Drainage pipe network pollution monitoring and advanced early warning system based on Internet of Things

Abstract

The invention relates to the technical field of pollution discharge intelligent sensing monitoring and early warning systems, in particular to a drainage pipe network pollution monitoring and advanced early warning system based on the Internet of Things. The method comprises the steps of firstly obtaining water quality data of a current monitoring period; further identifying a water quality fluctuation event according to fluctuation characteristics of data in each water quality data dimension, and extracting an event center moment and an event waveform characteristic vector; further constructing a matching deviation matrix; further obtaining a matching deviation parameter based on the optimal matching result, comparing the matching deviation parameter with a preset historical mode statistical baseline, and obtaining a mode deviation degree corresponding to the COD data dimension and each non-COD data dimension; and finally, based on the mode deviation degree, decomposing the COD data and extracting a real pollution load, and generating an early warning report based on the real pollution load, so that the false alarm rate is reduced, and the early warning accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing monitoring and early warning systems for sewage discharge, specifically to an Internet of Things-based system for monitoring and providing early warning of pollution in drainage pipe networks. Background Technology

[0002] In coastal cities employing combined sewer systems, the water quality and quantity of the pipe network are influenced by a combination of factors, including rainfall runoff and ocean tides. Existing technologies typically employ IoT-based online monitoring systems, deploying water quality sensors such as Chemical Oxygen Demand (COD) at pipe network nodes to monitor pollution levels.

[0003] In monitoring systems deployed in coastal combined sewer systems, a sharp rise in COD signal is detected when rainfall and tides occur simultaneously. Existing methods cannot effectively distinguish the cause of this signal. COD signal fluctuations may be caused by actual runoff pollution, primarily by measurement interference from brine backflow, or a combination of both. Due to the ambiguity in the fundamental causes of the pollution signal, the system's early warning signals lack decision-making value, potentially leading to overreactions to measurement interference or environmental impacts due to failure to identify actual pollution, ultimately resulting in false alarms and missed warnings. Summary of the Invention

[0004] To address the technical problem of false alarms and missed alarms in existing technologies due to their inability to effectively distinguish the causes of COD signals, the present invention aims to provide an IoT-based drainage network pollution monitoring and early warning system. The specific technical solution adopted is as follows: A drainage pipe network pollution monitoring and early warning system based on the Internet of Things, the system comprising: The data acquisition module is used to acquire water quality data such as COD, UV, and EC for the current monitoring period; The feature analysis module is used to identify water quality fluctuation events and obtain the event center time based on the fluctuation characteristics of data within each water quality data dimension; extract event waveform feature vectors based on the waveform asymmetry and rate of change of each water quality fluctuation event; the non-COD data dimension includes UV and EC data dimensions; construct a matching deviation matrix based on the difference between the event waveform feature vectors of the COD data dimension and each non-COD data dimension, combined with the time deviation between the water quality fluctuation events; obtain the matching deviation parameter between the COD data dimension and each non-COD data dimension based on the optimal matching result of the matching deviation matrix; and compare the matching deviation parameter with a preset historical pattern statistical baseline to obtain the pattern deviation degree corresponding to the COD data dimension and each non-COD data dimension. The analysis and early warning module is used to decompose COD data and extract the true pollution load based on the pattern deviation, and generate an early warning report based on the true pollution load.

[0005] Furthermore, the method for obtaining the event waveform feature vector includes: In each water quality data segment corresponding to the water quality fluctuation event, the peak tail length is obtained based on the duration difference on both sides of the event center time; the average rise rate from the start time to the event center time is obtained. An event waveform feature vector is constructed based on the peak tail and the average rise rate of each water quality fluctuation event.

[0006] Furthermore, the method for obtaining the matching deviation matrix includes: Select UV and EC data dimensions one by one as comparison dimensions; obtain the overall waveform difference parameters based on the overall difference between the event waveform feature vectors of the COD data dimension and the comparison dimensions. Based on the time deviation between the event center time of each water quality fluctuation event in the COD data dimension and the event center time of each water quality fluctuation event in the dimension to be compared, and combined with the overall waveform difference parameter, the time deviation cost is obtained; and a matching deviation matrix is ​​constructed based on all the time deviation costs.

[0007] Furthermore, the method for obtaining the pattern deviation includes: The preset historical pattern statistical baseline includes the mean and standard deviation of the matching deviation parameters corresponding to the two patterns: runoff flushing pattern and saltwater backflow interference pattern. Within the current monitoring period, based on the historical pattern statistical baseline of the runoff flushing pattern, the standard score of the matching deviation parameter between the COD data dimension and the UV data dimension is obtained as the pattern deviation degree. Based on the historical pattern statistical baseline of the brine backflow interference pattern, the standard score of the matching deviation parameter between the COD data dimension and the EC data dimension is used as the pattern deviation degree.

