A method and system for monitoring the entire wastewater treatment process based on the Internet of Things

Abstract

This invention discloses a method and system for monitoring the entire wastewater treatment process based on the Internet of Things (IoT), relating to the field of wastewater treatment monitoring technology. Based on a dynamically weighted isolated forest algorithm, this invention significantly enhances the ability to identify deliberate evasion behaviors by adjusting feature sensitivity in real time. A nighttime low-flow compensation strategy combined with concentration shift detection effectively solves the industry problem of enterprises illegally discharging wastewater in regulatory blind spots, penetrating disguised methods such as dilution discharge and intermittent discharge to directly pinpoint abnormal discharge periods. Furthermore, by introducing a Granger causality model and a collaborative interference removal mechanism, the invention aligns enterprise equipment start-up and shutdown events with the time sequence of abnormal water quality fluctuations. Through statistical significance analysis and time decay weight allocation, it accurately isolates the interference effects of enterprises discharging simultaneously. This fundamentally solves the problem of shirking responsibility for multi-source drainage, providing an undeniable chain of evidence for environmental law enforcement.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment monitoring technology, and in particular to a method and system for monitoring the entire wastewater treatment process based on the Internet of Things. Background Technology

[0002] Under the centralized wastewater treatment model in industrial parks, locating hidden pollution sources caused by multiple enterprises sharing the same pipe network has become a core challenge for full-process supervision. Especially in mixed-discharge parks such as chemical and dyeing industries, the drainage patterns of enterprises vary significantly. Although existing IoT-based monitoring systems have deployed a large number of water quality sensors, they generally rely on threshold alarm mechanisms. When enterprises take evasive measures, such as low-flow discharge at night or multiple enterprises coordinating to dilute pollutants, traditional methods are difficult to identify the responsible party. Recent research attempts to introduce machine learning to predict water quality fluctuations, but it can only issue early warnings of anomalies and cannot trace the source to specific enterprises, resulting in regulatory accountability becoming a mere formality.

[0003] The two newer solutions are: first, to install IoT monitoring equipment at the sewage outlets of enterprises, but this is constrained by enterprise resistance and high installation costs; second, to build a database of enterprise water use characteristics and trace the source by matching drainage patterns. Although this avoids hardware modifications, it has obvious defects. When heavy rain or production adjustments cause changes in drainage patterns, the false alarm rate surges. When the drainage time sequences of multiple enterprises overlap, the feature matching fails.

[0004] Existing algorithms only process data streams from a single sensor and lack the ability to perform multi-dimensional correlation analysis on enterprise equipment status, pipeline water quality, and discharge time series. Even if the latest deployed edge computing nodes can process data in real time, their analysis logic is still limited to preset rules and cannot dynamically capture unconventional drainage patterns deliberately created by enterprises. Therefore, a comprehensive wastewater treatment process monitoring solution is urgently needed to address these issues. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a method and system for monitoring the entire wastewater treatment process based on the Internet of Things (IoT) to solve the problems of existing monitoring relying on fixed thresholds, failing to identify deliberately evaded emissions; data silos causing delays in source tracing; and high false alarm rates under dynamic interference.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide a method for monitoring the entire wastewater treatment process based on the Internet of Things, comprising:

[0009] Step S1: Real-time collection of equipment status data, water quality data and drainage time sequence data of sewage outlets of various enterprises in the park, water quality data of key nodes of the pipeline network, and drainage time sequence data. The equipment status data includes pump start and stop signals and current fluctuation characteristics, and the water quality data includes COD, pH and flow rate.

[0010] Step S2: Through the dynamic behavior analysis unit, a database of enterprise drainage behavior characteristics is constructed based on historical data, and the isolated forest algorithm is used to detect abnormal drainage patterns in real-time data.

[0011] Step S3: When an anomaly is detected, the causal inference unit is triggered, and the time sequence of changes in enterprise equipment status and water quality anomalies is associated based on the Granger causal relationship model, and a list of confidence levels of pollution source enterprises is output.

[0012] Step S4: Push the list of polluting enterprises and the chain of evidence to the regulatory platform to generate disposal instructions.

