Intelligent monitoring device for urban drainage system

Abstract

The invention relates to the field of intelligent monitoring devices, in particular to an intelligent monitoring device of an urban drainage system, which is characterized by comprising a data acquisition module used for collecting sensor data in a drainage pipe network; the data preprocessing module is used for cleaning and standardizing the sensor data; the multi-source data fusion module is used for integrating heterogeneous data from different sensors; the space-time correlation analysis module is used for analyzing the time and space correlation of the heterogeneous data and extracting key features; the intelligent prediction module is used for predicting future conditions of the drainage system based on the key features; the decision support module is used for generating management suggestions according to future conditions; the visual display module is used for displaying the monitoring result and the prediction information; wherein the time-space correlation analysis module, the intelligent prediction module and the decision support module form a data processing closed loop, intelligent monitoring of the urban drainage system is achieved, and the prediction accuracy reaches 5.2% and is far better than that in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection systems, and more specifically, to an intelligent monitoring device for urban drainage systems. Background Art

[0003] At present, the following several methods are mainly used for monitoring urban drainage systems:

[0004] 1. Traditional SCADA system: This is the most common monitoring method, which can collect and display data in real time. However, the SCADA system mainly focuses on data collection and simple visualization, lacking intelligent analysis and prediction capabilities. In the face of complex dynamic changes in the drainage system, it is often unable to make accurate judgments in a timely manner, resulting in a lag in response.

[0005] 2. Simple machine learning models: Some cities have begun to try to apply machine learning technologies (such as random forests, support vector machines, etc.) to drainage system monitoring. These models can make certain predictions, but their prediction capabilities are limited. Especially when dealing with non-linear and highly dynamic drainage systems, the accuracy is often not satisfactory.

[0006] 3. Conventional deep learning models: In recent years, some studies have begun to explore the use of deep learning models (such as LSTM networks) for drainage system monitoring and prediction. These models have made significant progress compared with traditional methods, but there are still some key problems:

[0007] a) Insufficient handling of spatio-temporal correlations: The urban drainage system is a complex spatio-temporal system, and there are complex correlations between data in different regions and at different time points. Conventional deep learning models often have difficulty fully capturing these complex spatio-temporal relationships.

[0008] b) Lack of adaptability: The behavior of the drainage system varies greatly under different weather conditions. Existing models often have difficulty adapting to this highly dynamic environment, especially in extreme weather events, the prediction accuracy will drop significantly.

[0009] c) Weak ability to integrate multi-source heterogeneous data: The urban drainage system involves various types of data such as water level, flow rate, water quality, and meteorology. Existing methods have difficulties in integrating these heterogeneous data and cannot make full use of all available information.

[0010] d) Insufficient real-time performance and scalability: Many existing models are inefficient in processing large-scale real-time data and are difficult to meet the requirements of practical applications.

[0011] e) Limited decision-making support ability: Most existing systems only provide basic monitoring and early warning functions, lacking intelligent decision-making support ability and unable to provide specific operation suggestions for managers.

[0012] These problems severely restrict the intelligent management of urban drainage systems, increase the risk of urban waterlogging, and affect the quality of life and safety of urban residents. Therefore, it has become particularly necessary and urgent to develop an intelligent monitoring device that can comprehensively solve the above problems. Summary of the Invention

[0013] The purpose of the present invention is to provide an intelligent monitoring device for urban drainage systems, which is used to solve the problem of insufficient data processing accuracy in complex environments in the prior art. Especially under extreme weather conditions such as heavy rain, it can accurately identify and predict the water flow changes in the pipe network and respond to emergencies in a timely manner.

