A method and system for detecting congestion of a combined sewer network

Abstract

The application relates to the technical field of intelligent drainage, and discloses a rainwater and sewage pipe network congestion detection method and system. The method comprises the following steps: acquiring pipe network water level flow, historical water level flow and external environment working condition data of a pipe network key node; obtaining a hydraulic characteristic vector according to the pipe network water level flow and the historical water level flow; obtaining a hydraulic state change trajectory according to the hydraulic characteristic vector; obtaining a deviation quantification index according to the hydraulic state change trajectory and a pre-constructed pipe network benchmark model; obtaining a fluctuation classification label according to the deviation quantification index; obtaining a congestion credibility score according to the fluctuation classification label and pre-stored historical congestion records; obtaining an alarm instruction signal according to the congestion credibility score and the external environment working condition data; and obtaining a congestion detection result according to the alarm instruction signal. The method can improve the precision of pipe network congestion detection.

Description

Technical Field

[0001] This invention relates to the field of intelligent drainage technology, and in particular to a method and system for detecting congestion in stormwater and sewage pipe networks. Background Technology

[0002] Currently, with the acceleration of urbanization, the operating load of stormwater and sewage pipe networks continues to increase, and their hydraulic state exhibits significant dynamic time-varying characteristics, making it easy for discrepancies to arise between monitoring data and actual operating conditions. Relying solely on fixed thresholds for anomaly detection is insufficient to distinguish between normal operating condition fluctuations and genuine congestion risks, easily leading to false alarms or missed alarms, severely impacting the reliability of early warning systems. To achieve accurate identification of pipe network congestion under complex dynamic conditions, it is necessary to collect multi-node hydraulic data in real time based on high-density sensor network chips, and to construct an intelligent analysis method capable of integrating time-series characteristics, tracing the source of fluctuations, and assessing the credibility of risks, in order to support the generation of reliable alarm commands.

[0003] In one existing technology, sensors are deployed at pipeline nodes to collect water level and flow signals. A sliding window is used to extract real-time fluctuation amplitude and slope characteristics, which are then weighted and fused with historical data from the same period. Principal component analysis is then used to reduce the dimensionality and obtain a low-dimensional state description vector. Based on this vector, a Kalman filter algorithm is used to estimate the state trajectory, and a state difference sequence is generated by comparing it point-by-point with a preset benchmark model. When the Euclidean norm of this sequence exceeds a fixed threshold, it is considered an anomaly. The system further acquires upstream and downstream synchronous data of the anomaly point to analyze the fluctuation transmission pattern and matches external construction, rainfall, and other operating conditions based on simple rules. Finally, an alarm command is generated through weighted scoring. However, this existing technology relies on fixed thresholds and local monitoring, making it difficult to distinguish between dynamic hydraulic fluctuations and actual congestion, leading to frequent false alarms and missed alarms.

[0004] Therefore, existing technologies cannot improve the accuracy of pipeline congestion detection. Summary of the Invention

[0005] This invention provides a method for detecting congestion in stormwater and sewage pipe networks, thereby improving the accuracy of pipe network congestion detection.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for detecting congestion in stormwater and sewage pipe networks, comprising: Obtain pipeline water level and flow rate, historical water level and flow rate, and external environmental conditions data for key nodes in the pipeline network; Based on the water level and flow rate of the pipeline network and the historical water level and flow rate, time-series features are extracted, and the extraction results are weighted and fused and feature dimensionality reduced to obtain a hydraulic characteristic vector. Based on the hydraulic characteristic vector, state prediction, measurement residual calculation and state weighting correction are performed by Kalman filtering algorithm, and the correction results are connected in time sequence to obtain the hydraulic state change trajectory. Based on the hydraulic state change trajectory and the pre-built pipeline network benchmark model, the state difference is calculated point by point in time, and the calculation results are projected and the amplitude is calculated to obtain the deviation quantification index. Based on the aforementioned deviation quantification index, the similarity of fluctuation patterns and the classification of transmission patterns are performed to obtain fluctuation classification labels; Based on the fluctuation classification labels and pre-stored historical congestion records, the congestion probability is calculated, and the reliability of the calculation results is evaluated to obtain a congestion reliability score. Based on the congestion confidence score and the external environmental condition data, time alignment, condition feature extraction and alarm rule matching are performed to obtain an alarm command signal. Based on the alarm command signal, the pipeline spatial location and internal status are verified, and the verification results are used for congestion probability diagnosis to obtain the congestion detection result.

[0007] In one optional implementation, the step of extracting time-series features based on the pipeline water level-flow rate and the historical water level-flow rate, and then performing weighted fusion and feature dimensionality reduction on the extracted results to obtain a hydraulic characteristic vector, includes: The water level and flow rate of the pipeline network are captured according to a preset time window, and the water level change rate and flow rate change rate are calculated from the captured results to obtain the current time series feature matrix. The historical water level and flow rate are extracted according to the time window, and the water level change rate and flow rate change rate are calculated from the extracted results to obtain the historical time series feature matrix. The Pearson coefficient weights are obtained by calculating the ratio of the covariance to the standard deviation between the corresponding columns of the current time series feature matrix and the historical time series feature matrix. Based on the Pearson coefficient weights, the current time series feature matrix and the historical time series feature matrix are weighted and added together to obtain a fused feature set; Based on the fusion feature set, eigenvalues ​​and cumulative contribution rates are calculated using principal component analysis algorithm, and fusion features whose cumulative contribution rates exceed a preset contribution rate threshold are extracted to obtain the feature vector projection matrix. Multiplying the fused feature set with the feature vector projection matrix yields the hydraulic characteristic vector.

[0008] In one optional implementation, the step of calculating the state difference at each time point based on the hydraulic state change trajectory and a pre-built pipeline network benchmark model, and then projecting and calculating the magnitude of the calculation results to obtain a deviation quantification index, includes: Based on the hydraulic state change trajectory and the pre-built pipeline network benchmark model, a state difference sequence containing water level difference and flow rate difference is obtained by timestamp alignment and time-point difference calculation. The Euclidean norm is obtained by taking the square root of the sum of the squares of all water level differences and the squares of all flow rate differences in the state difference sequence. When the Euclidean norm is greater than the preset benchmark deviation safety threshold, the state difference sequence is projected onto the principal component direction of the pre-built pipeline benchmark model, and the projection component in the first principal component direction is extracted to obtain the main abnormal component containing the projection length. When the Euclidean norm is not greater than the preset benchmark deviation safety threshold, the congestion detection result is obtained directly; The monitoring deviation amplitude is obtained by calculating the ratio of the projected length of the main abnormal component to the historical normal fluctuation range pre-stored in the pipeline reference model; The monitoring deviation amplitude is linearly mapped to a preset standard scoring range using a minimum-maximum normalization algorithm to obtain a deviation quantification index.

[0009] In one optional implementation, the step of calculating the similarity of fluctuation patterns and classifying the transmission patterns based on the deviation quantification index to obtain fluctuation classification labels includes: Based on the aforementioned deviation quantification index, synchronous hydraulic data are obtained by extrapolating the time window using a pre-constructed pressure wave propagation delay model. Based on the synchronous hydraulic data, the fluctuation feature vector is obtained by calculating the amplitude change rate, fluctuation period and trend slope; Based on the fluctuation feature vector and the deviation quantification index, the morphological similarity value is obtained by calculating the weighted combination of the cosine value of the angle between the vectors and the amplitude difference. The morphological similarity value is matched with a preset fluctuation classification interval, and the fluctuation category is confirmed based on the matching result to obtain a fluctuation classification label.

