A full pipe operation early warning analysis method of a sewage pipe network

Abstract

The present application relates to the technical field of municipal pipeline detection, and discloses a full-pipe operation early warning analysis method for a sewage pipe network, comprising the following steps: S1, arranging sensors at key nodes of the sewage pipe network to collect real-time operation data of the sewage pipe network; S2, acquiring real-time meteorological data, including rainfall, rainfall intensity and other related meteorological factors; S3, performing data cleaning and standardization processing on the collected sewage pipe network data and meteorological data; S4, based on multivariate regression analysis, establishing a relationship model between flow, liquid level and full-pipe risk; S5, using a long short-term memory network to perform time series prediction to predict future flow and liquid level trends. Through time series prediction by the long short-term memory network, the early warning threshold is dynamically adjusted in combination with an adaptive algorithm, the operation state change of the sewage pipe network is realized in real time, the future flow and liquid level trends are accurately predicted, and the accuracy of early warning and the response speed to sudden full-pipe risk are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of municipal pipeline detection, in particular to a full-pipe operation early warning analysis method for a sewage pipe network. BACKGROUND

[0002] With the acceleration of urbanization and the increase of extreme weather events, the operation pressure of urban sewage pipe networks is increasing, and full-pipe operation and overflow problems occur from time to time, posing a serious challenge to the stable operation of urban drainage systems and environmental protection. The existing sewage pipe network monitoring and early warning system mostly uses a simple rule method based on static threshold. This method usually relies on fixed empirical values to set the early warning threshold, and does not fully consider the complexity and dynamic changes of the actual operating environment, resulting in false positives, false negatives and other phenomena when encountering sudden events or changes in operating conditions, and failing to timely and effectively warn of full-pipe risks. In addition, the traditional early warning method lacks consideration of meteorological factors, especially under extreme weather conditions (such as heavy rain and typhoon), the system is difficult to accurately assess the impact of meteorological changes on pipe network operation, thereby affecting the accuracy and reliability of the early warning.

[0003] The existing technology mainly uses simple threshold judgment or rule matching methods for abnormal detection. For complex and variable sewage pipe network operation data, these methods often cannot accurately identify potential operation abnormal states, especially for non-typical events such as illegal discharge and pipe blockage, the detection effect is poor, which seriously affects the intelligence level and practicality of the early warning system. The traditional method lacks effective algorithm support when processing multi-source data fusion, and the correlation and complex interaction between data cannot be fully utilized, resulting in large deviation of early warning analysis results. In addition, the existing technology is prone to overfitting problem in the process of multivariate modeling, which makes the model perform poorly when facing new data, reducing the robustness and adaptability of the early warning system.

[0004] Therefore, the present application provides a full-pipe operation early warning analysis method for a sewage pipe network to solve the problems raised in the background. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a full-pipe operation early warning analysis method for a sewage pipe network, which solves the problem of poor accuracy and adaptability of the existing early warning analysis method in use.

[0006] To achieve the above purpose, the present application is implemented by the following technical scheme: a full-pipe operation early warning analysis method for a sewage pipe network, comprising the following steps:

[0007] S1, sensors are arranged at key nodes of the sewage pipe network to collect real-time operation data of the sewage pipe network, including flow, liquid level and pressure data;

[0008] S2, acquiring real-time meteorological data, including rainfall, rainfall intensity and other related meteorological factors;

[0009] S3, data cleaning and standardization processing on the collected sewer network data and meteorological data;

[0010] S4, establishing a relationship model between flow, liquid level and full pipe risk based on multivariate regression analysis;

[0011] S5, using long short-term memory network for time series prediction to predict future flow and liquid level trend;

[0012] S6, dynamically adjusting the full pipe early warning threshold according to the time series prediction result and adaptive algorithm;

[0013] S7, fusing real-time meteorological data, using deep learning model to establish a nonlinear model of the influence of meteorology on sewer network;

[0014] S8, combining the dynamically adjusted threshold and the meteorological influence prediction result to calculate the final warning parameter;

[0015] S9, fusion analysis of multi-source data, using unsupervised learning algorithm to identify abnormal operation state in sewer network.

