An intelligent water management method and system based on artificial intelligence

Abstract

The application discloses an intelligent water management method and system based on artificial intelligence, relates to the technical field of intelligent water management, and comprises the following steps: initializing a pipe network digital twin model, loading pre-stored pipe network GIS data, collecting pipe pressure data, water quality parameters and user water consumption in real time, inputting a risk assessment model to calculate real-time risk probability, and generating a high-risk early warning signal when the real-time risk probability exceeds a warning threshold; based on the high-risk early warning signal, a pipe material state fractional order controller is used to calculate differential order parameters and integral gain parameters, generate a stable pressure control instruction and execute the same, and a control error signal in the stable pressure process is extracted; the control error signal and a water source type label are input into a meta-reinforcement learning model, the differential order parameters and the integral gain parameters are dynamically optimized, and a water source risk coefficient is obtained; and the application realizes personalized modeling and dynamic updating of risk assessment by outputting dynamic response characteristic data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent water management, and particularly relates to an intelligent water management method and system based on artificial intelligence. BACKGROUND

[0002] With the continuous advancement of urbanization, the contradiction between supply and demand of water resources is increasingly prominent, and water management is evolving from traditional extensive to intelligent and fine. In recent years, as an important part of smart city, intelligent water management relies on new generation information technologies such as Internet of Things, big data and artificial intelligence, and realizes real-time monitoring and intelligent control of the operation state of the water supply network system. Especially in the data acquisition layer, the application of distributed sensor network significantly improves the sensing accuracy of pipe network pressure, water quality parameters and user water consumption.

[0003] However, the existing intelligent water management still has obvious limitations in dealing with multi-source heterogeneous risk factors. On the one hand, the traditional PID control algorithm is difficult to adapt to the nonlinear and time-varying pipe aging characteristics in the pipe network, resulting in lagging adjustment response and insufficient control precision in the high-risk early warning state. On the other hand, most of the current risk assessment models rely on static threshold judgment mechanism, and lack of deep modeling of the difference of water source types and the dynamic evolution process of control error, thereby affecting the accuracy of risk coefficient calculation and the timeliness of decision-making. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides an intelligent water management method based on artificial intelligence to solve the problems of traditional PID control response lag and inaccurate static threshold evaluation mechanism.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides an intelligent water management method based on artificial intelligence, which comprises initializing a pipe network digital twin model, loading pre-stored pipe network GIS data, and collecting pipe pressure data, water quality parameters and user water consumption in real time, inputting a risk assessment model to calculate real-time risk probability, and generating a high-risk warning signal when the real-time risk probability exceeds a warning threshold; based on the high-risk warning signal, a pipe material state fractional order controller is used to calculate the differential order parameter and the integral gain parameter, generate a pressure stabilization control instruction and execute it, while extracting the control error signal in the pressure stabilization process; the control error signal and the water source type label are input into the meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and obtain the water source risk coefficient; based on the water source risk coefficient and the pipe network GIS data, the seepage critical threshold is calculated, and when the real-time risk probability exceeds the seepage critical threshold, a partition pressure reduction instruction is generated, and the state record is executed to the operation log record; the operation log record and the water source risk coefficient are written into the blockchain for notarization, and the pipe network digital twin model is updated.

[0008] As a preferred scheme of the intelligent water management method based on artificial intelligence, the high-risk warning signal is generated, and the specific steps are as follows,

[0009] The pipe network digital twin model is initialized, and the pre-stored pipe network GIS data is loaded from the blockchain distributed storage node to analyze the topological structure of the pipe network GIS data;

[0010] Based on the topological structure of the pipe network GIS data, the pipe pressure data, the water quality parameters and the user water consumption are collected;

[0011] The pipe pressure data, the water quality parameters and the user water consumption are input into the risk assessment model, and the real-time risk probability is calculated by a physical-data dual driving algorithm;

[0012] When the real-time risk probability exceeds the warning threshold, the high-risk warning signal is triggered.

[0013] As a preferred scheme of the intelligent water management method based on artificial intelligence, based on the high-risk warning signal, the pipe material state fractional order controller is used to calculate the differential order parameter and the integral gain parameter, generate a pressure stabilization control instruction and execute it, while extracting the control error signal in the pressure stabilization process, and the specific steps are as follows,

[0014] According to the high-risk warning signal, the pipe material quality characteristics, the pipe service life and the pipe corrosion degree in the warning area are analyzed to generate a three-dimensional pipe material state vector;

[0015] The three-dimensional pipe material state vector is input into a dynamic mapping function to calculate the differential order parameter and the integral gain parameter;

[0016] Based on the differential order parameter and the integral gain parameter, a partition pressure regulation instruction is generated through a fractional order control law, and is written into a blockchain smart contract for pipe network pressure control.

[0017] In the pipe network pressure control process, pressure control deviation time series data is collected, and a control error signal in the pipe network pressure control process is obtained through a multi-scale oscillation feature extraction algorithm.

[0018] As a preferred scheme of the intelligent water management method based on artificial intelligence, the water source risk coefficient is obtained by the following specific steps,

[0019] The control error signal and the water source type label are fused in multiple modes to generate a spatiotemporal coupling feature vector.

[0020] The spatiotemporal coupling feature vector is input into a meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and output dynamic response characteristic data in the optimization process.

