Intelligent water conservancy online monitoring system

Abstract

The invention relates to the technical field of water conservancy, and discloses an intelligent water conservancy online monitoring system, which comprises a data acquisition and preprocessing module for acquiring original data from a sensor, performing preprocessing operation, performing feature extraction and classification on the original data, and calculating an approximate leakage position by using time difference of arrival and a multi-sensor array technology; and the digital twin model building module is used for building a high-fidelity virtual model and simulating the running state of an actual pipeline system. The voiceprint recognition technology and the DRL algorithm are integrated, the response time is shortened, the positioning precision is improved, the system can dynamically adjust the state space and the action space according to the current water flow condition by calculating the instantaneous change rate of the water flow speed, the pressure and the acoustic signal and the average value of the sliding window through the self-adaptive regulation and control strategy, and the positioning accuracy is improved. The leakage point is accurately positioned, the safety risk caused by leakage is reduced, and the water resource management efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of water conservancy, and more specifically, to an intelligent water conservancy online monitoring system. Background Art

[0002] In water conservancy projects, pipeline leakage is one of the common safety hazards. Traditional detection methods mainly rely on manual inspection or simple sensor monitoring, which are often slow to respond, low in accuracy and high in cost.

[0003] Traditional methods are easily affected by environmental noise, leading to misjudgment. For long-distance pipeline networks, it is difficult to accurately locate leakage points, resulting in safety risks caused by leakage and affecting the efficiency of water resource management. Summary of the invention

[0004] The present invention provides a smart water conservancy online monitoring system to solve the technical problems in related technologies.

[0005] The present invention provides a smart water conservancy online monitoring system, comprising:

[0006] Data acquisition and preprocessing module: obtains raw data from sensors and performs preprocessing operations, extracts features and classifies the raw data, and calculates the approximate location of the leak using arrival time difference and multi-sensor array technology;

[0007] Digital twin model building module: build a high-fidelity virtual model to simulate the operating status of the actual pipeline system;

[0008] Simulation environment building module: Based on the digital twin model, a simulation environment is built to train and test the DRL agent in the simulation environment;

[0009] Agent training module: Train the DRL agent in a simulation environment so that it can make optimal decisions under different working conditions;

[0010] Adaptive control strategy module: dynamically adjusts the state space and action space of the DRL agent according to the current water flow conditions, and introduces an adaptive parameter adjustment mechanism;

[0011] Multi-round verification and precise positioning module: After multiple rounds of verification, the leak location is gradually narrowed down and the most accurate leak location is finally determined;

[0012] Result output and display module: outputs the final leak location and other related parameters, and displays them to the user in a visual way.

[0013] Furthermore, the following steps are performed in the data acquisition and preprocessing module:

[0014] S100, data collection and preprocessing: using acoustic sensors, water flow sensors and pressure sensors to collect sound, water flow and pressure data in the pipeline;

[0015] S200, feature extraction and classification: extract features from the collected sound signals and classify them using a trained machine learning model to determine whether there is a leak;

[0016] S300, preliminary positioning: using arrival time difference and multi-sensor array technology to calculate the approximate location of the leak.

[0017] Furthermore, step S300 also includes the following:

[0018] Assume that acoustic sensors, each of which records the arrival time of the leak sound signal ,in ;

[0019] For each pair of sensors , calculate the arrival time difference between them :

[0020] ;

[0021] in and are the arrival times recorded by the i-th and j-th sensors, respectively. is the arrival time difference between the ith and jth sensors;

[0022] Assume that the speed of sound wave propagation in the pipe is , based on the arrival time difference and the speed of sound, calculate the distance difference between each pair of sensors :

[0023] ;

[0024] in is the distance difference between the i-th and j-th sensors;

[0025] Assume the leak point is The positions of the sensors are , then construct the following system of equations:

[0026] ;

[0027] ;

[0028] ;

[0029] Write the above system of equations in matrix form:

[0030] ;

[0031] in is the coefficient matrix consisting of sensor positions and distance differences, is the position vector of the leak point, is a constant vector consisting of distance differences;

[0032] Solve the system of equations using the method of least squares or other optimization algorithms to get a preliminary estimate of the leak location:

[0033] ;

[0034] Assume that the sensor array is a uniform linear array consisting of sensors with a spacing of ;

[0035] The array response model is expressed as:

[0036] ;

[0037] in is the estimated azimuth angle of the leakage point relative to the array center, is the wavelength of the sound wave, Indicates the array response result;

[0038] Calculate the covariance matrix of the received signal :

[0039] ;

[0040] in is the received signal vector, represents the expectation operator, yes The conjugate transpose of ;

[0041] Use the MUSIC (Multiple Signal Classification) algorithm to perform spectrum estimation and find the direction angle estimate of the leakage point :

[0042] ;

[0043] in is the eigenvector matrix of the noise subspace, and the maximum value of the spectral estimate is selected. As the estimated value of the direction angle of the leak point , combined with the estimated direction angle And the known distance information can be used to further estimate the actual location of the leak point.

