Urban underground pipeline safety active prevention and control management system and method

Abstract

The invention discloses an urban underground pipeline safety active prevention and control management system and method, and mainly relates to the technical field of urban underground pipeline prevention and control management. The data acquisition module is used for acquiring multi-source sensing data of urban underground pipelines in real time; the data preprocessing module is used for performing noise filtering and normalization processing on the multi-source sensing data and extracting time domain and frequency domain features; the deep learning model module is composed of a time sequence feature extraction network and an anomaly detection network; the risk assessment module is used for calculating a pipeline health score based on the output of the deep learning model module; and the active prevention and control module triggers an early warning signal according to the risk level and is linked with a pressure regulating valve and a flow pump station to execute dynamic regulation and control. The method has the beneficial effects that fusion of multi-source data and accurate prevention and control and dynamic optimization of underground pipelines are realized, and the safety management efficiency of the pipelines is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of urban underground pipeline prevention and control management, and specifically to an active prevention and control management system and method for urban underground pipeline safety based on deep learning. Background Art

[0002] With the expansion of the scale and the increase in the complexity of urban underground pipelines, the demand for their safety prevention and control management is becoming more and more urgent. The current development of active prevention and control management systems and methods for underground pipelines presents the following status quo and problems: In terms of the system, existing technologies mostly rely on single or a small number of types of sensor data. For example, they only focus on pressure and flow monitoring, and there is insufficient fusion of multi-source information such as temperature, vibration signals, and pipeline materials, resulting in a single data dimension and making it difficult to comprehensively reflect the pipeline operation status. In the preprocessing link, simple filtering and standardization are often used, and it is not optimized for the complex environment of underground pipelines (such as electromagnetic interference and multiple noise sources), and the poor data quality affects subsequent analysis. In the application of deep learning models, basic LSTM networks are mostly used for time series feature extraction, and the attention mechanism is not combined to mine key time series correlations. The spatial feature fusion relies on traditional manual features or simple CNNs, and it is impossible to accurately capture the pipeline spatial topological relationship, and the accuracy of anomaly detection and risk assessment is limited. Risk prevention and control is mainly passive early warning, lacking an active regulation mechanism based on dynamic risk levels, and it is difficult to intervene in high-risk states in a timely manner; model updates rely on offline data and cannot adapt to the real-time changes of pipeline operation data. After long-term operation, the prediction accuracy decays, and it is difficult to ensure the continuous reliability of the system; At the method level, in the data acquisition and fusion stage, the spatio-temporal correlation mining of GIS geographical information and multi-source sensing data is insufficient, and a refined spatio-temporal data matrix is not constructed, making it difficult to support accurate feature extraction. Spatio-temporal feature extraction mostly separates time series and spatial analysis, and does not achieve deep fusion, losing key coupling information. Risk prediction only relies on a single scoring threshold, and does not consider the influence of dynamic factors such as seasons and pipeline loads on the threshold, resulting in a high risk of misjudgment and missed judgment. The generation of prevention and control strategies lacks quantitative regulation formulas and precise execution logics, relies on manual experience, has a lagging response and poor regulation effects. Model optimization does not design a loss function and an efficient update algorithm adapted to the underground pipeline scenario, the parameter update is slow and prone to overfitting, and it cannot quickly adapt to the changes in pipeline status.

[0003] These problems make it difficult for existing systems and methods to meet the requirements of urban safe operation in terms of the accuracy of pipeline risk identification, the initiative of prevention and control, and long-term reliability, and there is an urgent need for innovative technological breakthroughs. Summary of the Invention

[0004] The purpose of the present invention is to provide an active prevention and control management system and method for urban underground pipeline safety based on deep learning, which realizes the fusion of multi-source data, the precise prevention and control of underground pipelines, and dynamic optimization, and improves the pipeline safety management efficiency.