[0008] Furthermore, the method for obtaining the actual pollution load includes: Based on the relative distribution of the deviations of the two modes corresponding to the current cycle, the runoff scour contribution weight and the brine backflow interference weight are obtained. A runoff scour response model based on pre-established COD and UV data under the runoff scour mode, and a brine backflow interference response model based on pre-established COD and EC data under the brine backflow interference mode are obtained. Based on the runoff scour contribution weight, the brine backflow interference weight, the runoff scour response model, and the brine backflow interference response model, the COD data at each time point in the current monitoring period are dynamically decomposed to obtain the true pollution load at each time point.

[0009] Furthermore, based on the relative distribution of the deviations of the two modes corresponding to the current cycle, the method for obtaining the runoff scour contribution weight and the brine backflow interference weight includes: The mapping results of negatively correlated normalization of the squared values ​​of the deviations of the two modes are used as the runoff scour contribution weight and the brine backflow interference weight, respectively.

[0010] Furthermore, when the actual pollution load exceeds the preset actual pollution alarm threshold, an early warning is triggered, and the early warning report includes at least: the early warning status, the COD data at the time of the early warning, and the actual pollution load after decomposition.

[0011] Furthermore, the method for obtaining the water quality fluctuation event includes: Based on the preset minimum peak height and preset minimum peak spacing, in each dimension of water quality data, peaks are extracted and the peak points are used as fluctuation event points, and the event center time is marked. At each event center moment, the local minimum point with the nearest time domain and the corresponding peak amplitude of the data value is obtained on both sides of the time sequence, and used as the time endpoint to obtain the water quality fluctuation event.

[0012] Furthermore, the optimal matching result of the matching deviation matrix is ​​obtained through the Hungarian algorithm.

[0013] Furthermore, the method for obtaining the matching deviation parameter includes: In the optimal matching result of the matching deviation matrix, the average value of the corresponding element values ​​of all matching pairs in the matching deviation matrix is ​​used as the matching deviation parameter between the two corresponding data dimensions.

[0014] The present invention has the following beneficial effects: This invention first acquires water quality data for COD, UV, and EC during the current monitoring period, providing a data foundation for subsequent analysis. Further, based on the fluctuation characteristics of data within each water quality data dimension, it identifies water quality fluctuation events and obtains the event center time, automatically extracting the true fluctuation events from continuous monitoring data to avoid misjudgments caused by noise or weak disturbances. It further acquires event waveform feature vectors, capturing the fluctuation characteristics of the corresponding waveforms, providing a non-time-dimensional basis for distinguishing between the two causes. Finally, based on the differences between the event waveform feature vectors of the COD data dimension and each non-COD data dimension, combined with the time deviation between water quality fluctuation events, a matching deviation matrix is ​​constructed to characterize cross-dimensional fluctuations. The invention analyzes the combined deviation of morphological consistency and temporal synchronization among water quality events. Based on the optimal matching results, it obtains matching deviation parameters to characterize and quantify the consistency and coupling tightness of different water quality indicators' responses to the same water quality disturbance. These parameters are then compared with a preset historical pattern statistical baseline to obtain the pattern deviation between the COD data dimension and each non-COD data dimension, demonstrating the degree of deviation of water quality fluctuations in the current monitoring period from historical patterns in terms of the synergistic relationship between indicators. Finally, based on the pattern deviation, the COD data is decomposed and the true pollution load is extracted, removing the contribution of background interference such as brine backflow to the measurement signal. An early warning report is then generated based on the true pollution load. This invention analyzes the overall waveform morphology and temporal correlation of COD, UV, and EC fluctuation events, dynamically identifies the dominant causes of the signal, and decomposes and extracts the true pollution load, thereby reducing false alarms caused by brine backflow and making early warnings more targeted and reliable. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system block diagram of a drainage network pollution monitoring and early warning system based on the Internet of Things, provided in one embodiment of the present invention. Figure 2 This is a flowchart illustrating a method for obtaining actual pollution load according to an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an Internet of Things-based drainage network pollution monitoring and early warning system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details a specific solution for an IoT-based drainage network pollution monitoring and early warning system provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a system block diagram of an IoT-based drainage network pollution monitoring and early warning system according to an embodiment of the present invention. The system includes: a data acquisition module 101, a feature analysis module 102, and an analysis and early warning module 103.

[0021] The data acquisition module 101 is used to acquire water quality data such as COD, UV, and EC for the current monitoring period.

[0022] In one embodiment of the invention, data is processed in batches over a preset monitoring period (e.g., 24 hours). The data acquisition unit, deployed at the same water flow cross-section of the drainage network, is an integrated sensor group. In this sensor group, all sensor probes are physically installed adjacent to each other to ensure the consistency of all sensor data in time and space.

[0023] The system synchronously acquires raw readings for the following time series at a fixed sampling frequency (e.g., once per minute): raw readings from the Chemical Oxygen Demand (COD) sensor; raw readings from the Ultraviolet Absorbance at 254 nm (UV254, or UV) sensor; raw readings from the Electrical Conductivity (EC) sensor; and raw readings from the Water Level (L) sensor. Simultaneously, the system synchronously acquires local rainfall intensity sequences from external meteorological service data.