[0044] Furthermore, assuming that the state variables of the physical system are , then the state variables of the digital twin model are expressed as The goal of the digital twin model is to As close as possible ;

[0045] Perform the following steps in the Digital Twin Builder module:

[0046] Physical model building: Building a mathematical model of the system based on physical laws;

[0047] For water flow speed and pressure , using the following equation:

[0048] ;

[0049] in is the density of water, is the dynamic viscosity of water, is the gradient operator, is the Laplace operator, represents the partial derivative of velocity with respect to time, is the convection term, It is a pressure field. is the water velocity vector, It’s time;

[0050] Parameter calibration: Use historical data to calibrate the model to ensure that the output of the model is consistent with the output of the actual system;

[0051] The calibration objective function is expressed as:

[0052] ;

[0053] in is the sample size, is the actual system state, is the state of the digital twin model, are model parameters, represents the calibration objective function;

[0054] Real-time update: During operation, the status of the digital twin model is updated in real time based on the latest sensor data;

[0055] ;

[0056] in Indicates the current time The actual measurement data vector obtained from the sensor, represents the updated state vector of the digital twin model, Represents an update function.

[0057] Furthermore, assuming that the state of the simulation environment is , the action space is , the reward function is ;

[0058] Perform the following steps in the simulation environment building module:

[0059] State initialization: Initialize the state of the simulation environment:

[0060] ;

[0061] Action execution: The agent performs actions at each time step. Execute an action , and observe the response of the simulation environment:

[0062] ;

[0063] Reward calculation: Calculate the reward based on the current state and action:

[0064] ;

[0065] in , and They are rewards for reducing leakage risks, optimizing resource utilization, and maintaining system stability.

[0066] Furthermore, assuming that the policy network parameters of the agent are , then the policy function is expressed as:

[0067] ;

[0068] Perform the following steps in the Agent Training module:

[0069] Experience replay: store the experience of each step in an experience pool middle:

[0070] ;

[0071] Gradient update: randomly extract a batch of experience from the experience pool, calculate the loss function of the policy network, and update the parameters:

[0072] ;

[0073] in is the discount factor, is the Q-value function, are the parameters of the target network;

[0074] Parameter update:

[0075] Update the parameters of the policy network using gradient descent:

[0076] ;

[0077] in is the learning rate.

[0078] Furthermore, the following steps are performed in the adaptive control strategy module:

[0079] Capture rapid changes in water flow conditions and calculate the instantaneous rate of change of water velocity and pressure;

[0080] Use sliding window averages to calculate the mean of water velocity and pressure;

[0081] Analyze the collected water velocity and pressure data to identify abnormal conditions and assess their impact on the pipeline, use statistical methods to detect abnormal values ​​of water velocity and pressure, and use linear regression or other trend analysis methods to assess the long-term trend of water velocity and pressure;

[0082] The state space and action space of the DRL agent are dynamically adjusted according to the current water flow conditions, so that it can make optimal decisions under different working conditions.

[0083] Furthermore, the following steps are performed in the multi-round verification and precise positioning module:

[0084] An experience replay mechanism is introduced to store historical trajectories in an experience pool and randomly extract samples from it for training:

[0085] ;

[0086] in It's the experience pool. It is an experience sample;

[0087] Based on the preliminary positioning results, gradually adjust the gate opening of the relevant area, observe the changes in water flow, and conduct multiple rounds of verification until the most accurate leakage location is found;

[0088] Assume that the initial positioning result is , the final precise positioning is achieved through multiple rounds of verification:

[0089] ;

[0090] in is the final leak location. is the quality function of the prediction, is the actual mass function, is the total number of verifications.

[0091] Furthermore, the calculation formula for the leakage position finally determined is as follows:

[0092] ;

[0093] in Indicates the leak location, Indicates the leakage amount, Indicates the pipe diameter, Indicates the water flow velocity, Indicates the pressure difference, Indicates the pipe material. Indicates the pipe thickness;

[0094] ;

[0095] in Represents the transpose of a three-dimensional coordinate vector;

[0096] ;

[0097] in is the volume change of the leak, is the time interval;

[0098] ;

[0099] in is the cross-sectional area of ​​the pipe;

[0100] ;

[0101] in is the water flow rate;

[0102] ;

[0103] in and are the pressures upstream and downstream of the leak, respectively.

[0104] The present invention also proposes a storage medium storing non-temporary computer-readable instructions for executing steps corresponding to one or more modules in the aforementioned smart water conservancy online monitoring system.