[0014] To solve the above technical problems, the present invention adopts the following technical solutions: An intelligent monitoring device for urban drainage systems, comprising:

[0015] A data acquisition module, which is used to collect sensor data in the drainage pipe network;

[0016] A data preprocessing module, which is used to clean and standardize the sensor data;

[0017] A multi-source data fusion module, which is used to integrate heterogeneous data from different sensors;

[0018] A spatio-temporal correlation analysis module, which is used to analyze the temporal and spatial correlations of the heterogeneous data and extract key features;

[0019] An intelligent prediction module, which is used to predict the future conditions of the drainage system based on the key features;

[0020] A decision support module, which is used to generate management suggestions according to the future conditions;

[0021] A visualization display module, which is used to present the monitoring results and prediction information;

[0022] Among them, the spatio-temporal correlation analysis module, the intelligent prediction module and the decision support module form a data processing closed loop to realize the intelligent monitoring of urban drainage systems.

[0023] Specifically, the data acquisition module includes a distributed sensor network, which is used to collect water level, flow rate and water quality parameters in real time.

[0024] Specifically, the data preprocessing module uses a data filtering algorithm to remove outliers and performs data format unification processing.

[0025] Specifically, the multi-source data fusion module uses data alignment and feature extraction technologies to convert data from different sources into unified feature vectors.

[0026] Specifically, the spatio-temporal correlation analysis module uses an improved spatio-temporal convolutional network algorithm ST-CNN+ for analysis. The ST-CNN+ algorithm includes:

[0027] An adaptive time window mechanism, where the calculation formula for the time window size W(t) is

[0028] W(t) = W base ·exp(α·V(t))

[0029] where W(t) is the time window size at time t, W base is the basic window size, α is the adjustment factor, and V(t) is the data change rate at time t;

[0030] A spatial attention mechanism, where the calculation formula for the attention weight A(s) is:

[0031] A(s) = softmax(W a ·tanh(W s ·F(s) + b s ))

[0032] where F(s) is the spatial feature, W s and b s are learnable parameters, W a is the attention weight matrix, a network structure combining residual connection and dense connection, where the output X l of the l-th layer is calculated as:

[0033] X l = H l ([X 0 , X 1 ,..., X l-1 ) + X l-1

[0034] where H l is the non-linear transformation function, [X 0 , X 1 ,..., X l-1 represents the feature connection of all previous layers, and the multi-scale feature fusion function is:

[0035] F fusion = W 1 ·F small + W 2 ·F medium + W 3 ·F large

[0036] where F small , F medium , F largeRepresent features at small, medium, and large scales respectively, and W 1 , W 2 , W 3 are the corresponding weights.

[0037] Specifically, the execution steps of the ST-CNN+ algorithm include:

[0038] (1) Input multi-source heterogeneous data;

[0039] (2) Apply an adaptive time window mechanism to process time series data;

[0040] (3) Use a spatial attention mechanism to highlight important spatial features;

[0041] (4) Extract deep features through a network structure that combines residual connections and dense connections;

[0042] (5) Adopt a multi-scale feature fusion strategy to integrate spatio-temporal features at different scales;

[0043] (6) Output the spatio-temporal correlation analysis results.

[0044] Specifically, the intelligent prediction module uses a deep learning model for short-term and medium- to long-term predictions, and this deep learning model is trained and optimized based on the output results of the spatio-temporal correlation analysis module.

[0045] Specifically, the decision support module generates optimization suggestions by combining preset rules and machine learning algorithms, where the machine learning algorithms continuously learn and adjust based on historical decision data and actual effects.

[0046] Specifically, the visualization display module uses WebGL technology to achieve three-dimensional visualization display, including graphical presentation of the real-time status of the drainage pipe network, prediction results, and decision-making suggestions.

[0047] Specifically, it also includes a data storage module for storing historical monitoring data, analysis results, and decision-making records to support the continuous optimization and long-term analysis of the system.

[0048] The beneficial effects of the present invention include:

[0049] 1. High-precision prediction: Through the innovative ST-CNN+ algorithm, the present invention can effectively capture the complex spatio-temporal correlations of the drainage system, significantly improving the prediction accuracy. In actual tests, the prediction accuracy (MAPE) reaches 5.2%, far superior to the prior art.