[0010] In one optional implementation, the step of calculating the congestion probability based on the fluctuation classification label and pre-stored historical congestion records, and evaluating the reliability of the calculation result to obtain a congestion reliability score, includes: Based on the fluctuation classification labels and pre-stored historical congestion records, a conditional probability table is established by statistically analyzing the frequency of historical events, and the dependencies between nodes are determined according to the pipeline benchmark model to obtain a Bayesian network inference model. Based on the fluctuation classification label, the posterior probability value is obtained by probabilistic inference calculation through the Bayesian network inference model. The congestion confidence score is obtained by mapping the posterior probability value to a standard score range through a preset nonlinear transformation rule. The nonlinear transformation rule is determined by regression fitting of the posterior probability values ​​and scores of historical congestion events.

[0011] In one optional implementation, the step of performing time alignment, condition feature extraction, and alarm rule matching based on the congestion confidence score and the external environmental condition data to obtain an alarm command signal includes: When the congestion confidence score is higher than the preset confidence threshold, the timestamps are aligned uniformly according to the external environmental condition data, and the sliding average of rainfall intensity, the numerical coding of construction type and the gradient of soil moisture change are extracted as key condition indicators to obtain the condition feature vector. The operating condition feature vector is matched and logically determined with a pre-built fault operating condition association rule base to obtain the pipeline alarm scenario category; Based on the category of the pipeline alarm scenario, the alarm level is determined by querying a preset priority mapping table, and an alarm command signal is obtained.

[0012] In one optional implementation, the step of performing pipeline spatial positioning and internal status verification based on the alarm command signal, and then performing congestion probability diagnosis on the verification results to obtain congestion detection results includes: Based on the alarm command signal and the pipeline alarm scenario category, the target pipeline segment is obtained by performing geographical association matching through a pre-built pipeline topology database; Based on the target pipeline segment and the alarm command signal, a matrix is ​​constructed by calculating the liquid level rise rate and the upstream and downstream liquid level difference to obtain the pipeline internal operating status feature matrix; Based on the internal operating status feature matrix of the pipeline network, the congestion probability is calculated using a pre-constructed gradient boosting decision tree model to obtain the congestion probability value. The congestion probability value is compared with a preset congestion safety threshold to determine the congestion status and output the congestion detection result.

[0013] Secondly, the present invention provides a stormwater and sewage pipe network congestion detection system, comprising: The data acquisition module is used to acquire pipeline water level and flow rate, historical water level and flow rate, and external environmental conditions data. The feature description module is used to extract time-series features based on the water level and flow rate of the pipeline network and the historical water level and flow rate, and to perform weighted fusion and feature dimensionality reduction on the extracted results to obtain a hydraulic feature vector. The state prediction module is used to perform state prediction, measurement residual calculation and state weighting correction based on the hydraulic characteristic vector using the Kalman filter algorithm, and to construct the hydraulic state change trajectory by connecting the correction results in time sequence. The deviation quantification module is used to calculate the state difference at each time point based on the hydraulic state change trajectory and the pre-built pipeline reference model, and to project and calculate the magnitude of the calculation results to obtain the deviation quantification index. The fluctuation classification module is used to calculate the similarity of fluctuation patterns and classify the transmission patterns based on the deviation quantification index, so as to obtain fluctuation classification labels; The evaluation module is used to calculate the congestion probability based on the fluctuation classification label and the pre-stored historical congestion records, and to evaluate the reliability of the calculation results to obtain a congestion reliability score. The rule matching module is used to perform time alignment, condition feature extraction, and alarm rule matching based on the congestion confidence score and the external environmental condition data to obtain an alarm command signal. The output module is used to perform pipeline spatial positioning and internal status verification based on the alarm command signal, and to perform congestion probability diagnosis on the verification results to obtain congestion detection results.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention integrates the temporal features of current and historical water level and flow rate, and extracts hydraulic characteristic vectors by weighted fusion and principal component analysis, and constructs dynamic hydraulic state trajectory by Kalman filtering, thereby enhancing the continuity and anti-interference ability of state description and providing more stable and accurate input features for subsequent congestion identification.

[0015] (2) This invention quantifies the difference between the hydraulic state trajectory and the pipeline reference model, and introduces wave pattern similarity calculation and transmission pattern classification to achieve fine discrimination of abnormal wave patterns, effectively distinguishing normal fluctuations caused by the external environment from real congestion signals, and reducing false alarm rate and false alarm rate.

[0016] (3) This invention integrates fluctuation classification results, historical congestion records and external operating condition data, performs probabilistic reasoning and credibility assessment through Bayesian network, and combines multi-rule matching and spatial state verification to form a closed-loop detection process from signal perception, risk assessment to alarm decision-making, thereby comprehensively improving the accuracy and reliability of pipeline congestion detection. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the stormwater and sewage pipe network congestion detection method provided in the first embodiment of the present invention; Figure 2This is a schematic diagram of the structure of the stormwater and sewage pipe network congestion detection system provided in the second embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Figure 1 The first embodiment of the present invention provides a method for detecting congestion in stormwater and sewage pipe networks, comprising the following steps: S11, acquire network water level and flow rate, historical water level and flow rate, and external environmental conditions data of key nodes in the pipeline network; S12, based on the water level and flow rate of the pipeline network and the historical water level and flow rate, perform time-series feature extraction, and perform weighted fusion and feature dimensionality reduction on the extraction results to obtain the hydraulic characteristic vector; S13, Based on the hydraulic characteristic vector, state prediction, measurement residual calculation and state weighting correction are performed by Kalman filtering algorithm, and the correction results are connected in time sequence to obtain the hydraulic state change trajectory. S14. Based on the hydraulic state change trajectory and the pre-built pipeline network benchmark model, calculate the state difference at each time point, and project and calculate the amplitude of the calculation results to obtain the deviation quantification index. S15, Based on the aforementioned deviation quantification index, calculate the similarity of fluctuation patterns and classify the transmission patterns to obtain fluctuation classification labels; S16. Based on the fluctuation classification label and the pre-stored historical congestion records, the congestion probability is calculated, and the reliability of the calculation result is evaluated to obtain a congestion reliability score. S17. Based on the congestion confidence score and the external environmental condition data, perform time alignment, condition feature extraction and alarm rule matching to obtain an alarm command signal. S18. Based on the alarm command signal, perform pipeline spatial positioning and internal status verification, and perform congestion probability diagnosis on the verification results to obtain congestion detection results.

[0020] In step S11, it is necessary to obtain the network water level and flow rate, historical water level and flow rate, and external environmental conditions data of key nodes in the pipeline network.

[0021] Specifically, the water level and flow rate data in the pipeline network are sourced from monitoring and sensing devices deployed at key nodes such as pipeline inspection wells and booster pump station outlets. Water level data is measured using submersible pressure transmitters, which convert the sensed hydrostatic pressure into a standard current signal. This standard current signal is received by a data acquisition module, whose built-in conversion program converts the current signal value into the corresponding water level height value based on the pressure-water level correlation specified by the sensor at the factory. Flow rate data is measured using externally mounted ultrasonic flow meters. These flow meters use a pair of sensors installed outside the pipeline; one sensor emits an ultrasonic signal, and the other receives it. By comparing the time it takes for the ultrasonic waves to propagate downstream and upstream in the fluid, and using the known pipe dimensions, the flow velocity of the water in the pipe is determined. This velocity is then multiplied by the cross-sectional area of ​​the pipe to obtain the volume of water flowing through the pipe per unit time, i.e., the instantaneous flow rate. These monitoring devices operate continuously according to a pre-set sampling cycle, and the collected water level and instantaneous flow rate values ​​are transmitted to the central data platform via wired network or wireless communication modules.