[0024] Considering that the raw data collected by the sensor in the field environment inevitably contains high-frequency fluctuations caused by water turbulence in the pipe or electrical equipment, the raw data is preprocessed. Specifically, a center moving average filter is used with a window length of 5. That is, with each data point as the center, two adjacent data points are obtained on both sides of the time sequence of each data point, and the average of these 5 data points is used as the smoothing filter result for each data point. The raw COD, UV, EC, and water level readings for the current monitoring period were filtered by a central moving average to obtain water quality data for COD, UV, and EC, as well as water level data, providing a data foundation for subsequent analysis.

[0025] It should be noted that the working principles, installation, and settings of the various sensors are well-known technologies. The monitoring cycle, sampling frequency, and window length of the center moving average filter can be adjusted and set by the implementer. In other embodiments of the present invention, low-pass filters such as Savitzky-Golay can also be used for preprocessing, which will not be elaborated further.

[0026] The feature analysis module 102 is used to identify water quality fluctuation events and obtain the event center time based on the fluctuation characteristics of the data within each water quality data dimension; extract event waveform feature vectors based on the waveform asymmetry and rate of change of each water quality fluctuation event; the non-COD data dimension includes UV and EC data dimensions; construct a matching deviation matrix based on the difference between the event waveform feature vectors of the COD data dimension and each non-COD data dimension, combined with the time deviation between water quality fluctuation events; obtain the matching deviation parameter between the COD data dimension and each non-COD data dimension based on the optimal matching result of the matching deviation matrix; and compare the matching deviation parameter with the historical pattern statistical baseline to obtain the pattern deviation degree corresponding to the COD data dimension and each non-COD data dimension.

[0027] Existing methods for distinguishing different water quality events in a pipe network primarily rely on the temporal synchronicity of signals from different parameters. However, in complex scenarios where rainfall and tides may occur simultaneously, actual runoff scouring events and brine backflow disturbance events may highly overlap in time, rendering methods based solely on temporal synchronicity ineffective.

[0028] To this end, the feature analysis module 102 constructs an analysis framework based on the correlation between the event waveform feature vector and the event center time, ensuring that the attribution judgment not only utilizes the temporal correlation between events, but also supplements and corrects them through the overall difference in morphological features, thereby effectively overcoming the ambiguity of simple time synchronization analysis in composite events.

[0029] Considering that water quality data is a continuous time series containing environmental background, water quality fluctuation events are identified and the center time of the events are obtained based on the fluctuation characteristics of the data within each water quality data dimension. The real fluctuation events are automatically extracted from the continuous monitoring data to avoid misjudgment caused by noise or weak disturbances, so that the system can focus on the key information of significant data changes.

[0030] Preferably, in one embodiment of the present invention, considering that when a water quality fluctuation event occurs, it will cause water quality data fluctuation and present a peak shape, the find_peaks function of the SciPy library is used to extract the peak in each water quality data dimension based on the preset minimum peak height and preset minimum peak spacing, and the peak point is used as the fluctuation event point, and the event center time is marked. At the center of each event, the local minimum points in the time domain that are closest to the peak amplitude of the data value are obtained on both sides of the time sequence and used as time endpoints to obtain water quality fluctuation events.

[0031] As an example, with a preset ratio of 10%, taking any event center moment as an example, specifically searching from the event center moment in the negative direction of the time domain, the first point that satisfies the condition that the data value is a local minimum and the data value is less than 10% of the peak amplitude of the event center moment is obtained as a time endpoint; similarly, searching in the positive direction of the time domain yields another time endpoint. The data segment between the two time endpoints is regarded as a water quality fluctuation event, and the time point of the peak point (the maximum value point of the data segment) is the event center moment, using the peak to represent the fluctuation characteristics of the data.

[0032] It should be noted that the find_peaks function in the SciPy library is a commonly used peak detection tool for those skilled in the art. It provides rich parameter control, including height threshold, distance threshold, width threshold, etc. By taking the preset minimum peak height and preset minimum peak spacing as input control parameters, the filtered peaks can be obtained, and the time point of the peak point corresponding to the event center time can be obtained. After initial deployment, the system first uses water quality data collected during an initial learning period (e.g., two weeks) as a sample. It calculates the P-th percentile (e.g., P=75) of all potential peak amplitudes in the sample data as a reference value for the preset minimum peak height, and calculates the Q-th percentile (e.g., Q=50, i.e., the median) of the time intervals between all adjacent peaks as a reference value for the preset minimum peak spacing. Implementers can fine-tune the selection of percentiles and the setting of preset ratios based on the sensitivity requirements of specific monitoring scenarios. In other embodiments of the invention, half-peak width (i.e., the full width corresponding to half the peak height) or other event boundary delimitation methods known in the art can also be used to obtain water quality fluctuation events.

[0033] It should be noted that the analysis process is the same for each monitoring cycle, the same for each water quality data dimension, and the same for each water quality fluctuation event. Only one example will be described here, and it will not be repeated.