[0005] To achieve the above object, the present invention is implemented through the following technical solutions: On the one hand, a method for active prevention and control management of urban underground pipeline safety is provided, including the following steps: Step S1: Real-time collect pipeline operation data, and fuse pipeline GIS geographic information data to construct a spatio-temporal correlation data matrix , where is the number of pipeline nodes, is the time step, is the number of sensor channels; Step S2: Input the spatio-temporal correlation data matrix constructed in Step S1 into the time series feature extraction network to obtain time series features , where is the time series feature dimension; extract spatial topology features through the CNN of the anomaly detection network , where is the number of convolutional kernels; adopt feature splicing operation to fuse spatio-temporal features to obtain joint features , where represents feature splicing; Step S3: Input the joint feature in Step S2 into the fully connected layer, and calculate the health score through the Sigmoid function , where is the weight of the fully connected layer, is the bias term; divide the risk level according to the health score: Safe state: ; Warning state: ; High-risk state: ; Step S4: If it is judged as the high-risk state, generate a flow regulation instruction , where is the maximum safe flow of the pipeline, is the regulation coefficient; through the PID controller, adjust the opening of the pressure valve so that the actual flow satisfies .

[0006] Preferably, in the said Step S2, the structural parameters of the CNN network include: Convolutional layer: There are 3 layers set, the kernel sizes are 3*3, 5*5, 3*3 in sequence, and the number of channels are 32, 64, 128 in sequence; Pooling layer: Adopt max pooling, and the window size is 2*2; Activation function: Adopt LeakyReLU, and the negative slope coefficient is 0.01.

[0007] Preferably, in step S3, the risk level thresholds are specifically divided as follows: Initial threshold Determined through historical accident data statistics; The threshold is dynamically adjusted according to seasonal factors, including temperature changes and pipeline load fluctuations.

[0008] Preferably, in step S4, the control coefficient Dynamic adjustment based on pipeline material, specifically: When the pipeline material is cast iron, ; When the pipeline material is steel pipe, .

[0009] Preferably, it also includes: Step S5: Model online optimization, specifically: Calculate the loss function based on the newly added data: ; in, is the binary cross entropy, is the true label, 、 is the weight coefficient, are model parameters; Stochastic gradient descent is used to update the model parameters, and the learning rate .

[0010] On the other hand, an active prevention and control management system for urban underground pipeline safety is provided, based on the above-mentioned active prevention and control management method for urban underground pipeline safety, comprising: Data acquisition module: real-time acquisition of multi-source sensor data of urban underground pipelines, including pressure, temperature, flow, vibration signals and pipeline material information; Data preprocessing module: performs noise filtering and normalization processing on the multi-source sensor data, and extracts time domain and frequency domain features; Deep learning model module: It consists of a time series feature extraction network and an anomaly detection network. The time series feature extraction network uses a combination of LSTM network and attention mechanism to mine the time series correlation features of pipeline data. The anomaly detection network adopts a convolutional neural network (CNN) to fuse pipeline spatial topology features; Risk assessment module: Calculates pipeline health scores based on the output of the deep learning model module and generates three risk levels: safe, early warning, and high-risk based on preset thresholds; Active prevention and control module: triggers early warning signals based on risk levels, and links pressure regulating valves and flow pump stations to perform dynamic control; Feedback Optimization Module: Updates the model parameters through an online learning mechanism to continuously optimize the risk prediction accuracy.

[0011] Preferably, the normalization process of the data preprocessing module follows the following method: ; where is the mean of the sensor data, is the standard deviation, .

[0012] Preferably, the temporal feature extraction network performs the following calculation process: Input the temporal data sequence , and output the hidden state through the LSTM network, where ; Weight the historical hidden states through the attention weights to obtain the context vector , where is the weight matrix, is the bias term; Generate the final temporal feature through a linear transformation and the ReLU activation function: , where is the transformation matrix, is the bias term.

[0013] Preferably, the PID controller of the active prevention and control module is an incremental PID controller, and its control formula is: ; where is the control quantity at the th moment, is the flow difference at the th moment; The PID parameters are fixed as: , , .