[0105] The beneficial effects of the present invention are:

[0106] The present invention integrates voiceprint recognition technology and DRL algorithm, shortens the response time and improves positioning accuracy. By calculating the instantaneous change rate of water flow velocity, pressure and acoustic signals and the sliding window average value, the adaptive control strategy enables the system to dynamically adjust the state space and action space according to the current water flow conditions. The DRL intelligent agent can make intelligent decisions in complex and changeable environments through the adaptive parameter adjustment mechanism, accurately locate the leakage point, reduce the safety risks caused by leakage, and improve the efficiency of water resource management. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 It is a structural block diagram of a smart water conservancy online monitoring system proposed by the present invention;

[0108] Figure 2 It is a flow chart of a smart water conservancy online monitoring method proposed by the present invention.

[0109] In the figure: 101, data acquisition and preprocessing module; 102, digital twin model construction module; 103, simulation environment construction module; 104, intelligent agent training module; 105, adaptive control strategy module; 106, multi-round verification and precise positioning module; 107, result output and display module. DETAILED DESCRIPTION

[0110] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0111] like Figure 2 As shown, a smart water conservancy online monitoring method includes the following steps:

[0112] S100, data collection and preprocessing: using acoustic sensors, water flow sensors and pressure sensors to collect sound, water flow and pressure data in the pipeline;

[0113] S110, data collection: using a variety of sensors (such as acoustic sensors, water flow sensors, pressure sensors) to collect data such as sound, water flow velocity and pressure in the pipeline;

[0114] Acoustic sensor data: collects sound signals in the pipeline;

[0115] Water flow sensor data: measures the speed and flow of water;

[0116] Pressure sensor data: measures the pressure inside the pipe;

[0117] In one embodiment of the present invention, the calculation formula for data collection is as follows:

[0118] Assume that at time At that time, The data collected by the sensor is , then all sensor data vectors are expressed as:

[0119] ;

[0120] in It's time All sensor data vectors at time Indicates Sensors at time The readings, is the number of sensors.

[0121] S120, data preprocessing: cleaning the collected raw data, removing noise and outliers, normalizing the data from different sensors to the same scale, synchronizing the data so that the data from different sensors are consistent in time, and dividing the continuous data stream into time windows of fixed length;

[0122] The calculation formula for preprocessing is as follows:

[0123] Denoising:

[0124] ;

[0125] in is the denoised sensor data, represents a low-pass filter function;

[0126] Outlier Detection:

[0127] ;

[0128] in is the Z-score value, and are the mean and standard deviation of the denoised data respectively. If | ∣>threshold, the data point is considered an outlier and needs to be removed;

[0129] Normalization:

[0130] ;

[0131] in is the normalized sensor data, and are the minimum and maximum values ​​of the denoised data respectively;

[0132] Time Synchronization:

[0133] ;

[0134] in is the sensor data after time synchronization processing, It is the time synchronization protocol function;

[0135] Segmentation:

[0136] ;

[0137] in It is The data of the time window, is the window length;

[0138] S200, feature extraction and classification: extract features (such as MFCC) from the collected sound signals and classify them using the trained machine learning model to determine whether there is leakage;

[0139] S210, feature extraction: extract features from acoustic sensor data using Mel Frequency Cepstral Coefficients (MFCC), perform statistical feature extraction on the data of water flow sensor and pressure sensor, and fuse feature vectors of different sensors to form a comprehensive feature vector;

[0140] S220, classification: using the trained machine learning model to classify the fused feature vector to determine whether there is leakage;

[0141] Training model: Use historical data to train classification models. Commonly used classification algorithms include support vector machine (SVM), convolutional neural network (CNN), etc.

[0142] Model prediction: Use the trained model to classify and predict new feature vectors;

[0143] The calculation formula for its classification is as follows:

[0144] Assumptions is a trained classification model, the classification result can be expressed as:

[0145] ;

[0146] in Represents the classification result, usually a binary label (0 means no leakage, 1 means leakage), Represents the fused feature vector;

[0147] S300, preliminary positioning: using time difference of arrival (TDOA) or multi-sensor array technology to calculate the approximate location of the leak;

[0148] In one embodiment of the present invention, it is assumed that acoustic sensors, each of which records the arrival time of the leak sound signal ,in ;

[0149] For each pair of sensors , calculate the arrival time difference between them :

[0150] ;

[0151] in and are the arrival times recorded by the i-th and j-th sensors, respectively. is the arrival time difference between the ith and jth sensors;

[0152] Assume that the speed of sound wave propagation in the pipe is , based on the arrival time difference and the speed of sound, calculate the distance difference between each pair of sensors :

[0153] ;

[0154] in is the distance difference between the i-th and j-th sensors;

[0155] Assume the leak point is The positions of the sensors are , then construct the following system of equations:

[0156] ;

[0157] ;

[0158] ;

[0159] Write the above system of equations in matrix form:

[0160] ;

[0161] in is the coefficient matrix consisting of sensor positions and distance differences, is the position vector of the leak point, is a constant vector consisting of distance differences.