[0021] Based on the dynamic response characteristic data, the water source pollution characteristic distribution and the pipe network material state distribution are combined to calculate the water source risk coefficient through three-dimensional coupling.

[0022] As a preferred scheme of the intelligent water management method based on artificial intelligence, the water source risk coefficient and the pipe network GIS data are used to calculate the seepage critical threshold value, and the specific steps are as follows,

[0023] The water source risk coefficient is mapped to the topological structure of the pipe network GIS data to generate a dynamic risk field distribution.

[0024] The dynamic risk field distribution is subjected to three-dimensional spatial integral operation to generate an initial seepage reference value.

[0025] The initial seepage reference value is dynamically corrected using the water source risk coefficient, and the seepage critical threshold value is calculated in combination with the structure compensation factor of the pipe network GIS data.

[0026] As a preferred scheme of the intelligent water management method based on artificial intelligence, when the real-time risk probability exceeds the seepage critical threshold value, a partition pressure reduction instruction is generated, and a state record is executed to an operation log, and the specific steps are as follows,

[0027] When the real-time risk probability exceeds the seepage critical threshold value, a partition pressure reduction instruction is generated based on the topological structure of the pipe network GIS data and the pipe pressure data, and pressure regulation is performed according to the partition pressure reduction instruction.

[0028] The equipment state and pressure change data in the pressure regulation process are collected and integrated into an operation log record in combination with the partition pressure reduction instruction.

[0029] As a preferred scheme of the intelligent water management method based on artificial intelligence, the operation log record and the water source risk coefficient are written into the blockchain for notarization, and the pipe network digital twin model is updated, and the specific steps are as follows,

[0030] The operation log record and the water source risk coefficient are written into the blockchain smart contract for notarization to generate a notarization identifier.

[0031] Based on the notarization identifier, the quantum entangled state is triggered, and the pipe network digital twin model is updated by the quantum-fluid holographic mapping method.

[0032] In a second aspect, the present application provides an intelligent water management system based on artificial intelligence, comprising a twin initialization module, a pressure control module, a parameter optimization module, a seepage warning module and a blockchain notarization module. The twin initialization module is used to initialize the pipe network digital twin model, load the pre-stored pipe network GIS data, and real-time collect the pipe pressure data, water quality parameters and user water consumption, input the risk assessment model to calculate the real-time risk probability, and generate a high-risk warning signal when the real-time risk probability exceeds the warning threshold. The pressure control module is used to calculate the differential order parameter and integral gain parameter based on the high-risk warning signal through the pipe material state fractional order controller, generate a pressure control instruction and execute it, and extract the control error signal in the pressure control process. The parameter optimization module is used to input the control error signal and the water source type label into the meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and obtain the water source risk coefficient. The seepage warning module is used to calculate the seepage critical threshold based on the water source risk coefficient and the pipe network GIS data, and generate a partition pressure reduction instruction when the real-time risk probability exceeds the seepage critical threshold, and execute the state record to the operation log record. The blockchain notarization module is used to write the operation log record and the water source risk coefficient into the blockchain for notarization, and update the pipe network digital twin model.

[0033] In a third aspect, the present application provides a computer device comprising a memory and a processor, the memory storing a computer program, wherein the computer program is executed by the processor to implement any step of the intelligent water management method based on artificial intelligence according to the first aspect of the present application.

[0034] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by the processor to implement any step of the intelligent water management method based on artificial intelligence according to the first aspect of the present application.

[0035] The present application has the beneficial effects that: through the fractional order controller parameter adaptive calculation of the pipe state, high-precision pressure stabilization control of the nonlinear and time-varying pipe network is realized, and the response speed and robustness in the high-risk early warning state are improved; further, through the dynamic optimization of the meta-reinforcement learning model, dynamic response characteristic data is output, and personalized modeling and dynamic updating of risk assessment are realized. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0037] Fig. 1 The flowchart of the intelligent water management method based on artificial intelligence.

[0038] Fig. 2 For generating a high-risk early warning signal.

[0039] Fig. 3 The flowchart for generating pressure stabilization control instructions and extracting errors.

[0040] Fig. 4 The flowchart for calculating the water source risk coefficient. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0042] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0043] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0044] REFERENCE Figs. 1-4 For one embodiment of the present application, the embodiment provides an intelligent water management method based on artificial intelligence, comprising the following steps:

[0045] S1: initialize the digital twin model of the pipe network, load the pre-stored pipe network GIS data, and collect real-time pipe pressure data, water quality parameters and user water consumption, input the risk assessment model to calculate the real-time risk probability, and generate a high-risk warning signal when the real-time risk probability exceeds the warning threshold.

[0046] S1.1: initialize the digital twin model of the pipe network, and load the pre-stored pipe network GIS data from the blockchain distributed storage node, and parse the topological structure of the pipe network GIS data.

[0047] The specific process includes, when initializing the digital twin model of the pipe network, establishing a data connection channel with the blockchain distributed storage node, obtaining the pre-stored pipe network GIS data from the blockchain distributed storage node, and the pipe network GIS data includes node coordinates, elevations, pipe diameters, material properties and pipe segment connection relationship information of the pipe network. When parsing the pipe network GIS data, the graph theory algorithm is used to build the node and initialize the pipe network digital twin model, and the pre-stored pipe network GIS data is loaded from the blockchain distributed storage node, and the adjacency matrix and spatial coordinate mapping in graph theory are used to automatically identify the node connection relationship and associate attribute parameters. The topological structure of the pipe network GIS data stores the corresponding spatial coordinates and attribute parameters of each node, and each pipe network record connection relationship and hydraulic characteristic parameters.