[0034] Since runoff scouring is a process in which pollutants gradually accumulate and slowly dilute with the runoff, while brine backflow is a process in which a high-salinity water front rapidly advances, the waveform morphology (symmetry) and rate of change of the corresponding water quality fluctuation events are different. Therefore, based on the asymmetry and rate of change of the waveform of each water quality fluctuation event, the event waveform feature vector is extracted to capture the fluctuation characteristics of the corresponding waveform, providing a non-time dimension basis for subsequent differentiation of the two causes.

[0035] Preferably, in one embodiment of the present invention, it is considered that the signal waveform of the event corresponding to runoff scour often exhibits an asymmetrical characteristic of a gentle rise and a trailing fall; while the signal waveform of the event corresponding to brine backflow often exhibits a symmetrical characteristic of a rapid rise and a rapid fall. The greater the difference in duration between the two sides of the event center time, the stronger the asymmetry of the waveform (peak shape). Therefore, in the water quality data segment corresponding to each water quality fluctuation event, the peak tail is obtained based on the difference in duration between the two sides of the event center time to characterize the waveform asymmetry. Considering the gradual convergence of runoff and the different rise rates of the waveform caused by the rapid advance of the high salinity water front, the average rise rate from the start time to the event center time is obtained to reflect the rate of change of the waveform. The event waveform feature vector is constructed based on the peak tail and average rise rate of each water quality fluctuation event.

[0036] As an example, in the water quality data segment corresponding to each water quality fluctuation event, the time interval (time length) between the event center time and the event end time is used as the numerator, and the time interval between the event start time and the event center time is used as the denominator. The ratio of the fractions is used as the peak tail degree. The larger the numerator is compared with the denominator, the more obvious the tail feature of the waveform, the stronger the asymmetry, and the greater the peak tail degree. The average rise rate from the start time to the event center time is obtained based on the two-point slope formula (two-point slope method). After linear normalization in the corresponding data dimension, the normalized results of the peak tail and the average rise rate are used to construct the event waveform feature vector of the corresponding water quality fluctuation event. Both elements in the event waveform feature vector are dimensionless.

[0037] It should be noted that the minimum denominator is the unit time interval, which is definitely not zero; the data dimension composed of the same type of parameters of all monitoring cycles is used as the data dimension used for the normalization of the corresponding parameters; it can also be limited to the data dimension composed of parameters obtained in an initial learning period (e.g., two weeks), and when the maximum value of the parameters exceeds the initial learning period, it is directly set to 1; linear normalization is a well-known technique in the art and will not be elaborated further.

[0038] In another embodiment of the present invention, the implementer may also use the absolute value of the difference in the time length of the water quality data segments on both sides of the event center time as the numerator, the time length of the water quality data segments as the denominator, and the ratio of the fractions as the peak shape asymmetry, to show the asymmetry of the waveform, and replace the peak shape tailing degree to construct the event waveform feature vector.

[0039] COD characterizes the overall oxidative load of organic pollutants in water bodies, while UV and EC correspond to the light absorption characteristics of organic matter and changes in ion concentration in water bodies, respectively. Water quality disturbances often exhibit "synchronous but not completely consistent" responses across multiple dimensions. Therefore, comparing COD with UV and COU with EC allows for the analysis of the true pollution patterns corresponding to COD data. Non-COD data dimensions include UV and EC data dimensions. Considering that the event waveform feature vector characterizes the waveform change mechanism of water quality fluctuation events and reflects the change mechanism of pollutant concentration, the differences between event waveform feature vectors reflect the different responses of different dimensions to the same pollution disturbance, and the time deviation between water quality fluctuation events corresponds to the response delay or actual asynchrony of different sensors, a matching deviation matrix is ​​constructed based on the differences between the event waveform feature vectors of the COD data dimension and each non-COD data dimension, combined with the time deviation between water quality fluctuation events. This matrix characterizes the degree of joint deviation between the morphological consistency and time synchronization of cross-dimensional events, preparing for subsequent analysis by combining multiple dimensions to obtain the optimal matching results.

[0040] Preferably, in one embodiment of the present invention, UV and EC data dimensions are selected one by one as the dimensions to be compared, so as to facilitate the comparison of COD with UV and COU with EC one by one; Considering that the greater the overall difference between events across different data dimensions, the more significant the differences in their physicochemical reaction mechanisms, waveform evolution patterns, and other factors, this overall morphological difference provides prior knowledge for assessing the reliability of time matching between events: the greater the morphological difference, the higher the probability of accidental time synchronization, and the higher the reliability of their time correlation should be, while the cost of time deviation should be increased accordingly. Therefore, based on the overall difference between the event waveform feature vectors of the COD data dimension and the dimension to be compared, the overall waveform difference parameter is obtained.

[0041] As an example, take any event waveform feature vector in the COD data dimension and any event waveform feature vector in the dimension to be compared, forming a vector pair. The water quality fluctuation event corresponding to each vector pair forms an event pair. Calculate the Euclidean distance d between the two vectors in the vector pair, representing the difference between the two vectors. Use the average value of d for all vector pairs as the overall waveform difference parameter, reflecting the degree of difference in the overall waveform morphology between events in the COD dimension and events in the dimension to be compared within the current monitoring period.