[0050] 2. Quick response: Thanks to the efficient data processing and analysis capabilities, the present invention can respond to abnormal situations within 2 minutes, greatly improving the system's ability to handle emergencies.

[0051] 3. Low false alarm rate and missed alarm rate: While maintaining high sensitivity, the present invention controls the false alarm rate and missed alarm rate at 3.5% and 1.2% respectively, significantly reducing resource waste and potential safety hazards.

[0052] 4. Strong adaptability: Through the adaptive time window mechanism and multi-scale feature fusion, the present invention can adapt to various weather conditions and drainage scenarios, especially performing excellently in extreme weather events.

[0053] 5. Multi-source data integration: The present invention can effectively integrate various types of data such as water level, flow velocity, water quality, and meteorology, making full use of all available information and improving the comprehensiveness of monitoring and prediction.

[0054] 6. Intelligent decision support: Through deep reinforcement learning technology, the present invention not only provides monitoring and early warning, but also generates specific operation suggestions for managers, greatly improving the decision-making efficiency.

[0055] 7. Visualization ability: By adopting advanced 3D visualization technology, the present invention can intuitively present the complex state of the drainage system and prediction results, facilitating managers to quickly understand and make decisions.

[0056] 8. High stability: The system has only 0.2 failures per week on average, demonstrating excellent long-term operation ability and ensuring continuous monitoring and early warning.

[0057] 9. Scalability: The present invention adopts a modular design and is easy to be customized and expanded according to the needs of different cities.

[0058] In summary, the intelligent monitoring device for urban drainage systems provided by the present invention comprehensively solves the problems existing in the prior art through innovative algorithms and system designs, providing strong technical support for the intelligent management of urban drainage systems. It can not only effectively reduce the risk of urban waterlogging and improve the urban disaster prevention and mitigation ability, but also provide new solutions for the construction of smart cities and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is the framework schematic diagram of an intelligent monitoring device for urban drainage systems of the present invention.

[0060] Figure 2 is the framework schematic diagram of the spatio-temporal correlation analysis module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0061] The scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. The embodiments in the present invention and all other embodiments obtained by ordinary persons in the art without making creative work are within the scope of protection of the present invention.

[0062] Please refer to Figure 1-2 The present invention provides an intelligent monitoring device for urban drainage system, comprising a data acquisition module 1, a data preprocessing module 2, a multi-source data fusion module 3, a spatiotemporal correlation analysis module 4, an intelligent prediction module 5, a decision support module 6 and a visualization display module 7. These modules work together to form a complete data processing closed loop to realize intelligent monitoring of urban drainage system.

[0063] The present invention is described in detail below in conjunction with specific embodiments. It should be understood by those skilled in the art that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0064] In a typical application scenario, the urban drainage system intelligent monitoring device of the present invention is deployed in a coastal city with a population of about 5 million. The city faces the threat of typhoons and rainstorms all year round, and the intelligent management of the drainage system is particularly important.

[0065] 1. Data acquisition module 1, data acquisition module 1 includes a sensor network distributed throughout the city. These sensors mainly include:

[0066] Water level sensor: installed in main drainage pipes and rainwater wells to monitor water level changes in real time.

[0067] Flow rate sensor: installed at key nodes to monitor water flow rate.

[0068] Water quality sensors: monitor parameters such as pH, turbidity, dissolved oxygen, etc.

[0069] Weather station: provides real-time rainfall, wind speed and other meteorological data.

[0070] These sensors transmit data to the central processing system in real time through IoT technologies such as NB-IoT or LoRa. The frequency of data collection can be dynamically adjusted according to weather conditions, usually once every 5 minutes, and can be increased to once a minute during heavy rain.

[0071] 2. Data preprocessing module 2: Data preprocessing module 2 cleans and standardizes the original data. The specific steps include:

[0072] (1) Outlier detection and processing: Use the improved Z-score algorithm to detect outliers.