[0022] Historical water level and flow rate data are stored in the time-series database of the central data platform. This database persistently stores historical monitoring data records of all monitoring nodes over a relatively long period of time, according to the field structure of monitoring node identifier, timestamp, water level value, and flow rate value. The sampling period of the historical monitoring data is consistent with that of real-time monitoring.

[0023] External environmental data includes rainfall data, construction activity data, and soil moisture data. Rainfall data originates from the meteorological department's automatic rain gauge network. Measuring devices within these stations record the cumulative rainfall per unit time, forming a time-series of rainfall intensity. Construction activity data is sourced from the municipal engineering management department's construction permit database. The system periodically retrieves structured data records from this database, containing information such as the coordinates of the construction site, project type, and planned start and completion dates. Soil moisture data is derived from moisture sensors deployed along the pipeline network on the surface or in shallow underground layers. These sensors measure the soil's water content and output moisture content values.

[0024] During data acquisition, the system preprocesses the received raw data. For pipeline water level and flow rate data, preprocessing includes adding the water level height value to the sensor installation reference elevation to obtain the absolute water level value, and performing a rationality check based on the sensor range and pipeline design parameters. For external environmental condition data, preprocessing includes converting the timestamps of data from different sources to the system reference time, and performing data association and matching based on the geographical location of pipeline nodes and the spatial location relationships of rain gauges, construction areas, and humidity monitoring points. All preprocessed data are aligned and integrated according to their time attributes to form a unified time series dataset for subsequent analysis and processing.

[0025] In step S12, time-series features are extracted based on the pipeline water level and flow rate and the historical water level and flow rate. The extracted results are then weighted, fused, and feature dimensionality reduced to obtain a hydraulic characteristic vector.

[0026] In one specific implementation, the water level and flow rate of the pipeline network are intercepted according to a preset time window, and the water level change rate and flow rate change rate are calculated on the intercepted results to obtain the current time series feature matrix; The historical water level and flow rate are extracted according to the time window, and the water level change rate and flow rate change rate are calculated from the extracted results to obtain the historical time series feature matrix. The Pearson coefficient weights are obtained by calculating the ratio of the covariance to the standard deviation between the corresponding columns of the current time series feature matrix and the historical time series feature matrix. Based on the Pearson coefficient weights, the current time series feature matrix and the historical time series feature matrix are weighted and added together to obtain a fused feature set; Based on the fusion feature set, eigenvalues ​​and cumulative contribution rates are calculated using principal component analysis algorithm, and fusion features whose cumulative contribution rates exceed a preset contribution rate threshold are extracted to obtain the feature vector projection matrix. Multiplying the fused feature set with the feature vector projection matrix yields the hydraulic characteristic vector.

[0027] Specifically, the system sets a preset time window length, which is determined based on the hydraulic response time of the pipeline network, for example, including data points for a continuous 24 hours. Starting from the current moment of the pipeline water level and flow rate data sequence, the system extracts all water level and flow rate data points within this time window. For the extracted water level sequence, the system calculates its rate of change by calculating the difference between each data point in the sequence and its preceding data point, then dividing by the sampling time interval. The flow rate sequence undergoes the same processing. The calculated water level rate of change sequence and flow rate of change sequence are arranged in chronological order, together forming the current time-series feature matrix.

[0028] The system retrieves historical data records from the historical water level and flow rate database that have the same calendar date and duration as the current time window. Using the same steps as processing the current data, this historical data is extracted, and the rates of change in water level and flow rate are calculated to form a historical time-series feature matrix.

[0029] The Pearson coefficient weights are calculated between corresponding columns of the current time-series feature matrix and the historical time-series feature matrix. These weights measure the statistical linear correlation between the current volatility pattern and historical patterns. Specifically, for each feature column in the matrix (e.g., the water level change rate series), the covariance of the current series and the historical series are calculated, then the standard deviation of the current series and the historical series are calculated, and finally, the covariance value is divided by the product of the two standard deviations to obtain the Pearson coefficient weight value for that feature column.

[0030] Based on the calculated Pearson coefficient weights, the current time-series feature matrix and the historical time-series feature matrix are weighted and summed. The weighted summation process involves multiplying each element in the current time-series feature matrix by its corresponding Pearson coefficient weight, then multiplying the corresponding element in the historical time-series feature matrix by the same weight, and finally summing the corresponding elements in the two weighted matrices to obtain the fused feature set. This set is a matrix with the number of rows equal to the number of sampling points within the time window and the number of columns equal to the number of features.

[0031] Principal component analysis (PCA) is applied to the fused feature set for feature dimensionality reduction. First, the covariance matrix of the fused feature set is calculated. Then, the eigenvalues ​​and corresponding eigenvectors of this covariance matrix are calculated. The eigenvalues ​​are arranged in descending order, and the proportion of each eigenvalue to the sum of all eigenvalues ​​is calculated to obtain the contribution rate. The contribution rates are accumulated in this order to obtain the cumulative contribution rate. A preset contribution rate threshold is used to determine how many principal components to retain; this threshold is set to 85%, based on the premise that the retained principal components can explain most of the variance information in the original data. All eigenvectors whose cumulative contribution rate first exceeds this threshold are selected. These eigenvectors are then used to form the eigenvector projection matrix according to their original sorting order.

[0032] The fused feature set is multiplied by the eigenvector projection matrix. This multiplication involves multiplying the fused feature set matrix by the transpose of the eigenvector projection matrix. The result of this matrix multiplication is a new matrix with the same number of rows as the fused feature set, but with fewer columns (the number of retained principal components). This new matrix is ​​the hydraulic characteristic vector, where each row represents the dimensionality-reduced hydraulic state feature at a given sampling time.

[0033] In step S13, based on the hydraulic characteristic vector, state prediction, measurement residual calculation, and state weighting correction are performed using the Kalman filter algorithm, and the correction results are connected in time sequence to construct the hydraulic state change trajectory.

[0034] Specifically, the system applies the Kalman filter algorithm to recursively estimate the hydraulic state. The algorithm first requires initialization: the state vector is set to a zero vector with the same dimension as the hydraulic characteristic vector, and the state covariance matrix is ​​set to a large identity matrix, indicating significant uncertainty in the initial estimate. The algorithm also requires presetting two key parameter matrices. The process noise covariance matrix characterizes the uncertainty of the system state's evolution. This matrix is ​​determined by analyzing the statistical properties of a large number of historical hydraulic characteristic vectors at adjacent time points; specifically, it involves calculating the covariance matrix of these difference sequences. The observation noise covariance matrix characterizes the noise level contained in the hydraulic characteristic vectors as observed values. This matrix is ​​determined by analyzing the statistical properties of historical hydraulic characteristic vector sequences obtained under stable network operation conditions; specifically, it involves calculating the residual covariance matrix between the stable sequence and its moving average sequence.