[0016] Preferably, the relationship model established by multivariate regression analysis in step S4 is:

[0017]

[0018] wherein, is the full pipe risk index, is the flow of different monitoring points, is the liquid level of different monitoring points, is the pressure of different monitoring points, , , is the model parameter, which is optimized by fitting the historical data under full pipe operation condition.

[0019] Preferably, the specific model of long short-term memory network for time series prediction in step S5 is:

[0020]

[0021] wherein, is the predicted next time flow or liquid level for full pipe state prediction; is the current hidden state, reflecting the current operation situation of the pipe network; is the current input, representing real-time monitoring data, , is the weight matrix, is the bias term, is an activation function.

[0022] Preferably, the dynamic adjustment of the full-pipe early warning threshold in step S6 adopts an adaptive algorithm, and the adjustment formula is:

[0023]

[0024] wherein, is the dynamically adjusted early warning threshold, reflecting the real-time response to the full-pipe early warning; is the basic threshold; is the adjustment factor, and the adjustment amplitude matches the change of the operation state of the pipe network; is the difference from the predicted value, directly affecting the full-pipe risk.

[0025] Preferably, the deep learning model in step S7 adopts a hybrid model of convolutional neural network and long short-term memory network, which is specifically used to calculate the correction of weather on the full-pipe risk, and the expression is:

[0026]

[0027]

[0028] wherein, is the weather influence correction coefficient, used to correct the full-pipe early warning threshold; , is the model weight, reflecting the influence of weather data on the load capacity of the sewage pipe network.

[0029] Preferably, the calculation formula of the final early warning parameter in step S8 is:

[0030]

[0031] wherein, is the final early warning threshold, directly used for triggering the full-pipe early warning; is the adjustment weight, controlling the influence range of weather correction on the threshold; is the amplification factor, adjusting the sensitivity of the threshold response.

[0032] Preferably, the model parameters of the multivariate regression analysis are By fitting the historical data of the full-pipe operation, the least square method and regularization optimization are adopted, and the objective function is:

[0033]

[0034] wherein, is the sample number, is the full-pipe risk predicted by the model, is a regularization parameter, controlling the model complexity to prevent overfitting.

[0035] Preferably, the identification of the abnormal operating state in step S9 adopts the unsupervised learning algorithm Isolation Forest, and the abnormal detection score function is:

[0036]

[0037] wherein, is an abnormal score, and a lower score value indicates a possible full-pipe risk; is the average path length of the data points, reflecting the degree of abnormal operation of the sewer network.

[0038] Preferably, the loss function of the LSTM model is the weighted mean square error, and the calculation formula is:

[0039]

[0040] wherein, is a sample weight, which is inversely adjusted by the abnormal score S related to the full-pipe risk to enhance the sensitivity of the model to the full-pipe event.

[0041] Preferably, the early warning threshold is When the actual monitoring data exceeds the early warning threshold, the system automatically triggers a full-pipe early warning signal and starts the corresponding emergency response scheme to prevent sewage overflow and environmental pollution.

[0042] The present application provides a full-pipe operation early warning analysis method for a sewer network.

[0043] 1. The present application uses long short-term memory network for time series prediction, combines with adaptive algorithm to dynamically adjust the early warning threshold, so that the system can respond to the changes of the operation state of the sewer network in real time, accurately predict the future flow and liquid level trend, thereby significantly improving the accuracy of the early warning and the response speed to the sudden full-pipe risk.

[0044] 2. The present application uses a hybrid model of convolutional neural network and LSTM to include the nonlinear influence of real-time weather data in full-pipe risk assessment, and realizes multi-source data fusion analysis and intelligent abnormal detection through the unsupervised learning algorithm Isolation Forest, effectively identifies the abnormal state in the operation of the pipe network, and significantly improves the accuracy and reliability of the early warning system, especially under extreme weather conditions.