[0013] As a preferred embodiment of the IoT-based wastewater treatment process monitoring method described in this invention, step S2 employs the isolated forest algorithm to detect abnormal drainage patterns in real-time data, including:

[0014] Constructing real-time feature vectors:

[0015]

[0016] Where t is the sampling time number, c t The concentration of COD at that moment is expressed in mg·L⁻¹. -1 φ t f is the pH value at that moment. t Instantaneous flow rate, unit: m³ / s 3 ·h -1 i t The amplitude of the current fluctuation, in amperes (A). This is the start / stop indicator for the water pump, and can be either 0 or 1. t The vector represents the original feature columns in real time, and T represents the transpose operation;

[0017] Calibration weighted feature weight:

[0018] w c +w φ +w f +w i +w s =1,

[0019] in, For the weighted features, w c ,w φ ,w f ,w i ,ws The weights are COD, pH, flow rate, current, and start / stop signal, respectively. Zero elements in the matrix mark unweighted channels.

[0020] In the formula:

[0021]

[0022] in, For the k-th dimension weight at the next time step, Here, λ represents the current weights, and λ represents the adaptive learning rate. Let be the partial derivative of the isolated score with respect to the k-th dimension feature, and j be the summation index;

[0023] Conduct isolated forest training within a historical window:

[0024]

[0025] in, Let l be the path length of the sample in a tree, l be the tree level number, and d be the path length of the sample in a tree. t Let I be the depth from the sample to the leaf node, and let I(·) be the indicator function. θ is the feature value for segmentation of this layer. l The threshold for random segmentation;

[0026]

[0027] Where c(ψ) is the expected path length of the sample, ψ is the subsample size, and H ψ-1 It is the harmonic number;

[0028] Isolated score and dynamic threshold are:

[0029]

[0030] Among them, s t For isolated scores;

[0031]

[0032]

[0033] τ t =μ s +ασ s ,

[0034] Where L is the length of the sliding window, s j For historical scores within the window, μ s σ is the window mean. s Let α be the standard deviation of the window, α be the coefficient of variation, and τ be the standard deviation of the window. t For dynamic thresholds;

[0035] Determine if low flow rate is detected at night:

[0036]

[0037] Among them, flag t As an anomaly marker, For window average flow, For low flow threshold, All windows are COD. For concentration deviation, the logical symbol ∨ represents OR, and ∧ represents AND; if flag t If true, then lock 30 minutes of continuous data and send them to the subsequent causal inference unit.

[0038] As a preferred embodiment of the IoT-based wastewater treatment process monitoring method of the present invention, wherein: in step S3, the step of associating changes in enterprise equipment status and the time sequence of water quality anomalies based on the Granger causality model is as follows:

[0039] Constructing the alignment sequence:

[0040]

[0041] Among them, w t For comprehensive water quality indicators, c t Where COD is η, pH sensitivity coefficient is φ t pH For window-averaged pH, a t This is an abnormal indication. σ w Let I(·) represent the window mean and standard deviation, and let I(·) be the indicator function.

[0042] Aligning Enterprise Event Flow:

[0043] u j,t =∑ q δ(t-τ j,q ), DJ t =u j,t *g κ (t),

[0044] Among them, u j,t Let τ be the start / stop pulse sequence for enterprise j. j,q Let d be the timestamp of the q-th start / stop, δ(·) be the Dirac pulse, and d be the timestamp of the start / stop. j,t g represents the event intensity after convolution smoothing. κ (t)=e -t / κ Let κ = 5min, and let κ be the exponential kernel, which is the decay constant of the exponential kernel.

[0045] Establish the VAR(p) model:

[0046]

[0047] Where P is the lag order, α k β is the autoregressive coefficient. j,k Cross-sequence coefficients, Unbiased residuals;

[0048] In the formula, the sum of squares of the constrained and unconstrained residuals are respectively:

[0049]

[0050] Among them, SSR u SSR is the sum of squared residuals of the unconstrained model. r To constrain the sum of squared residuals of the model, T is the total sample length;

[0051] Perform Granger F statistic and confidence level:

[0052]

[0053] p j =1-F P,T-2P-1 (F j ),

[0054] γ j =1-p j ,

[0055] Among them, F j Let F be the causal statistic for firm j. P,T-2P-1 (·) is the cumulative distribution function of the F-distribution, p j γ represents the significance probability. j Confidence level;

[0056] Eliminating collaborative interference:

[0057]

[0058] Among them, Γ j To eliminate post-confidence, Δt k,j Let θ be the time difference between events k and j for firms, and θ = 10 min. Time weighting, m represents the index of enterprises other than j;

[0059] When Γ j When β > 0.9, enterprise j will be included in the pollution source list; β = 0.9. j The confidence matrix is ​​generated by sorting the data in descending order and then transmitted to the regulatory platform.