[0162] Solve the system of equations using the method of least squares or other optimization algorithms to get a preliminary estimate of the leak location:

[0163] ;

[0164] Assume that the sensor array is a uniform linear array (ULA) consisting of sensors, with a spacing of ;

[0165] The array response model is expressed as:

[0166] ;

[0167] in is the estimated azimuth angle of the leakage point relative to the array center, is the wavelength of the sound wave;

[0168] Calculate the covariance matrix of the received signal :

[0169] ;

[0170] in is the received signal vector, represents the expectation operator, yes The conjugate transpose of ;

[0171] Use the MUSIC (Multiple Signal Classification) algorithm to perform spectrum estimation and find the direction angle estimate of the leakage point :

[0172] ;

[0173] in is the eigenvector matrix of the noise subspace, represents the spectral estimate;

[0174] Select the maximum value of the spectral estimate corresponding to As the estimated value of the direction angle of the leak point ;

[0175] Combining the direction angle estimates and known distance information (such as the distance between the sensor and the pipeline), to further estimate the actual location of the leak;

[0176] S400, simulation training and digital twin: Use digital twin technology to conduct simulation experiments, simulate the control effects under various working conditions, and train the intelligent agent in advance to ensure that it can still operate efficiently under dynamic water flow conditions;

[0177] S410, Build a digital twin model: Create a virtual, high-fidelity water conservancy project system model that can simulate the operating status and response of the actual system.

[0178] Assume that the state variables of the physical system are , then the state variables of the digital twin model can be expressed as .

[0179] The goal of the digital twin model is to As close as possible .

[0180] The process of building a digital twin model typically involves the following steps:

[0181] Physical model establishment:

[0182] Build a mathematical model of the system based on the laws of physics, such as the equations of fluid dynamics.

[0183] For example, for water velocity and pressure , using the Navier-Stokes equations:

[0184] ;

[0185] in is the density of water, is the dynamic viscosity of water, is the gradient operator, is the Laplace operator, represents the partial derivative of velocity with respect to time, is the convection term, It is a pressure field. is the water velocity vector, It's time.

[0186] Parameter calibration: Use historical data to calibrate the mathematical model to ensure that the output of the mathematical model is consistent with the output of the actual system.

[0187] The calibration objective function can be expressed as:

[0188] ;

[0189] in is the sample size, is the actual system state, is the state of the digital twin model, are model parameters, represents the calibration objective function.

[0190] Real-time update: During operation, the status of the digital twin model is updated in real time based on the latest sensor data.

[0191] ;

[0192] in Indicates the current time The actual measurement data vector obtained from the sensor, represents the updated state vector of the digital twin model, represents the update function, using Kalman filter.

[0193] S420, simulation environment construction: Based on the digital twin model, a simulation environment is built so that the intelligent agent can be trained and tested in the simulation environment.

[0194] Assume that the state of the simulation environment is , the action space is , the reward function is .

[0195] State initialization: Initialize the state of the simulation environment:

[0196] ;

[0197] Action execution: The agent performs an action at each time step t , and observe the response of the simulation environment:

[0198] ;

[0199] Reward calculation: Calculate the reward based on the current state and action:

[0200] ;

[0201] in , and Rewards for reducing leakage risks, optimizing resource utilization, and maintaining system stability;

[0202] S430, Agent Training: Train the agent in a simulation environment so that it can make optimal decisions under different working conditions.

[0203] Assume that the policy network parameters of the agent are , then the policy function can be expressed as:

[0204] ;

[0205] Experience replay: store the experience of each step in an experience pool middle:

[0206] ;

[0207] in is the current state, is the action performed, It is the reward obtained. is the next state, where the experience pool Contains status ,action ,award and the next state ;

[0208] Gradient update: randomly extract a batch of experience from the experience pool, calculate the loss function of the policy network, and update the parameters:

[0209] ;

[0210] in is the discount factor, is the Q-value function, are the parameters of the target network, Express expectations, Indicates the current state. Indicates the current action. Indicates immediate reward, Indicates the next state, Indicates the next action. It is used to estimate the next action to take in the next state. long-term returns.

[0211] Parameter update:

[0212] Update the parameters of the policy network using gradient descent:

[0213] ;

[0214] in is the learning rate, is the loss function About parameters The gradient of is the policy network parameter at time t, is the policy network parameter at time t+1, represents the loss function, Indicates the parameters Find the partial derivative.

