Municipal water distribution pipe network data analysis system and method and electronic equipment

Abstract

The invention relates to a municipal water distribution pipe network data analysis system and method and electronic equipment, and relates to the technical field of municipal engineering technology.The municipal water distribution pipe network data analysis system comprises a data acquisition and preprocessing module, a state recognition module, an anomaly detection module and an intelligent scheduling module; and the long and short-term memory network is used for analyzing the pipe network time sequence data and predicting the future pipe network state. According to the method, dynamic prediction of the pipe network state is realized by introducing the gated neural network, anomaly recognition is performed in combination with the statistical residual threshold, and the prediction precision and the anomaly detection capability are effectively improved. Meanwhile, self-adaptive generation of a scheduling strategy is realized by adopting a particle swarm optimization algorithm, and water supply safety and energy efficiency balance are ensured. The modular design of the electronic equipment integrates data acquisition, intelligent analysis and control output, an integrated closed-loop system is constructed, and high responsiveness and deployment flexibility are achieved.

Description

Technical Field

[0001] The present application relates to the field of municipal engineering technology, and in particular to a municipal water distribution network data analysis system, method, and electronic equipment. Background Art

[0002] In the operation and management of urban water distribution systems, traditional condition monitoring methods are often based on fixed thresholds, using empirically defined ranges to determine anomalies in sensor data. This approach is incapable of responding to sudden disturbances or seasonal fluctuations, often leading to misjudgments or missed detections, impacting the accuracy of scheduling decisions. In particular, in real-world scenarios with large volumes of data and frequent changes, static rules struggle to adapt to the dynamic nature of complex system operations.

[0003] Existing forecasting methods typically use simple models such as linear regression and moving average to construct time series trends. These methods have limited ability to handle nonlinear relationships between variables and are unable to accurately capture the multivariable coupling behavior of pipeline network operations. Faced with high-frequency, asynchronous multi-source data, forecast results lag and often deviate significantly from actual conditions, making them difficult to use for forward-looking scheduling.

[0004] In the dispatch and execution phase, many systems still rely on pre-set policy libraries, rigid control logic, and a lack of data-driven dynamic adjustment mechanisms. Dispatching instructions often lag behind abnormal changes, making it difficult to achieve real-time trade-offs between multiple objectives, such as energy optimization and water pressure stability. This rigidity of policy limits overall operational efficiency.

[0005] At the system deployment level, there's a lack of unified design for communication and collaboration between different functional units. Data collection, analysis, and control equipment are often provided by different vendors, with inconsistent interface protocols, making system integration difficult. This results in low collaboration efficiency between modules, high maintenance costs, and limited reliability and real-time performance to meet the requirements for intelligent urban water supply upgrades. Summary of the Invention

[0006] The purpose of this application is to provide a municipal water distribution network data analysis system, method and electronic equipment to solve the problems of inaccurate state prediction, untimely anomaly identification and unintelligent scheduling response in existing water distribution networks.

[0007] In a first aspect, the municipal water distribution network data analysis system provided in this application adopts the following technical solution: the municipal water distribution network data analysis system includes a data acquisition and preprocessing module for collecting data from various sensors, and cleaning, standardizing and fusing the data to generate standardized data; A state recognition module, based on a deep learning algorithm, receives the standardized data and outputs a state prediction result of the pipeline network; an anomaly detection module receives the state prediction result of the pipeline network and the real-time observation data, compares the difference between the two, and generates an anomaly alarm; An intelligent scheduling module receives the abnormal information and the pipe network state prediction result output by the abnormality detection module, and generates a scheduling optimization scheme for water resources. The deep learning algorithm is a long short-term memory network, which is used for analyzing pipe network time series data and predicting future pipe network states.

[0008] Preferably, the data acquisition and preprocessing module comprises: A data cleaning unit is configured to perform interpolation processing on the collected data to fill in missing values in the data. A data standardization unit is configured to perform standardization processing on the data, and convert the data to a unified scale by using a Z-score standardization method. A data fusion unit is configured to fuse data from different sensors by using a Kalman filtering algorithm.

[0009] Preferably, the state recognition module is modeled by using a long short-term memory network, and the long short-term memory network comprises: A forget gate is configured to control the forgetting degree of input information. An input gate is configured to control the writing degree of new information. An output gate is configured to control the output of the hidden state at the current time. The long short-term memory network is trained by using a back propagation algorithm to optimize the parameters of the network.

