Secondary water supply full-chain digital management system

Abstract

This invention discloses a full-chain digital management system for secondary water supply, belonging to the field of water supply supervision technology. By constructing a non-contact multimodal sensing network, full-area sensing of the secondary water supply system can be achieved. The generated dataset can be directly used for subsequent pollution identification model training. By updating the digital twin model of the secondary water supply network, simulating theoretical water quality parameters, and comparing them with the sensing data to generate calibration coefficients, the sensing results can be corrected, effectively reducing water quality parameter errors and meeting the accuracy requirements of secondary water supply safety monitoring. Dynamic calibration reduces the number of on-site sensor calibrations, and based on the calibrated sensing data, the pollution identification accuracy of the twin model can be effectively improved. Through pollution trend prediction to lock in risks in advance, threshold triggering mechanisms to ensure precise intervention timing, and multi-measure adaptation to improve intervention efficiency, a complete closed loop of data monitoring, risk prediction, proactive intervention, and effect verification is ultimately formed.

Description

Technical Field

[0001] This invention relates to the field of water supply monitoring technology, specifically to a digital management system for the entire secondary water supply chain. Background Technology

[0002] Digital management of the entire secondary water supply chain is a modern management model that uses next-generation information technologies such as the Internet of Things, big data, and artificial intelligence to intelligently upgrade the entire process of secondary water supply systems, from planning and construction, operation monitoring, maintenance management to user services. Its core objective is to break down the information silos and manual labor-dependent extensive management model in traditional management, and achieve safe, efficient, energy-saving, and refined operation of the "last mile" of water supply.

[0003] In secondary water supply systems, residual chlorine sensors may misinterpret as "excessive residual chlorine" when rust or algae adhere to their surface. Turbidity sensors may continuously output false signals indicating "clear water" when their lenses are obscured by scale. Such issues are rare in industrial settings because industrial fluids are typically clean, at constant pressure, and flow at high speeds. Secondary water supply systems lack automatic cleaning and self-calibration mechanisms, resulting in both false alarms and missed alarms caused by pollution and scaling. Summary of the Invention

[0004] The purpose of this invention is to provide a digital management system for the entire secondary water supply chain, which can solve the technical problems mentioned in the background art.

[0005] The objective of this invention can be achieved through the following technical solutions: A fully digital management system for the secondary water supply chain, including: Non-contact multimodal sensing network construction module: Deploy a hyperspectral imaging module, an ultrasonic guided wave detection unit, and a micro-meteorological sensor. Based on the acquired and processed spectral feature matrix, ultrasonic attenuation coefficient sequence, and environmental parameter vector, generate a multimodal sensing dataset. Digital twin dynamic calibration module: Based on the generated multimodal sensing dataset, update the digital twin model of the secondary water supply network, calculate theoretical water quality parameters through LBM lattice Boltzmann fluid simulation, and compare them with the sensing data to generate calibration coefficients to correct the sensing results; Pollution prediction and proactive intervention optimization module: Based on the corrected water quality parameters, an input feature set is constructed and input into a pre-trained deep Q-network model to predict the probability of contamination on the sensor surface. Based on the analysis results of the sensor surface contamination probability, proactive intervention measures and closed-loop optimization are triggered.

[0006] Furthermore, signal transmission and reception are performed according to a preset cycle, the received echo signal is preprocessed, and the amplitude of the transmitted signal and the amplitude of the received signal are extracted based on the preprocessed echo signal. The temperature compensation coefficient is calculated by collecting water temperature data from the micro-meteorological sensors in the pipeline network. Based on the obtained transmitted signal amplitude, received signal amplitude, and temperature compensation coefficient, the attenuation coefficient is calculated; the calculated attenuation coefficients are arranged and combined in chronological order to obtain the ultrasonic attenuation coefficient sequence.

[0007] Furthermore, when updating the digital twin model of the secondary water supply network, the generated multimodal sensing data is converted into input parameters that the digital twin model can recognize, including water quality parameter mapping, pipeline status parameter mapping, and environmental parameter mapping.