[0215] S500, Dynamic Water Flow Condition Monitoring and Analysis: Real-time monitoring of water flow and pressure changes, using adaptive algorithms to adjust the state space and action space of the DRL agent, and introducing an adaptive parameter adjustment mechanism in the DRL agent to enable it to dynamically adjust its strategy according to the current water flow conditions;

[0216] In one embodiment of the present invention, the specific steps are as follows:

[0217] S510 captures rapid changes in water flow conditions and calculates the instantaneous rate of change (i.e. derivative) of water flow velocity and pressure:

[0218] ;

[0219] ;

[0220] in is the real-time water flow velocity, is the real-time pressure value, is the instantaneous rate of change of water velocity, is the instantaneous rate of change of pressure;

[0221] S520, using the sliding window average to calculate the mean of water flow velocity and pressure:

[0222] ;

[0223] ;

[0224] in is the time length of the sliding window, and are the sliding window averages of water velocity and pressure, respectively;

[0225] S530, analyzing the collected water flow velocity and pressure data, identifying abnormal conditions and evaluating their impact on the pipeline, using statistical methods (such as Z-score or IQR) to detect abnormal values ​​of water flow velocity and pressure, and using linear regression or other trend analysis methods to evaluate the long-term trend of water flow velocity and pressure;

[0226] S540, dynamically adjusts the state space and action space of the DRL agent according to the current water flow conditions, so that it can make optimal decisions under different working conditions;

[0227] Specifically, the water flow velocity, pressure and their rate of change are used as state variables to update the state space of the DRL agent:

[0228] ;

[0229] in is the state at time t;

[0230] have There are adjustable gates, and the opening change of each gate is , then the action vector It can be expressed as:

[0231] ;

[0232] in It is The change in the opening of each gate;

[0233] Dynamically adjust the DRL agent’s action space based on changing water flow conditions; for example, when water flow speed increases, gate openings may need to be adjusted more frequently:

[0234] ;

[0235] in It's in time The actions taken when is the policy function (defined by the DRL agent), is an adaptive parameter;

[0236] An adaptive parameter adjustment mechanism is introduced to enable the DRL agent to dynamically adjust its strategy according to the current water flow conditions:

[0237] ;

[0238] in is the learning rate, Is a performance indicator About parameters The gradient of

[0239] Performance Indicators Usually defined as cumulative rewards:

[0240] ;

[0241] in is a trajectory, i.e. a series of states and actions from the initial state to the final state, is a discount factor that balances the importance of immediate rewards and future rewards, is the length of the trajectory.

[0242] S600, multiple rounds of verification and precise positioning: Collect new water flow data and compare it with the prediction model, and perform multiple rounds of verification until the most accurate leak location is found.

[0243] Introduce the experience replay mechanism, which stores historical trajectories in an experience pool (Replay Buffer) and randomly extracts samples from it for training:

[0244] ;

[0245] in It's the experience pool. It is an experience sample;

[0246] Based on the preliminary positioning results, the gate openings in the relevant areas are gradually adjusted, the water flow changes are observed, and multiple rounds of verification are carried out until the most precise leakage location is found.

[0247] Assume that the initial positioning result is , the final precise positioning can be achieved through multiple rounds of verification:

[0248] ;

[0249] in is the final leak location. is the quality function of the prediction, is the actual mass function, is the total number of verifications, is a positional parameter;

[0250] In one embodiment of the present invention, the calculation formula for the leakage position finally determined is as follows:

[0251] ;

[0252] Leak location ( ): The most accurate position estimate obtained after multiple rounds of verification.

[0253] Expressed as a three-dimensional coordinate vector:

[0254] ;

[0255] Leakage volume ( ): describes the flow or volume rate of a leak.

[0256] The formula means:

[0257] ;

[0258] in is the volume change of the leak, is the time interval.

[0259] Pipe diameter ( ): The inner diameter of the pipe, which affects the flow characteristics of the fluid.

[0260] The formula means:

[0261] ;

[0262] in is the cross-sectional area of ​​the pipe.

[0263] Water flow rate ( ): The water flow velocity in the pipe affects the sensitivity of leak detection.

[0264] The formula means:

[0265] ;

[0266] in is the water flow rate, is the cross-sectional area of ​​the pipe.

[0267] Pressure difference ( ): The change in pressure before and after the leak point is used to assist in positioning.

[0268] The formula means: ;

[0269] in and are the pressures upstream and downstream of the leak, respectively.

[0270] Pipe material and thickness ( / ): The material (such as PVC, steel pipe, etc.) and wall thickness of the pipe affect the development and spread of the leak.

[0271] In one embodiment of the present invention, the final positioning result and analysis report are sent to relevant personnel, and an early warning system is triggered to notify management personnel to take corresponding measures.