[0048] Further, the pre-training process of the pipe network digital twin model needs to collect historical pipe network operation data, including pressure monitoring records, flow variation curves, water quality parameter time series data and user water consumption statistical information. After cleaning and standardizing the historical pipe network operation data, a basic model architecture is constructed using a spatio-temporal graph convolutional neural network, and the model is trained through self-supervised learning to capture the spatio-temporal feature dependency relationship in the pipe network. Specifically, node attribute prediction and edge connection reconstruction are used as pre-training tasks to learn the general representation of the pipe network topological structure on a large amount of unlabeled data. Then, the pipe network digital twin model parameters obtained by pre-training are used as initialization weights to fine-tune on specific task labeled data, and finally a pipe network digital twin model with pipe network state prediction and anomaly detection capabilities is formed.

[0049] S1.2: Based on the topological structure of the pipe network GIS data, collect pipe pressure data, water quality parameters and user water consumption.

[0050] The specific process includes that when the topological structure of the pipe network GIS data is dynamically bound to the monitoring equipment, the node coordinates in the pipe network GIS data are mapped with the pressure sensors, water quality detectors and intelligent water meters in the physical space, the data channel of the equipment and the topological network node is established through the Internet of Things communication protocol, the topological position identifier of the corresponding node is recorded synchronously when the pipe pressure data is collected in real time, the pipe segment material and the connection relationship in the pipe network GIS data are associated in the water quality parameter collection process, the user water consumption data is uploaded through the intelligent water meter and matched with the end user node in the topological network, and the pipe pressure data, the water quality parameter and the user water consumption all have the spatial position label and the timestamp information in the topological network structure.

[0051] S1.3: inputting the pipe pressure data, the water quality parameter and the user water consumption into a risk assessment model, and calculating a real-time risk probability through a physical-data double-driven algorithm, and the expression is as follows:

[0052] ;

[0053] Among them, represents the real-time risk probability, represents a Sigmoid function, represents a pressure risk weight (0.4≤ ≤0.7), represents a pipe real-time pressure detection value, represents a safety benchmark pressure, represents a water quality risk weight (0.2≤ ≤0.5), represents a user behavior risk value, represents a differential of the water quality parameter, represents a differential of time, represents a user water consumption time sequence, represents an abnormal score of the user water consumption time sequence , represents a unit value of the safety benchmark pressure , represents a unit value of the water quality parameter change rate.

[0054] It should be explained that the ratio of the pressure deviation and the unit value of the safety benchmark pressure converts the pressure feature into a relative deviation coefficient, the ratio of the water quality parameter change rate and the unit value of the standard change rate converts it into a relative change intensity, and the user behavior score itself is dimensionless and remains unchanged, all input items are converted into relative values without dimension, so that they can be weighted and summed on a unified scale, the expression processed in this way not only maintains the physical meaning of each item, but also meets the requirement of the mathematical model on the consistency of the input dimension, and finally the Sigmoid function maps the comprehensive score into a risk probability value between 0 and 1.

[0055] The specific process includes that after the pipeline pressure data, water quality parameters and user water consumption are input into the risk assessment model, the physical-data dual driving algorithm realizes the transient hydraulic analysis of pipeline pressure by coupling the fluid mechanics equation with real-time sensor data and adopting dynamic data assimilation technology to correct the model parameters, calculates the pressure fluctuation propagation characteristics combined with the topological network structure of the pipe network, and obtains the chemical compatibility evaluation of the corrosion tendency index (CTI) through the multi-parameter coupling analysis of water quality parameters (pH value, ion concentration) and pipeline pressure data. The user water consumption data is detected for abnormal water consumption behavior through a pattern recognition algorithm, and the abnormal patterns of the calculated pipe material stress distribution and corrosion rate are fused in multiple dimensions, and the physical simulation results and data characteristics are integrated through Bayesian probability, and finally the real-time risk probability reflecting the overall safety state of the pipe network is output.

[0056] The abnormal score of the user water consumption time series is obtained by comparing the real-time water consumption data with the historical mode or the machine learning prediction value.

[0057] Further, the training process of the risk assessment model needs to prepare a data set containing historical pipeline pressure data, water quality parameters, user water consumption and corresponding accident records, adopt a physically constrained deep learning architecture, jointly train the physical features such as pressure gradient distribution and pipe material corrosion rate prediction obtained by hydraulic calculation with data-driven time series anomaly detection features, simultaneously optimize the physical consistency loss and accident classification loss through a multi-task learning framework, capture the risk propagation path in the pipe network topology structure using a graph attention mechanism, and finally obtain a risk assessment model that can fuse the features of physical simulation results and real-time monitoring data.

[0058] S1.4: When the real-time risk probability exceeds the warning threshold, a high-risk warning signal is triggered.

[0059] The specific process includes that when the real-time risk probability exceeds the warning threshold, the risk assessment model immediately generates a high-risk warning signal containing the risk level and the risk location, the high-risk warning signal is distributed to the pipe network monitoring center and the related control unit through the message queue, the high-risk warning signal records the risk occurrence time, the pipe network location coordinates and the risk type classification in detail, and at the same time triggers the sound and light alarm device and the visual interface warning mark.