[0008] Furthermore, when loading the digital twin model of the secondary water supply network, the network topology, material parameters, and boundary conditions are imported. The network topology includes pipe length, pipe diameter, and node location; the material parameters include pipe roughness and pipe wall friction coefficient; and the boundary conditions include inlet flow rate and outlet pressure. Based on the state parameters of the twin model, the core boundary conditions for LBM simulation are set.

[0009] Furthermore, when performing simulation calculations of theoretical water quality parameters, the particle distribution function is updated, and based on the flow field velocity distribution, the LBM passive scalar model is used to solve for residual chlorine decay and turbidity diffusion.

[0010] Furthermore, when simulating residual chlorine concentration, an evolution equation for the residual chlorine scalar distribution function is established, convection, diffusion, and attenuation processes are introduced, and the macroscopic residual chlorine concentration is extracted from the residual chlorine particle distribution function.

[0011] Furthermore, when performing turbidity diffusion simulation, a scalar evolution equation similar to that of residual chlorine is used, the diffusion coefficient is adjusted to the diffusion coefficient of turbidity particles, and the macroscopic turbidity value is extracted.

[0012] Furthermore, the least squares method is used to calculate the calibration coefficient between the perceived data and the theoretical data. The actual perceived water quality parameter matrix is ​​corrected using the calibration coefficient of the perceived data, and the corrected water quality parameters are fed back to the digital twin model to update the water quality distribution status within the model.

[0013] Furthermore, the corrected water quality parameters are obtained, and combined with historical sensor operating data, an input feature set is constructed and the data is normalized. This data is then input into a pre-trained deep Q-network model, and forward propagation is performed to calculate the Q-value.

[0014] Furthermore, the Q-value output by the model is converted into an interpretable pollution probability, and the obtained pollution probability is analyzed. Active intervention measures are dynamically triggered based on the analysis results. Several hours after the intervention is completed, the sensor status is verified and the model parameters are updated.

[0015] Compared to existing solutions, the beneficial effects achieved by this invention are: This invention constructs a non-contact multimodal sensing network, enabling full-area sensing of the secondary water supply system without modifying the secondary water supply pipeline structure. The generated dataset can be directly used for subsequent pollution identification model training. Compared with traditional single-sensor data, it can effectively improve the model's recognition accuracy and provide core data support for the digital management of the entire secondary water supply chain.

[0016] This invention updates the digital twin model of the secondary water supply network, simulates and calculates theoretical water quality parameters, and compares them with the sensing data to generate calibration coefficients, thereby correcting the sensing results. This can effectively reduce water quality parameter errors and meet the accuracy requirements for secondary water supply safety monitoring. Through dynamic calibration, the number of on-site sensor calibrations can be reduced. Based on the calibrated sensing data, the accuracy of pollution identification in the twin model can be effectively improved, enabling early warning of secondary water supply safety.

