected to variational mode decomposition again using the optimal number of decomposed modes and penalty factor to obtain multiple essential mode functions with different center frequencies. Then, a screening stage is initiated. First, a preset mapping table between particle size and mode function center frequency is used. This table is established through experimental calibration. For example, the table shows that particles with a diameter of 1 to 10 micrometers mainly excite essential mode functions with center frequencies between 10 and 100 kHz. Therefore, essential mode functions with center frequencies falling within this range are selected to form a primary subset, i.e., the essential mode function subset. Next, the Pearson correlation coefficient between each mode function in this subset and the original normalized waveform time series data is calculated. For example, the preset correlation coefficient threshold is 0.6. Only mode functions with correlation coefficients greater than this threshold are retained. The final target essential mode function set is then used for subsequent reconstruction.

[0061] Step S3: Superimpose the essential mode functions in the target essential mode function set to generate particulate matter characteristic waveform signals.

[0062] Specifically, in step S3, the superposition is linear superposition, and the correlation coefficient between each essential mode function participating in the superposition and the standardized waveform time series data is normalized and used as the weight coefficient of the superposition.

[0063] In the specific implementation process, the target essential mode function set is first preprocessed to ensure that all essential mode functions participating in the superposition have the same length and sampling time point to achieve waveform alignment. Then, based on the Pearson correlation coefficient between each essential mode function in the target essential mode function set and the original standardized waveform time series data, a set of weight coefficients is obtained through normalization. Finally, a weighted linear superposition is performed, multiplying each data point of each essential mode function by its corresponding normalized weight coefficient. Then, the values ​​of all weighted essential mode functions at the same time are added to generate the final particulate feature waveform signal. This weighted superposition process can be expressed as the value of the new signal at each time point being equal to the sum of the values ​​of each essential mode function at the same time point multiplied by its weight. This highlights the modal components with high correlation to the original signal and suppresses the components with low correlation, ultimately outputting a particulate feature waveform signal with enhanced signal-to-noise ratio for subsequent processing.

[0064] Step S4: Identify the particulate matter characteristic waveform signal using the dynamic threshold peak detection algorithm to obtain multiple independent signal events, and extract the multidimensional feature vector corresponding to each independent signal event.

[0065] Specifically, the following steps are included before step S4:

[0066] Calculate the average signal-to-noise ratio of the particulate matter characteristic waveform signal within a preset signal-to-noise ratio time window;

[0067] If the average signal-to-noise ratio is higher than the preset first signal-to-noise ratio threshold, then dynamic threshold peak detection is performed;

[0068] If the average signal-to-noise ratio is lower than the preset first signal-to-noise ratio threshold but higher than the preset second signal-to-noise ratio threshold, then the particulate matter characteristic waveform signal is subjected to noise reduction preprocessing based on wavelet transform before dynamic threshold peak detection is performed.

[0069] If the average signal-to-noise ratio is lower than the preset second signal-to-noise ratio threshold, the data within the preset signal-to-noise ratio time window is determined to be invalid and discarded. At the same time, a sensor cleaning and calibration prompt message is generated, wherein the first signal-to-noise ratio threshold is greater than the second signal-to-noise ratio threshold.

[0070] Specifically, in step S4, the dynamic threshold in the dynamic threshold peak detection algorithm is calculated based on the local statistical characteristics of the particulate matter characteristic waveform signal within a preset sliding time window.

[0071] In the specific implementation process, firstly, the signal quality of the particulate matter characteristic waveform signal is assessed. For example, the average signal-to-noise ratio (SNR) of the particulate matter characteristic waveform signal within a preset SNR time window is calculated. Specifically, this is done by calculating the ratio of the root mean square (RMS) value of the effective signal component to the RMS value of the noise component within the preset SNR time window. For example, the preset SNR time window is 200ms. If the calculated average SNR is higher than a preset first SNR threshold, the dynamic threshold peak detection process is directly initiated. If the average SNR is lower than the preset first SNR threshold but higher than a preset second SNR threshold, the signal undergoes wavelet transform-based noise reduction preprocessing, followed by peak detection. If the average SNR is lower than the preset second SNR threshold, the data for that time period is deemed invalid, and a sensor cleaning prompt is generated. The first signal-to-noise ratio (SNR) threshold is greater than the second SNR threshold. In the dynamic threshold peak detection, a 20ms sliding time window is set. The local mean and standard deviation of the particulate matter characteristic waveform signal within the sliding time window are calculated in real time. The dynamic threshold is set as the local mean plus three times the standard deviation. When the amplitude of the particulate matter characteristic waveform signal exceeds this dynamic threshold and meets the condition that the pulse width is between 0.1ms and 5ms, it is identified as a valid independent signal event. For each detected independent signal event, its multidimensional feature vector is extracted, including nine feature parameters in the time domain: peak value, amplitude pulse width, rise time, fall time, and integral area, as well as in the frequency domain: spectral centroid, spectral bandwidth, and spectral entropy obtained through fast Fourier transform. These form a complete feature description of the independent signal event for subsequent classification and recognition.