[0035] Each iteration of the Kalman filter algorithm consists of two main steps. The first step is the prediction step. Based on the optimal state estimate obtained after the previous sampling time, the algorithm uses a pre-defined state transition matrix to predict the state at the current sampling time. This state transition matrix is ​​set as an identity matrix, based on the assumption that the basic characteristics of the hydraulic state remain unchanged within a short sampling interval. Simultaneously, the algorithm updates the uncertainty of the state estimate based on this matrix and a pre-defined process noise covariance matrix.

[0036] The second step is the update step. The algorithm reads the hydraulic characteristic vector at the current sampling time and uses it as the observation value at the current time. It calculates the difference between this observation value and the predicted state vector to obtain the measurement residual. The covariance of the measurement residual is calculated; this value is jointly determined by the prediction covariance matrix and the observation noise covariance matrix. Based on the prediction covariance matrix and the residual covariance, the Kalman gain matrix is ​​calculated. The Kalman gain matrix determines the algorithm's confidence weight for the predicted and observed values ​​in this iteration. The predicted state is corrected using the Kalman gain matrix by adding the product of the Kalman gain matrix and the measurement residual to the predicted state vector, obtaining the optimal state estimate for the current time. Simultaneously, based on the Kalman gain matrix and the prediction covariance matrix, the uncertainty of the state estimate is updated to prepare for the next iteration.

[0037] The system sequentially performs the above prediction and update steps for each row of the hydraulic characteristic vector matrix. The optimal state estimation vector obtained in each iteration is saved in chronological order. All these optimal state estimation vectors arranged in chronological order together constitute the hydraulic state change trajectory. This trajectory is a matrix with the same number of rows as the time series length and the same number of columns as the hydraulic characteristic vector dimension, describing the dynamic evolution of the network hydraulic state after filtering out observation noise.

[0038] In step S14, based on the hydraulic state change trajectory and the pre-built pipeline reference model, the state difference is calculated point by point in time, and the calculation results are projected and the magnitude is calculated to obtain the deviation quantification index.

[0039] In one specific implementation, based on the hydraulic state change trajectory and the pre-built pipeline reference model, a state difference sequence containing water level difference and flow rate difference is obtained by timestamp alignment and time-point difference calculation. The Euclidean norm is obtained by taking the square root of the sum of the squares of all water level differences and the squares of all flow rate differences in the state difference sequence. When the Euclidean norm is greater than the preset benchmark deviation safety threshold, the state difference sequence is projected onto the principal component direction of the pre-built pipeline benchmark model, and the projection component in the first principal component direction is extracted to obtain the main abnormal component containing the projection length. When the Euclidean norm is not greater than the preset benchmark deviation safety threshold, the congestion detection result is obtained directly; The monitoring deviation amplitude is obtained by calculating the ratio of the projected length of the main abnormal component to the historical normal fluctuation range pre-stored in the pipeline reference model; The monitoring deviation amplitude is linearly mapped to a preset standard scoring range using a minimum-maximum normalization algorithm to obtain a deviation quantification index.

[0040] Specifically, the pre-constructed pipeline benchmark model is a mathematical representation describing the normal hydraulic behavior of the pipeline network under abnormal conditions such as no congestion. The model adopts a multivariate time series model structure, which is constructed based on a large amount of historical water level and flow monitoring data collected during the long-term stable operation of the pipeline network. Specifically, a sufficiently long period of historical monitoring data under normal operating conditions, which has been manually confirmed to be free of any congestion events, is selected as the training set. The water level and flow data in the training set undergo the same time series feature extraction and dimensionality reduction processing as in step S12 to obtain a sequence of hydraulic characteristic vectors under historical normal operating conditions. Principal component analysis is applied to this sequence to calculate its covariance matrix and extract eigenvectors. The top few eigenvectors with the largest eigenvalues ​​are taken as the principal component directions of the benchmark model. These principal component directions constitute a projection subspace, representing the main patterns of normal hydraulic fluctuations. At the same time, the statistical distribution of the projected length of the hydraulic characteristic vectors under historical normal operating conditions on this projection subspace is calculated, and its 95th percentile value is taken as the upper limit threshold of the historical normal fluctuation range.

[0041] During the model usage phase, the system first aligns the hydraulic state change trajectory with the pipeline network baseline model in the time dimension. The alignment is based on the timestamps of both models, matching the state vector at each sampling moment in the trajectory with the normal state vector (i.e., the baseline vector) at the same or nearest moment in the historical data on which the baseline model is based.

[0042] For each aligned sampling time, the difference between the state vector in the hydraulic state change trajectory and the corresponding reference vector is calculated. This difference is a vector whose components represent the differences in water level characteristics and flow rate characteristics, respectively. This calculation is performed for all sampling times to obtain a state difference sequence, which is a matrix where each row represents the water level and flow rate difference vector at a given time.

[0043] Calculate the Euclidean norm of the state difference sequence. The calculation process involves: for each row vector in the state difference sequence, calculating the sum of squares of all its components, then taking the square root of this sum to obtain a scalar, which represents the degree of overall deviation from the normal state at that moment. Averaging or taking the maximum value of the scalars over all moments yields the Euclidean norm, representing the overall deviation degree throughout the monitoring period.

[0044] The preset baseline deviation from the safety threshold is used to determine whether further analysis is needed. This threshold is determined by analyzing historical normal operating condition data. Specifically, the method involves calculating the Euclidean norm of the state difference sequence for each time period under historical normal operating conditions, arranging these norm values ​​in ascending order, and taking the 95th percentile value after arranging as the baseline deviation from the safety threshold.

[0045] The calculated Euclidean norm is compared with a baseline deviation from a safe threshold. When the Euclidean norm is not greater than this threshold, the system determines that the current hydraulic state is within the normal fluctuation range, does not trigger subsequent congestion analysis procedures, directly outputs a no-congestion detection result, and ends the current detection process.

[0046] When the Euclidean norm exceeds a certain threshold, the system determines that a significant anomaly exists, requiring further analysis of the anomaly's main patterns. At this point, the entire state difference sequence is projected onto the principal component directions of a pre-constructed pipeline benchmark model. The projection process involves multiplying the state difference sequence matrix by the transpose of the projection matrix formed by the principal component directions of the benchmark model. The projection component along the first principal component direction is extracted from the projection result. This component is a time series, and its value at each moment represents the intensity of the anomaly in the most dominant fluctuation direction, constituting the main anomaly component.

[0047] The projected length of the main anomaly component is calculated; this length is the root mean square value of the component's time series. The ratio of this projected length to the upper limit threshold of the historical normal fluctuation range pre-stored in the pipeline network baseline model is then calculated to obtain the monitoring deviation magnitude. This ratio quantifies the extent to which the current main anomaly intensity exceeds the maximum range of historical normal fluctuations.

[0048] Finally, the monitoring deviation amplitude is linearly mapped to a preset standard scoring interval using a min-max normalization algorithm. The upper and lower limits of the preset scoring interval are set according to the requirements for classifying the severity of anomalies in actual applications. The mapping process of this algorithm is as follows: First, obtain the monitoring deviation amplitude values ​​corresponding to all confirmed congestion events in the historical records, and find the maximum and minimum values ​​among them; then, subtract the historical minimum value from the currently calculated monitoring deviation amplitude value to obtain a difference value A; at the same time, calculate the difference B between the historical maximum value and the historical minimum value; next, calculate the ratio of difference A to difference B; finally, multiply this ratio by the width of the standard scoring interval, and add the lower limit of the scoring interval. The result is the mapped standard score. The value obtained after this mapping is the deviation quantification index, and the larger the value, the more serious the deviation from the normal state.