[0014] Preferably, the feedback optimization module updates the model parameters using the stochastic gradient descent algorithm with momentum, and the specific formula is: ; where is the momentum term at the th step, which is used to accelerate the convergence of the model, and the momentum coefficient ; is the gradient of the loss function at , is the learning rate.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The data acquisition module covers multi-source information such as pressure, temperature, flow rate, vibration, and material. Compared with the single data dimension of the existing system, it can comprehensively depict the operation status of the pipeline, laying a foundation for accurate risk identification. The preprocessing module solves the data quality problems in the complex environment of underground pipelines through noise filtering, normalization, and feature extraction, improving the data availability; 2. The deep learning model module integrates the time series network of LSTM + attention and the CNN spatial network. In time series feature extraction, the attention mechanism focuses on key historical states, strengthening the mining of effective time series correlations; in spatial feature fusion, CNN is used to accurately capture the pipeline topology relationship. Compared with traditional models, it significantly improves the accuracy of anomaly detection and risk assessment, reducing false positives and false negatives.

[0016] 3. The risk assessment module outputs a dynamic health score and a three-level risk level. The active prevention and control module controls and regulates based on the level linkage hardware, changing from passive warning to active intervention to promptly contain high-risk hazards. The feedback optimization module continuously updates the model through online learning to adapt to the real-time changes of pipeline data, ensuring the long-term prediction accuracy and reliability of the system and solving the problem of accuracy decay in existing systems; 4. A spatio-temporal correlation data matrix is constructed, integrating multi-source sensing and GIS geographical information to achieve fine-grained organization of data in spatio-temporal dimensions; features are extracted and spliced and fused through the time series network and CNN respectively to mine spatio-temporal coupling information. Compared with the existing method of separating spatio-temporal analysis, it is more in line with the spatio-temporal correlation characteristics of underground pipelines, improving the feature representation ability; 5. The health score is calculated based on the Sigmoid function, and the risk level is divided in combination with a dynamic threshold (historical accident statistics + seasonal factor adjustment). It not only uses historical data to ensure the rationality of the threshold but also adapts to dynamic factors such as seasons, reducing the false positive and false negative rates and improving the accuracy of risk prediction; 6. A quantitative flow regulation formula is generated for high-risk states, clarifying parameters such as regulation coefficients and maximum safe flow rates, and accurately executing in combination with a PID controller to replace manual experience regulation, realizing scientific quantification and rapid response of prevention and control strategies and improving the efficiency of high-risk hazard disposal; 7. A loss function containing binary cross-entropy and L2 regularization is designed to balance the classification loss and model complexity. Stochastic gradient descent is used to update the parameters online to adapt to the scenario of incremental update of underground pipeline data, ensuring continuous optimization of the model and solving the problems of lagging model update and overfitting in existing methods. Description of the Drawings

[0017] Figure 1 is a schematic structural diagram of an active prevention and control management system for urban underground pipeline safety in an embodiment of the present invention; Figure 2This is a flow chart of an active prevention and control management method for urban underground pipeline safety in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used in limiting the scope of the present invention.In addition, should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the application equally.

[0019] In the present invention, terms such as "upper", "lower", "left", "right", "front", "back", "vertical", "horizontal", "side", "bottom", etc. indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are relational words determined only for the convenience of describing the structural relationships of the various parts or elements of the present invention, and do not specifically refer to any part or element in the present invention, and should not be understood as limiting the present invention.

[0020] Example: like Figure 1 As shown, this embodiment provides an urban underground pipeline safety active prevention and control management system, including: Data acquisition module: real-time acquisition of multi-source sensor data of urban underground pipelines, including pressure, temperature, flow, vibration signals and pipeline material information; Data preprocessing module: performs noise filtering and normalization processing on the multi-source sensor data, and extracts time domain and frequency domain features; Deep learning model module: It consists of a time series feature extraction network and an anomaly detection network. The time series feature extraction network uses a combination of LSTM network and attention mechanism to mine the time series correlation features of pipeline data. The anomaly detection network adopts a convolutional neural network (CNN) to fuse pipeline spatial topology features; Risk assessment module: Calculates pipeline health scores based on the output of the deep learning model module and generates three risk levels: safe, early warning, and high-risk based on preset thresholds; Active prevention and control module: triggers early warning signals based on risk levels and links pressure regulating valves, flow pump stations and other equipment to perform dynamic control; Feedback optimization module: Updates model parameters through online learning mechanism to continuously optimize risk prediction accuracy.