[0042] Based on the time deviation between the event center time of each water quality fluctuation event in the COD data dimension and the event center time of each water quality fluctuation event in the dimension to be compared, and combined with the overall waveform difference parameter, the time deviation cost is obtained; and a matching deviation matrix is ​​constructed based on all time deviation costs.

[0043] As an example, the time interval between the event center times of the two events in the event tuple is taken as the time deviation, the sum of constant 1 and the overall waveform difference parameter is taken as the time deviation correction coefficient, and the product of the time deviation and the corresponding time deviation correction coefficient is taken as the time deviation cost of the corresponding event tuple. Sort the events in each dimension according to the event order to obtain the event index. Use the event index i of the COD dimension as the row coordinate of the matrix and the event index k of the dimension to be compared as the column coordinate of the matrix. Fill in the time deviation cost corresponding to each (i,k) position to construct the matching deviation matrix.

[0044] The constant 1 is used to ensure that the time deviation correction coefficient is greater than 1, thus creating a benchmark amplification effect. The larger the overall waveform difference parameter, the more significantly the time deviation cost is amplified, thereby reducing the priority of matching event pairs with such large morphological differences in subsequent global matching.

[0045] In another embodiment of the present invention, the sum of the Euclidean distance d of the vector binary of each event binary and the constant 1 can be used as its own time deviation correction coefficient. The product of the time deviation of each event binary and the corresponding time deviation correction coefficient can be used as the time deviation cost of the corresponding event binary, and weighted correction can be performed from the individual perspective.

[0046] Furthermore, considering that the optimal matching result of the matching deviation matrix represents the degree of inherent response coupling of different water quality indicators under the same water disturbance, the matching deviation parameter between the COD data dimension and each non-COD data dimension is obtained based on the optimal matching result of the matching deviation matrix. This characterizes and quantifies the consistency and coupling tightness of the responses of different water quality indicators to the same water quality disturbance, providing a basis for subsequent comparison with historical model baselines. By comparing the matching deviation parameter with the preset historical model statistical baseline, it is analyzed whether the current disturbance deviates from the existing multi-indicator collaborative response pattern, and the model deviation degree corresponding to the COD data dimension and each non-COD data dimension is obtained, showing the degree of deviation of the water quality fluctuation in the current monitoring period from the historical model in terms of the collaborative relationship between indicators.

[0047] Preferably, in one embodiment of the present invention, since a monitoring period may contain multiple independent events, it is necessary to find a globally optimal event matching scheme to explain the observed phenomena. Therefore, the optimal matching result of the matching deviation matrix is ​​obtained by using the Hungarian algorithm.

[0048] In the optimal matching result of the matching deviation matrix, the average value of the element values ​​(time deviation cost) corresponding to all matching pairs in the matching deviation matrix is ​​used as the matching deviation parameter between the two corresponding data dimensions.

[0049] During the initial operation phase of the system, there is no historical data, and it is impossible to obtain a historical pattern statistical baseline. At this time, the system calculates a pattern matching deviation ratio R. The matching deviation parameters corresponding to COD and UV are used as the numerator, and the matching deviation parameters corresponding to COD and EC are used as the denominator. The ratio of the fractions is R. When R is less than 1, it indicates that the time deviation of the runoff flushing pattern (COD and UV matching) is less than the time deviation of the brine backflow interference pattern (COD and EC matching), indicating that the dominant process of the cycle is runoff flushing. Conversely, when R is greater than or equal to 1, it indicates that the dominant process of the cycle is brine backflow. When the judgment period is runoff scouring, the COD data is directly used as the actual pollution load. When the actual pollution load is greater than the preset actual pollution alarm threshold, an early warning is triggered, but no early warning report is provided.

[0050] It should be noted that when the matching deviation parameter corresponding to COD and EC is 0, it is directly determined to be runoff scouring.

[0051] The system includes a background learning module that automatically identifies and labels the monitoring period as either a "runoff flushing period" or a "saltwater intrusion period." When the monitoring period shows only rainfall and no EC fluctuations, it is likely that the system is simply affected by runoff flushing, corresponding to a runoff flushing period. When the monitoring period shows only EC fluctuations and no rainfall, it is likely that the system is simply affected by saltwater intrusion, corresponding to a saltwater intrusion period.

[0052] Specifically, if there is rainfall during the monitoring period, but no (water quality fluctuation) events in the EC data dimension, or the number of events is small (e.g., less than 5 events in total), the standard can be adjusted according to the cycle length and the actual scenario, and it is marked as a runoff flushing cycle; if there is no rainfall during the monitoring period, but the number of events in the EC data dimension is large (e.g., no less than 5 events), it is marked as a saline backflow cycle.