[0072] Step S5: Input all multidimensional feature vectors into a pre-trained event classifier to obtain target independent signal events labeled as target suspended particulate matter, and calculate the equivalent particle size and relative mass of the target independent signal events based on a regression model.

[0073] In the specific implementation process, firstly, the multidimensional feature vector corresponding to each independent signal event is input into a pre-trained event classifier. This event classifier adopts a support vector machine model based on radial basis function kernel function. Its training data comes from a large number of historical signal events of known categories, including target suspended particulate matter events and various common interference events, for example, including but not limited to bubble vibration noise. During training, grid search is used to optimize the penalty parameters and kernel function parameters to maximize the classification accuracy. The classifier outputs the probability value of each event belonging to target suspended particulate matter. When the probability value is higher than the preset probability threshold, the event is officially marked as a target independent signal event. Subsequently, for each marked target independent signal event, The same multidimensional feature vectors are input into a pre-trained regression model, which employs the support vector regression algorithm. This model also uses a radial basis function kernel and is trained on a standard particulate matter event dataset with known equivalent particle size and relative mass. During training, the feature vectors are used as input, and the logarithms of particle size and mass are used as outputs to better fit the nonlinear relationship. The regression model directly calculates the equivalent particle size value (in micrometers) and relative mass value (dimensionless) for each target independent signal event through the learned mapping relationship, thus completing the accurate inversion from signal features to physical parameters. The entire classification and regression process ensures that only high-confidence target independent signal events can be used for physical quantity calculations, guaranteeing the reliability of the final results.

[0074] Step S6: Within a preset independent time window, count the number of all target independent signal events and accumulate the corresponding relative mass. Perform correction and compensation through a counting correction algorithm, and output the particulate matter number concentration and total mass concentration of suspended matter within the preset independent time window.

[0075] Specifically, in step S6, the counting correction algorithm is based on the Poisson distribution theory, and its correction factor is calculated based on the average time interval of the target independent signal events.

[0076] In the specific implementation process, a fixed-length independent time window is first set, for example, the length of the independent time window is 1 minute. Within this independent time window, all target independent signal events marked as target suspended particulate matter are counted one by one to obtain the original count results. At the same time, the relative mass values ​​corresponding to these target independent signal events are summed to obtain the initial value of the total mass. Then, a counting correction algorithm based on Poisson distribution theory is applied for compensation calculation. First, the time points of occurrence of all target independent signal events within the independent time window are counted, the interval between adjacent time points is calculated, and their average value is obtained to obtain the average time interval. Then, based on the characteristics of Poisson distribution, the number of actually observed events is regarded as the realization value of a random process. The event occurrence rate parameter of this process is inversely estimated through maximum likelihood estimation, and then the event occurrence rate parameter is calculated using the same independent time window. The expected total number of events that should theoretically occur within an independent time window is calculated. The difference between the theoretical expected value and the actual observed value is then added to the original count as a correction to obtain the corrected number of events. Similarly, based on the relative mass of each independent signal event, the mass of the correction portion is added to the initial total mass value to obtain the corrected total mass value. The corrected number of events is divided by the water volume corresponding to the independent time window to obtain the particulate matter number concentration (in units per milliliter). The corrected total mass is divided by the water volume to obtain the total mass concentration (in milligrams per liter). Finally, these two concentration values ​​are output as the accurate measurement results within the independent time window. The entire correction process effectively compensates for the missed count error caused by signal overlap and indistinguishability due to the simultaneous passage of particulate matter through the detection area, thus improving the measurement accuracy under high concentration conditions.