[0060] Secondly, this invention provides an Internet of Things-based wastewater treatment process monitoring system, including:

[0061] The edge sensing layer is deployed at enterprise sewage outlets and pipeline nodes, including multi-protocol IoT sensors for collecting water quality data and equipment status monitoring terminals for collecting water pump current signals.

[0062] The edge computing layer includes:

[0063] The dynamic behavior analysis unit is used to build a database of enterprise drainage behavior characteristics based on historical data and to use the isolated forest algorithm to detect abnormal drainage patterns in real-time data.

[0064] The causal inference unit, when an anomaly is detected, correlates the changes in the state of enterprise equipment with the time sequence of water quality anomalies based on the Granger causal relationship model, and outputs a confidence list of pollution source enterprises.

[0065] The cloud-based monitoring platform is used to push the list of polluting enterprises and the chain of evidence to the monitoring platform to generate disposal instructions.

[0066] As a preferred embodiment of the IoT-based wastewater treatment process monitoring method described in this invention, step S2, which involves constructing a database of enterprise wastewater discharge behavior characteristics, includes:

[0067] Extract the periodic characteristics, flow fluctuation amplitude, and pollutant concentration change gradient of each enterprise's wastewater discharge sequence;

[0068] Drainage behavior during nighttime hours is modeled separately, and anomalous concentration shifts during low-flow phases are identified through sliding window analysis.

[0069] As a preferred embodiment of the IoT-based wastewater treatment process monitoring method described in this invention, the input data association method for the Granger causality model in step S3 is as follows:

[0070] Align the timestamp sequence of the target company's pump start / stop events with the start time of abnormal fluctuations in the pipe network water quality;

[0071] Compare equipment status data from other companies during the same time period to eliminate interference from collaborative dilution.

[0072] As a preferred embodiment of the IoT-based wastewater treatment process monitoring system described in this invention, the edge computing layer further includes a data verification module for performing:

[0073] When a company is identified as a source of pollution, its video surveillance data is retrieved, and the changes in the liquid level of the dosing device are analyzed to see if they are consistent with the current signal.

[0074] If the current signal indicates that the equipment is running but the liquid level does not drop, the probability weight of fraud for that company will be increased.

[0075] As a preferred embodiment of the IoT-based wastewater treatment process monitoring system described in this invention, the device status monitoring terminal of the edge sensing layer integrates a current spectrum analysis unit to verify the actual operating status of the equipment by analyzing the motor's operating frequency characteristics.

[0076] As a preferred embodiment of the IoT-based wastewater treatment process monitoring system described in this invention, the cloud monitoring platform is connected to a meteorological data interface. When a rainstorm event is predicted, it sends instructions to the edge computing layer to adjust the anomaly detection threshold of the isolated forest algorithm to adapt to the influent dilution effect.

[0077] As a preferred embodiment of the IoT-based wastewater treatment process monitoring method of the present invention, wherein the evidence chain in step S4 includes:

[0078] Time period identifier for abnormal drainage patterns;

[0079] The sequence of related enterprise equipment events output by the causal inference unit;

[0080] Fraud probability assessment results from the data verification module.

[0081] The beneficial effects of this invention are as follows: Based on the dynamically weighted isolated forest algorithm, this invention significantly enhances the ability to identify deliberate evasion behavior by adjusting feature sensitivity in real time; the nighttime low flow compensation strategy combined with concentration offset detection effectively solves the industry problem of enterprises illegally discharging under regulatory blind spots; compared with the traditional threshold alarm mechanism, this invention can penetrate the disguised means such as dilution discharge and intermittent discharge, and directly lock the abnormal discharge period.

[0082] This invention introduces a Granger causal model and a collaborative interference elimination mechanism to align enterprise equipment start-up and shutdown events with the time sequence of abnormal water quality fluctuations. Through statistical significance analysis and time decay weight allocation, it accurately isolates the interference effects of enterprises that discharge simultaneously. This fundamentally solves the problem of shirking responsibility for multi-source drainage and provides an undeniable chain of evidence for environmental law enforcement.

[0083] This invention integrates a multi-source verification module, current spectrum analysis, and video liquid level comparison at the edge computing layer. When the system detects that the equipment current signal is normal but the water quality has not improved, it automatically triggers a fraud probability assessment. This design can effectively detect cheating behavior by enterprises that falsify equipment operating status and fill the gap in the data authenticity verification process of the existing regulatory system.