[0045] 3. The present application introduces a regularization technique to optimize the multivariate regression analysis model, prevent overfitting, and combine comprehensive analysis and real-time adjustment of multi-dimensional data. The system can automatically adapt to complex and variable operating environments to ensure the robustness and consistency of the early warning analysis results, thereby providing reliable technical support for the efficient management and safe operation of the urban drainage system. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A flow chart of the method of the present application. DETAILED DESCRIPTION

[0047] The technical solutions of the present application will be described clearly and completely below in combination with the drawings of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0048] Embodiment 1:

[0049] Please refer to the accompanying Figure 1 The embodiment of the present application provides a full-pipe operation early warning analysis method for a sewage pipe network, which comprises the following steps:

[0050] S1, sensors are arranged at key nodes of the sewage pipe network to collect real-time operation data of the sewage pipe network, including flow rate, liquid level and pressure data;

[0051] Specifically, the types of sensors include high-precision ultrasonic flow meters, electromagnetic liquid level meters and pressure sensors, etc., which are respectively used for monitoring the flow rate, water level and internal pressure of sewage in the pipe network. The positions of each sensor should be determined through detailed system analysis and modeling to ensure the comprehensiveness of data coverage and the key monitoring of key nodes. The sensor network is connected to the central control system through a wireless communication module to realize remote transmission of real-time data. In order to improve the timeliness of data, the system adopts a high-frequency data acquisition mode combined with advanced data compression and transmission technology to reduce the risk of transmission delay and data loss. During the data acquisition process, the system will automatically clean and filter invalid data caused by sensor failure or external interference through an automated data cleaning and outlier filtering mechanism to ensure the reliability of the data. Real-time operation data will be automatically stored by the central control system and used as a basic data source for subsequent hydraulic model calculation, early warning analysis and emergency response. Through this data acquisition process, not only the real-time operation status of the sewage pipe network can be accurately mastered, but also scientific data support can be provided for the maintenance, optimization and emergency management of the pipe network, effectively improving the operation efficiency and safety level of the urban sewage treatment system. The implementation of this data acquisition helps to realize comprehensive monitoring of the complex operation status of the sewage pipe network, quickly respond to potential operation abnormalities, and ultimately improve the fine level of urban water environment management.

[0052] S2, real-time meteorological data, including rainfall, rainfall intensity and other related meteorological factors, are obtained;

[0053] Specifically, these meteorological data not only include rainfall and rainfall intensity, but also cover key meteorological parameters that may affect the operation of the sewer network, such as wind speed, temperature, humidity, atmospheric pressure, etc. By accessing real-time data from local weather stations, remote sensing satellites and high-precision radars, the system can obtain full-coverage meteorological information. In addition, the system uses a real-time weather data update mechanism based on API interface to synchronize the obtained meteorological data with the monitoring data of the sewer network, ensuring the timeliness and consistency of the data. In terms of data processing, the system uses data fusion and correction techniques to integrate and verify multi-source meteorological data, eliminate outliers and perform necessary smoothing to reduce the impact of sensor errors and data noise on the analysis results. The acquisition and processing of real-time meteorological data provide the necessary boundary conditions for the hydraulic model of the sewer network, enabling the full-pipe operation warning analysis to more accurately reflect the actual situation. Especially in extreme weather events such as heavy rain and typhoons, through the sensitive capture and rapid response of meteorological data, the system can timely adjust the warning parameters, significantly improve the prediction ability of the sewer network full-pipe risk and the accuracy of the warning response, and provide a scientific basis for the emergency dispatch and operation and maintenance of the urban drainage system. Through the collection and application of real-time meteorological data, the system can achieve rapid response to weather changes, further improving the overall intelligent level and reliability of the sewer network warning analysis.

[0054] S3, data cleaning and standardization processing of collected sewer network data and meteorological data;

[0055] Specifically, in the data cleaning process, the system first performs preliminary integrity checks on the collected sewer network data (including flow, liquid level and pressure) and meteorological data (including rainfall, rainfall intensity, wind speed, etc.), excluding missing, abnormal or obviously unreasonable data points caused by sensor failure, network transmission problems or other unexpected reasons. For detected outliers and noise data, the system uses statistical analysis methods such as mean filtering, time series analysis, etc. to smooth the data or interpolate the missing data, thereby eliminating short-term fluctuations and ensuring data continuity and reasonableness.