[0017] This invention identifies risks in advance through pollution trend prediction, ensures precise intervention timing through threshold triggering mechanisms, and improves intervention efficiency through the adaptation of multiple measures. Ultimately, it forms a complete closed loop of data monitoring, risk prediction, proactive intervention, and effect verification. It can intervene in sensor contamination in advance, avoid water quality exceedances caused by data distortion, and fundamentally solve the water quality monitoring problem caused by secondary water supply sensor contamination. It can effectively improve the reliability of secondary water supply monitoring and the water quality compliance rate. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the operation of the digital management system for the entire secondary water supply chain of this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, the present invention is a digital management system for the entire chain of secondary water supply, including a non-contact multimodal sensing network construction module, a digital twin dynamic calibration module, and a pollution prediction and proactive intervention optimization module; The non-contact multimodal sensing network construction module deploys a hyperspectral imaging module, an ultrasonic guided wave detection unit, and a micro-meteorological sensor. Based on the acquired and processed spectral feature matrix, ultrasonic attenuation coefficient sequence, and environmental parameter vector, a multimodal sensing dataset is generated. Specific steps include: When deploying hyperspectral imaging modules, ultrasonic guided wave detection units, and micro-meteorological sensors, hyperspectral imaging modules are deployed at key nodes such as the outlet of the secondary water supply tank and the user's inlet, using a push-broom imaging method. Ultrasonic guided wave detection units are attached to the outer wall every 100 meters along the main water supply pipeline. Piezoelectric ceramic transducers are used and are tightly bonded to the pipeline wall with a coupling agent. Micro-meteorological sensors are deployed on the top of the secondary water supply tank to collect ambient temperature, relative humidity, and atmospheric pressure. In addition, all devices are connected to the network through an industrial-grade LoRaWAN gateway with a communication distance of ≥5km and a packet loss rate of ≤0.1%, meeting the communication requirements of complex underground environments. Furthermore, the IEEE 1588 PTPv2 precise time protocol is used to achieve time synchronization of all sensing nodes, which is an existing conventional technical solution. The specific implementation steps will not be elaborated here. The hyperspectral imaging module starts acquisition through a preset triggering mechanism. Specifically, it is automatically triggered once every 10 minutes, or an emergency acquisition is triggered when the water level in the tank changes by ≥5%. Acquire a black field image, which is an image with the light source off, and subtract dark current noise from the raw data: ;in, To correct the data; This is the raw hyperspectral data; This is black field image data; The corrected data is centered by band so that the mean of each band is 0: ;in, For centralized data; This is the mean vector of pixel values ​​for each band; When extracting core spectral features using principal component analysis, the covariance matrix is ​​calculated from the centered data: ;in, Let covariance matrix be the variance matrix. H×W represents the total number of pixels in the image, specifically 768×1024=786432. Singular value decomposition (SVD) is performed on the covariance matrix to obtain eigenvalues. and eigenvectors This is an existing conventional technical solution, and the specific implementation steps will not be elaborated here; Calculate the cumulative contribution rate And select the top 256 principal components with a cumulative contribution rate ≥ 95%; where j is the principal component index; k is the number of principal components selected, with a value of 256; Projecting the centered data onto the principal component space yields the dimensionality-reduced spectral feature matrix: ;in, This is the final output spectral feature matrix; It is a matrix composed of the first 256 eigenvectors; The ultrasonic guided wave detection unit performs signal transmission and reception according to a preset cycle, and preprocesses the received echo signal. The preprocessing includes, but is not limited to, noise reduction and enhancement, which are existing conventional technical solutions. The specific implementation steps will not be elaborated here. Based on the preprocessed echo signal, the amplitude of the transmitted signal is extracted. and received signal amplitude ; The temperature compensation coefficient is calculated by collecting the water temperature T from the pipeline micro-meteorological sensor. : ;in, The reference water temperature is 20°C by default. Based on the obtained transmitted signal amplitude Received signal amplitude and temperature compensation coefficient Calculate the attenuation coefficient: ;in, The attenuation coefficient; To detect the length of the pipe; The calculated attenuation coefficients are arranged and combined in chronological order to obtain the ultrasonic attenuation coefficient sequence. ; The micro-meteorological sensor continuously collects environmental parameters. When the collected environmental parameters are filtered by a sliding window to eliminate random noise, the current collection time is used as the end point of the window, and nine historical data points are taken backward to form a window containing 10 data points. Calculate the arithmetic mean of the data within the window: ;in, is the filtered environmental parameter value at time t, representing the standardized environmental parameter after sliding window filtering at time t, which can correspond to any one of temperature, relative humidity, or atmospheric pressure; t is the current acquisition time; i is the summation index variable; The original sampled value at time i is the unfiltered raw environmental parameters collected by the micro-meteorological sensor at time i, which includes sensor electronic noise and random environmental interference. The filtered environmental parameters are output in a time-series structure to obtain an environmental parameter vector M, which includes timestamp, temperature value, humidity value, and atmospheric pressure value. The obtained spectral feature matrix will be processed. Ultrasonic attenuation coefficient sequence The environmental parameter vector M is used to achieve time alignment of multi-source data through high-precision time synchronization, ultimately generating a structured multimodal sensing dataset. ; Specifically, a hyperspectral imaging module at the outlet of the secondary water supply tank is selected as the master clock node, with its built-in GPS module providing a UTC time reference; the remaining sensing devices are used as slave clock nodes, connected to the master clock node via industrial Ethernet, and time synchronization is achieved using the IEEE 1588 PTPv2 precision time protocol.