[0073]

[0074] Among them, X is the original data, μ is the moving average, and σ is the standard deviation. If |Z| > 3, then this data point is considered an outlier

[0075] value, and it is replaced with the average value of the data before and after.

[0076] (2) Data standardization: The Min-Max standardization method is adopted.

[0077]

[0078] Among them, X is the original data, X min and X max are the minimum and maximum values of this type of data respectively.

[0079] (3) Time alignment: Align the data of different sensors according to the timestamps to ensure the time consistency of the data.

[0080] 3. Multi-source data fusion module 3. The multi-source data fusion module 3 integrates heterogeneous data from different sensors. The specific steps are as follows:

[0081] (1) Feature extraction: Extract features from each type of sensor data. For example, extract features such as mean, variance, and change rate from water level data.

[0082] (2) Data alignment: Use the dynamic time warping (DTW) algorithm to perform time alignment on data from different sources.

[0083] (3) Feature vector construction: Combine the features of various sensors into a unified feature vector.

[0084] F = [F 水位 , F 流速 , F 水质 , F 气象

[0085] 4. Spatiotemporal correlation analysis module 4. The spatiotemporal correlation analysis module 4 is the core innovation of the present invention and uses an improved spatiotemporal convolutional network algorithm (ST-CNN+) for analysis. This algorithm includes the following key components:

[0086] (1) Adaptive time window mechanism 41

[0087] W(t) = W base ·exp(α·V(t))

[0088] Among them, W(t) is the size of the time window at time t, W base ​Based on the basic window size (usually set to 1 hour), α is the adjustment factor (the empirical value range is 0.1 - 0.5), and V(t) is the data change rate at time t. This mechanism allows the algorithm to automatically shrink the time window when the data changes violently (such as at the beginning of a heavy rain) to improve the analysis accuracy; and expand the window when the data is relatively stable to reduce the calculation amount.

[0089] (2) Spatial attention mechanism 42

[0090] A(s) = softmax(W a ·tanh(W s ·F(s) + b s ))

[0091] Among them, F(s) is the spatial feature, W s and b s are learnable parameters, and W a is the attention weight matrix. The spatial attention mechanism can automatically identify and highlight important spatial features. For example, during heavy rain, the data of certain waterlogging-prone points will receive higher attention.

[0092] (3) Network structure combining residual connection and dense connection 43

[0093] X l = H l ([X 0 , X 1 , …, X l-1 ) + X l-1

[0094] Among them, X l is the output of the l-th layer, H l is the non-linear transformation function, and [X 0 , X 1 , …, X l-1 represents the feature connection of all previous layers.

[0095] This structure can effectively improve the expression ability of the network, and at the same time alleviate the problem of gradient disappearance, enabling the model to better capture long-term dependence relationships.

[0096] (4) Multi-scale feature fusion

[0097] F fusion = W 1 ·F small + W 2 ·F medium + W 3 ·F large

[0098] Among them, F small 、F medium 、Flarge represent small, medium, and large-scale features respectively, and W 1 , W 2 , W 3 are the corresponding weights.

[0099] Multi-scale feature fusion allows the model to consider features at different spatio-temporal scales simultaneously. For example, small-scale features may reflect local emergencies, while large-scale features may represent the drainage trend of the entire city.

[0100] The specific implementation steps of the ST-CNN+ algorithm are as follows:

[0101] Step 1: Input multi-source heterogeneous data.

[0102] Step 2: Apply an adaptive time window mechanism to process time series data.

[0103] Step 3: Use a spatial attention mechanism to highlight important spatial features.

[0104] Step 4: Extract deep features through a network structure that combines residual connections and dense connections. In this step, we constructed a network with 10 convolutional layers. The output of each convolutional layer is connected not only to the next layer but also to the outputs of all previous layers. This structure allows information to flow more freely in the network, helping to capture complex spatio-temporal patterns.