[0049] In step S15, based on the deviation quantification index, the similarity of the fluctuation pattern and the classification of the transmission pattern are performed to obtain the fluctuation classification label.

[0050] In one specific implementation, synchronous hydraulic data is obtained by extrapolating the time window using a pre-constructed pressure wave propagation delay model based on the deviation quantification index. Based on the synchronous hydraulic data, the fluctuation feature vector is obtained by calculating the amplitude change rate, fluctuation period and trend slope; Based on the fluctuation feature vector and the deviation quantification index, the morphological similarity value is obtained by calculating the weighted combination of the cosine value of the angle between the vectors and the amplitude difference. The morphological similarity value is matched with a preset fluctuation classification interval, and the fluctuation category is confirmed based on the matching result to obtain a fluctuation classification label.

[0051] Specifically, the system first uses a pre-built pressure wave propagation delay-based model to analyze the time series of deviation quantification indicators to estimate the impact range and time window of hydraulic fluctuations propagating in the pipe network. The model is constructed based on the pipe network topology, pipe material parameters, and fluid mechanics principles. The model training process is a data-driven parameter calibration process. Using multi-node water level and flow time series data from historical normal operation, optimization algorithms are used to adjust model parameters (such as the equivalent wave velocity in the pipe) to minimize the error between the model's predicted pressure wave propagation time delay and the actual observed time difference of fluctuations at each node under known disturbance events. When a significant deviation quantification indicator is detected at a node, the trained model calculates the sets of upstream and downstream nodes that the fluctuation may propagate to based on the node's location, pressure wave propagation velocity, and pipe network flow direction. It then estimates the theoretical time delay range of the fluctuation reaching these nodes, thereby determining a synchronous analysis time window that includes the source node and potentially affected nodes.

[0052] The pressure wave propagation delay model employs a simplified formula based on one-dimensional water hammer theory. Its core is calculating the propagation time of the pressure wave along the pipe, obtained by dividing the pipe length by the pressure wave velocity within the pipe; the wave velocity is obtained through… The estimate is that, The bulk modulus of water. The density of water, The elastic modulus of the pipe. For pipe diameter, The model parameters are calibrated by collecting known instantaneous disturbance events (such as the sudden start-up and shutdown of a pumping station) and the response time difference of water levels at upstream and downstream nodes to the disturbance from historical data. The equivalent wave velocity of all pipe segments in the network is used as the optimization variable, and the wave velocity to be determined is solved by the gradient descent algorithm, thereby completing the model calibration.

[0053] Based on the time window calculated by the model, the system extracts the water level and flow time series data of relevant nodes within the time window from the water level and flow rate of the pipeline network. These multi-node data after time alignment constitute synchronous hydraulic data.

[0054] Feature extraction is performed on synchronous hydraulic data to obtain a fluctuation feature vector. Specifically, for each node's water level and flow sequence, three features are calculated: amplitude change rate, fluctuation period, and trend slope. The amplitude change rate is obtained by calculating the root mean square value of the sequence, representing the average intensity of the fluctuation. The fluctuation period is estimated by calculating the autocorrelation function of the sequence and finding the time delay corresponding to the first significant peak, representing the repeatability time scale of the fluctuation. The trend slope is obtained by performing linear regression analysis on the sequence and taking its regression coefficient, representing the overall trend of fluctuation within the time window. These three feature values ​​from all nodes and all indicators are concatenated in a fixed order to form a high-dimensional fluctuation feature vector.

[0055] Based on the time series of fluctuation feature vectors and deviation quantification indicators, a morphological similarity value is calculated. This value quantifies the degree of morphological similarity between the current fluctuation pattern and known congestion patterns. The calculation consists of two parts. The first part calculates the cosine of the angle between the vectors, which is obtained by calculating the cosine of the angle between the fluctuation feature vector and a pre-stored reference feature vector representing a typical congestion pattern, reflecting their directional similarity in the feature space. The second part calculates the amplitude difference, which is obtained by dividing the absolute value of the difference between the root mean square value of the current deviation quantification indicator sequence and the root mean square value of the pre-stored typical congestion deviation sequence by the root mean square value of the typical congestion deviation sequence, reflecting the relative difference in fluctuation intensity. The morphological similarity value is obtained by weighted summing the cosine of the angle between the vectors and an amplitude similarity value, where the amplitude similarity value is calculated by subtracting the amplitude difference.

[0056] The preset weighting coefficients are used to balance the importance of directional similarity and intensity similarity. These coefficients are determined by analyzing the contribution of the two types of similarity in historical events to the final classification accuracy. Specifically, the system knows the true category label of each historical fluctuation event in advance from the historical dataset. For each combination of weighting coefficients to be tried, the system first uses this weighting combination to calculate the morphological similarity value of all historical events, and then classifies them into the preset fluctuation classification interval according to the value, thereby obtaining the predicted category label.

[0057] Subsequently, the predicted category labels are compared with the actual category labels, and the proportion of correctly predicted historical events to the total number of historical events is calculated. This proportion represents the classification accuracy under that weight combination. The system attempts various combinations of weight coefficients and ultimately selects the set of weight coefficients that yields the highest classification accuracy as the preset value.

[0058] The calculated morphological similarity values ​​are matched with preset fluctuation classification intervals to obtain fluctuation classification labels. These preset intervals are derived during model training by clustering morphological similarity values ​​from a large number of historical fluctuation events. Each cluster interval corresponds to a fluctuation category, such as "local transient congestion," "upstream and downstream conduction congestion," "rainfall inflow disturbance," and "construction drainage disturbance." The system compares the morphological similarity value of the current event with the center values ​​of these cluster intervals and assigns it to the category corresponding to the interval with the closest Euclidean distance. This category identifier is the fluctuation classification label.

[0059] In step S16, the congestion probability is calculated based on the fluctuation classification label and the pre-stored historical congestion records, and the reliability of the calculation result is evaluated to obtain a congestion reliability score.

[0060] In one specific implementation, a conditional probability table is established by statistically analyzing the frequency of historical events based on the fluctuation classification labels and pre-stored historical congestion records, and the dependencies between nodes are determined based on the pipeline benchmark model to obtain a Bayesian network inference model. Based on the fluctuation classification label, the posterior probability value is obtained by probabilistic inference calculation through the Bayesian network inference model. The congestion confidence score is obtained by mapping the posterior probability value to a standard score range through a preset nonlinear transformation rule. The nonlinear transformation rule is determined by regression fitting of the posterior probability values ​​and scores of historical congestion events.

[0061] Specifically, the system first constructs a Bayesian network inference model based on historical congestion records and a pipeline benchmark model. The model construction process consists of two steps. First, a conditional probability table is established by statistically analyzing the frequency of various fluctuation classification labels in historical congestion records and the conditional frequency of congestion at different pipeline nodes. This table quantifies the probability that different fluctuation patterns will cause congestion at different locations in the pipeline. Second, based on the hydraulic connections between nodes described by the pipeline benchmark model, the directed connections between nodes in the Bayesian network are determined, forming the network topology. Nodes represent key monitoring points or pipe segments in the pipeline, and node states represent random events of congestion occurring at that location. The final Bayesian network inference model includes the network structure and the conditional probability distribution parameters of all nodes. Specifically, firstly, the network topology is determined based on the physical flow diagram of the pipeline, with upstream nodes serving as parent nodes for downstream nodes. Secondly, historical congestion records are used to statistically analyze the frequency of congestion in child nodes under different parent node states (normal / abnormal). For sparse data, Laplace smoothing is used to ensure the stability of the conditional probability estimation.