[0022] It should be noted that by using layered deployment and differentiated preprocessing, full-area perception of the secondary water supply system can be achieved, with a data collection coverage of 100%, which can effectively improve the signal-to-noise ratio of the preprocessed data. By using the IEEE 1588 PTPv2 protocol to achieve time synchronization accuracy, the problem of feature mismatch caused by time misalignment of multi-source data can be solved, thus improving the data alignment accuracy.

[0023] In this embodiment of the invention, by constructing a non-contact multimodal sensing network, the entire secondary water supply system can be perceived without modifying the secondary water supply pipeline structure. The generated dataset can be directly used for subsequent pollution identification model training. Compared with traditional single-sensor data, it can effectively improve the model's recognition accuracy and provide core data support for the digital management of the entire secondary water supply chain.

[0024] Digital Twin Dynamic Calibration Module: Based on the generated multimodal sensing dataset, the digital twin model of the secondary water supply network is updated. Theoretical water quality parameters are calculated using LBM lattice Boltzmann fluid simulation, and the results are compared with the sensing data to generate calibration coefficients, thus correcting the sensing results. Specific steps include: When updating the digital twin model of the secondary water supply network, the generated multimodal sensing data is converted into input parameters that the digital twin model can recognize, including water quality parameter mapping, pipeline status parameter mapping and environmental parameter mapping; In the process of mapping water quality parameters, a pre-trained convolutional neural network model is used to map the spectral feature matrix. Converted to actual water quality parameters: ;in, This is a matrix of actually perceived water quality parameters, including residual chlorine concentration and turbidity, corresponding to the first and second rows, respectively. N is the number of data collection nodes; For a pre-trained CNN model, the input is a spectral feature matrix of 256×1024, and the output is the normalized water quality parameters. It should be noted that the pre-trained convolutional neural network model includes an input layer, convolutional layer 1, pooling layer 1, convolutional layer 2, pooling layer 2, a fully connected layer, and an output layer. The input layer consists of a single-channel feature matrix containing 2048 spectral wavelength points from a single sample. Convolutional layer 1: 32 3×1 convolutional kernels, stride 1, padding mode "same", activation function ReLU; Pooling layer 1: 2×1 max pooling window, stride 2, fill method "same"; Convolutional layer 2: 64 3×1 convolutional kernels, stride 1, padding "same", activation function ReLU; Pooling layer 2: 2×1 max pooling window, stride 2, fill method "same"; Fully connected layer: 128 neurons, ReLU activation function; Output layer: 2 neurons, no activation function, corresponding to the two output variables of residual chlorine concentration and turbidity; The convolutional neural network model was trained and validated using a training set, a validation set, and a test set, which is a conventional technical solution. The specific implementation steps are not detailed here. Specifically, the training set consisted of collecting water sample spectra with different pollution levels using a standard-configured spectral sensor on a secondary water supply simulation experimental platform. Water quality parameters were simultaneously calibrated using national standard methods, resulting in a total of 12,000 samples. The national standard methods included: residual chlorine: N,N-diethyl-p-phenylenediamine spectrophotometry; turbidity: formalazine standard solution turbidimetric method. Additionally, 8,000 valid spectral-water quality parameter pairs from the most recent year were selected from 10 secondary water supply monitoring stations in 3 cities. The validation set was randomly selected from the laboratory calibration data and field historical data of the training set in a stratified ratio of 1:6. The test set consists of calibrated data from an independent laboratory, consisting of 2,000 sets of new water samples collected on a secondary water supply simulation platform, covering novel pollutants not included in the training set. When implementing pipeline state parameter mapping, it is based on the ultrasonic attenuation coefficient sequence. Correct pipe roughness parameters: ;in, For the updated pipe roughness; The original roughness of the model; This is an empirical coefficient used to quantify the impact of changes in the attenuation coefficient on roughness; the default value is 0.01. This is the difference between the current and previous cycle's ultrasonic attenuation coefficient; When implementing environmental parameter mapping, the temperature, humidity, and atmospheric pressure values ​​in the environmental parameter vector M are directly used as the boundary condition inputs of the twin model; It should be noted that by implementing the above steps, a seamless connection between the perceived data and the virtual model parameters is achieved, ensuring the accuracy of the twin model input.