[0056] Standardization processing is to eliminate the dimensional differences between different data types and monitoring points, so that all input data can be compared and analyzed on a unified scale. The system uses normalization or standard normal distribution transformation methods to convert various data into a relatively stable range (e.g. between 0 and 1 or a distribution with a mean of 0 and a standard deviation of 1), to avoid excessive bias of a certain type of data on the analysis results. Through standardization processing, flow, liquid level, pressure and meteorological data can be processed under the same weight in subsequent model calculation and machine learning processes, thereby improving the robustness of the model and the efficiency of the calculation.

[0057] After data cleaning and standardization, the system will verify the processed data to ensure that the cleaning and standardization steps have not introduced new abnormalities or distortions. After these processing steps, the data set will have high quality, uniformity and consistency, providing a solid data foundation for subsequent multivariate regression analysis, time series prediction, dynamic threshold adjustment and anomaly detection analysis modules. Finally, through rigorous cleaning and standardization, the system can significantly improve the accuracy and response speed of the full-pipe operation warning of the sewer network, reduce the probability of false positives and false negatives, and provide reliable data support for the stable operation of the urban sewage treatment and drainage system.

[0058] S4, based on multivariate regression analysis, a relationship model between flow, liquid level and full-pipe risk is established; the relationship model established by multivariate regression analysis is:

[0059]

[0060] wherein, is the full-pipe risk index, is the flow of different monitoring points, is the liquid level of different monitoring points, is the pressure of different monitoring points, , , is a model parameter, which is obtained by fitting historical data under full-pipe operation conditions;

[0061] Model parameters of multivariate regression analysis By fitting the historical data of full-pipe operation, the least squares method and regularization optimization are used to obtain, and the objective function is:

[0062]

[0063] wherein, is the number of samples, is the full-pipe risk predicted by the model, is a regularization parameter, which controls the complexity of the model to prevent overfitting.

[0064] Specifically, first, the method uses the collected multi-dimensional data, including the flow, liquid level and pressure of different monitoring points, as input variables, and sets the full-pipe risk index as the output target. Through regression analysis, the method can determine the influence degree and relative importance of each input variable on the full-pipe risk, thereby establishing a statistical relationship model reflecting the relationship between the operation state of the pipe network and the full-pipe risk.

[0065] To ensure the accuracy and adaptability of the model, the method uses historical operation data for fitting when establishing the model, and applies the least squares method to minimize the error between the model prediction value and the actual full pipe risk index. In addition, to prevent overfitting when dealing with complex data, the method introduces regularization techniques to reduce model complexity, improve model adaptability and stability to new data.

[0066] Through this multivariate regression analysis and parameter optimization method, a precise full pipe risk prediction model can be established. This model not only can evaluate the current operation state of the sewer network in real time, but also can predict the future possible full pipe risk according to real-time data, providing a scientific basis for early warning.

[0067] S5, using long short-term memory network for time series prediction, predicting future flow and liquid level trend; the specific model of long short-term memory network for time series prediction is:

[0068]

[0069] Among them, is the predicted next time flow or liquid level, used for full pipe state prediction; is the current hidden state, reflecting the current operation trend of the pipe network; is the current input, representing real-time monitoring data, , is the weight matrix, is the bias term, is the activation function;

[0070] The loss function of the LSTM model is the weighted mean square error, and its calculation formula is:

[0071]

[0072] Among them, is the sample weight, which is inversely adjusted by the abnormal score S related to the full pipe risk to enhance the sensitivity of the model to full pipe events.

[0073] Specifically, the method fully utilizes the advantages of LSTM model in processing time series data by constructing LSTM model, captures the time dependence and complex nonlinear relationship of sewage pipe network flow and liquid level change. The LSTM model adopts a gating mechanism that can remember long-term historical information and effectively suppress irrelevant information, making the prediction of future flow and liquid level trend more accurate and stable.

[0074] In this method, the input to the LSTM model includes the current flow rate, liquid level, and other relevant real-time monitoring data. This input data is processed through the embedding layer and the hidden layer of the network, and the output is the predicted flow rate or liquid level value for the next time step. The hidden states used in the model can reflect the current operating status of the pipeline network, while the activation function (such as the Sigmoid or Tanh function) is used to introduce nonlinear features, enabling the model to capture complex dynamic features in the data.