[0084] The cloud monitoring platform of this invention links with a meteorological data interface to dynamically adjust algorithm thresholds before heavy rain, avoiding false alarms caused by rainwater dilution. Simultaneously, the enterprise drainage behavior feature database supports sliding window updates, ensuring the model automatically evolves with changes in production patterns and maintains high accuracy over the long term. From real-time data collection at the edge perception layer to on-site analysis at the computing layer, and then to platform instruction generation, a minute-level response closed loop of monitoring-source tracing-disposal is formed. The structured output of enterprise pollution discharge evidence chains significantly improves the enforcement efficiency of environmental protection departments. Attached Figure Description

[0085] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0086] Figure 1 This is a schematic diagram of a wastewater treatment process monitoring method based on the Internet of Things in Example 1.

[0087] Figure 2 This is a schematic diagram of the framework of an IoT-based wastewater treatment process monitoring system in Example 1. Detailed Implementation

[0088] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0089] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0090] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0091] Example 1, referring to Figure 1 and Figure 2 This embodiment provides a method for monitoring the entire wastewater treatment process based on the Internet of Things, including the following steps:

[0092] Step S1: Real-time collection of equipment status data, water quality data and drainage time sequence data of sewage outlets of various enterprises in the park, water quality data of key nodes of the pipeline network, and equipment status data including pump start and stop signals and current fluctuation characteristics, and water quality data including COD, pH and flow rate.

[0093] Step S2: Through the dynamic behavior analysis unit, a database of enterprise drainage behavior characteristics is constructed based on historical data, and the isolated forest algorithm is used to detect abnormal drainage patterns in real-time data.

[0094] Step S2, which involves constructing the enterprise drainage behavior feature database, includes:

[0095] Extract the periodic characteristics, flow fluctuation amplitude, and pollutant concentration change gradient of each enterprise's wastewater discharge sequence;

[0096] Drainage behavior during the nighttime period (20:00 to 6:00 the next day) is modeled separately, and anomalous concentration shifts during low flow phases are identified through sliding window analysis.

[0097] Step S2 uses the Isolation Forest algorithm to detect abnormal drainage patterns in real-time data, including:

[0098] Constructing real-time feature vectors:

[0099]

[0100] Where t is the sampling time number, c t The concentration of COD at that moment is expressed in mg·L⁻¹. -1 φ t f is the pH value at that moment. t Instantaneous flow rate, unit: m³ / s 3 ·h -1 i t The amplitude of the current fluctuation, in amperes (A). This is the start / stop indicator for the water pump, and can be either 0 or 1. t The vector represents the original feature columns in real time, and T represents the transpose operation;

[0101] Calibration weighted feature weight:

[0102] w c +w φ +w f +w i +w s =1,

[0103] in, For the weighted features, w c ,w φ ,w f ,w i ,w sThe weights are COD, pH, flow rate, current, and start / stop signal, respectively. Zero elements in the matrix mark unweighted channels.

[0104] In the formula:

[0105]

[0106] in, For the k-th dimension weight at the next time step, Here, λ represents the current weights, and λ is the adaptive learning rate, set to 0.05. Let be the partial derivative of the isolated score with respect to the k-th dimension feature, and j be the summation index;

[0107] Conduct isolated forest training within a historical window:

[0108]

[0109] in, Let l be the path length of the sample in a tree, l be the tree level number, and d be the path length of the sample in a tree. t Let I be the depth from the sample to the leaf node, and let I(·) be the indicator function. θ is the feature value for segmentation of this layer. l The threshold for random segmentation;

[0110]

[0111] Where c(ψ) is the expected path length of the sample, ψ is the subsample size, taken as 256, and H ψ-1 It is the harmonic number;

[0112] Isolated score and dynamic threshold are:

[0113]

[0114] Among them, s t For isolated scores;

[0115]

[0116]

[0117] τ t =μ s +ασ s ,

[0118] Where L is the length of the sliding window, which is 144 s. j For historical scores within the window, μ s σ is the window mean. s Let τ be the standard deviation of the window, α be the coefficient of variation, taken as 2.5, and τ be the standard deviation of the window. t For dynamic thresholds;

[0119] Determine if low flow rate is detected at night:

[0120]