[0105] Step 5: Adopt a multi-scale feature fusion strategy to integrate spatio-temporal features at different scales. Specifically, we used three scales of convolutional kernels: 3x3 (small scale), 5x5 (medium scale), and 7x7 (large scale). These features at different scales are fused by weighted summation, and the weights are automatically learned through backpropagation.

[0106] Step 6: Output the spatio-temporal correlation analysis results. The final output is a high-dimensional feature vector that contains the status information and future trends of key points in the drainage system.

[0107] In practical applications, the ST-CNN+ algorithm can effectively process complex spatio-temporal data in urban drainage systems. For example, when predicting whether there will be waterlogging in a specific area, the algorithm not only considers the real-time water level and rainfall in that area but also takes into account the drainage conditions in the upstream area, terrain features, and drainage patterns under similar weather conditions in historical data. This comprehensive analysis greatly improves the accuracy of the prediction.

[0108] 5. Intelligent Prediction Module 5. Based on the output results of the Spatiotemporal Association Analysis Module, the Intelligent Prediction Module 5 uses a deep learning model for short-term (1 - 3 hours) and medium- and long-term (6 - 24 hours) predictions. This module adopts a method that combines the Long Short-Term Memory Network (LSTM) and the attention mechanism.

[0109] The specific steps are as follows:

[0110] (1) Data Preparation: Use the output results of the Spatiotemporal Association Analysis Module as input features, and at the same time introduce external data such as weather forecast information.

[0111] (2) Model Structure:

[0112] Input Layer: Receive spatiotemporal feature vectors

[0113] LSTM Layer: Capture long-term dependencies, including 128 hidden units

[0114] Attention Layer: Calculate the importance weights of different time steps

[0115] Fully Connected Layer: Map the output of the LSTM to the prediction results

[0116] (3) Training Process:

[0117] Use historical data for training, and adopt the sliding window method to generate training samples

[0118] Loss Function: Mean Squared Error (MSE)

[0119] Optimizer: Adam, the initial learning rate is set to 0.001, and the learning rate decay strategy is used

[0120] Number of Training Epochs: Dynamically adjusted according to the performance of the validation set, usually between 100 - 200 epochs

[0121] (4) Prediction:

[0122] Short-Term Prediction: Update every 15 minutes, and predict the state of the drainage system in the next 3 hours

[0123] Medium- and Long-Term Prediction: Update every hour, and predict the trend in the next 24 hours

[0124] The prediction results include the predicted water levels and flow velocities at each key point, as well as the assessment of potential overflow risks.

[0125] 6. Decision Support Module 6. The Decision Support Module 6 combines preset rules and machine learning algorithms to generate optimization suggestions. The core of this module is a decision-making system based on reinforcement learning, and the specific implementation is as follows:

[0126] (1) State Space: Includes the current water level, flow velocity, prediction results, etc.;

[0127] (2) Action space: including adjusting the operation of the pumping station, opening / closing the gate, issuing early warnings, etc.;

[0128] (3) Reward function: based on the overall system performance score, such as waterlogging area, duration of waterlogging, etc.;

[0129] The decision-making process adopts the Deep Q-Network (DQN) algorithm:

[0130] Q(s,a) = R(s,a) + γ·max(Q(s′,a′))

[0131] Among them, s is the current state, a is the action, R is the immediate reward, γ is the discount factor (usually set to 0.95), and s′ is the next state.

[0132] DQN network structure:

[0133] Input layer: state vector

[0134] Hidden layer: two fully connected layers, with 256 neurons in each layer, and the activation function is ReLU

[0135] Output layer: Q values corresponding to each possible action

[0136] Training process:

[0137] Use the experience replay buffer to store the transition samples (s,a,r,s′), and at each time step, randomly sample batches from the buffer for training.

[0138] Calculation of the target Q value: y = r + γ·max(Q(s′,a′;θ-))

[0139] Loss function: L = (y - Q(s,a;θ)) 2

[0140] Regularly update the target network parameters θ.