[0060] The warning threshold is a dynamic critical value determined based on the statistical analysis of historical accident data combined with the pipe network material characteristic parameters, through the fitting of the probability density function and the calibration of engineering practice.

[0061] S2: Based on the high-risk warning signal, the differential order parameter and the integral gain parameter are calculated through the pipe material state fractional order controller, the stable pressure control instruction is generated and executed, and the control error signal in the stable pressure process is extracted.

[0062] S2.1: According to the high-risk early warning signal, analyze the pipe material characteristics, pipe service life and pipe corrosion degree in the early warning area, and generate a three-dimensional pipe material state vector.

[0063] The specific process includes: according to the high-risk early warning signal, positioning the spatial range of the early warning area in the pipe network topology network structure, extracting the pipe material characteristics of all pipe sections in the spatial range including cast iron, PE or PVC and the like, obtaining the pipe service life record and analyzing the corrosion degree index, converting the pipe material characteristics into a material strength coefficient, converting the pipe service life into an aging attenuation factor, quantifying the pipe corrosion degree into a corrosion rate parameter, and arranging and combining the three parameters in order of spatial coordinates to form a three-dimensional pipe material state vector. Each element in the three-dimensional pipe material state vector corresponds to the material state characteristics of a specific node in the pipe network topology network structure. Finally, a three-dimensional pipe material state vector containing complete pipe material information of the early warning area is output.

[0064] S2.2: Input the three-dimensional pipe material state vector into the dynamic mapping function to calculate the differential order parameter and the integral gain parameter, and the expression is:

[0065] ;

[0066] ;

[0067] wherein, represents the differential order parameter (0.2≤ ≤1.5) of the fractional order controller, represents the gamma function, represents the pipe service life, represents the aging time constant, represents the oscillation coupling operator, represents the spatial gradient of the pipe corrosion degree, represents the reference unit value of the spatial gradient of the pipe corrosion degree, represents the smoothing coefficient of the pipe corrosion degree (0.01≤ ≤0.1), represents the pipe material adaptation operator, represents the Gaussian error function, represents the pipe material sensitivity coefficient (0.1≤ ≤0.9), represents the pipe material characteristics, represents the integral gain parameter of the fractional order controller, represents the gain global scaling coefficient (1.5≤ ≤2.2), represents the pipe corrosion index coefficient (0.6≤ ≤1.7), represents the attenuation coupling operator, represents the pipeline usage attenuation coefficient (0.1 ≤ 0.3), represents the pipeline material periodicity coefficient (1.0 ≤ 1.5).

[0068] It should be noted that for parameters with clear physical units such as corrosion degree spatial gradient, service life, etc., the corresponding reference unit value (such as dividing the corrosion gradient by the standard corrosion rate unit, and dividing the time integral term by the time reference unit) needs to be divided first to convert it into a dimensionless relative value. At the same time, the input of mathematical functions (such as gamma function, , exponential function, etc.) must be strictly dimensionless, so the dimension must be eliminated through parameter definition (such as smoothing coefficient, attenuation coefficient, etc.) or additional normalization factor. Finally, all sub-terms are converted to dimensionless form, so that the operation and function combination of the coupling operator meet the principle of dimensional consistency.

[0069] The specific process includes that after the three-dimensional pipe material state vector is input into the dynamic mapping function, the pipeline material characteristics, pipeline service life and pipeline corrosion degree in the three-dimensional pipe material state vector are normalized. The dynamic mapping function is based on the pre-established pipe material characteristics and control parameter relationship matrix, and the three-dimensional pipe material state vector is converted into initial differential order parameters and integral gain parameters through matrix multiplication operation. At the same time, the parameters are dynamically corrected by combining the real-time collected pipeline pressure fluctuation data. The differential order parameter adjusts the control response speed according to the pipeline material characteristics and the pipeline corrosion degree, and the integral gain parameter adjusts the steady-state accuracy according to the change trend of the pipeline service life and the pipeline pressure data. Finally, the optimized differential order parameter and integral gain parameter combination are output, which directly acts on the fractional order controller to generate control instructions.

[0070] S2.3: Based on the differential order parameter and the integral gain parameter, a partition pressure regulation instruction is generated through the fractional order control law, and is written into a blockchain smart contract for pipeline pressure control.

[0071] The specific process includes that based on the differential order parameter and the integral gain parameter, the fractional order control law analyzes the pressure regulation amount of each partition of the pipeline network. The fractional order control law uses the Riemann-Liouville fractional differential operator to process the pressure deviation signal to generate a partition pressure regulation instruction containing a target pressure value and a regulation rate. The partition pressure regulation instruction is transmitted to the blockchain smart contract through an encrypted channel. The blockchain smart contract records in the distributed ledger after verifying the instruction signature, and triggers the preset valve control logic at the same time. The valve actuator receives the encrypted instruction issued by the blockchain smart contract and executes it after decryption, accurately adjusts the valve opening degree according to the partition pressure regulation instruction, realizes the closed-loop control of the pipeline network pressure, and the valve state change and pressure feedback data in the execution process are real-time fed back to the blockchain smart contract to form a complete control record.