[0121] Among them, flag t As an anomaly marker, For window average flow, For low flow threshold, All windows are COD. For concentration deviation, the logical symbol ∨ represents OR, and ∧ represents AND; if flag t If true, then lock 30 minutes of continuous data and send them to the subsequent causal inference unit;

[0122] Specifically, the above process highlights key pollutants through vector weighting, and the isolated forest maintains its separation efficiency for high-dimensional features through the dual effects of subsampling and random segmentation; the weight autoregression update stage uses the score gradient to adjust the sensitivity of each feature in real time, and reacts quickly to sudden drainage behavior; the dynamic threshold adapts to seasonal fluctuations in flow based on window statistics, reducing misjudgments caused by long-term drift; the nighttime compensation strategy superimposes concentration deviation testing in the low flow range, strengthening the detection of illegal discharge at night; the overall mechanism completes training, judgment and weight correction in minutes, providing the causal inference unit with high confidence and accurate time period anomaly segments;

[0123] Step S3: When an anomaly is detected, the causal inference unit is triggered, and the time sequence of changes in enterprise equipment status and water quality anomalies is associated based on the Granger causal relationship model, and a list of confidence levels of pollution source enterprises is output.

[0124] The input data association method for the Granger causal relationship model in step S3 is as follows:

[0125] Align the timestamp sequence of the target company's pump start / stop events with the start time of abnormal fluctuations in the pipe network water quality;

[0126] Compare equipment status data from other companies during the same time period to eliminate interference from collaborative dilution.

[0127] In step S3, the steps for linking changes in enterprise equipment status with the time series of water quality anomalies based on the Granger causality model are as follows:

[0128] Constructing the alignment sequence:

[0129]

[0130] Among them, w t For comprehensive water quality indicators, c t Where COD is denoted as η, pH sensitivity coefficient is taken as 0.8, and φ is the pH value. t pH For window-averaged pH, at This is an abnormal indication. σ w Let I(·) represent the window mean and standard deviation, and let I(·) be the indicator function.

[0131] Aligning Enterprise Event Flow:

[0132] u j,t =∑ q δ(t-τ j,q ), d j,t =u j,t *g κ (t),

[0133] Among them, u j,t Let τ be the start / stop pulse sequence for enterprise j. j,q Let d be the timestamp of the q-th start / stop, δ(·) be the Dirac pulse, and d be the timestamp of the start / stop. j,t g represents the event intensity after convolution smoothing. κ (t)=e -t / κ Let κ = 5min, and let κ be the exponential kernel, which is the decay constant of the exponential kernel.

[0134] Establish the VAR(p) model:

[0135]

[0136] Where P is the lag order, α k β is the autoregressive coefficient. j,k Cross-sequence coefficients, Unbiased residuals;

[0137] In the formula, the sum of squares of the constrained and unconstrained residuals are respectively:

[0138]

[0139]

[0140] Among them, SSR u SSR is the sum of squared residuals of the unconstrained model. r To constrain the sum of squared residuals of the model, T is the total sample length;

[0141] Perform Granger F statistic and confidence level:

[0142]

[0143] p j =1-F P,T-2P-1 (F j ),

[0144] γ j =1-p j ,

[0145] Among them, F j Let F be the causal statistic for firm j. P,T-2P-1 (·) is the cumulative distribution function of the F-distribution, p j γ represents the significance probability. j Confidence level;

[0146] Eliminating collaborative interference:

[0147]

[0148] Among them, Γ j To eliminate post-confidence, Δt k,j Let θ be the time difference between events k and j for firms, and θ = 10 min. Time weighting, m represents the index of enterprises other than j;

[0149] When Γ j When β > 0.9, enterprise j will be included in the pollution source list; β = 0.9. j The confidence matrix is ​​generated by sorting the data in descending order and then transmitted to the regulatory platform.

[0150] Specifically, the comprehensive water quality index couples concentration and pH shift to first eliminate small fluctuations; pulse convolution transforms discrete start and stop events into regressible smooth intensity, improving the stability of VAR fitting; the F statistic accurately assesses the explanatory gain of the event flow on the abnormal sequence, and the confidence score is mapped to γ ​​based on significance. j Quantification; the collaborative interference elimination process redistributes influence based on time decay, effectively reducing false pointing in synchronous dilution scenarios; the final output prioritizes highly trustworthy enterprises with β as the threshold.