[0062] During the model usage phase, the system inputs the currently detected fluctuation classification labels and their corresponding pipeline node location information as evidence into the Bayesian network inference model. The model performs probabilistic inference calculations, which are based on Bayes' theorem. By combining the known conditional probability relationships of each node in the network and considering the impact of the current input evidence on the state of all relevant nodes, the probability value of congestion at each node in the network is gradually updated. The calculation process decomposes the joint probability distribution of all variables and iteratively calculates and adjusts the probabilities of non-evidence variables based on the evidence variables until the probability estimates of all nodes reach a stable state. Finally, the updated probability value of the target node experiencing congestion, i.e., the posterior probability value, is extracted.

[0063] After obtaining the posterior probability value, the system maps it to a preset standard score range using a pre-defined nonlinear transformation rule, thus obtaining a congestion confidence score. The process of establishing this nonlinear transformation rule involves collecting the posterior probability values ​​calculated by the model from historical congestion events, along with the reference scores given by the system based on the actual severity and impact of the events. Using these data pairs of posterior probability values ​​and corresponding scores, regression analysis is used to fit a mapping curve or function from the posterior probability to the standard score. This curve is the pre-defined nonlinear transformation rule, whose purpose is to convert the probability values ​​output by the model into scores that better reflect actual risk assessment practices and have clear comparability.

[0064] The congestion confidence score obtained through this mapping indicates that the higher the value, the more closely the current situation matches the characteristics of historical congestion events, and the higher the confidence that real congestion has occurred. This score serves as a key input for subsequent alarm decisions.

[0065] In step S17, based on the congestion confidence score and the external environmental condition data, time alignment, condition feature extraction, and alarm rule matching are performed to obtain an alarm command signal.

[0066] In one specific implementation, when the congestion confidence score is higher than a preset confidence threshold, the timestamps are uniformly aligned according to the external environmental condition data, and the sliding average of rainfall intensity, the numerical coding of construction type, and the gradient of soil moisture change are extracted as key condition indicators to obtain a condition feature vector. The operating condition feature vector is matched and logically determined with a pre-built fault operating condition association rule library to obtain the pipeline alarm scenario category; Based on the category of the pipeline alarm scenario, the alarm level is determined by querying a preset priority mapping table, and an alarm command signal is obtained.

[0067] Specifically, the system compares the congestion confidence score with a preset confidence threshold. The preset confidence threshold is used to determine whether the congestion risk has reached a level that requires fine-grained alarm decision-making in conjunction with external operating conditions. This threshold is determined by analyzing historical alarm records and feedback data. Specifically, the congestion confidence scores corresponding to historically confirmed valid alarm events are sorted in ascending order, and the 10th percentile value after sorting is taken as the confidence threshold, that is, the value at the 10th percentile after ascending order.

[0068] When the congestion confidence score exceeds the confidence threshold, the system determines that the congestion risk is significant and requires the introduction of external environmental factors for comprehensive judgment. The system first performs a time alignment operation, aligning the timestamps of the external environmental operating condition data with the time period corresponding to the current congestion assessment (i.e., the time window calculated by the pressure wave model in step S15) to ensure that the time base for analysis is consistent.

[0069] Based on the time-aligned data, the system extracts three key operating condition indicators to form an operating condition feature vector. The moving average of rainfall intensity is obtained by calculating the arithmetic mean of the rainfall intensity time series within the aligned period, representing the average rainfall level during that period. The numerical coding of construction type, based on a predefined mapping table, converts the project type in the construction activity data into a corresponding numerical code, representing the type of external construction activity. The gradient of soil moisture change is obtained by calculating the slope of the linear regression of the soil moisture content series within the aligned period, representing the trend of soil moisture content change.

[0070] The obtained operating condition feature vectors are matched and logically determined with a pre-built fault operating condition association rule base. This rule base is constructed based on case analysis summaries of historical fault events. Each rule contains one or more operating condition feature conditions and their corresponding pipeline alarm scenario category. Specifically, historical fault events and their operating condition feature vectors at the time of occurrence, such as the moving average of rainfall intensity, construction type coding, and soil moisture change gradient, are used as a transaction database. The FP-Growth algorithm is used to mine frequent itemsets and generate association rules in the form of {operating condition feature condition A, operating condition feature condition B} → {pipeline abnormal operating condition category C}. The generated rules are filtered based on lift and confidence, retaining strong rules with a lift greater than 1 and a confidence level higher than 70%, forming the final fault operating condition association rule base. The system compares each indicator value in the current operating condition feature vector with the conditions of each rule in the rule base one by one. If all indicators in the current vector satisfy all conditions of a certain rule, the match is considered successful, and the pipeline alarm scenario category corresponding to that rule is the output result.

[0071] Based on the obtained pipeline alarm scenario categories, the system queries a pre-defined priority mapping table to determine the alarm level. The priority mapping table is a pre-defined list based on the severity of the abnormal scenarios' impact on pipeline operational safety and public safety, where each pipeline alarm scenario category is mapped to a specific alarm level. The system retrieves the corresponding alarm level from this table based on the matched scenario category. This alarm level is encapsulated as an alarm command signal, which is a structured data object containing information such as the alarm level, target pipeline location, and suggested response measures.

[0072] If the congestion confidence score is not higher than the preset confidence threshold, the system determines that the current congestion risk is low and does not generate an alarm command signal.

[0073] In step S18, based on the alarm command signal, the pipeline spatial location and internal status are verified, and the verification results are used for congestion probability diagnosis to obtain congestion detection results.

[0074] In one specific implementation, the target pipeline segment is obtained by performing geographical association matching through a pre-built pipeline topology database based on the alarm command signal and the pipeline alarm scenario category. Based on the target pipeline segment and the alarm command signal, a matrix is ​​constructed by calculating the liquid level rise rate and the upstream and downstream liquid level difference to obtain the pipeline internal operating status feature matrix; Based on the internal operating status feature matrix of the pipeline network, the congestion probability is calculated using a pre-constructed gradient boosting decision tree model to obtain the congestion probability value. The congestion probability value is compared with a preset congestion safety threshold to determine the congestion status and output the congestion detection result.

[0075] Specifically, the system first queries the pipeline topology database based on the target pipeline location information and pipeline alarm scenario category contained in the alarm command signal. Through geographical association matching, the system identifies the sets of upstream and downstream pipeline segments directly connected to the alarm location, as well as the key monitoring nodes on these segments, thereby locating the target pipeline segment that needs to be focused on for review. This segment typically includes the alarm point and several adjacent pipeline segments.

[0076] The pipeline topology database is a structured database that stores the geographic coordinates, connection relationships, pipe diameter, slope, and other attribute information of all pipe segments, nodes, and manholes in a pipeline network. The database construction process is as follows: First, authoritative electronic data of the pipeline network as-built drawings or geographic information system (GIS) layer data are obtained from the municipal drainage management department. This data contains basic attributes such as the plane coordinates, elevation, pipe diameter, material, and construction year of all pipe segments, manholes, pumping stations, and other facilities. Then, this raw data undergoes data cleaning, including correcting obvious coordinate errors, supplementing missing attribute fields, and standardizing the coding specifications of different data sources. Next, based on the coordinate connection relationships between pipe segments and manholes, the topological connection relationships of the pipeline network are automatically established using spatial analysis algorithms, clarifying the upstream and downstream directions of each pipeline. Finally, the cleaned spatial coordinate data, attribute data, and generated topological connection relationship data are imported into a relational database management system according to a pre-defined database table structure to form the pipeline topology database.