[0272] like Figure 1 As shown, at least one embodiment disclosed in the present invention provides a smart water conservancy online monitoring system, including:

[0273] Data acquisition and preprocessing module: obtains raw data from sensors and performs preprocessing operations, extracts features and classifies the raw data, and calculates the approximate location of the leak using arrival time difference and multi-sensor array technology;

[0274] Digital twin model building module: build a high-fidelity virtual model to simulate the operating status of the actual pipeline system;

[0275] Simulation environment building module: Based on the digital twin model, a simulation environment is built to train and test the DRL agent in the simulation environment;

[0276] Agent training module: Train the DRL agent in a simulation environment so that it can make optimal decisions under different working conditions;

[0277] Adaptive control strategy module: dynamically adjusts the state space and action space of the DRL agent according to the current water flow conditions, and introduces an adaptive parameter adjustment mechanism;

[0278] Multi-round verification and precise positioning module: After multiple rounds of verification, the leak location is gradually narrowed down and the most accurate leak location is finally determined;

[0279] Result output and display module: outputs the final leak location and other related parameters, and displays them to the user in a visual way.

[0280] At least one embodiment disclosed in the present invention provides a storage medium storing non-temporary computer-readable instructions for executing steps corresponding to one or more modules in the aforementioned smart water conservancy online monitoring system.

[0281] The computer program may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims should not be construed as limiting the scope.

[0282] Example 1: Pipeline Leakage Detection in Urban Water Supply System

[0283] Background: A city water supply system consists of multiple branch pipes covering a wide area. Traditional methods rely on manual inspections and simple sensor monitoring, which have slow response speed and low accuracy, making it difficult to detect and deal with leakage problems in a timely manner.

[0284] The modules of this system perform the following:

[0285] Real-time data acquisition and preprocessing module:

[0286] Install high-precision water flow sensors, pressure sensors and acoustic sensors at key nodes;

[0287] Transmit data to a central control system using high-speed communication protocols such as EtherCAT or Profinet;

[0288] Voiceprint recognition and feature extraction module:

[0289] Use voiceprint recognition technology to analyze acoustic signals and extract features such as frequency and amplitude;

[0290] Use machine learning models (such as support vector machines or convolutional neural networks) to classify the extracted features and distinguish between normal operation and leakage states;

[0291] Digital twin model building blocks:

[0292] Build a digital twin model of the water supply system to simulate the operating status of the actual pipeline system;

[0293] Calibrate model parameters based on historical data to ensure that model output is consistent with the actual system;

[0294] Simulation environment building module:

[0295] Build a simulation environment based on the digital twin model to train the DRL agent;

[0296] The agent performs actions in the simulation environment (such as adjusting the gate opening) and observes the system response to optimize the strategy;

[0297] Agent training and adaptive control module:

[0298] Use the experience replay mechanism to store the experience of each step, and randomly extract a batch of experience from the experience pool to update the strategy;

[0299] Introducing an adaptive parameter adjustment mechanism that enables the agent to dynamically adjust its strategy based on current water flow conditions;

[0300] Multi-round verification and precise positioning module:

[0301] After multiple rounds of verification, the leak location was gradually narrowed down and the most accurate leak location was finally determined;

[0302] Calculate the final leak location using the error minimization method .

[0303] The system can detect leaks and sound an alarm within a few minutes, and the location error of the leak point is controlled within ±1 meter, reducing the number of manual inspections and lowering maintenance costs.

[0304] Example 2: Leakage detection of diversion tunnels in large hydropower stations

[0305] Background: The water diversion tunnel of a large hydropower station is several thousand meters long. The traditional detection method relies on regular manual inspection, but the internal environment of the tunnel is complex, and manual inspection is difficult and inefficient.

[0306] The modules of this system perform the following:

[0307] Real-time data acquisition and preprocessing module:

[0308] Install water flow sensors, pressure sensors and acoustic sensors at key locations in the diversion tunnel;

[0309] Use wireless sensor networks (WSNs) to transmit data to a central control system;

[0310] Voiceprint recognition and feature extraction module:

[0311] Use voiceprint recognition technology to analyze acoustic signals, extract features and classify them;

[0312] Use deep learning models (such as long short-term memory networks (LSTMs)) to analyze time series data to improve the accuracy of leak detection;

[0313] Digital twin model building blocks:

[0314] Build a digital twin model of the water diversion tunnel to simulate the operating status of the actual system;

[0315] Calibrate model parameters based on historical data to ensure that model output is consistent with the actual system;

[0316] Simulation environment building module:

[0317] Build a simulation environment based on the digital twin model to train the DRL agent;

[0318] The agent performs actions in the simulation environment (such as adjusting the power of a pump) and observes the system response to optimize the strategy;