[0021] Among them, the normalization processing of the data preprocessing module follows the following method: ; in, is the mean of the sensor data, is the standard deviation, to adapt to the input requirements of the deep learning model; The time series feature extraction network performs the following calculation process: Input time series data sequence , and the hidden state is output through the LSTM network , where ; Weight the historical hidden states through the attention weights to obtain the context vector , where is the weight matrix, is the bias term; Generate the final time series features through linear transformation and ReLU activation function: , where is the transformation matrix, is the bias term; The PID controller of the active prevention and control module described above is an incremental PID controller, and its control formula is: ; where, is the control quantity at the th moment, is the flow difference at the th moment ; The PID parameters are fixed as: , , (optimal control parameters verified by simulation experiments); The feedback optimization module updates the model parameters using the stochastic gradient descent algorithm with momentum, and the specific formula is: ; where, is the momentum term at the th step, which is used to accelerate the convergence of the model, and the momentum coefficient ; is the gradient of the loss function at , is the learning rate; In this embodiment, the vibration sensor of the data acquisition module is a fiber Bragg grating vibration sensor, which has the characteristics of anti-electromagnetic interference and adaptability to complex underground environments; the pipeline material information is obtained by non-contact reading of RFID tags, supporting fast and batch material identification.

[0022] As Figure 2 shown, this embodiment also provides a method for actively preventing and controlling the safety of urban underground pipelines applied to the above urban underground pipeline safety active prevention and control management system, including the following steps: Step S1: Real-time collect the pipeline operation data, and fuse the pipeline GIS geographic information data to construct a spatio-temporal correlation data matrix , where is the number of pipeline nodes, is the time step, is the number of sensor channels; Step S2: Input the spatio-temporal correlation data matrix constructed in Step S1 into the time series feature extraction network to obtain time series features , where is the time series feature dimension; Extract the spatial topology features through the CNN of the anomaly detection network , where is the number of convolutional kernels; Adopt the feature splicing operation to fuse the spatio-temporal features to obtain the joint features , where represents feature splicing; Step S3: Input the joint feature in Step S2 into the fully connected layer, and calculate the health score through the Sigmoid function , where is the weight of the fully connected layer, is the bias term; Divide the risk level according to the health score: Safe state: ; Warning state: ; High-risk state: ; Step S4: If it is judged to be in a high-risk state, generate a flow regulation instruction , where is the maximum safe flow of the pipeline, is the regulation coefficient; Through the PID controller, adjust the opening of the pressure valve so that the actual flow satisfies ; Step S5: Online optimization of the model, specifically: Calculate the loss function based on the new data: ; Among them, is the binary cross-entropy, is the true label, , are the weight coefficients, are the model parameters; Adopt stochastic gradient descent to update the model parameters, and the learning rate .

[0023] Among them, the structural parameters of the NN network include: Convolutional layer: There are 3 layers, with kernel sizes of 3*3, 5*5, and 3*3 in sequence, and the number of channels is 32, 64, and 128 in sequence; Pooling layer: Max pooling is adopted, and the window size is 2*2; Activation function: LeakyReLU is adopted, and the negative slope coefficient is 0.01; The threshold division of the risk level is specifically as follows: Initial threshold Determined by statistical analysis of historical accident data; The threshold is dynamically adjusted according to seasonal factors, and the seasonal factors include temperature changes and pipeline load fluctuations; Regulation coefficient Dynamically adjusted according to the pipeline material, specifically: When the pipeline material is cast iron, ; When the pipeline material is steel pipe, 。

[0024] In this embodiment, taking the safety prevention and control of the winter underground water supply pipeline in City A as an example, it is set that the main water supply pipeline is made of cast iron, with a pipe diameter of DN800, and the historical freeze-thaw accidents are most frequent from December to February of the following year.