[0077] The working principle of the water quality suspended solids detection method based on sensor data provided by this invention is as follows:

[0078] This invention first utilizes the inherent backscattered spontaneous emission noise within the sensor as an intrinsic reference. By monitoring its power changes in real time and comparing them with a preset calibration curve, the system gain is dynamically calibrated, thereby eliminating signal baseline drift caused by environmental fluctuations and device aging, and generating standardized waveform data with stable amplitude. Then, an adaptive variational mode decomposition technique is employed to adaptively decompose the standardized signal into a series of essential mode functions. The number of modes and penalty factors in this decomposition process are automatically determined by an optimization algorithm targeting modal component sparsity to ensure optimal decomposition results. Next, based on the physical mapping relationship between particle size and modal center frequency, and the correlation between each mode and the original signal, a dual screening process is performed to accurately extract the signal components dominated by the target particles. Subsequently, a particle characteristic waveform signal with significantly improved signal-to-noise ratio is reconstructed through linear superposition with correlation coefficient weights. Finally, an intelligent signal quality gating mechanism is introduced in the event detection stage based on real-time signal... The system employs an automatic noise ratio selection strategy, employing direct detection followed by noise reduction or data discarding to ensure that subsequent processing is only performed when the signal quality is reliable. A dynamic threshold peak detection algorithm automatically adjusts the detection threshold by analyzing the local statistical characteristics of the signal within a sliding window, accurately identifying the transient pulse events generated by each particle and extracting its multidimensional time-frequency features. The feature vector of each event is input into a pre-trained classifier to distinguish target particles from interfering substances. A regression model is then used to calculate the equivalent particle size and relative mass of each event. Finally, the number of valid events is counted within a fixed time window, and their masses are accumulated. A Poisson-based counting correction algorithm compensates for missed count errors caused by signal overlap. Ultimately, the system synchronously outputs the particle number concentration and total mass concentration, forming a complete closed loop from signal perception, feature extraction, event recognition, physical inversion to concentration calculation. This enables high-precision online monitoring of suspended solids concentration and particle size distribution in water bodies.

[0079] Example 2

[0080] A water quality suspended solids detection system based on sensor data, in its specific implementation process, such as... Figure 2 As shown, it illustrates a modular structure diagram of a water quality suspended solids detection system based on sensor data according to the present invention, comprising:

[0081] The signal standardization module 100 is used to acquire the mixed optical signal containing the main return optical signal and the backspontaneous emission noise collected by the sensor; based on the power of the backspontaneous emission noise, the mixed optical signal is used to generate standardized waveform timing data through dynamic gain compensation.

[0082] The mode decomposition screening module 200 is used to perform adaptive variational mode decomposition on standardized waveform time series data to obtain multiple essential mode functions. Based on the preset particle size and mode function center frequency mapping table and the correlation coefficient between the essential mode functions and the standardized waveform time series data, the target essential mode function set is screened out.

[0083] The signal waveform reconstruction module 300 is used to superimpose the essential mode functions in the target essential mode function set to generate particulate characteristic waveform signals;

[0084] The pulse event detection module 400 is used to identify particulate matter characteristic waveform signals through a dynamic threshold peak detection algorithm, obtain multiple independent signal events, and extract the multi-dimensional feature vector corresponding to each independent signal event.

[0085] The target recognition and inversion module 500 is used to input all multidimensional feature vectors into a pre-trained event classifier to obtain target independent signal events labeled as target suspended particulate matter, and calculate the equivalent particle size and relative mass of the target independent signal events based on a regression model.

[0086] The result calculation and output module 600 is used to count the number of all target independent signal events and accumulate the corresponding relative mass within a preset independent time window, perform correction and compensation through a counting correction algorithm, and output the particulate number concentration and total mass concentration of suspended matter within the preset independent time window.