[0319] Agent training and adaptive control module:

[0320] Use a distributed reinforcement learning framework such as Ray RLlib to speed up the training process;

[0321] Introducing an adaptive parameter adjustment mechanism that enables the agent to dynamically adjust its strategy based on current water flow conditions;

[0322] Multi-round verification and precise positioning module:

[0323] After multiple rounds of verification, the leak location was gradually narrowed down and the most accurate leak location was finally determined;

[0324] Use methods such as Kalman filter to fuse sensor data to further improve positioning accuracy;

[0325] This system can remotely monitor the status of the water diversion tunnel, reducing the need for manual entry into dangerous areas. The position error of the leak point is controlled within ±0.5 meters, and the leak point is quickly located, which shortens the maintenance time and improves the operating efficiency of the power station.

[0326] Example 3: Pipeline Leakage Detection in Agricultural Irrigation Systems

[0327] Background: A large agricultural irrigation system covers a vast area of ​​farmland. Traditional methods rely on manual inspection and simple flow meter monitoring, but the irrigation pipes are widely distributed and manual inspection is inefficient.

[0328] The modules of this system perform the following:

[0329] Real-time data acquisition and preprocessing module:

[0330] Install water flow sensors, pressure sensors and acoustic sensors at key nodes of irrigation pipes;

[0331] Use low-power wide-area network technologies such as LoRaWAN to transmit data to a central control system;

[0332] Voiceprint recognition and feature extraction module:

[0333] Use voiceprint recognition technology to analyze acoustic signals, extract features and classify them;

[0334] Classify the extracted features using lightweight machine learning models such as random forests, suitable for resource-constrained edge devices;

[0335] Digital twin model building blocks:

[0336] Build a digital twin model of the irrigation system to simulate the operating status of the actual pipeline system;

[0337] Calibrate model parameters based on historical data to ensure that model output is consistent with the actual system;

[0338] Simulation environment building module:

[0339] Build a simulation environment based on the digital twin model to train the DRL agent;

[0340] The agent performs actions in the simulation environment (such as adjusting the valve opening) and observes the system response to optimize the strategy;

[0341] Agent training and adaptive control module:

[0342] Use reinforcement learning frameworks such as TensorFlow Agents for agent training;

[0343] Introducing an adaptive parameter adjustment mechanism that enables the agent to dynamically adjust its strategy based on current water flow conditions;

[0344] Multi-round verification and precise positioning module:

[0345] After multiple rounds of verification, the leak location was gradually narrowed down and the most accurate leak location was finally determined;

[0346] Use methods such as sliding window average to smooth sensor data and improve positioning accuracy;

[0347] This system can automatically monitor the status of irrigation pipes, reducing the need for manual inspections and high-precision positioning: the position error of the leak point is controlled within ±1 meter, and the leak point is quickly located, reducing water waste and improving irrigation efficiency.

[0348] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A smart water conservancy online monitoring system, characterized in that: include: Data acquisition and preprocessing module: obtains raw data from sensors and performs preprocessing operations, extracts features and classifies the raw data, and calculates the approximate location of the leak using arrival time difference and multi-sensor array technology; Digital twin model building module: build a high-fidelity virtual model to simulate the operating status of the actual pipeline system; Simulation environment building module: Based on the digital twin model, a simulation environment is built, and the DRL agent can be trained and tested in the simulation environment; Agent training module: Train the DRL agent in a simulation environment so that it can make optimal decisions under different working conditions; Adaptive control strategy module: dynamically adjusts the state space and action space of the DRL agent according to the current water flow conditions, and introduces an adaptive parameter adjustment mechanism; Multi-round verification and precise positioning module: After multiple rounds of verification, the leak location is gradually narrowed down and the most accurate leak location is finally determined; Result output and display module: outputs the final leak location and other related parameters, and displays them to the user in a visual way.

2. According to claim 1, a smart water conservancy online monitoring system is characterized in that: Perform the following steps in the data acquisition and preprocessing module: S100, data collection and preprocessing: using acoustic sensors, water flow sensors and pressure sensors to collect sound, water flow and pressure data in the pipeline; S200, feature extraction and classification: extract features from the collected sound signals and classify them using a trained machine learning model to determine whether there is a leak; S300, preliminary positioning: using arrival time difference and multi-sensor array technology to calculate the approximate location of the leak.