[0025] If ultrasonic data detects a pipeline leak, the specific leak can be determined by the change in attenuation coefficient ≥ 0.2dB / m. The leak location is automatically marked and the model topology is updated, enabling the addition of leak point nodes. Furthermore, based on the temperature value in the environmental parameter vector, the dynamic viscosity coefficient of the water flow within the pipe network is corrected. ;in, The dynamic viscosity at the current water temperature; The dynamic viscosity at the reference water temperature; It should be noted that by implementing the above steps, the twin model can reflect the changes in the physical state of the pipeline network in real time, thereby improving the consistency between the model and the actual pipeline network.

[0026] When loading the digital twin model of the secondary water supply network, import the network topology, material parameters, and boundary conditions. The network topology includes pipe length, pipe diameter, and node location. The material parameters include pipe roughness and pipe wall friction coefficient. The boundary conditions include inlet flow rate and outlet pressure. Based on the state parameters of the twin model, the core boundary conditions for LBM simulation are set, specifically: The lattice model adopts the D2Q9 two-dimensional nine-directional lattice model to simulate water flow and water quality diffusion within the pipe network; When setting boundary conditions: The inlet boundary is set as a velocity inlet, and the flow rate data comes from the flow meter in the secondary water supply pump station; The outlet boundary is set as a pressure outlet, and the pressure value comes from the pipeline pressure sensor. The wall boundary adopts a bounce boundary condition to simulate the no-slip characteristics of the pipe wall; The simulation parameters are specifically: time step. Spatial step size Total number of simulation steps ; It should be noted that by constructing a simulation environment that matches the actual pipe network conditions, the authenticity of theoretical water quality parameters can be ensured.

[0027] When performing theoretical water quality parameter simulation calculations, LBM simulation is run to calculate the residual chlorine concentration decay and turbidity diffusion in the secondary water supply network. Specifically, the particle distribution function is updated using the BGK (Bhatnagar-Gross-Krook) collision model: ;in, Let x be the time and position x. The particle distribution function in the direction; x is the two-dimensional spatial position vector; These are the nine discrete velocity directions of the D2Q9 model, such as (0, 0), (1, 0), (−1, 0), etc. The relaxation time is set to 0.6 by default. The equilibrium distribution function is calculated from the macroscopic velocity: , This refers to the macroscopic density of the fluid, specifically the density of water within the secondary water supply network. These are discrete directional weighting coefficients, such as the stationary direction. axial direction u is the macroscopic velocity vector, that is, the actual water flow velocity at a certain point within the secondary water supply network. ; For virtual speed of sound within the LBM model, ; It should be noted that simulating macroscopic flow fields through the microscopic evolution of discrete particles can effectively improve calculation accuracy compared to the traditional finite volume method. It can accurately capture complex flow fields at pipe bends and diameter changes, providing precise velocity field data for the calculation of water quality parameters.