[0072] The preset valve control logic is a fuzzy rule base and a PID parameter combination trained based on a pipe network hydraulic balance equation and pipe material pressure-bearing characteristics.

[0073] S2.4: In the process of executing the pipe network pressure control, pressure control deviation time series data is collected, and a control error signal in the pipe network pressure control process is obtained through a multi-scale oscillation feature extraction algorithm.

[0074] The specific process includes that, in the process of executing the pipe network pressure control, a pressure sensor continuously monitors the deviation of the actual pressure value from the target pressure value to form pressure control deviation time series data, a multi-scale oscillation feature extraction algorithm performs wavelet transform decomposition on the pressure control deviation time series data, extracts fluctuation components at different time scales, then analyzes the energy distribution and frequency characteristics of each fluctuation component, obtains instantaneous amplitude and phase information through Hilbert transform, and finally fuses high-frequency oscillation features and low-frequency trend components to identify abnormal fluctuation patterns in the pipe network pressure control process, output a control error signal representing control accuracy, and the control error signal contains quantitative indicators of steady-state deviation and transient oscillation.

[0075] S3: Input the control error signal and the water source type label into the meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and obtain the water source risk coefficient.

[0076] S3.1: Perform multi-modal fusion of the control error signal and the water source type label to generate a spatio-temporal coupling feature vector.

[0077] The specific process includes that, in the multi-modal fusion of the control error signal and the water source type label, the steady-state deviation component and the transient oscillation feature in the control error signal are normalized, the water source type label is converted into a one-hot encoding vector, the time domain features of the control error signal and the category features of the water source type label are subjected to tensor product operation through a feature cross layer to generate an intermediate feature matrix with spatio-temporal correlation, the intermediate feature matrix is analyzed by a multi-head attention mechanism to determine the weight distribution between different modalities, and finally a spatio-temporal coupling feature vector that fuses the water source characteristics and the control performance is output, the spatio-temporal coupling feature vector contains both the pressure control accuracy and the influence of the water source physical properties on the stability of the pipe network.

[0078] The water source type label is an encoded label obtained by classifying water quality detection data, including three categories of surface water, groundwater and reclaimed water.

[0079] S3.2: Input the spatio-temporal coupling feature vector into the meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and output dynamic response characteristic data in the optimization process.

[0080] The specific process includes that after the space-time coupling feature vector is input into the meta-reinforcement learning model, the reinforcement learning model decouples the space-time dimensions of the feature vector through the space-time attention mechanism, respectively extracts the associated weight matrix of the control performance indicators (such as the regulation rate and the overshoot) and the water source characteristic parameters (such as the turbidity fluctuation and the ion concentration), generates the parameter adjustment decision, the strategy network generates the adjustment amount of the differential order parameter and the integral gain parameter by using the double-delay deep deterministic policy gradient algorithm, the actuator network updates the parameter configuration of the fractional order controller in real time according to the adjustment amount, and the value network quantitatively evaluates the control effect by analyzing the dynamic error integral (such as the IAE index) of the actual control curve and the target curve and combining the pipe network stability coefficient (such as the Lyapunov index), and feeds back to the value network for strategy optimization, and by analyzing the time sequence difference error of the state-action value function and combining the strategy gradient rising method, the action selection probability is dynamically adjusted to maximize the long-term cumulative reward, so as to determine the parameter optimization direction. The meta-reinforcement learning model records the change trajectory of the differential order parameter and the integral gain parameter and the corresponding pipe network response data during the optimization process, and forms the dynamic response characteristic data containing the adjustment process of the differential order parameter and the integral gain parameter and the control effect.

[0081] Further, the training process of the meta-reinforcement learning model, the experience replay pool containing the pipe network control state, the action instruction and the reward signal is constructed, the strategy network and the value network parameters are updated synchronously by using the double-layer optimization framework, the strategy network outputs the continuous adjustment action of the differential order parameter and the integral gain parameter by using the deterministic policy gradient algorithm, the value network analyzes the estimated value of the state-action value function based on the time difference error, the algorithm stability is maintained by using the target network delay update during the training process, and the priority experience replay mechanism is used to learn the key control scene, so that the meta-reinforcement learning model can quickly adapt to the control requirements under different water source types and pipe network states.

[0082] S3.3: Based on the dynamic response characteristic data, the water source risk coefficient is calculated by coupling the pipe network material state distribution and the water source pollution characteristic distribution through three-dimensional coupling, and the expression is:

[0083] ;

[0084] Among them, the water source risk coefficient (0≤ ≤1), denotes the time length of the control window, and Δμ denotes the differential order parameter correction amount of the fractional order controller, denotes the integral gain parameter correction amount of the fractional order controller, denotes the Kullback-Leibler divergence, denotes the pipe network material state distribution, denotes the water source pollution characteristic distribution, denotes the spatial gradient weight coefficient (0.2≤ ≤0.6), denotes the spatial gradient operator, denotes the water quality chemical parameter, denotes the unit value of time integral, denotes the unit value of differential order correction, denotes the unit value of water quality gradient.