3. According to claim 2, a smart water conservancy online monitoring system is characterized in that: Step S300 also includes the following: have acoustic sensors, each of which records the arrival time of the leak sound signal ,in ; For each pair of sensors , calculate the arrival time difference between them ; Assume that the speed of sound wave propagation in the pipe is , based on the arrival time difference and the speed of sound, calculate the distance difference between each pair of sensors ; Assume the leak point is The positions of the sensors are , then construct the following system of equations: ； ； ； Write the above system of equations in matrix form: ； in is the coefficient matrix consisting of sensor positions and distance differences, is the position vector of the leak point, is a constant vector consisting of distance differences; Solve the system of equations using the least squares method or other optimization algorithms to obtain a preliminary location estimate of the leak; Assume that the sensor array is a uniform linear array consisting of sensors, with a spacing of ; The array response model is expressed as: ； in is the estimated azimuth angle of the leakage point relative to the array center, is the wavelength of the sound wave, Indicates the array response result; Calculate the covariance matrix of the received signal : ； in is the received signal vector, represents the expectation operator, yes The conjugate transpose of ; Use multiple signal classification algorithm to perform spectrum estimation and find the direction angle estimate of the leakage point .

4. According to claim 3, a smart water conservancy online monitoring system is characterized in that: Assume that the state variables of the physical system are , then the state variables of the digital twin model are expressed as The goal of the digital twin model is to As close as possible ; Perform the following steps in the Digital Twin Builder module: Physical model building: Building a mathematical model of the system based on physical laws; For water flow speed and pressure , the equation of the mathematical model is as follows: ； in is the density of water, is the dynamic viscosity of water, is the gradient operator, is the Laplace operator, represents the partial derivative of velocity with respect to time, is the convection term, It is a pressure field. is the water velocity vector, It’s time; Parameter calibration: Use historical data to calibrate the mathematical model to ensure that the output of the mathematical model is consistent with the output of the actual system; The calibration objective function is expressed as: ； in is the sample size, is the actual system state, is the state of the digital twin model, are model parameters, represents the calibration objective function; Real-time update: During operation, the status of the digital twin model is updated in real time based on the latest sensor data.

5. The smart water conservancy online monitoring system according to claim 4 is characterized in that: Assume that the state of the simulation environment is , the action space is , the reward function is ; Perform the following steps in the simulation environment building module: State initialization: Initialize the state of the simulation environment; Action execution: The agent performs actions at each time step. Execute an action , and observe the response of the simulation environment; Reward calculation: Calculate the reward based on the current state and action: ； in , and They are rewards for reducing leakage risks, optimizing resource utilization, and maintaining system stability.

6. The smart water conservancy online monitoring system according to claim 5 is characterized in that: Assume that the policy network parameters of the agent are , then the policy function is expressed as: ； in is the policy function, is the policy network parameter at the current moment; Perform the following steps in the Agent Training module: Experience replay: store the experience of each step in an experience pool middle: ； The experience pool Contains status ,action ,award and the next state ; Gradient update: randomly extract a batch of experience from the experience pool, calculate the loss function of the policy network, and update the parameters; Parameter update: Use gradient descent to update the parameters of the policy network: ； in is the learning rate, is the loss function About parameters The gradient of is the policy network parameter at time t, is the policy network parameter at time t+1, represents the loss function, Indicates the parameters Find the partial derivative.

7. The smart water conservancy online monitoring system according to claim 6 is characterized in that: The following steps are performed in the adaptive control strategy module: Capture rapid changes in water flow conditions and calculate the instantaneous rate of change of water velocity and pressure; Use sliding window averages to calculate the mean of water velocity and pressure; Analyze the collected water velocity and pressure data to identify abnormal conditions and assess their impact on the pipeline, use statistical methods to detect abnormal values ​​of water velocity and pressure, and use linear regression or other trend analysis methods to assess the long-term trend of water velocity and pressure; The state space and action space of the DRL agent are dynamically adjusted according to the current water flow conditions, so that it can make optimal decisions under different working conditions.

8. The smart water conservancy online monitoring system according to claim 7 is characterized in that: The following steps are performed in the multi-round verification and precise positioning module: Introducing an experience replay mechanism to store historical trajectories in an experience pool and randomly extract samples from it for training; Based on the preliminary positioning results, gradually adjust the gate opening of the relevant area, observe the changes in water flow, and conduct multiple rounds of verification until the most accurate leakage location is found; Assume that the initial positioning result is , the final precise positioning is achieved through multiple rounds of verification: ； in is the final leak location. is the quality function of the prediction, is the actual mass function, is the total number of verifications, is a positional parameter, Indicates action, is the state at time t.

9. The smart water conservancy online monitoring system according to claim 8 is characterized in that: The calculation formula for the final leakage position is as follows: ； in Indicates the leak location, Indicates the leakage amount, Indicates the pipe diameter, Indicates the water flow velocity, Indicates the pressure difference, Indicates the pipe material. Indicates the pipe thickness.

10. A storage medium, characterized in that: Non-temporary computer-readable instructions are stored for executing steps corresponding to one or more modules in a smart water conservancy online monitoring system as described in any one of claims 1-9.

Images