[0010] Preferably, the abnormality detection module is based on a residual calculation method, and determines whether an abnormality exists by calculating the residual between the pipe network state prediction result and the actual observation data, specifically as follows: The residual is calculated, and if the residual exceeds a preset threshold, it is determined that an abnormal state exists. The threshold is dynamically calculated according to the mean and standard deviation of the residual.

[0011] Preferably, the intelligent scheduling module schedules water resources by using a particle swarm optimization algorithm, and the update formula of the particle swarm optimization algorithm is as follows: wherein, is the speed of the i-th particle at the k-th generation, is the current position of the i-th particle, is the historical best position of the particle itself, g * is the global optimal position, c1 and c2 are acceleration constants, r1 and r2 are random numbers, and w is the inertia weight.

[0012] In a second aspect, the municipal water distribution pipe network data analysis method provided by the present application adopts the following technical solution: A municipal water distribution network data analysis method comprises the following steps: S1. Collecting data from multiple sensors and performing cleaning, standardization and fusion processing on the data; S2. Analyzing the processed data based on a long short-term memory network to predict future network state; S3. Comparing the difference between the network state prediction result and the actual observation data, and judging whether there is an anomaly by calculating the residual error; S4. Generating a water resource scheduling optimization scheme according to the anomaly detection result and the network state prediction result.

[0013] Preferably, the cleaning step includes filling in missing data using an interpolation method, the standardization step includes converting data using a Z-score standardization method, and the fusion step includes fusing data by a Kalman filtering algorithm.

[0014] Preferably, the long short-term memory network is used to analyze network time series data, the training process of the network optimizes network parameters through a back propagation algorithm, and the network controls information transmission and update through a gating mechanism.

[0015] In a third aspect, the municipal water distribution network data analysis electronic device provided by the present application adopts the following technical solution: a municipal water distribution network data analysis electronic device comprises: a data interface module for accessing data from sensors or other protocol devices, supporting access of multiple communication protocols; a cache and processing unit for cleaning and synchronously processing collected data; an edge computing module for model inference and prediction based on input data, outputting a prediction result of network state; an anomaly analysis and optimization control module for identifying anomalies by calculating the residual error between the predicted value and the actual value, and generating an optimized scheduling strategy in combination with a particle swarm optimization algorithm; a communication and control output module for transmitting the optimized scheduling command to an execution terminal for control.

[0016] Preferably, the data interface module supports real-time access and protocol conversion of sensor data, supports compatible access of multiple industrial communication protocols, and ensures seamless connection of data interaction between devices.

[0017] In summary, the present application has at least one of the following beneficial technical effects: 1. This invention utilizes a state prediction model based on a gated neural network (such as LSTM) and a multi-dimensional time series input fusion mechanism to dynamically estimate the future operating state of the pipeline network, achieving the technical benefits of early warning of operational anomalies and rationally arranging dispatch resources. Compared to existing prediction solutions that rely on static rules or traditional regression methods, this solves the problems of delayed prediction and slow response in nonlinear fluctuation scenarios.

[0018] 2. This invention significantly improves the robustness and positioning accuracy of anomaly detection by introducing a statistical residual dynamic threshold judgment mechanism and combining spatial structure information to cluster and associate outliers. Compared with traditional anomaly recognition technologies based on fixed thresholds or single-point judgments, it effectively addresses the shortcomings of high false alarm rates and lack of spatial perception.

[0019] 3. This invention utilizes a particle swarm optimization algorithm to construct a water resource scheduling optimization module, achieving comprehensive optimization scheduling objectives: supply and demand balance, energy consumption control, and anomaly avoidance. Based on the flexible combination of scheduling constraints, the system can dynamically adjust control strategies. Compared to existing fixed scheduling strategies or manual adjustment mechanisms, this solves the problems of slow response and rigid scheduling.

[0020] 4. The analytical electronic equipment provided by this invention utilizes a modular design to integrate multiple sensor data access, status prediction, anomaly detection, and control output functions, forming a complete closed loop. Compared to traditional hardware solutions with decentralized deployment and independent operation of multiple modules, this solution addresses the technical shortcomings of poor system interface compatibility, low device integration, and complex on-site deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the overall system architecture of the present invention; Figure 2 is a flow chart of the data analysis method of the present invention; Figure 3 This is a diagram showing the functional modules of the electronic device of the present invention; Figure 4 This is the scheduling optimization flow chart of the present invention (particle swarm schematic). DETAILED DESCRIPTION

[0022] The following is combined with Figure 1 -Attached Figure 4 , further details of this application are given.