[0085] It should be noted that first, the differential order correction is dimensionless, eliminating its physical units; second, the time integral term is divided by the characteristic time scale, converting the dynamic response into a dimensionless relative change intensity; at the same time, the spatial gradient term is divided by its typical variation amplitude, converted into a dimensionless relative spatial heterogeneity index. Through this hierarchical normalization method, both the dynamic response term (including time derivative and KL divergence) and the spatial gradient term are converted into dimensionless quantities, which not only preserves the contribution proportion of each physical characteristic, but also meets the requirements of Sigmoid function for dimensionless input, finally ensuring that the output water source risk coefficient is a standardized probability value strictly within the range of 0 to 1.

[0086] The specific process includes: based on the dynamic response characteristic data, extracting the pressure fluctuation characteristics and valve adjustment frequency in the pipe network control process, combining the pipe corrosion rate and service life parameters in the pipe network material state distribution, and the pollutant concentration and diffusion trend in the water source pollution characteristic distribution, through three-dimensional tensor operation, the dynamic response characteristic data, pipe network material state distribution and water source pollution characteristic distribution are fused, the spatial convolution operation is used to capture the nonlinear relationship between different dimensional features, and the water source risk coefficient reflecting the comprehensive risk level of the pipe network is calculated. The water source risk coefficient quantifies the coupling influence degree of control performance, pipe material degradation and water quality deterioration.

[0087] The pipe network material state distribution is constructed by pipe corrosion detection data and service life records to reflect the spatial probability distribution of material degradation of each part of the pipe network.

[0088] The water source pollution characteristic distribution is the pollutant diffusion trend spatial distribution obtained based on the pollutant concentration data of water quality monitoring sites and fluid dynamics simulation.

[0089] S4: Based on the water source risk coefficient and the pipe network GIS data, calculate the seepage critical threshold, when the real-time risk probability exceeds the seepage critical threshold, generate partition pressure reduction instructions, and execute state record to operation log record.

[0090] S4.1: Map the water source risk coefficient to the topological structure of the pipe network GIS data to generate a dynamic risk field distribution.

[0091] The specific process includes that when the water source risk coefficient is mapped to the topology structure of the pipe network GIS data, the water source risk coefficient is spatially corresponded to the node coordinates in the pipe network GIS data, the discrete water source risk coefficient is converted into a continuous spatial field through the Kriging spatial interpolation algorithm, the gradient diffusion analysis is performed in combination with the connection relationship and the pipe section attribute in the pipe network topology structure, the three-dimensional dynamic risk field distribution covering the entire pipe network is generated, the risk value of each grid point in the dynamic risk field distribution reflects the comprehensive risk level of the corresponding position, and the dynamic risk field distribution is updated in real time with the change of the water source risk coefficient.

[0092] S4.2: Perform three-dimensional spatial integral operation on the dynamic risk field distribution to generate an initial seepage reference value.

[0093] The specific process includes that when the three-dimensional spatial integral operation is performed on the dynamic risk field distribution, the three-dimensional voxel grid is established within the spatial range defined by the pipe network GIS data, each voxel unit stores the risk value of the corresponding position in the dynamic risk field distribution, the risk volume integral of each voxel unit is analyzed by using the Gauss quadrature method, the integral result is accumulated layer by layer along the flow direction in the pipe network topology structure, the weight correction is performed in combination with the pipe section diameter and the connection relationship, and finally the initial seepage reference value reflecting the overall seepage risk level of the pipe network is output.

[0094] S4.3: Dynamically correct the initial seepage reference value by using the water source risk coefficient, and calculate the seepage critical threshold in combination with the structure compensation factor of the pipe network GIS data, and the expression is as follows:

[0095] ;

[0096] Among them, represents the seepage critical threshold, represents the initial seepage reference value, represents the natural exponential function, represents the risk change sensitivity coefficient, represents the structure compensation intensity coefficient, represents the pipe network graph structure, represents the pipe network graph topology diameter, represents the normalized reference length of the pipe network graph topology diameter.

[0097] It should be noted that in the formula for calculating the seepage critical threshold, in order to ensure the dimensional consistency, the dynamic correction term and the structure compensation term are normalized respectively. First, by introducing the characteristic time scale, the risk change rate is converted into a dimensionless relative change intensity, so that the input of the natural exponential function meets the mathematical requirements; at the same time, the characteristic length reference is introduced to the pipe network topological diameter, which is converted into a dimensionless relative value, to ensure the legality of the denominator operation. Through this double normalization processing, the dynamic correction term and the structure compensation term are both converted into dimensionless quantities, which not only retains the physical meaning of each term, but also meets the requirement of mathematical operation for dimensional consistency. Finally, the output seepage critical threshold and the initial reference value have the same physical dimension.

[0098] The specific process includes that when the initial seepage reference value is dynamically corrected using the water source risk coefficient, the time change rate of the water source risk coefficient is analyzed as a dynamic adjustment factor, the dynamic adjustment factor and the initial seepage reference value are weighted and fused, and the pipe segment diameter, connection density and node elevation difference are extracted from the pipe network GIS data as structure compensation factors. The structure compensation factors are converted into compensation coefficients through the pipe network topological characteristic function, the compensation coefficients and the dynamically corrected seepage reference value are convolved, and finally the seepage critical threshold which comprehensively reflects the real-time risk change and the pipe network structure characteristics is output. The seepage critical threshold is used for subsequent risk warning judgment.

[0099] S4.4: When the real-time risk probability exceeds the seepage critical threshold, a partition pressure reduction instruction is generated based on the topological structure of the pipe network GIS data and the pipe pressure data, and the pressure is adjusted according to the partition pressure reduction instruction.