[0053] When there are at least 3 runoff flushing cycles and brines backflow cycles, the initial operation phase of the system (the phase for calculating the pattern matching deviation ratio) is terminated, and the baseline of each pattern is acquired. Considering the mean and standard deviation of the matching deviation parameters for each pattern, which represent the central tendency and normal fluctuation range of the matching deviation parameters under the ideal situation of being affected by only one factor; the standard score can linearly map the actual deviation of the current period to the degree of deviation relative to the historical normal level, making deviations of different time periods and different magnitudes comparable on the same scale, and can highlight the abnormal degree of deviation exceeding the normal fluctuation range. Based on this, the mean and standard deviation of the matching deviation parameters for each mode are used as the preset historical mode statistical baseline. The preset historical mode statistical baseline includes the mean and standard deviation of the matching deviation parameters for the two modes, runoff flushing mode and saltwater backflow interference mode.

[0054] Within the current monitoring period, based on the historical pattern statistical baseline of the runoff flushing pattern, the standard score of the matching deviation parameter between the COD data dimension and the UV data dimension is obtained as the pattern deviation degree. Based on the historical model statistical baseline of the brine backflow interference pattern, the standard score of the matching deviation parameter between the COD data dimension and the EC data dimension is used as the model deviation degree.

[0055] The greater the deviation of the standard score from 0, the greater the deviation between the current cycle and the corresponding pattern. The greater the pattern deviation, the more it provides a basis for subsequent decomposition of COD data and extraction of the true pollution load.

[0056] It should be noted that the historical mode statistical baseline is updated periodically, such as every 30 days, and each update is based on the data from the most recent 2 years (or all data if less than 2 years).

[0057] It should be noted that the rainfall information can be obtained from the local meteorological data of the sensor installation location. The Euclidean distance between vectors and the Hungarian algorithm for solving the cost matrix (matching deviation matrix) are all well-known techniques in the art and will not be elaborated further.

[0058] The analysis and early warning module 103 is used to decompose COD data and extract the true pollution load based on the pattern deviation, and generate an early warning report based on the true pollution load.

[0059] Considering that COD signal fluctuations may be caused by actual runoff pollution or mainly by measurement interference from brine backflow, and that model deviation reflects the co-variation relationship between each non-COD data dimension and the COD signal, as well as the degree of deviation from historical prior models, the COD data is ultimately decomposed and the actual pollution load is extracted based on model deviation. This effectively removes the contribution of background interference such as brine backflow to the measurement signal, so that the output pollution load only represents the part caused by pollutants in the runoff itself. Based on the actual pollution load, early warning reports are generated, which significantly reduces the false alarm rate, makes the early warning more targeted and reliable, and can accurately point to the actual pollution events that require intervention.

[0060] Preferably, in one embodiment of the present invention, please refer to Figure 2 The diagram illustrates a flowchart of a method for obtaining actual pollution load according to an embodiment of the present invention, specifically including: Step S301: Based on the relative distribution of the deviation of the two modes corresponding to the current cycle, obtain the runoff scour contribution weight and the brine backflow interference weight.

[0061] Considering that the relative distribution of the deviations of the two models corresponding to the current cycle reflects the degree of conformity of the current cycle to the historical prior model, the greater the deviation of the model deviation from 0, the greater the degree of conformity of the current cycle to the corresponding model, and the lower the weight, the negative correlation normalization mapping result of the square values ​​of the model deviations corresponding to the two models is used as the runoff scour contribution weight and the brine backflow interference weight, respectively.

[0062] As an example, the reciprocal of the sum of the squared value of the pattern deviation of each pattern and the constant 1 is taken as the pattern conformity. The constant 1 is used to prevent the denominator from being zero, and the reciprocal corresponds to a negative correlation mapping. Then, the ratio of the pattern conformity of each pattern to the sum of the conformities of the two patterns is taken as the causal weight of each pattern. The causal weight of the runoff flushing pattern is the runoff flushing contribution weight, and the causal weight of the brine backflow interference pattern is the brine backflow interference weight.

[0063] The larger the weight of runoff scouring contribution, the more likely the COD data for the current monitoring period is contributed by runoff scouring; the larger the weight of brine backflow interference, the more likely the COD data for the current monitoring period is contributed by brine backflow.

[0064] In other embodiments of the present invention, the squared value of the mode deviation can also be negatively correlated by an exponential function exp(-x) with the natural constant e as the base, where x corresponds to the independent variable.

[0065] Step S302: Obtain the runoff scour response model of COD data and UV data pre-established under the runoff scour mode, and the brine backflow interference response model of COD data and EC data pre-established under the brine backflow interference mode; Based on the runoff scour contribution weight, brine backflow interference weight, runoff scour response model and brine backflow interference response model, dynamically decompose the COD data at each time point in the current monitoring period to obtain the true pollution load at each time point.

[0066] While learning and setting the preset historical pattern statistical baseline, the background learning module also constructs a training dataset from the data sequence binary pairs of COD water quality data and UV data of the monitoring period under the marked runoff flushing pattern, performs regression analysis (e.g., linear regression), and establishes a response model between the two data as a runoff flushing response model. The model predicts the theoretical COD value corresponding to the organic matter concentration indicated by UV absorbance under the condition of no saline interference.