[0141] In practical applications, the decision support module 6 will give specific operation suggestions according to the current situation and prediction results. For example, when it is predicted that waterlogging may occur in a certain area, the system will suggest turning on the pumping station in that area in advance and adjusting the opening degree of the upstream gate at the same time to disperse the water flow.

[0142] 7. Visualization display module 7. The visualization display module 7 uses WebGL technology to achieve three-dimensional visualization display, including the graphical presentation of the real-time state of the drainage pipe network, prediction results, and decision-making suggestions. Specific functions include:

[0143] (1) Three-dimensional city model: display the city terrain and main buildings;

[0144] (2) Visualization of the drainage pipe network: Represent the water level and flow velocity status of the pipe network with different colors;

[0145] (3) Marking of early warning areas: Mark the areas prone to waterlogging prominently;

[0146] (4) Display of decision-making suggestions: Display specific decision-making suggestions in the form of icons and text descriptions;

[0147] (5) Timeline control: Allow users to view historical data and future prediction results;

[0148] Implementation details:

[0149] Use the Three.js library to build a 3D scene;

[0150] Adopt hierarchical loading technology to ensure smooth display of large-scale data;

[0151] Implement interactive functions such as zooming, rotating, and clicking to view details.

[0152] 8. Data storage module 8, which is used to store historical monitoring data, analysis results, and decision-making records to support the continuous optimization and long-term analysis of the system. This module adopts a distributed storage architecture, specifically including:

[0153] (1) Time series database (such as InfluxDB): Store raw sensor data and calculation results

[0154] (2) Relational database (such as PostgreSQL): Store structured data such as system configurations and device information

[0155] (3) Document database (such as MongoDB): Store unstructured data such as decision-making records and system logs

[0156] Data storage strategy:

[0157] Real-time data: Retain high-precision data for the most recent 30 days (sampling interval of 5 minutes);

[0158] Historical data: Downsample data over 30 days and retain the hourly average, with a storage period of 5 years.

[0159] To verify the superiority of the present invention, a series of comparative experiments were conducted. The following are the detailed descriptions and comparison results of the embodiment and three comparative examples.

[0160] Example 1: Use the intelligent monitoring device for urban drainage systems of the present invention

[0161] Deploy the intelligent monitoring device of the present invention in a coastal city with an area of about 500 square kilometers and a population of about 5 million. The annual average rainfall in this city is 1500 millimeters, and it is vulnerable to typhoons. The system includes 500 water level sensors, 200 flow velocity sensors, 100 water quality sensors and 50 weather stations, covering the city's main drainage pipelines and flood-prone areas.

[0162] Comparative Example 1: Traditional SCADA system

[0163] Use the traditional Supervisory Control and Data Acquisition (SCADA) system to monitor the drainage system. This system can collect and display data in real time, but lacks intelligent analysis and prediction functions.

[0164] Comparative Example 2: Simple machine learning model

[0165] On the basis of the traditional SCADA system, a machine learning model based on random forest is added for simple prediction. This model can make short-term predictions, but does not have complex spatio-temporal analysis capabilities.

[0166] Comparative Example 3: Conventional deep learning model

[0167] Adopt a conventional deep learning model (such as the LSTM network) to monitor and predict the drainage system. This model has certain prediction capabilities, but lacks full consideration of complex spatio-temporal relationships.

[0168] The detection standards and methods are as follows. In order to comprehensively evaluate the performance of each system, we set the following detection indicators:

[0169] 1. Prediction accuracy: Use the Mean Absolute Percentage Error (MAPE) to evaluate the prediction accuracy.

[0170] MAPE = (1 / n) * Σ|(actual value - predicted value) / actual value| * 100%

[0171] 2. Response time: The time interval from receiving abnormal data to generating an early warning.

[0172] 3. False alarm rate: The percentage of the number of false alarms in the total number of early warnings.

[0173] 4. Missed alarm rate: The percentage of actual waterlogging events that failed to be warned in the total number of waterlogging events.

[0174] 5. System stability: The number of system failures per week.