[0025] First, configure the sensors: Pressure sensor (range 0-2.5MPa, accuracy ±0.5%), vibration sensor (sampling rate 1kHz), temperature sensor (buried depth 1.5m, monitoring soil temperature), and perform data preprocessing: # Example of normalization (pressure data sequence): raw_data = [1.8, 1.9, 2.1, 1.7] # Unit: MPa, μ = np.mean(raw_data) # Calculation result: 1.875, σ = np.std(raw_data) # Calculation result: 0.158, x_norm = [(x - μ) / σ for x in raw_data] # Result: [-0.47, 0.16, 1.42, -1.11].

[0026] Secondly, perform spatio-temporal feature extraction: The model input is: (100 nodes × 60 minutes × 5 types of sensors); LSTM-attention layer: # LSTM outputs the hidden state h_t (dimension 128): lstm_layer = LSTM(units=128, return_sequences=True), h_t = lstm_layer(D), # Attention weight calculation: attention = Dense(1, activation='tanh')(h_t), alpha_t = Softmax(axis=1)(attention) # Example of weight distribution: [0.02, 0.11, 0.63,0.24], c = tf.reduce_sum(alpha_t * h_t, axis=1) # Context vector; CNN spatial feature extraction: Convolution kernel parameters: 3×3 (32 channels) → 5×5 (64 channels) → 3×3 (128 channels), Output spatial features 。

[0027] Next, perform dynamic threshold risk determination: Dynamic calculation of the threshold: The current date is set to December 15, 2023, environmental parameters: (Historical average in December in City A), parameter value (Cast iron pipe), (Accident rate in winter in the past 3 years is 0.18%): ; Due to low temperature, the threshold floats up to 2.45 (original threshold 0.7), greatly improving the alarm sensitivity; Health score and alarm: Model output (Due to detected abnormal vibration + sudden drop in pressure); Risk determination: → High-risk status (red alarm).

[0028] When performing dynamic threshold risk determination, it involves the generation of flow regulation instructions, specifically: ; ( , ); PID pressure regulation is: % PID controller parameters (Kp = 1.2, Ki = 0.01, Kd = 0.1), target_pressure = 1.2; % Target pressure value (MPa), current_pressure = 2.1; % Current pressure value (MPa), error = target_pressure - current_pressure; valve_opening=Kp*error+Ki*integral(error) + Kd*derivative(error); The calculation of the loss function is specifically as follows: ; The parameter update is expressed as: optimizer = SGD(learning_rate=0.001), optimizer.minimize(loss) # Update weights by gradient descent.

[0029] The above has specifically described the preferred embodiments of the present invention. However, the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of this application.

Claims

1. An active prevention and control management method for the safety of urban underground pipelines, characterized in that, It includes the following steps: Step S1: Collect pipeline operation data in real time and fuse pipeline GIS geographic information data to construct a spatio-temporal correlation data matrix , where is the number of pipeline nodes, is the time step, is the number of sensor channels; Step S2: Input the spatio-temporal correlation data matrix constructed in Step S1 into the temporal feature extraction network to obtain temporal features , where is the dimension of the temporal features; Extract spatial topological features through CNN of the anomaly detection network , where is the number of convolutional kernels; Fusing spatio-temporal features by using a feature splicing operation to obtain a joint feature , where represents feature splicing; Step S3: Input the combined features in Step S2 into the fully-connected layer, and calculate the health score through the Sigmoid function , where is the weight of the fully-connected layer, is the bias term; divide the risk level according to the health score: Safe state: ; Warning status: ; High-risk state: ; Step S4: If it is determined to be in a high-risk state, generate a flow regulation instruction , where is the maximum safe flow rate of the pipeline, is the regulation coefficient; through the PID controller, adjust the opening of the pressure valve so that the actual flow rate satisfies .