[0028] Based on the flow field velocity distribution, the LBM passive scalar model is used to solve for residual chlorine decay and turbidity diffusion; specifically: When simulating residual chlorine concentration, a scalar distribution function of residual chlorine is established. The evolution equation, incorporating convection, diffusion, and attenuation processes, has the following expression: in, Discrete direction The residual chlorine particle distribution function at time t and position x; The relaxation time of the residual chlorine scalar, i.e., the characteristic time for residual chlorine particles to recover from a non-equilibrium state to an equilibrium state, is given by the residual chlorine diffusion coefficient. It is deduced that, ; Let be the equilibrium distribution function of residual chlorine; The residual chlorine decay coefficient is the natural decay rate of residual chlorine per unit time, which is affected by water temperature and pH. A typical value at 20℃ is used, and the value is taken as [value missing]. ; This refers to the macroscopic residual chlorine concentration; Extracting macroscopic residual chlorine concentration from residual chlorine particle distribution function: ; When performing turbidity diffusion simulation, a scalar evolution equation similar to that used for residual chlorine is employed, and the diffusion coefficient is adjusted to match the diffusion coefficient of the turbidity particles. And extract the macroscopic turbidity value. The diffusion coefficient is adjusted because turbidity particles are larger than residual chlorine molecules, resulting in slower diffusion. The distribution function of the turbidity scalar; After every 100 steps of LBM flow field calculation, the updated velocity field is substituted into the water quality scalar evolution equation to achieve synchronous iteration of flow field and water quality parameters. The total number of simulation steps is 10,000, corresponding to 100 seconds of actual time.

[0029] It should be noted that the above steps achieve bidirectional coupling between the flow field and water quality parameters, which can effectively reduce the error in calculating residual chlorine concentration and turbidity, and accurately simulate the spatiotemporal changes in water quality within the pipe network.

[0030] The calibration coefficients between the sensed data and the theoretical data were calculated using the least squares method. ;in, Calibrate coefficients for perceived data; The operator for finding the minimum value; For node index variables; For the first Theoretical residual chlorine concentration at the node; For the first Nodes sense residual chlorine concentration; The actual sensed water quality parameter matrix is ​​corrected using calibration coefficients derived from the sensed data. ;in, This is the corrected water quality parameter matrix; The corrected water quality parameters are fed back into the digital twin model to update the water quality distribution within the model, providing the latest input for the next round of simulation.

[0031] It should be noted that through the synergy of the above steps, the dynamic correction of the sensing data is achieved. The dynamic update of the twin model provides the latest physical state of the pipeline network for LBM simulation. The theoretical water quality parameters generated by the simulation are compared with the sensing data to obtain calibration coefficients. The corrected sensing data is then used to update the twin model, forming a closed loop of sensing, simulation, calibration, and updating, ensuring real-time synchronization between the twin model and the physical pipeline network.

[0032] In this embodiment of the invention, the synergy of the above steps can effectively reduce water quality parameter errors and meet the accuracy requirements of secondary water supply safety monitoring; through dynamic calibration, the number of on-site sensor calibrations can be reduced, and based on the calibrated sensing data, the pollution identification accuracy of the twin model can be effectively improved, enabling early warning of secondary water supply safety.