[0100] The specific process includes that when the real-time risk probability exceeds the seepage critical threshold, the topological range of the high-risk area is located in the pipe network GIS data, the connection relationship and pressure sensor data of the pipe segment in the topological range of the high-risk area are extracted, based on the node flow conservation equation and the pipe resistance characteristics, the pressure adjustment amount of each node is iteratively analyzed to realize the pipe network hydraulic balance, the partition pressure reduction instruction containing the target pressure value and the adjustment rate is generated, the partition pressure reduction instruction is transmitted to the execution mechanism through the encryption communication protocol, the execution mechanism parses the instruction to accurately adjust the valve opening, realizes the hierarchical control of the pipe network pressure, and the valve state and pressure change data in the pressure adjustment process are fed back to the control center in real time to form a closed loop.

[0101] S4.5: Collect the equipment state and pressure change data in the pressure adjustment process, and integrate them into the operation log record in combination with the partition pressure reduction instruction.

[0102] The specific process includes recording the valve opening change curve and pressure sensor readings when collecting the equipment state and pressure change data during the pressure regulation process, aligning the equipment state data and pressure change data in time sequence, merging and processing the target pressure value and regulation rate parameter in the partition pressure reduction instruction with the real-time collected data, forming a complete operation log record through timestamp association, and the operation log record contains instruction issuing time, execution process data and final regulation result.

[0103] S5: Write the operation log record and water source risk coefficient into the blockchain for notarization, and update the digital twin model of the pipe network.

[0104] S5.1: Write the operation log record and water source risk coefficient into the blockchain smart contract for notarization, and generate a notarization identification.

[0105] The specific process includes that after the operation log record and water source risk coefficient generate digital fingerprints through encryption algorithm, calling the notarization interface of the blockchain smart contract to write into the distributed ledger, the blockchain smart contract verifies the data integrity to generate a notarization identification containing timestamp and transaction hash, the notarization identification establishes an unalterable mapping relationship with the operation log record and water source risk coefficient, and the notarization identification serves as the unique voucher for subsequent query verification.

[0106] S5.2: Trigger quantum entangled state based on notarization identification, and update the digital twin model of the pipe network through quantum-fluid holographic mapping method.

[0107] The specific process includes that when the quantum entangled state is triggered based on the notarization identification, the transaction hash value in the notarization identification is encoded into a quantum bit sequence, and an entangled state particle pair is generated through quantum gate operation, one state is retained in the local node and the other state is transmitted to the digital twin model, the quantum-fluid holographic mapping method converts the pipe network state change information carried by the entangled state into a holographic projection field, the holographic projection field is reconstructed into a three-dimensional pressure distribution and flow field of the pipe network through inverse Fourier transform, and finally the digital twin model is driven to complete the state update completely synchronized with the physical pipe network.

[0108] The embodiment also provides a smart water management system based on artificial intelligence, which comprises a twin initialization module, a stable voltage control module, a parameter optimization module, a seepage early warning module and a blockchain storage module. The twin initialization module is used for initializing a pipe network digital twin model, loading pre-stored pipe network GIS data, and collecting pipe pressure data, water quality parameters and user water consumption in real time, inputting a risk assessment model to calculate a real-time risk probability, and generating a high-risk early warning signal when the real-time risk probability exceeds a warning threshold. The stable voltage control module is used for calculating differential order parameters and integral gain parameters based on the high-risk early warning signal through a pipe material state fractional order controller, generating a stable voltage control instruction and executing the same, and extracting a control error signal in the stable voltage process. The parameter optimization module is used for inputting the control error signal and a water source type label into an element reinforcement learning model, dynamically optimizing the differential order parameters and the integral gain parameters, and obtaining a water source risk coefficient. The seepage early warning module is used for calculating a seepage critical threshold based on the water source risk coefficient and the pipe network GIS data, generating a partition pressure reduction instruction when the real-time risk probability exceeds the seepage critical threshold, and executing state recording to an operation log record. The blockchain storage module is used for writing the operation log record and the water source risk coefficient into a blockchain for storage, and updating the pipe network digital twin model.

[0109] The embodiment also provides a computer device suitable for the smart water management method based on artificial intelligence, which comprises a memory and a processor. The memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the smart water management method based on artificial intelligence as proposed in the above embodiment.

[0110] The computer device can be a terminal, which comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used for providing computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used for wired or wireless communication with external terminals. The wireless communication can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. The input device can also be an external keyboard, touchpad or mouse, etc.

[0111] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement the method for intelligent water management based on artificial intelligence proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0112] To sum up, the application realizes high-precision pressure stabilization control of a nonlinear and time-varying pipe network by adaptive calculation of the fractional order controller parameters of the pipe state, thereby improving the response speed and robustness in a high-risk early warning state; further, the dynamic response characteristic data are output by dynamic optimization of the meta-reinforcement learning model, thereby realizing personalized modeling and dynamic updating of risk assessment.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the application but not limit the application, and although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the application, and all of them should be covered in the scope of the claims of the application.