[0175] 6. Computational efficiency: The average time required to process data per hour.

[0176] Detection method:

[0177] During the 6 - month test period, including normal weather and 3 typhoon weather conditions, system operation data was collected.

[0178] An independent monitoring device was used to verify the prediction results.

[0179] The response time and system stability were analyzed through manual records and system logs.

[0180] Professional water engineers were invited to evaluate the early - warning effect of the system and the quality of decision - making suggestions.

[0181] The test results are shown in the following table. The following are the average performance indicators of each system during the 6 - month test period:

[0182] Index Example 1 Comparative Example 1 Comparative Example 2 Comparative Example 3 Prediction Accuracy Rate (MAPE) 5.2％ Not applicable 15.8％ 9.7％ Response Time (minutes) 215 105 Not applicable Not applicable False Alarm Rate 3.5％ 12.5％ 8.7％ 6.2％ Missed Alarm Rate 1.2％ 18.3％ 7.5％ 4.8％ System Stability (faults / week) 0.2 1.5 0.8 0.5 Computing Efficiency (seconds / hour of data) 45 50 30 60

[0183] Based on the above test results, the present invention (Example 1) is significantly superior to other comparative systems in most key indicators and can therefore be considered the best embodiment. Now, an in - depth analysis of the test results is carried out:

[0184] 1. Prediction accuracy: The MAPE of the present invention is 5.2%, far superior to other solutions. This is mainly due to the effective modeling of complex spatio - temporal relationships by the ST - CNN+ algorithm. High accuracy means that the system can more accurately predict potential waterlogging risks, providing a reliable basis for decision - makers.

[0185] 2. Response time: The present invention only needs 2 minutes to respond to abnormal situations, which is particularly important when dealing with extreme weather events such as sudden heavy rain. A quick response can gain precious preparation time for flood control departments.

[0186] 3. False - alarm rate and missed - alarm rate: The present invention performs well in both of these two indicators, being 3.5% and 1.2% respectively. A low false - alarm rate reduces unnecessary resource waste, while a low missed - alarm rate ensures that potential safety hazards can be detected in a timely manner. This balance reflects that the system has achieved a good trade - off between sensitivity and specificity.

[0187] 4. System stability: There are only 0.2 failures per week on average, indicating that the system has a high degree of reliability. In long - term operation, high stability can ensure continuous monitoring and early - warning capabilities.

[0188] 5. Computational efficiency: Although the present invention takes a relatively long time (45 seconds) to process hourly data, considering its complex analysis process and high - precision output, this time is acceptable. Moreover, this processing time is still much less than the data collection interval (1 hour), so it will not affect real - time performance.

[0189] These results fully verify the superiority of the present invention, especially in the following aspects:

[0190] a) Complex environmental adaptability: Through the ST-CNN+ algorithm, the present invention can effectively process complex spatio-temporal data in urban drainage systems and adapt to various weather conditions and drainage scenarios.

[0191] b) Prediction accuracy: Significantly higher than the prediction accuracy of other methods, providing a solid foundation for accurate risk assessment and decision-making.

[0192] c) Real-time performance: Fast response time and low computational overhead enable the system to respond promptly to emergencies.

[0193] d) Reliability: Low false alarm rate and missed alarm rate, as well as high system stability, ensure the reliability of long-term operation.

[0194] In summary, the intelligent monitoring device for urban drainage systems of the present invention realizes efficient and accurate monitoring and prediction of complex urban drainage systems by integrating advanced data processing technologies and deep learning algorithms. This not only improves the urban flood control and disaster reduction capabilities but also provides strong support for the construction of smart cities. In future practical applications, this system is expected to further optimize urban water resource management.

[0195] However, the protection scope of the present invention is not limited thereto; anyone familiar with the art within the scope disclosed by the present invention; according to the solution of the present invention and its improved conceptions for substitution or change; should be covered within the protection scope of the present invention.