[0087] The working principle of the water quality suspended solids detection system based on sensor data provided by this invention is as follows:

[0088] First, the signal standardization module 100 receives the mixed optical signal collected by the sensor, converts it into a voltage signal through its built-in photoelectric conversion circuit, and then obtains a digital waveform sequence through a high-speed analog-to-digital converter. A specific filtering algorithm is used to extract the backscattered spontaneous emission noise component from the digital waveform sequence, and its power value is calculated in real time. Then, the calibration curve stored in memory is consulted to obtain the current actual gain value of the system. Finally, a dynamic gain compensation algorithm is used to output standardized waveform time-series data with stable amplitude. The mode decomposition and filtering module 200 receives the standardized data and calls an adaptive variational mode decomposition algorithm to decompose the standardized waveform time-series data, generating multiple essential mode functions (EMFs). Then, a preliminary screening is performed according to a preset particle size and center frequency mapping table. A second screening is performed by calculating the Pearson correlation coefficient between each EMF and the original standardized waveform time-series data, ultimately obtaining the target essential mode function set. Signal waveform reconstruction is then performed. Module 300 performs weighted linear superposition of the selected target modal functions to generate a particulate matter feature waveform signal with enhanced signal-to-noise ratio. The pulse event detection module 400 uses a dynamic threshold peak detection algorithm to detect the particulate matter feature waveform signal, thereby accurately identifying the independent signal events generated by each particulate matter and extracting its multidimensional feature vector. The target recognition inversion module 500 includes a pre-trained event classifier and a regression model classifier. It uses a support vector machine algorithm to classify the input multidimensional feature vector and identify the target independent signal events of the target suspended particulate matter. The regression model calculates the equivalent particle size and relative mass of each target independent signal event. The result calculation and output module 600 counts the number of target events and accumulates their relative masses within a set independent time window. Finally, it applies a counting correction algorithm based on Poisson distribution theory to compensate for possible missed counts and outputs accurate particulate matter number concentration and total mass concentration.

[0089] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0090] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0091] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for detecting suspended solids in water based on sensor data, characterized in that, The method for detecting suspended solids in water includes: S1: Acquire the mixed optical signal containing the main return optical signal and the backspontaneous emission noise collected by the sensor; based on the power of the backspontaneous emission noise, generate standardized waveform timing data of the mixed optical signal through dynamic gain compensation; S2: Perform adaptive variational mode decomposition on the standardized waveform time series data to obtain multiple essential mode functions. Based on the preset mapping table of particle size and mode function center frequency and the correlation coefficient between the essential mode functions and the standardized waveform time series data, select the target essential mode function set. S3: Superimpose the essential mode functions in the target essential mode function set to generate particulate matter characteristic waveform signals; S4: The dynamic threshold peak detection algorithm is used to identify the characteristic waveform signal of particulate matter, obtain multiple independent signal events, and extract the multi-dimensional feature vector corresponding to each independent signal event; S5: Input all multidimensional feature vectors into a pre-trained event classifier to obtain target independent signal events labeled as target suspended particulate matter, and calculate the equivalent particle size and relative mass of the target independent signal events based on a regression model; S6: Within a preset independent time window, count the number of all target independent signal events and accumulate the corresponding relative mass. Then, perform correction and compensation through a counting correction algorithm, and output the particulate number concentration and total mass concentration of suspended matter within the preset independent time window. In step S1, the steps for generating standardized waveform timing data include: The mixed optical signal, which includes the main return optical signal and the backscattered spontaneous emission noise, is converted into a voltage signal. The voltage signal is sampled and converted into a discrete-time digital waveform sequence; The discrete-time digital waveform sequence is processed by a digital bandpass filter to extract the backscattered spontaneous noise component. The power of the back-emitted spontaneous noise is calculated and compared with the preset calibration curve to obtain the actual gain. Based on the preset target gain and actual gain, the gain compensation coefficient is calculated, and the gain compensation coefficient is used to perform gain compensation on the discrete-time digital waveform sequence to generate standardized waveform timing data.

2. The method for detecting suspended solids in water based on sensor data according to claim 1, characterized in that, In step S2, the number of decomposition modes and the penalty factor in adaptive variational mode decomposition are parameter combinations adaptively determined by an optimization algorithm.

3. The method for detecting suspended solids in water based on sensor data according to claim 2, characterized in that, In step S2, the parameter combination of the number of decomposed modes and the penalty factor in the adaptive variational mode decomposition takes the sparsity of multiple essential mode functions as the optimization objective.