[0075] LSTM models learn the relationship between input and output by continuously updating the weight matrix and bias terms. To improve the model's sensitivity to full-pipe events, this method uses weighted mean squared error (WMSE) as the loss function. In the design of the loss function, the error of each sample is assigned a different weight, which is inversely adjusted by the anomaly score associated with the risk of full-pipe events. By giving higher weights to high-risk samples, the model can focus more on data that may trigger full-pipe events during training, thereby enhancing its ability to predict full-pipe states.

[0076] By using LSTM for time series prediction, this method can not only accurately predict the future flow and level trends of sewage pipe networks, providing reliable basic data for early warning of full pipe conditions, but also respond to dynamic changes in sewage pipe networks in a timely manner.

[0077] S6. The full-pipe warning threshold is dynamically adjusted based on the time-series prediction results and the adaptive algorithm; the dynamic adjustment of the full-pipe warning threshold adopts an adaptive algorithm, and its adjustment formula is as follows:

[0078]

[0079] in, The dynamically adjusted warning threshold reflects the real-time response to the full-pipe warning. Basic threshold; As an adjustment factor, the adjustment range is adjusted to match changes in the pipeline network's operating status; The difference between the predicted and current values ​​directly affects the risk of full pipe.

[0080] Specifically, the adaptive algorithm assesses changes in the pipeline network's operating status by comparing predicted values ​​with current actual values, and adjusts the warning threshold according to the degree of change. Specifically, the adaptive algorithm first sets a basic warning threshold, determined based on historical data and experience, to reflect the risk of the pipeline network being full under standard operating conditions. Then, based on the difference between the real-time predicted flow rate or liquid level and the current value, the algorithm calculates an adjustment factor that reflects the magnitude of the change in the pipeline network's operating status. By incorporating the adjustment factor into the threshold calculation, the adaptive algorithm can dynamically amplify or reduce the warning threshold, making it more closely reflect the current operational reality.

[0081] The dynamically adjusted early warning threshold can not only respond to changes in the state of the pipe network in real time, but also can warn of potential full-pipe risks. Through this adaptive dynamic adjustment method, the early warning system can significantly reduce the incidence of false positives and false negatives, and improve the ability to respond to sudden situations.

[0082] S7, fuse real-time weather data, and use a deep learning model to establish a nonlinear model of the influence of weather on the sewage pipe network; the deep learning model uses a hybrid model of convolutional neural network and long short-term memory network, which is specifically used to calculate the correction of the influence of weather on the full-pipe risk, and the expression is:

[0083]

[0084]

[0085] wherein, is the weather influence correction coefficient, used to correct the full-pipe early warning threshold; , is the model weight, reflecting the influence of weather data on the load capacity of the sewage pipe network.

[0086] Specifically,

[0087] S8, combine the dynamically adjusted threshold and the weather influence prediction result to calculate the final early warning parameter; the calculation formula of the final early warning parameter is:

[0088]

[0089] wherein, is the final early warning threshold, directly used for full-pipe early warning triggering; is the adjustment weight, controlling the influence range of weather correction on the threshold; is the amplification factor, adjusting the sensitivity of the threshold response; when the early warning threshold exceeds the actual monitoring data, the system automatically triggers the full-pipe early warning signal and starts the corresponding emergency response scheme to prevent sewage overflow and environmental pollution.

[0090] Specifically, the method uses real-time weather data such as rainfall, rainfall intensity, wind speed, temperature, etc., and combines these weather factors with the operating conditions of the sewage pipe network through a deep learning model to establish a nonlinear influence model of weather on full-pipe risk. In order to more accurately reflect the dynamic influence of weather changes on the load capacity of the pipe network, the method uses a hybrid model of convolutional neural network (CNN) and long short-term memory network (LSTM).

[0091] Convolutional Neural Networks (CNNs) are primarily used to extract spatial features from meteorological data, enabling the identification of complex patterns in sewage pipe network loads under different meteorological conditions. Through multi-level convolution and pooling operations in convolutional layers, CNNs can effectively capture the local correlations and spatial distribution characteristics of meteorological data. These features are further input into an LSTM network, which excels at processing time-series data and can learn the temporal dynamics of meteorological factors and their long-term impact on pipe network conditions. Therefore, a hybrid model combining CNNs and LSTMs can fully leverage their respective strengths, establishing a comprehensive meteorological impact assessment model through joint learning of the spatiotemporal features of meteorological data.