[0077] Based on the identified target pipeline segments, the system extracts water level time series data for relevant monitoring nodes within the most recent analysis period from real-time monitoring data. Using this data, the system calculates two key hydraulic state indicators. The first indicator is the rate of water level rise, obtained by calculating the slope of a linear regression of the water level sequence at each monitoring node within the analysis period, characterizing the speed of water level change at that point. The second indicator is the upstream-downstream water level difference; for each pipeline segment, the difference in average water level between the upstream and downstream monitoring nodes within the analysis period is calculated, characterizing the hydraulic gradient of that segment. The rate of water level rise and the upstream-downstream water level difference for all monitoring nodes across all target pipeline segments are organized sequentially by segment and node, constructing a network internal operating state characteristic matrix. Each row of this matrix represents a monitoring node or a pipeline segment, and each column represents a characteristic indicator.

[0078] The internal operating status feature matrix of the pipeline network is input into a pre-constructed gradient boosting decision tree model to calculate the congestion probability. The model is built based on historical event data; training data comes from historically occurring and verified congestion event records, as well as a large amount of data from normal operating periods. For each historical event or normal period, its corresponding internal operating status feature matrix is ​​extracted as input features and labeled as "congested" or "normal" as the target label. The model is trained using a gradient boosting decision tree algorithm, iteratively constructing multiple decision trees. Each tree learns to correct the prediction residual of the previous tree, and finally, the outputs of all trees are combined for prediction. After training, the model can output a congestion probability value between zero and one based on the input feature matrix.

[0079] The congestion probability value output by the model is compared with a preset congestion safety threshold. This preset threshold is used to ultimately determine the congestion status. It is determined by analyzing the model's predicted probability values ​​for normal operating conditions in historical data. Specifically, all probability values ​​output by the model under historical normal operating conditions are sorted in ascending order, and the 95th percentile of this sorted value is used as the congestion safety threshold. If the congestion probability value is greater than this threshold, the system determines that the target pipeline segment is congested and outputs a congestion detection result containing the specific location and probability value. This result will guide on-site troubleshooting and maintenance decisions. If the congestion probability value is not greater than this threshold, the system determines that no congestion evidence was found during the review and outputs a no-congestion detection result.

[0080] In summary, this invention discloses a method for detecting congestion in stormwater and sewage pipe networks, including a system for detecting congestion in stormwater and sewage pipe networks. This invention improves the accuracy of pipe network congestion detection through this method.

[0081] Reference Figure 2 The second embodiment of the present invention provides a stormwater and sewage pipe network congestion detection system, comprising: The data acquisition module is used to acquire pipeline water level and flow rate, historical water level and flow rate, and external environmental conditions data. The feature description module is used to extract time-series features based on the water level and flow rate of the pipeline network and the historical water level and flow rate, and to perform weighted fusion and feature dimensionality reduction on the extracted results to obtain a hydraulic feature vector. The state prediction module is used to perform state prediction, measurement residual calculation and state weighting correction based on the hydraulic characteristic vector using the Kalman filter algorithm, and to construct the hydraulic state change trajectory by connecting the correction results in time sequence. The deviation quantification module is used to calculate the state difference at each time point based on the hydraulic state change trajectory and the pre-built pipeline reference model, and to project and calculate the magnitude of the calculation results to obtain the deviation quantification index. The fluctuation classification module is used to calculate the similarity of fluctuation patterns and classify the transmission patterns based on the deviation quantification index, so as to obtain fluctuation classification labels; The evaluation module is used to calculate the congestion probability based on the fluctuation classification label and the pre-stored historical congestion records, and to evaluate the reliability of the calculation results to obtain a congestion reliability score. The rule matching module is used to perform time alignment, condition feature extraction, and alarm rule matching based on the congestion confidence score and the external environmental condition data to obtain an alarm command signal. The output module is used to perform pipeline spatial positioning and internal status verification based on the alarm command signal, and to perform congestion probability diagnosis on the verification results to obtain congestion detection results.

[0082] It should be noted that the stormwater and sewage pipe network congestion detection system provided in this embodiment of the invention is used to execute all the process steps of the stormwater and sewage pipe network congestion detection method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0083] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0084] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for detecting congestion in stormwater and sewage pipe networks, characterized in that, include: Obtain pipeline water level and flow rate, historical water level and flow rate, and external environmental conditions data for key nodes in the pipeline network; Based on the pipeline water level and flow rate and the historical water level and flow rate, time-series features are extracted, and the extracted results are weighted, fused, and feature dimensionality reduced to obtain a hydraulic characteristic vector. This includes: intercepting the pipeline water level and flow rate according to a preset time window, and calculating the water level change rate and flow rate change rate of the intercepted results to obtain a current time-series feature matrix; intercepting the historical water level and flow rate according to the time window, and calculating the water level change rate and flow rate change rate of the intercepted results to obtain a historical time-series feature matrix; obtaining Pearson coefficient weights by calculating the ratio of the covariance to the standard deviation between corresponding columns of the current time-series feature matrix and the historical time-series feature matrix; weighting and summing the current time-series feature matrix and the historical time-series feature matrix according to the Pearson coefficient weights to obtain a fused feature set; and calculating eigenvalues ​​and cumulative contribution rates using principal component analysis based on the fused feature set, and extracting fused features whose cumulative contribution rates exceed a preset contribution rate threshold to obtain a feature vector projection matrix. Multiplying the fused feature set with the feature vector projection matrix yields the hydraulic characteristic vector; Based on the hydraulic characteristic vector, state prediction, measurement residual calculation and state weighting correction are performed by Kalman filtering algorithm, and the correction results are connected in time sequence to obtain the hydraulic state change trajectory. Based on the hydraulic state change trajectory and the pre-built pipeline network benchmark model, time-by-time state difference calculations are performed, and the calculation results are projected and amplitude calculated to obtain a deviation quantification index, including: based on the hydraulic state change trajectory and the pre-built pipeline network benchmark model, through timestamp alignment and time-by-time difference calculation, a state difference sequence containing water level difference and flow rate difference is obtained; the square root of the sum of the squares of all water level differences and the squares of all flow rate differences in the state difference sequence is calculated to obtain the Euclidean norm; when the Euclidean norm is greater than a preset benchmark deviation safety threshold, [further action is taken]. The state difference sequence is projected onto the principal component direction of a pre-constructed pipeline benchmark model, and the projected components in the first principal component direction are extracted to obtain the main abnormal components containing the projection length. When the Euclidean norm is not greater than the preset benchmark deviation safety threshold, the congestion detection result is directly obtained. The monitoring deviation amplitude is obtained by calculating the ratio of the projection length of the main abnormal components to the historical normal fluctuation range pre-stored in the pipeline benchmark model. The monitoring deviation amplitude is linearly mapped to the preset standard scoring range through the minimum-maximum normalization algorithm to obtain the deviation quantification index. Based on the aforementioned deviation quantification index, the similarity of fluctuation patterns and the classification of transmission patterns are performed to obtain fluctuation classification labels; Based on the fluctuation classification labels and pre-stored historical congestion records, the congestion probability is calculated, and the reliability of the calculation results is evaluated to obtain a congestion reliability score. Based on the congestion confidence score and the external environmental condition data, time alignment, condition feature extraction, and alarm rule matching are performed to obtain an alarm command signal. This includes: when the congestion confidence score is higher than a preset confidence threshold, aligning timestamps uniformly based on the external environmental condition data, and extracting the moving average of rainfall intensity, the numerical code of construction type, and the gradient of soil moisture change as key condition indicators to obtain a condition feature vector; performing condition feature vector condition matching and logical judgment with a pre-built fault condition association rule library to obtain the pipeline network alarm scenario category; and determining the alarm level by querying a preset priority mapping table based on the pipeline network alarm scenario category to obtain the alarm command signal. Based on the alarm command signal, the pipeline spatial location and internal status are verified, and the verification results are used for congestion probability diagnosis to obtain the congestion detection result.