4. The method for detecting suspended solids in water based on sensor data according to claim 3, characterized in that, In step S2, the screening steps for the target essential mode function set include: Based on a pre-defined mapping table between particle size and modal function center frequency, a subset of essential modal functions whose center frequencies match the target particle size range are selected from multiple essential modal functions. The correlation coefficient between each essential mode function in the essential mode function subset and the standardized waveform time series data is calculated, and a second screening is performed using a preset correlation coefficient threshold to obtain the target essential mode function set.

5. The method for detecting suspended solids in water based on sensor data according to claim 4, characterized in that, In step S3, the superposition is linear superposition, and the correlation coefficient between each essential mode function participating in the superposition and the standardized waveform time series data is normalized and used as the weight coefficient of the superposition.

6. The method for detecting suspended solids in water based on sensor data according to claim 5, characterized in that, The following steps are included before step S4: Calculate the average signal-to-noise ratio of the particulate matter characteristic waveform signal within a preset signal-to-noise ratio time window; If the average signal-to-noise ratio is higher than the preset first signal-to-noise ratio threshold, then dynamic threshold peak detection is performed; If the average signal-to-noise ratio is lower than the preset first signal-to-noise ratio threshold but higher than the preset second signal-to-noise ratio threshold, then the particulate matter characteristic waveform signal is subjected to noise reduction preprocessing based on wavelet transform before dynamic threshold peak detection is performed. If the average signal-to-noise ratio is lower than the preset second signal-to-noise ratio threshold, the data within the preset signal-to-noise ratio time window is determined to be invalid and discarded. At the same time, a sensor cleaning and calibration prompt message is generated, wherein the first signal-to-noise ratio threshold is greater than the second signal-to-noise ratio threshold.

7. The method for detecting suspended solids in water based on sensor data according to claim 6, characterized in that, In step S4, the dynamic threshold in the dynamic threshold peak detection algorithm is calculated based on the local statistical characteristics of the particulate matter characteristic waveform signal within a preset sliding time window.

8. The method for detecting suspended solids in water based on sensor data according to claim 7, characterized in that, In step S6, the counting correction algorithm is based on the Poisson distribution theory, and its correction factor is calculated based on the average time interval of the target independent signal events.

9. A water quality suspended solids detection system based on sensor data, characterized in that, The method for detecting suspended solids in water based on sensor data, as described in any one of claims 1 to 8, comprises: The signal standardization module is used to acquire the mixed optical signal, which includes the main return optical signal and the backspontaneous emission noise, collected by the sensor; based on the power of the backspontaneous emission noise, the mixed optical signal is used to generate standardized waveform timing data through dynamic gain compensation. The mode decomposition and screening module is used to perform adaptive variational mode decomposition on standardized waveform time series data to obtain multiple essential mode functions. Based on the preset mapping table of particle size and mode function center frequency and the correlation coefficient between essential mode functions and standardized waveform time series data, the target essential mode function set is screened out. The signal waveform reconstruction module is used to superimpose the essential mode functions in the target essential mode function set to generate particulate characteristic waveform signals; The pulse event detection module is used to identify particulate matter characteristic waveform signals through a dynamic threshold peak detection algorithm, obtain multiple independent signal events, and extract the multi-dimensional feature vector corresponding to each independent signal event. The target recognition and inversion module is used to input all multidimensional feature vectors into a pre-trained event classifier to obtain target-independent signal events labeled as target suspended particulate matter, and calculate the equivalent particle size and relative mass of the target-independent signal events based on a regression model. The result calculation and output module is used to count the number of all target independent signal events and accumulate the corresponding relative mass within a preset independent time window. The counting correction algorithm is used for correction and compensation, and the particulate matter number concentration and total mass concentration of suspended matter within the preset independent time window are output. In the signal normalization module, the steps for generating normalized waveform time-series data include: The mixed optical signal, which includes the main return optical signal and the backscattered spontaneous emission noise, is converted into a voltage signal. The voltage signal is sampled and converted into a discrete-time digital waveform sequence; The discrete-time digital waveform sequence is processed by a digital bandpass filter to extract the backscattered spontaneous noise component. The power of the back-emitted spontaneous noise is calculated and compared with the preset calibration curve to obtain the actual gain. Based on the preset target gain and actual gain, the gain compensation coefficient is calculated, and the gain compensation coefficient is used to perform gain compensation on the discrete-time digital waveform sequence to generate standardized waveform timing data.

Images