[0092] In the model, the feature maps extracted by the CNN are processed by an activation function and then fed into an LSTM network. The LSTM network calculates the meteorological impact correction coefficient. This correction coefficient is used to dynamically adjust the full-pipeline warning threshold, ensuring that the threshold reflects the actual impact of current meteorological conditions on pipeline operation. The corrected warning threshold can respond more flexibly to extreme weather events, such as heavy rain or high temperatures, significantly improving the warning system's ability to identify these sudden risks.

[0093] The application of this deep learning model enables the early warning analysis method for sewage pipe networks to not only rely on static operational data but also dynamically incorporate the influence of meteorological factors, forming a comprehensive assessment of the risk of pipe overflow. Through nonlinear modeling and correction of meteorological data, the system can more accurately predict the risk of pipe overflow under different weather conditions, thereby providing more reliable decision support for urban sewage treatment and drainage management.

[0094] S9. Perform fusion analysis on multi-source data and utilize unsupervised learning algorithms to identify abnormal operating states in the sewage pipe network. The identification of abnormal operating states employs the Isolation Forest unsupervised learning algorithm, whose anomaly detection scoring function is:

[0095]

[0096] in, An abnormal score is indicated; a lower score suggests a potential risk of the pipe being full. The average path length of the data points reflects the degree of operational anomalies in the sewage pipe network.

[0097] Specifically, in order to identify abnormal operating states in the pipeline network, the method adopts the unsupervised learning algorithm Isolation Forest. This algorithm is specifically designed to detect outliers in the data. It separates data points by constructing multiple random binary trees and uses the ease with which data points are separated in the trees to measure anomalies.

[0098] The core of the Isolation Forest algorithm lies in constructing Isolation Trees by randomly selecting features and splitting points, and evaluating the anomaly of data points by the path length required for their separation in these trees. Specifically, the algorithm calculates the average path length of each data point in all Isolation Trees, and a shorter path means that the data point is more easily separated, i.e., it may be an outlier. Conversely, if the average path length of a data point is longer, it indicates that it is more difficult to isolate in the dataset, and it may be a normal point.

[0099] The anomaly detection scoring function is used to quantify the degree of anomaly of each data point, where a lower score value indicates that the data point is more abnormal, suggesting a potential full pipe risk. The advantage of the Isolation Forest algorithm lies in its sensitivity to multi-dimensional data anomalies, which can effectively handle complex, non-linear sewer network operation data and quickly identify abnormal operation patterns. Through this method, potential problems in the sewer network, such as pipe blockage, illegal discharge or other abnormal situations, can be detected early, so that preventive and response measures can be taken in a timely manner.

[0100] Example 2:

[0101] A specific implementation of early warning analysis according to Example 1:

[0102] Background:

[0103] In the sewer network system of a city, multiple key nodes are equipped with flow meters, level meters and pressure sensors to monitor the operation status of the network. This paper will take the actual collected data as an example to demonstrate how the method of the invention can be applied to the full pipe operation early warning analysis of the sewer network.

[0104] Data collection

[0105] Node 1 (main pipeline):

[0106] Flow h

[0107] Liquid level m

[0108] Pressure bar

[0109] Node 2 (branch pipeline):

[0110] Flow h

[0111] Liquid level m

[0112] Pressure bar

[0113] Weather data:

[0114] Current rainfall mm

[0115] Rainfall intensity mm / h

[0116] Implementation steps

[0117] 1. Multivariate regression analysis to establish relationship model

[0118] According to historical data and the above real-time data, a full-pipe risk model is established using multivariate regression analysis:

[0119] 2. Time series prediction and dynamic adjustment

[0120] The LSTM model is used to predict the flow and liquid level at the next moment, and the output is:

[0121] Predicted flow ;

[0122] According to the prediction results, the adaptive algorithm is applied to adjust the full-pipe warning threshold: 、 .

[0123] 3. Weather influence fusion and correction

[0124] A hybrid model of convolutional neural network and LSTM is used to calculate the weather influence correction coefficient , and the output of the weather model is: .