[0067] The COD water quality data and EC data of the monitoring period under the labeled brine backflow interference mode were used to form a training dataset for regression analysis. A response model between the two datasets was established as the brine backflow interference response model. The model predicted the theoretical interference value of the chloride ion concentration indicated by conductivity to the COD sensor under the condition of no real pollution.

[0068] Furthermore, the runoff scour contribution weight and the brine backflow interference weight reflect the contribution proportion of COD data in the current monitoring period, while the runoff scour response model and the brine backflow interference response model reflect the estimated COD data corresponding to the data value at each time point under the two modes, respectively. Therefore, based on the runoff scour contribution weight, the brine backflow interference weight, the runoff scour response model, and the brine backflow interference response model, the COD data at each time point in the current monitoring period is dynamically decomposed to obtain the true pollution load at each time point.

[0069] Specifically, due to fitting errors between the runoff scour response model and the brine backflow disturbance response model, potential drift from long-term sensor operation, and some unmodeled environmental factors, there is usually a systematic proportional deviation between the sum of the weighted model estimates and the sum of the original COD data. To ensure the numerical closure and physical rationality of the decomposition model, a model closure correction coefficient is set to correct for this systematic deviation. The formula for calculating the model closure correction coefficient includes: ; Where t represents the sequence number of the target time point; This represents COD data at time point t. Indicates the weight of runoff scour contribution; Indicates the weight of the brine backflow interference; This represents UV data at time point t; Represents EC data at time point t; A function representing the runoff scour response model. This represents the theoretical estimate calculated using the runoff scour response model at time point t. A function representing the brine backflow disturbance response model; The theoretical estimate at time point t is calculated using the brine backflow disturbance response model. Indicates the model closure correction coefficient. T represents the total number of all sampling time points in the current monitoring period.

[0070] Actual pollution load , Chloride ion interference load , .

[0071] It should be noted that the methods for establishing the runoff scour response model and the brine backflow disturbance response model are well-known technologies in the field. Specifically, by selecting representative samples from historical data and using mature mathematical methods such as regression analysis to establish the mapping relationship between input variables (UV or EC) and output variables (COD), it is easy for those skilled in the art to implement, and therefore will not be elaborated here. The two response models can be updated along with the preset historical pattern statistical baseline. In other embodiments of the present invention, linear regression, multinomial regression, or other applicable data fitting methods can also be used to construct the quantitative relationship between variables.

[0072] In another embodiment of the present invention, considering that signal smoothing processing may also filter out some rapid COD signal fluctuations caused by real and severe pollution discharge events while suppressing noise, the original readings of water quality data such as COD, UV and EC can be used to fully extract the original COD signal when extracting the real pollution load.

[0073] In a preferred embodiment of the present invention, considering that the larger the actual pollution load, the more likely actual pollution is to occur, an early warning is triggered when the actual pollution load exceeds a preset actual pollution alarm threshold. The early warning report should include at least: Warning status: Actual pollution levels exceed standards; COD data at the time of the alert: ; Actual pollution load after decomposition: ; Chloride ion interference load after decomposition: ; Model contribution weight: and ; Evidence for event attribution: Pattern deviation between the two patterns.

[0074] When multiple real pollution loads trigger an early warning, the data for all corresponding times will be displayed in the early warning report.

[0075] It should be noted that the actual pollution alarm threshold is preset based on the environmental protection standards or water quality discharge limits of the area where the drainage network is located. Those skilled in the art can determine the specific value of this threshold according to specific regulatory requirements, and this invention will not limit it further.

[0076] In summary, to address the technical problem of false alarms and missed alarms in existing technologies due to the inability to effectively distinguish the causes of COD signals, this invention provides an IoT-based drainage network pollution monitoring and early warning system. This invention first acquires water quality data for COD, UV, and EC during the current monitoring period; further, based on the fluctuation characteristics of data within each water quality data dimension, it identifies water quality fluctuation events and obtains the event center time and event waveform feature vector; further, based on the differences between the event waveform feature vectors of the COD data dimension and each non-COD data dimension, combined with the time deviation between water quality fluctuation events, it constructs a matching deviation matrix; further, based on the optimal matching result, it obtains matching deviation parameters, and then compares these parameters with a preset historical pattern statistical baseline to obtain the pattern deviation degree between the COD data dimension and each non-COD data dimension; finally, based on the pattern deviation degree, it decomposes the COD data and extracts the true pollution load, and generates an early warning report based on the true pollution load. This invention analyzes the overall waveform morphology and temporal correlation of COD, UV, and EC fluctuation events, dynamically identifies the dominant causes of the signals, and decomposes and extracts the true pollution load, thereby reducing false alarms caused by brine backflow and making early warnings more targeted and reliable.