2. The active prevention and control management method for urban underground pipeline safety according to claim 1, wherein, In the step S2, the structural parameters of the CNN network include: Convolutional layer: There are 3 layers, the kernel sizes are 3*3, 5*5, 3*3 in sequence, and the number of channels are 32, 64, 128 in sequence; Pooling layer: Max pooling is adopted, and the window size is 2*2; Activation function: LeakyReLU is adopted, and the negative slope coefficient is 0.

01.

3. The active prevention and control management method for urban underground pipeline safety according to claim 1, characterized in that In the step S3, the threshold division of the risk level is specifically: Initial threshold Determined by statistical analysis of historical accident data; The threshold is dynamically adjusted according to seasonal factors, and the seasonal factors include temperature changes and pipeline load fluctuations.

4. A method for actively preventing and controlling the safety of urban underground pipelines according to claim 1, characterized in that, In the step S4, the regulation coefficient is dynamically adjusted according to the pipeline material, specifically as follows: When the pipeline material is cast iron, ; When the pipeline material is steel pipe, .

5. The active prevention and control management method for urban underground pipeline safety according to claim 1, characterized in that It also includes: Step S5: Online optimization of the model, specifically: Calculating the loss function based on the new data: ； Among them, is the binary cross-entropy, is the true label, , are the weight coefficients, are the model parameters; Update the model parameters using stochastic gradient descent, with the learning rate .

6. An active prevention and control management system for urban underground pipeline safety, based on an active prevention and control management method for urban underground pipeline safety as described in claim 1, characterized in that, Including: Data acquisition module: Real-time acquisition of multi-source sensing data of urban underground pipelines, and the multi-source sensing data includes pressure, temperature, flow rate, vibration signals and pipeline material information; Data preprocessing module: Performing noise filtering, normalization processing on the multi-source sensing data, and extracting time-domain and frequency-domain features; Deep learning model module: It consists of a time-series feature extraction network and an anomaly detection network. Among them: the time-series feature extraction network adopts a combined architecture of an LSTM network and an attention mechanism, and is used to mine the time-series correlation features of pipeline data; The anomaly detection network adopts a convolutional neural network (CNN) and is used to fuse pipeline spatial topology features; Risk assessment module: Calculating the pipeline health score based on the output of the deep learning model module, and generating three-level risk levels of safety, warning, and high risk in combination with the preset threshold; Active prevention and control module: Triggering a warning signal according to the risk level, and linking with a pressure regulating valve and a flow pumping station to perform dynamic regulation; Feedback optimization module: Updating the model parameters through an online learning mechanism to continuously optimize the risk prediction accuracy.

7. The active prevention and control management system for urban underground pipeline safety according to claim 6, characterized in that, The normalization processing of the data preprocessing module follows the following method: ； Among them, is the mean value of the sensor data, is the standard deviation, .

8. The active prevention and control management system for urban underground pipeline safety according to claim 6, characterized in that The time-series feature extraction network performs the following calculation process: Input time series data sequence , the hidden state is output by the LSTM network , where ; Through attention weights weight the historical hidden state to obtain a context vector , where is the weight matrix, is the bias term; Generate the final temporal features through linear transformation and ReLU activation function: , where is the transformation matrix, is the bias term.

9. The active prevention and control management system for urban underground pipeline safety according to claim 6, characterized in that The PID controller of the active prevention and control module is an incremental PID controller, and its control formula is: ； Among them, is the control quantity at the moment, is the flow difference at the moment; The PID parameters are fixed as: , , .

10. The active prevention and control management system for urban underground pipeline safety according to claim 6, characterized in that, The feedback optimization module updates the model parameters by using the stochastic gradient descent algorithm with momentum, and the specific formula is: ； Among them, is the momentum term of the step, which is used to accelerate the convergence of the model, and the momentum coefficient ; is the gradient of the loss function at , and is the learning rate.

Images