Claims

1. An intelligent monitoring device for urban drainage system, characterized in that: include: Data acquisition module, used to collect sensor data in the drainage network; A data preprocessing module, used for cleaning and standardizing the sensor data; Multi-source data fusion module, used to integrate heterogeneous data from different sensors; A spatiotemporal correlation analysis module, used to analyze the temporal and spatial correlation of the heterogeneous data and extract key features; An intelligent prediction module for predicting the future condition of the drainage system based on the key characteristics; a decision support module for generating management recommendations based on the future conditions; Visual display module, used to present monitoring results and prediction information; Among them, the spatiotemporal correlation analysis module, the intelligent prediction module and the decision support module form a data processing closed loop to realize intelligent monitoring of the urban drainage system.

2. The intelligent monitoring device for urban drainage system according to claim 1 is characterized in that: The data acquisition module includes a distributed sensor network for real-time acquisition of water level, flow rate and water quality parameters.

3. The intelligent monitoring device for urban drainage system according to claim 1 is characterized in that: The data preprocessing module uses a data filtering algorithm to remove abnormal values ​​and performs data format unification processing.

4. The intelligent monitoring device for urban drainage system according to claim 1, characterized in that: The multi-source data fusion module uses data alignment and feature extraction technology to transform data from different sources into a unified feature vector.

5. The intelligent monitoring device for urban drainage system according to claim 1 is characterized in that: The spatiotemporal correlation analysis module uses an improved spatiotemporal convolutional network algorithm ST-CNN+ for analysis. The ST-CNN+ algorithm includes: Adaptive time window mechanism, where the calculation formula of the time window size W(t) is: Among them, W(t) is the time window size at time t, W base is the basic window size, α is the adjustment factor, and V(t) is the data change rate at time t; Spatial attention mechanism, where the calculation formula of attention weight A(s) is: A(s)=softmax(W a ·tanh(W s ·F(s)+b s )) Among them, F(s) is the spatial feature, W s and b s is a learnable parameter, W a is the attention weight matrix, the network structure combining residual connection and dense connection, where the lth layer outputs X l The calculation formula is: X l =H l ([X0,X1,...,X l-1 )]+X l-1 Among them, H l is a nonlinear transformation function, [X0,X1,...,X l-1 ] indicates that the features of all previous layers are connected to the multi-scale feature fusion, and its fusion function is: F fusion =W1·F small +W2·F medium +W3·F large Among them, F small、 F medium 、F large They represent small, medium and large scale features respectively, and W1, W2 and W3 are the corresponding weights.

6. The intelligent monitoring device for urban drainage system according to claim 5 is characterized in that: The execution steps of the ST-CNN+ algorithm include: (1) Input multi-source heterogeneous data; (2) Applying the adaptive time window mechanism to process time series data; (3) Use spatial attention mechanism to highlight important spatial features; (4) Extract deep features through a network structure that combines residual connections with dense connections; (5) Adopting multi-scale feature fusion strategy to integrate spatiotemporal features of different scales; (6) Output the spatiotemporal correlation analysis results.

7. The intelligent monitoring device for urban drainage system according to claim 1 is characterized in that: The intelligent prediction module uses a deep learning model to perform short-term and medium- to long-term predictions, and the deep learning model is trained and optimized based on the output results of the spatiotemporal correlation analysis module.

8. The intelligent monitoring device for urban drainage system according to claim 1, characterized in that: The decision support module generates optimization suggestions in combination with preset rules and machine learning algorithms, wherein the machine learning algorithms continuously learn and adjust based on historical decision data and actual results.

9. The intelligent monitoring device for urban drainage system according to claim 1, characterized in that: The visualization display module uses WebGL technology to achieve three-dimensional visualization, including the real-time status of the drainage network, the graphical presentation of the prediction results and the decision-making suggestions.

10. The intelligent monitoring device for urban drainage system according to claim 1, characterized in that: It also includes a data storage module for storing historical monitoring data, analysis results and decision records to support continuous optimization and long-term analysis of the system.

Images