[0023] Example 1: Municipal water distribution network data analysis system, please refer to the attached Figure 1 ,include, The data acquisition and preprocessing module is used to collect data from various sensors, clean, standardize and fuse the data, and generate standardized data; In this paper, the data acquisition and preprocessing module is a crucial component of the municipal water distribution network data analysis system, responsible for the real-time acquisition, cleaning, standardization, and integration of sensor data. This module provides high-quality input data for subsequent state recognition, anomaly detection, and scheduling optimization modules, and therefore plays a crucial role in the overall system.

[0024] Typically, the data acquisition and preprocessing module involves several key steps: first, collecting data from a variety of sensors; second, processing the raw data to remove missing values ​​and outliers; and finally, applying standardized methods to the data and combining data from different sensors through multi-source data fusion to ensure consistency and reliability. This series of steps ensures that the system receives high-quality data input to support subsequent deep learning model training and anomaly detection.

[0025] In this embodiment, the municipal water distribution network data acquisition module is primarily responsible for collecting real-time operational data from various sensors. These sensors are typically installed at key locations in the network, including water supply inlets, pipeline branches, and outlets. Data comes from a wide range of sources, including but not limited to flow meters, pressure sensors, temperature sensors, turbidity sensors, and water quality sensors.

[0026] In some embodiments, the data collection system transmits data collected by each monitoring node to a central server or cloud platform via wireless communication protocols (such as LoRa and NB-IoT). This type of network communication technology supports low power consumption and wide coverage, adapting to the complex geographical distribution and needs of municipal pipe networks.

[0027] The data acquisition and preprocessing module includes a data cleaning unit, which interpolates the collected data and fills in missing values. In this embodiment, data cleaning is a key step in preprocessing the collected raw data. Sensor data often contains noise, missing values, or outliers. These inaccuracies directly affect the accuracy of subsequent analysis and prediction results. Therefore, data cleaning is a very important step.

[0028] For missing data, this implementation adopts interpolation method to fill it. Generally, missing data can be filled by linear interpolation or spline interpolation. The basic formula of linear interpolation method is: Where x(t) is the interpolation result at time t, x(t1) is the known data point, t0 and t1 are the timestamps of the known data points, and t is the time point to be interpolated.

[0029] The system identifies outliers by setting reasonable upper and lower limits. Data values ​​outside this range are considered abnormal and require removal or further processing. In some embodiments, outlier detection and removal utilizes a rule-based algorithm that incorporates historical equipment operating data and expert experience to set upper and lower limits.

[0030] The data acquisition and preprocessing module includes a data standardization unit, which standardizes data using the Z-score method to convert the data to a uniform scale. Data standardization is another important step in data preprocessing. In municipal water distribution networks, measurement data from different sensor types may have different dimensions and scales. Therefore, standardization is necessary to ensure that all input data is compared and analyzed at the same scale.

[0031] This implementation uses the Z-score standardization method to uniformly standardize the data to ensure that each type of data has a consistent impact on the subsequent model. The Z-score standardization formula is as follows: Among them, x is the original data, μ is the mean of the data, σ is the standard deviation of the data, and x′ is the standardized data.

[0032] Through standardization, the data will have the same mean and standard deviation, so that the sensor data can be processed at the same scale, thus avoiding the uneven impact of data of different scales on the model.

[0033] The data acquisition and preprocessing module also includes a data fusion unit, which uses a Kalman filter algorithm to fuse data from different sensors. In municipal water distribution networks, due to the simultaneous operation of multiple sensors from different manufacturers and types, data discrepancies between sensors are inevitable. Data fusion technology can effectively integrate this data from different sensors, eliminating discrepancies and improving the reliability and accuracy of the entire system.

[0034] This implementation uses a Kalman filter to fuse multi-source data. The Kalman filter takes into account the measurement error of each sensor and the dynamic characteristics of the system, and performs a weighted estimate on the data of multiple sensors to obtain more accurate results. The recursive formula of the Kalman filter is as follows: P k =AP k-1 A T +Q; in, represents the state estimate at time k, represents the real state at time k-1, A is the state transfer matrix, B is the control input matrix, which represents the influence of external control on the system state, u k represents the control input vector, the control signal at time k, P k is the covariance matrix, P k-1 is the state estimation error covariance at time k-1, Q is the process noise covariance, A T is the transpose of the state transfer matrix A.