[0151] Step S4: Push the list of polluting enterprises and the chain of evidence to the regulatory platform to generate disposal instructions;

[0152] The chain of evidence in step S4 includes:

[0153] Time period identifier for abnormal drainage patterns;

[0154] The sequence of related enterprise equipment events output by the causal inference unit;

[0155] Fraud probability assessment results from the data verification module.

[0156] This embodiment also provides an IoT-based wastewater treatment process monitoring system, including:

[0157] The edge sensing layer is deployed at enterprise sewage outlets and pipeline nodes, including multi-protocol IoT sensors for collecting water quality data and equipment status monitoring terminals for collecting water pump current signals.

[0158] The edge computing layer includes:

[0159] The dynamic behavior analysis unit is used to build a database of enterprise drainage behavior characteristics based on historical data and to use the isolated forest algorithm to detect abnormal drainage patterns in real-time data.

[0160] The causal inference unit, when an anomaly is detected, correlates the changes in the state of enterprise equipment with the time sequence of water quality anomalies based on the Granger causal relationship model, and outputs a confidence list of pollution source enterprises.

[0161] The edge computing layer also includes a data verification module for performing:

[0162] When a company is identified as a source of pollution, its video surveillance data is retrieved, and the changes in the liquid level of the dosing device are analyzed to see if they are consistent with the current signal.

[0163] If the current signal indicates that the equipment is running but the liquid level does not drop, the probability weight of fraud for that company will be increased.

[0164] The device status monitoring terminal of the edge sensing layer integrates a current spectrum analysis unit, which verifies the actual operating status of the device by analyzing the motor's operating frequency characteristics (normal range is 5-10Hz).

[0165] The cloud-based monitoring platform is used to push the list of polluting enterprises and the chain of evidence to the monitoring platform to generate disposal instructions;

[0166] The cloud monitoring platform connects to the meteorological data interface. When a rainstorm event is predicted, it sends instructions to the edge computing layer to adjust the anomaly detection threshold of the isolated forest algorithm to adapt to the water inflow dilution effect.

[0167] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A sewage treatment whole-process supervision method based on Internet of Things, characterized in that, Comprising, Step S1, real-time acquisition of equipment state data of each enterprise sewage outlet in the park, water quality data of key nodes of pipe network and drainage timing data, the equipment state data including water pump start-stop signal and current fluctuation characteristics, the water quality data including COD, pH and flow; Step S2, constructing an enterprise drainage behavior feature library according to historical data through a dynamic behavior analysis unit, and detecting abnormal drainage patterns in real-time data by using an isolation forest algorithm; Step S3, when an anomaly is detected, triggering a causal inference unit, correlating enterprise equipment state changes and water quality abnormal timing based on a Granger causal relationship model, and outputting a pollution source enterprise confidence list; Step S4, pushing the pollution source enterprise list and evidence chain to a supervision platform to generate disposal instructions; In step S2, the isolation forest algorithm is used to detect abnormal drainage patterns in real-time data, including: Constructing a real-time feature vector: ， wherein, is the sampling time index, is the COD concentration at this time, unit is mg / L, , is the pH value at this time, is the instantaneous flow, unit is m3 / h, , is the current fluctuation amplitude, unit A, is the water pump start-stop flag, taking 0 or 1, is the real-time original feature column vector, denotes the transpose operation; Calibrating weighted feature weights: ， ， wherein, is the weighted feature, are the weights of COD, pH, flow, current, start-stop signal, respectively, and zero elements in the matrix mark unweighted channels; In the formula: ， wherein, is the kth dimension weight for the next time instant, is the current weight, is the adaptive learning rate, is the partial derivative of the isolation score with respect to the kth dimension feature, j is the summation index; Performing isolation forest training on a historical window: ， wherein, is the path length of a sample in a tree, is the tree level number, is the depth of a sample to a leaf node, is an indicator function, is the split feature value of the layer, is a random split threshold; ， wherein, is the expected path length for the sample, is the sub-sampling amount, is the harmonic number; Isolation score and dynamic threshold are: ， wherein, is the isolated score; ， ， ， wherein L is a sliding window length, is a historical score within the window, is a window mean, is a window standard deviation, is a coefficient of difference, is a dynamic threshold value; Performing night low-flow compensation judgment: ， wherein, is an anomaly flag, is a window average flow, is a low flow threshold, is a window average COD, is a concentration bias, logical symbols denotes or, denotes and; if is true, lock consecutive 30 min data to subsequent causal inference unit; In step S3, the step of correlating enterprise equipment state changes and water quality abnormal timing based on a Granger causal relationship model is: Constructing an alignment sequence: ， ， wherein, is the integrated water quality index, is the COD, is the pH sensitivity coefficient, is the pH, is the windowed mean pH, is the anomaly indication, is the windowed mean and standard deviation, is the indicator function; Aligning enterprise event streams: ， ， wherein, is the start-stop sequence for the enterprise is the start-stop timestamp for the enterprise is the start-stop sequence for the enterprise is the start-stop timestamp for the enterprise is the Dirac impulse is the event intensity smoothed by convolution is the exponential kernel is the exponential kernel decay constant Establishing a VAR(p) model: ， where P is the lag order, is an autoregressive coefficient, is a cross-sectional coefficient, is an unbiased residual; In the formula, the constrained and unconstrained residual sum of squares are: ， ， wherein, is the unconstrained model residual sum of squares, is the constrained model residual sum of squares, T is the total length of the sample; Performing Granger F statistics and confidence: ， ， ， wherein, for an enterprise a causal statistic, for a distribution cumulative distribution function, a significance probability, a confidence level; Eliminating synergistic interference: ， wherein, is the posterior confidence after rejection, is the enterprise k to j event time difference, , is the time weight, is the enterprise index other than j; When Enterprise j is included in the pollution source list, ; according to decreasing order to generate a confidence matrix transmitted to the regulatory platform.