[0033] Pollution prediction and proactive intervention optimization module: Based on the corrected water quality parameters, an input feature set is constructed and input into a pre-trained deep Q-network model to predict the probability of contamination on the sensor surface. Based on the analysis results of the sensor surface contamination probability, proactive intervention measures and closed-loop optimization are triggered. Specific steps include: Obtain the corrected water quality parameters and construct the input feature set by combining them with historical sensor operating data; the historical sensor operating data includes the signal-to-noise ratio. Historical cleaning records runtime ; A 10-step sliding window is used to extract temporal features and generate the feature vector at time t. : ; Let be the sequence of environmental parameters at time t. , These represent the temperature, relative humidity, and atmospheric pressure at time t, respectively. The generated feature vectors are subjected to Min-Max normalization, which maps all features to the [0,1] interval to obtain the normalized feature vectors. Min-Max normalization is a conventional technical solution, and the specific implementation steps will not be elaborated here. Load the pre-trained deep Q-network model. The model structure specifically includes: Input layer: 26 dimensions, strictly matching the dimensions of the temporal feature vector; Hidden layers: 2 fully connected layers, each with 128 neurons, using ReLU activation function; Output layer: 1-dimensional, corresponding to the Q-value output of a single action based on the pollution probability prediction; The normalized feature vector Input a DQN model, which is a deep Q-network model, and perform forward propagation to calculate the Q-value, specifically including: During input layer processing, the 26-dimensional normalized feature vector is input into the model input layer to verify that the input dimension matches the model input layer dimension. If they do not match, an exception warning is triggered and the inference is terminated; if they match, the next layer of processing is performed. During hidden layer processing, linear computation is performed through the first fully connected layer: ;in, This is the linear calculation result of the first hidden layer; These are the weight matrix and bias vector of the first fully connected layer. right Each element undergoes ReLU activation. Elements less than 0 are set to 0, while positive eigenvalues ​​are retained; Activate the ReLU output for the first hidden layer; Linear computation is performed through a second fully connected layer: ;in, This is the linear calculation result of the second hidden layer; For the weight matrix and bias vector of the second fully connected layer; right Each element undergoes ReLU activation. ; Activate the ReLU output for the second hidden layer; When processing the output layer, based on the output Perform Q-value calculation: ;in, The state-action value function output by the DQN model represents the state in the current state. Next action Value; Features of the current state ; For actions that predict the probability of contamination, the model only outputs the Q value corresponding to that action; The weight matrix and bias vector of the output layer; When converting the model's output Q-value into an interpretable contamination probability, the Q-value is mapped to the [0,1] interval using the Sigmoid function, and then multiplied by 100% to obtain the contamination probability: ;in, Let be the probability of sensor surface contamination at time t; Use the Sigmoid activation function; During the training phase, the Q value ranges from [-5, 5], corresponding to a pollution probability range of 0.67% to 99.33%, covering the full pollution state of the secondary water supply sensor. Based on pollution probability The sensor status is divided into 3 status levels: At this time, the corresponding cleanliness level indicates that cleaning is not required and normal monitoring should continue. At that time, corresponding to a light pollution level, cleaning and maintenance are planned to be carried out within 72 hours; At that time, corresponding to a severe pollution level, a cleaning command is immediately triggered for automatic cleaning; specifically, automatic cleaning can be implemented based on existing pulse airflow purging equipment; in addition, the analysis involves The values ​​can be customized and adjusted according to the application requirements of the actual application scenario; Several hours after the intervention, specifically one hour later, while verifying the sensor status and updating the model parameters, the sensor signal-to-noise ratio recovery rate is calculated. ;in, For sensor signal-to-noise ratio recovery rate; Signal-to-noise ratio before intervention; Signal-to-noise ratio after intervention; Signal-to-noise ratio under normal conditions; like If the rate is ≥90%, the intervention is marked as successful, and the data from this intervention is added to the model training set as a positive sample. The intervention data includes feature vectors, contamination probability, intervention measures, and recovery rate. like If the percentage is less than 90%, the intervention is marked as ineffective. The reward function weights of the model are adjusted, and the model is retrained. The reward function weights of the model can be adjusted, for example, by increasing the penalty coefficient for intervention failure under high pollution probability.

[0034] It should be noted that by using a deep reinforcement learning model to analyze multi-source data in real time after calibration, the passive mode of monitoring and maintenance is broken, and an active closed loop of prediction, intervention and verification is constructed, providing dynamic and accurate technical support for the safety of secondary water supply.

[0035] In this embodiment of the invention, risks are identified in advance by predicting pollution trends, intervention timing is ensured by a threshold triggering mechanism, and intervention efficiency is improved by adapting multiple measures. Ultimately, a complete closed loop of data monitoring, risk prediction, proactive intervention, and effect verification is formed. This allows for early intervention in sensor contamination, avoiding water quality exceedances caused by data distortion. It addresses the water quality monitoring problem caused by secondary water supply sensor contamination at its root, and can effectively improve the reliability of secondary water supply monitoring and the water quality compliance rate.