[0077] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0078] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A drainage pipe network pollution monitoring and early warning system based on the Internet of Things, characterized in that, The system includes: The data acquisition module is used to acquire water quality data such as COD, UV, and EC for the current monitoring period; The feature analysis module is used to identify water quality fluctuation events and obtain the event center time based on the fluctuation characteristics of data within each water quality data dimension; extract event waveform feature vectors based on the waveform asymmetry and rate of change of each water quality fluctuation event; the non-COD data dimension includes UV and EC data dimensions; construct a matching deviation matrix based on the difference between the event waveform feature vectors of the COD data dimension and each non-COD data dimension, combined with the time deviation between the water quality fluctuation events; obtain the matching deviation parameter between the COD data dimension and each non-COD data dimension based on the optimal matching result of the matching deviation matrix; and compare the matching deviation parameter with a preset historical pattern statistical baseline to obtain the pattern deviation degree corresponding to the COD data dimension and each non-COD data dimension. The analysis and early warning module is used to decompose COD data and extract the true pollution load based on the pattern deviation, and generate an early warning report based on the true pollution load.

2. The drainage pipe network pollution monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The method for obtaining the event waveform feature vector includes: In each water quality data segment corresponding to the water quality fluctuation event, the peak tail length is obtained based on the duration difference on both sides of the event center time; the average rise rate from the start time to the event center time is obtained. An event waveform feature vector is constructed based on the peak tail and the average rise rate of each water quality fluctuation event.

3. The drainage pipe network pollution monitoring and early warning system based on the Internet of Things as described in claim 1, characterized in that, The method for obtaining the matching deviation matrix includes: Select UV and EC data dimensions one by one as comparison dimensions; obtain the overall waveform difference parameters based on the overall difference between the event waveform feature vectors of the COD data dimension and the comparison dimensions. Based on the time deviation between the event center time of each water quality fluctuation event in the COD data dimension and the event center time of each water quality fluctuation event in the dimension to be compared, and combined with the overall waveform difference parameter, the time deviation cost is obtained; and a matching deviation matrix is ​​constructed based on all the time deviation costs.

4. The drainage pipe network pollution monitoring and early warning system based on the Internet of Things as described in claim 1, characterized in that, The method for obtaining the pattern deviation includes: The preset historical pattern statistical baseline includes the mean and standard deviation of the matching deviation parameters corresponding to the two patterns: runoff flushing pattern and saltwater backflow interference pattern. Within the current monitoring period, based on the historical pattern statistical baseline of the runoff flushing pattern, the standard score of the matching deviation parameter between the COD data dimension and the UV data dimension is obtained as the pattern deviation degree. Based on the historical pattern statistical baseline of the brine backflow interference pattern, the standard score of the matching deviation parameter between the COD data dimension and the EC data dimension is used as the pattern deviation degree.

5. A drainage pipe network pollution monitoring and early warning system based on the Internet of Things as described in claim 4, characterized in that, The method for obtaining the actual pollution load includes: Based on the relative distribution of the deviations of the two modes corresponding to the current cycle, the runoff scour contribution weight and the brine backflow interference weight are obtained. A runoff scour response model based on pre-established COD and UV data under the runoff scour mode, and a brine backflow interference response model based on pre-established COD and EC data under the brine backflow interference mode are obtained. Based on the runoff scour contribution weight, the brine backflow interference weight, the runoff scour response model, and the brine backflow interference response model, the COD data at each time point in the current monitoring period are dynamically decomposed to obtain the true pollution load at each time point.

6. The IoT-based drainage network pollution monitoring and early warning system according to claim 5, characterized in that, The method for obtaining the runoff scour contribution weight and the brine backflow interference weight based on the relative distribution of the deviations of the two modes corresponding to the current cycle includes: The mapping results of negatively correlated normalization of the squared values ​​of the deviations of the two modes are used as the runoff scour contribution weight and the brine backflow interference weight, respectively.

7. The drainage pipe network pollution monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, When the actual pollution load exceeds the preset actual pollution alarm threshold, an early warning is triggered. The early warning report includes at least: the early warning status, the COD data at the time of the early warning, and the actual pollution load after decomposition.

8. The IoT-based drainage network pollution monitoring and early warning system according to claim 1, characterized in that, The methods for obtaining the water quality fluctuation events include: Based on the preset minimum peak height and preset minimum peak spacing, in each dimension of water quality data, peaks are extracted and the peak points are used as fluctuation event points, and the event center time is marked. At each event center moment, the local minimum point with the nearest time domain and the corresponding peak amplitude of the data value is obtained on both sides of the time sequence, and used as the time endpoint to obtain the water quality fluctuation event.

9. A drainage pipe network pollution monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that, The optimal matching result of the matching deviation matrix is ​​obtained by using the Hungarian algorithm.

10. A drainage pipe network pollution monitoring and early warning system based on the Internet of Things as described in claim 1, characterized in that, The method for obtaining the matching deviation parameter includes: In the optimal matching result of the matching deviation matrix, the average value of the corresponding element values ​​of all matching pairs in the matching deviation matrix is ​​used as the matching deviation parameter between the two corresponding data dimensions.

Images