2. The sewage treatment whole-process supervision method based on the Internet of Things according to claim 1, characterized in that, In step S2, the construction of the enterprise drainage behavior feature library includes: Extracting the periodical characteristics, flow fluctuation amplitude and pollutant concentration change gradient of each enterprise drainage timing; Modeling the drainage behavior of the night period separately, and identifying abnormal concentration deviation in the low-flow stage through sliding window analysis. 3.The sewage treatment whole-process monitoring method based on the Internet of Things according to claim 1, characterized in that, In step S3, the input data correlation method of the Granger causal relationship model is: Aligning the timestamp sequence of the target enterprise water pump start / stop event with the start time of the pipe network water quality abnormal fluctuation; Comparing the equipment state data of other enterprises in the same time period to exclude synergistic dilution interference. 4.The sewage treatment whole-process monitoring method based on the Internet of Things according to claim 1, wherein, The evidence chain in step S4 includes: Time period identification of abnormal drainage patterns; Enterprise equipment event sequence output by the causal inference unit; Fraud probability evaluation results of the data verification module.

5. A sewage treatment whole-process monitoring system based on Internet of Things, based on the sewage treatment whole-process monitoring method based on Internet of Things in any one of claims 1 to 4, characterized in that, Comprising, An edge perception layer deployed at enterprise sewage outlets and pipe network nodes, including multi-protocol Internet of Things sensors for collecting water quality data and equipment state monitoring terminals for collecting water pump current signals; An edge computing layer, including: A dynamic behavior analysis unit for constructing an enterprise drainage behavior feature library according to historical data, and detecting abnormal drainage patterns in real-time data by using an isolation forest algorithm; And a causal inference unit that, when an anomaly is detected, correlates enterprise equipment state changes and water quality abnormal timing based on a Granger causal relationship model, and outputs a pollution source enterprise confidence list; A cloud supervision platform for pushing the pollution source enterprise list and evidence chain to a supervision platform to generate disposal instructions. 6.The sewage treatment whole-process monitoring system based on Internet of Things according to claim 5, wherein, The edge computing layer further includes a data verification module for performing: When a certain enterprise is listed as a pollution source, video monitoring data thereof is called to analyze whether the liquid level change of a dosing device is consistent with the current signal; If the current signal indicates that the equipment is running but the liquid level does not drop, the enterprise fraud probability weight is increased. 7.The sewage treatment whole-process monitoring system based on the Internet of Things according to claim 5, wherein, The edge-aware layer integrates a current spectrum analysis unit in the equipment state monitoring terminal, and verifies the real running state of the equipment by analyzing the motor working frequency characteristics. 8.The sewage treatment whole-process monitoring system based on the Internet of Things according to claim 5, wherein, The cloud monitoring platform is connected with a meteorological data interface, and when a rainstorm event is predicted, an instruction is sent to the edge computing layer to adjust the abnormality judgment threshold of the isolation forest algorithm to adapt to the water dilution effect.

Images