Claims

1. An artificial intelligence-based intelligent water management method, characterized in that: comprising, initializing the pipe network digital twin model, loading the pre-stored pipe network GIS data, and collecting pipe pressure data, water quality parameters and user water consumption in real time, inputting the risk assessment model to calculate the real-time risk probability, and generating a high-risk warning signal when the real-time risk probability exceeds the warning threshold; based on the high-risk warning signal, calculating the differential order parameter and integral gain parameter through the pipe material state fractional order controller, generating a stable pressure control instruction and executing it, while extracting the control error signal in the stable pressure process, the specific steps are as follows, According to the high-risk warning signal, analyze the pipe material characteristics, pipe service life and pipe corrosion degree in the warning area to generate a three-dimensional pipe material state vector; input the three-dimensional pipe material state vector into the dynamic mapping function to calculate the differential order parameter and integral gain parameter; based on the differential order parameter and integral gain parameter, generate a partition pressure regulation instruction through the fractional order control law, and write it into the smart contract of the blockchain for pipe network pressure control; In the process of executing the pipe network pressure control, collect the pressure control deviation time series data, and obtain the control error signal in the pipe network pressure control process through the multi-scale oscillation feature extraction algorithm; input the control error signal and the water source type label into the meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and obtain the water source risk coefficient, the specific steps are as follows, fuse the control error signal and the water source type label to generate a spatiotemporal coupling feature vector; input the spatiotemporal coupling feature vector into the meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and output the dynamic response characteristic data in the optimization process; based on the dynamic response characteristic data, combine the pipe network material state distribution and the water source pollution characteristic distribution, and calculate the water source risk coefficient through three-dimensional coupling; based on the water source risk coefficient and the pipe network GIS data, calculate the seepage critical threshold, when the real-time risk probability exceeds the seepage critical threshold, generate a partition pressure reduction instruction, and execute the state record to the operation log record; write the operation log record and the water source risk coefficient into the blockchain for notarization, and update the pipe network digital twin model, the specific steps are as follows, write the operation log record and the water source risk coefficient into the smart contract of the blockchain for notarization, and generate a notarization identifier; trigger the quantum entanglement state based on the notarization identifier, and update the pipe network digital twin model through the quantum-fluid holographic mapping method.

2. The artificial intelligence-based intelligent water management method of claim 1, wherein: The specific steps of generating a high-risk warning signal are as follows, initialize the pipe network digital twin model, and load the pre-stored pipe network GIS data from the blockchain distributed storage node to analyze the topological structure of the pipe network GIS data; based on the topological structure of the pipe network GIS data, collect pipe pressure data, water quality parameters and user water consumption; input the pipe pressure data, water quality parameters and user water consumption into the risk assessment model to calculate the real-time risk probability through the physical-data dual driving algorithm; when the real-time risk probability exceeds the warning threshold, trigger the high-risk warning signal. 3.The AI-based intelligent water management method of claim 2, wherein: The specific steps of calculating the seepage critical threshold based on the water source risk coefficient and the pipe network GIS data are as follows, map the water source risk coefficient to the topological structure of the pipe network GIS data to generate a dynamic risk field distribution; Performing three-dimensional space integral operation on the dynamic risk field distribution to generate an initial seepage benchmark value; Using the water source risk coefficient to dynamically correct the initial seepage benchmark value, and combining the structure compensation factor of the pipe network GIS data to calculate the seepage critical threshold.

4. The artificial intelligence-based intelligent water management method of claim 3, wherein: When the real-time risk probability exceeds the seepage critical threshold, a partition pressure reduction instruction is generated, and a state record is executed to an operation log record, and the specific steps are as follows, When the real-time risk probability exceeds the seepage critical threshold, a partition pressure reduction instruction is generated based on the topological structure of the pipe network GIS data and the pipe pressure data, and pressure regulation is performed according to the partition pressure reduction instruction; Collecting equipment state and pressure change data during pressure regulation, and combining the partition pressure reduction instruction to integrate into an operation log record.

5. An intelligent water management system based on artificial intelligence, based on the artificial intelligence-based intelligent water management method of any one of claims 1-4, characterized in that: It includes a twin initialization module, a stable voltage control module, a parameter optimization module, a seepage early warning module and a blockchain storage module, The twin initialization module is used to initialize the pipe network digital twin model, load the pre-stored pipe network GIS data, and input the real-time risk probability calculated by the risk assessment model by real-time collection of pipe pressure data, water quality parameters and user water consumption, and generate a high-risk early warning signal when the real-time risk probability exceeds the early warning threshold; The stable voltage control module is used to calculate the differential order parameter and integral gain parameter based on the high-risk early warning signal through the pipe material state fractional order controller, generate a stable voltage control instruction and execute it, and extract the control error signal during the stable voltage process; The parameter optimization module is used to input the control error signal and the water source type label into the meta-reinforcement learning model to dynamically optimize the differential order parameter and the integral gain parameter, and obtain the water source risk coefficient; The seepage early warning module is used to calculate the seepage critical threshold based on the water source risk coefficient and the pipe network GIS data, and generate a partition pressure reduction instruction when the real-time risk probability exceeds the seepage critical threshold, and execute a state record to an operation log record; The blockchain storage module is used to write the operation log record and the water source risk coefficient into the blockchain for storage, and update the pipe network digital twin model. 6.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the intelligent water management method based on artificial intelligence in any one of claims 1-4.

7. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the intelligent water management method based on artificial intelligence in any one of claims 1-4.

Images