2. The method for detecting congestion in stormwater and sewage pipe networks according to claim 1, characterized in that, The step of calculating the similarity of fluctuation patterns and classifying the transmission patterns based on the deviation quantification index to obtain fluctuation classification labels includes: Based on the aforementioned deviation quantification index, synchronous hydraulic data are obtained by extrapolating the time window using a pre-constructed pressure wave propagation delay model. Based on the synchronous hydraulic data, the fluctuation feature vector is obtained by calculating the amplitude change rate, fluctuation period and trend slope; Based on the fluctuation feature vector and the deviation quantification index, the morphological similarity value is obtained by calculating the weighted combination of the cosine value of the angle between the vectors and the amplitude difference. The morphological similarity value is matched with a preset fluctuation classification interval, and the fluctuation category is confirmed based on the matching result to obtain a fluctuation classification label.

3. The method for detecting congestion in stormwater and sewage pipe networks according to claim 1, characterized in that, The process involves calculating the congestion probability based on the fluctuation classification label and pre-stored historical congestion records, and then evaluating the reliability of the calculation results to obtain a congestion reliability score, including: Based on the fluctuation classification labels and pre-stored historical congestion records, a conditional probability table is established by statistically analyzing the frequency of historical events, and the dependencies between nodes are determined according to the pipeline benchmark model to obtain a Bayesian network inference model. Based on the fluctuation classification label, the posterior probability value is obtained by probabilistic inference calculation through the Bayesian network inference model. The congestion confidence score is obtained by mapping the posterior probability value to a standard score range through a preset nonlinear transformation rule. The nonlinear transformation rule is determined by regression fitting of the posterior probability values ​​and scores of historical congestion events.

4. The method for detecting congestion in stormwater and sewage pipe networks according to claim 1, characterized in that, The process involves performing pipeline spatial positioning and internal status verification based on the alarm command signal, and then using the verification results for congestion probability diagnosis to obtain congestion detection results, including: Based on the alarm command signal and the pipeline alarm scenario category, the target pipeline segment is obtained by performing geographical association matching through a pre-built pipeline topology database; Based on the target pipeline segment and the alarm command signal, a matrix is ​​constructed by calculating the liquid level rise rate and the upstream and downstream liquid level difference to obtain the pipeline internal operating status feature matrix; Based on the internal operating status feature matrix of the pipeline network, the congestion probability is calculated using a pre-constructed gradient boosting decision tree model to obtain the congestion probability value. The congestion probability value is compared with a preset congestion safety threshold to determine the congestion status and output the congestion detection result.

5. A system for detecting congestion in stormwater and sewage pipe networks, characterized in that, include: The data acquisition module is used to acquire pipeline water level and flow rate, historical water level and flow rate, and external environmental conditions data. The feature description module is used to extract time-series features based on the pipeline water level and flow rate and the historical water level and flow rate, and to perform weighted fusion and feature dimensionality reduction on the extracted results to obtain a hydraulic characteristic vector. This includes: intercepting the pipeline water level and flow rate according to a preset time window, and calculating the water level change rate and flow rate change rate on the intercepted results to obtain a current time-series feature matrix; intercepting the historical water level and flow rate according to the time window, and calculating the water level change rate and flow rate change rate on the intercepted results to obtain a historical time-series feature matrix; obtaining Pearson coefficient weights by calculating the ratio of the covariance to the standard deviation between corresponding columns of the current time-series feature matrix and the historical time-series feature matrix; weighting and summing the current time-series feature matrix and the historical time-series feature matrix according to the Pearson coefficient weights to obtain a fused feature set; and calculating eigenvalues ​​and cumulative contribution rates using principal component analysis algorithm based on the fused feature set, and extracting fused features whose cumulative contribution rates exceed a preset contribution rate threshold to obtain a feature vector projection matrix. Multiplying the fused feature set with the feature vector projection matrix yields the hydraulic characteristic vector; The state prediction module is used to perform state prediction, measurement residual calculation and state weighting correction based on the hydraulic characteristic vector using the Kalman filter algorithm, and to construct the hydraulic state change trajectory by connecting the correction results in time sequence. The deviation quantification module is used to calculate the state difference at each time point based on the hydraulic state change trajectory and a pre-built pipeline network benchmark model, and to project and calculate the magnitude of the calculation results to obtain a deviation quantification index. This includes: obtaining a state difference sequence containing water level differences and flow rate differences based on the hydraulic state change trajectory and the pre-built pipeline network benchmark model through timestamp alignment and time-point difference calculation; calculating the square root of the sum of the squares of all water level differences and the squares of all flow rate differences in the state difference sequence to obtain the Euclidean norm; and determining the deviation safety when the Euclidean norm is greater than a preset benchmark deviation threshold. At the threshold, the state difference sequence is projected onto the principal component direction of the pre-constructed pipeline reference model, and the projected components in the first principal component direction are extracted to obtain the main abnormal component containing the projection length; when the Euclidean norm is not greater than the preset reference deviation safety threshold, the congestion detection result is directly obtained; the monitoring deviation amplitude is obtained by calculating the ratio of the projection length of the main abnormal component to the historical normal fluctuation range pre-stored in the pipeline reference model; the monitoring deviation amplitude is linearly mapped to the preset standard scoring range through the minimum-maximum normalization algorithm to obtain the deviation quantification index; The fluctuation classification module is used to calculate the similarity of fluctuation patterns and classify the transmission patterns based on the deviation quantification index, so as to obtain fluctuation classification labels; The evaluation module is used to calculate the congestion probability based on the fluctuation classification label and the pre-stored historical congestion records, and to evaluate the reliability of the calculation results to obtain a congestion reliability score. The rule matching module is used to perform time alignment, condition feature extraction, and alarm rule matching based on the congestion confidence score and the external environmental condition data to obtain an alarm command signal. This includes: when the congestion confidence score is higher than a preset confidence threshold, aligning timestamps uniformly based on the external environmental condition data, and extracting the moving average of rainfall intensity, the numerical code of construction type, and the gradient of soil moisture change as key condition indicators to obtain a condition feature vector; performing condition matching and logical judgment between the condition feature vector and a pre-built fault condition association rule library to obtain the pipeline network alarm scenario category; and determining the alarm level by querying a preset priority mapping table based on the pipeline network alarm scenario category to obtain the alarm command signal. The output module is used to perform pipeline spatial positioning and internal status verification based on the alarm command signal, and to perform congestion probability diagnosis on the verification results to obtain congestion detection results.

Images