[0036] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0037] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0038] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0039] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A fully digital management system for secondary water supply, characterized in that: include: Non-contact multimodal sensing network construction module: Deploy a hyperspectral imaging module, an ultrasonic guided wave detection unit, and a micro-meteorological sensor. Based on the acquired and processed spectral feature matrix, ultrasonic attenuation coefficient sequence, and environmental parameter vector, generate a multimodal sensing dataset. Digital twin dynamic calibration module: Based on the generated multimodal sensing dataset, update the digital twin model of the secondary water supply network, calculate theoretical water quality parameters through LBM lattice Boltzmann fluid simulation, and compare them with the sensing data to generate calibration coefficients to correct the sensing results; Pollution prediction and proactive intervention optimization module: Based on the corrected water quality parameters, an input feature set is constructed and input into a pre-trained deep Q-network model to predict the probability of contamination on the sensor surface. Based on the analysis results of the sensor surface contamination probability, proactive intervention measures and closed-loop optimization are triggered.

2. The digital management system for the entire secondary water supply chain according to claim 1, characterized in that, Signal transmission and reception are performed according to a preset cycle. The received echo signal is preprocessed, and the amplitude of the transmitted signal and the amplitude of the received signal are extracted based on the preprocessed echo signal. The temperature compensation coefficient is calculated by collecting water temperature data from the micro-meteorological sensors in the pipeline network. Based on the obtained transmitted signal amplitude, received signal amplitude, and temperature compensation coefficient, the attenuation coefficient is calculated; the calculated attenuation coefficients are arranged and combined in chronological order to obtain the ultrasonic attenuation coefficient sequence.

3. The digital management system for the entire secondary water supply chain according to claim 2, characterized in that, When updating the digital twin model of the secondary water supply network, the generated multimodal sensing data is converted into input parameters that the digital twin model can recognize, including water quality parameter mapping, pipeline status parameter mapping, and environmental parameter mapping.

4. The digital management system for the entire secondary water supply chain according to claim 3, characterized in that, When loading the digital twin model of the secondary water supply network, import the network topology, material parameters, and boundary conditions. The network topology includes pipe length, pipe diameter, and node location. The material parameters include pipe roughness and pipe wall friction coefficient. The boundary conditions include inlet flow rate and outlet pressure. Based on the state parameters of the twin model, the core boundary conditions for LBM simulation are set.

5. The digital management system for the entire secondary water supply chain according to claim 4, characterized in that, When performing theoretical water quality parameter simulation calculations, the particle distribution function is updated, and the LBM passive scalar model is used to solve for residual chlorine decay and turbidity diffusion based on the flow field velocity distribution.

6. The digital management system for the entire secondary water supply chain according to claim 5, characterized in that, When simulating residual chlorine concentration, an evolution equation for the residual chlorine scalar distribution function is established, convection, diffusion and attenuation processes are introduced, and the macroscopic residual chlorine concentration is extracted from the residual chlorine particle distribution function.

7. The digital management system for the entire secondary water supply chain according to claim 6, characterized in that, When performing turbidity diffusion simulation, a scalar evolution equation similar to that of residual chlorine is used, the diffusion coefficient is adjusted to the diffusion coefficient of turbidity particles, and the macroscopic turbidity value is extracted.

8. The digital management system for the entire secondary water supply chain according to claim 7, characterized in that, The least squares method is used to calculate the calibration coefficient between the perceived data and the theoretical data. The calibration coefficient of the perceived data is used to correct the actual perceived water quality parameter matrix. The corrected water quality parameters are fed back to the digital twin model to update the water quality distribution status in the model.

9. The digital management system for the entire secondary water supply chain according to claim 8, characterized in that, The corrected water quality parameters are obtained, and combined with historical sensor operation data, an input feature set is constructed and the data is normalized. This data is then input into a pre-trained deep Q-network model, and forward propagation is performed to calculate the Q-value.

10. The digital management system for the entire secondary water supply chain according to claim 9, characterized in that, The Q-value output by the model is converted into an interpretable pollution probability, and the obtained pollution probability is analyzed. Active intervention measures are dynamically triggered based on the analysis results. Several hours after the intervention is completed, the sensor status is verified and the model parameters are updated.

Images