[0125] 4. Final warning parameter calculation

[0126] The final warning threshold is calculated by combining the dynamically adjusted threshold and the weather influence correction coefficient:

[0127]

[0128]

[0129] 5. Anomaly detection and warning

[0130] Anomaly detection is performed using the Isolation Forest algorithm, resulting in an anomaly score S\approx0.15S≈0.15, which is much lower than the warning threshold (e.g. set to 0.5). Therefore, the system triggers a full-pipe warning signal and notifies the operation and maintenance personnel through the warning platform to take emergency response.

[0131] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for sewer network full pipe operation early warning analysis, characterized in that, The method comprises the following steps: S1, laying sensors at key nodes of the sewer network to collect real-time operation data of the sewer network, including flow, liquid level and pressure data; S2, obtaining real-time meteorological data, including rainfall, rainfall intensity and other related meteorological factors; S3, performing data cleaning and standardization processing on the collected sewer network data and meteorological data; S4, establishing a relationship model between flow, liquid level and full pipe risk based on multivariate regression analysis; The relationship model established by the multivariate regression analysis in step S4 is: ； wherein, is the full pipe risk index, is the flow rate at different monitoring points, is the liquid level at different monitoring points, is the pressure at different monitoring points, , , are model parameters, optimized by fitting historical data under full pipe operating conditions; S5, using a long short-term memory network for time series prediction to predict future flow and liquid level trends; The specific model of the long short-term memory network for time series prediction in step S5 is: ； wherein, is the predicted next time flow or level for full pipe state prediction; is the current hidden state, reflecting the current operation situation of the pipe network; is the current input, representing real-time monitoring data, , is the weight matrix, is the bias term, is the activation function; S6, dynamically adjusting the full pipe early warning threshold according to the time series prediction result and an adaptive algorithm; The dynamic adjustment of the full pipe early warning threshold in step S6 uses an adaptive algorithm, and the adjustment formula is: ； wherein, is a dynamically adjusted early warning threshold, reflecting the real-time response to the full pipe early warning; is a basic threshold value; is an adjustment factor, adjusting the amplitude to match the changes in the pipe network operating state; is the difference between the prediction and the current value, directly affecting the full pipe risk; S7, fusing real-time meteorological data and using a deep learning model to establish a nonlinear model of the influence of meteorological factors on the sewer network; The deep learning model in step S7 uses a hybrid model of convolutional neural network and long short-term memory network, which is specifically used to calculate the correction of the influence of meteorological factors on the full pipe risk, and the expression is: ； ； wherein, is a weather influence correction factor for correcting the full pipe early warning threshold; , is a model weight embodying the influence of weather data on the load capacity of the sewer network. S8, combining the dynamically adjusted threshold and the meteorological influence prediction result to calculate the final early warning parameter; The calculation formula of the final early warning parameter in step S8 is: ； wherein, is the final warning threshold, directly used for full pipe warning triggering; is the adjustment weight, controlling the influence range of weather correction on the threshold; is the amplification factor, adjusting the sensitivity of the threshold response; S9, performing fusion analysis on multi-source data and using an unsupervised learning algorithm to identify abnormal operation states in the sewer network; The identification of the abnormal operation state in step S9 uses an unsupervised learning algorithm called isolation forest, and the anomaly detection score function is: ； wherein, is the anomaly score, lower score values indicate possible full pipe risk; is the average path length of data points, reflecting the degree of operational anomaly of the sewer network.

2. The full pipe operation early warning analysis method of a sewage pipe network according to claim 1, characterized in that, where the model parameters of the multivariate regression analysis By fitting the historical data of full pipe operation, the least square method and regularization optimization are adopted, and the objective function is: ； where, is the number of samples, is the full pipe risk predicted by the model, is the regularization parameter, controlling the model complexity to prevent overfitting.

3. The full pipe operation early warning analysis method of a sewage pipe network according to claim 1, characterized in that, The loss function of the long short-term memory network is weighted mean square error, and the calculation formula is: ； wherein, is the sample weight, adjusted inversely with the anomaly score S related to the full pipe risk, to enhance the model sensitivity to full pipe events.

Images