[0035] The basic idea of ​​the Kalman filter is to update the optimal estimate of the system state in real time by weightedly fusing a priori estimates with actual measurement data. For municipal water distribution network data in a multi-sensor environment, the Kalman filter can effectively reduce the errors of individual sensors, thereby improving data accuracy and consistency.

[0036] After data cleaning, standardization, and fusion, the resulting data serves as input for subsequent modules such as state recognition and anomaly detection. The data obtained through these steps has high accuracy and consistency, providing effective support for subsequent deep learning models and optimization algorithms.

[0037] Specifically, after data cleaning and standardization, various sensor data types are fused into a unified, standardized dataset, providing a stable and consistent data foundation for subsequent deep learning models. This fused data effectively improves data credibility in multi-sensor environments, reducing the impact of errors caused by sensor bias, ensuring the system can accurately predict pipeline network status and promptly identify potential problems.

[0038] In this embodiment, the design of the data acquisition and preprocessing module ensures the data quality of the municipal water distribution network during operation. Through real-time data collection, cleaning, standardization, and integration, the module provides high-quality, reliable data input for subsequent status identification, anomaly detection, and intelligent scheduling. These technical details provide a solid foundation for the efficient operation of the system and further promote the intelligent and automated management of municipal water services.

[0039] The state recognition module, based on a deep learning algorithm, receives standardized data and outputs the state prediction results of the pipeline network; The state recognition module plays a key role in the municipal water distribution network data analysis system. Its primary task is to analyze and predict the network's operating status based on real-time sensor data using deep learning techniques (particularly long short-term memory networks (LSTMs)). Accurate state recognition not only provides real-time feedback on the network's operating status but also predicts potential risks, providing a basis for subsequent anomaly detection and scheduling optimization.

[0040] Typically, the state recognition module receives high-quality, standardized data from the data acquisition and preprocessing module and feeds this data into the LSTM model for processing. The LSTM model is particularly well-suited for analyzing time series data, capturing long-term dependencies within the data and adapting to dynamic changes in the network's operating status. This model enables the system to accurately predict the network's operating status and identify future trends.

[0041] In this embodiment, the state recognition module uses a deep learning-based LSTM model for data analysis and state prediction. The LSTM model, through its unique gating mechanism, can effectively handle the long-term dependencies existing in pipeline network time series data.

[0042] Specifically, the core structure of the LSTM model includes a forget gate, an input gate, candidate memory cells, and an output gate. These gating mechanisms dynamically adjust the information flow to determine which information should be remembered and which should be forgotten.

[0043] The function of the forget gate is to determine how much historical information should be retained in the current state. The specific formula is as follows: t =σ(W f ·[h t-1 ,x t ]+b f ); Among them, f t is the output of the forget gate, h t-1 is the hidden state of the previous moment, x t is the input data at the current moment, W f and b f are the weight matrix and bias term respectively, and σ is the sigmoid activation function.

[0044] The input gate controls the impact of the current input information on the memory unit. The formula is as follows: i t =σ(W i ·[h t-1 ,x t ]+b i ); Among them, i t is the output of the input gate, W i and b i are the weight matrix and bias term respectively.

[0045] The candidate memory unit calculates the new candidate memory value based on the current input and the state of the previous moment: in, is the output of the candidate memory unit, W C and bC are the weight matrix and bias term, and tanh is the hyperbolic tangent activation function.

[0046] The update of the memory cell combines the outputs of the forget gate and the input gate, and finally obtains the memory value at the current time: where C t is the memory cell at the current time, C t-1 is the memory cell at the previous time.

[0047] The output gate controls the output of the hidden state at the current time, and the formula is as follows: o t = σ(W o · [h t-1 , x t ] + b o ); h t = o t · tanh(C t ); where o t is the output of the output gate, and h t is the hidden state at the current time.

[0048] In this embodiment, the training of the LSTM model uses a supervised learning method. The input of the model is the historical data of the municipal water distribution network, which includes monitoring values such as flow, pressure, temperature, etc. By inputting these data into the LSTM network, the model is trained through the backpropagation algorithm (Backpropagation Through Time, BPTT), gradually adjusting the weights and biases in the network to minimize the error between the predicted results and the actual values.

[0049] Generally, the LSTM network will learn data in multiple time steps until the error between the predicted results of the model and the actual values is small enough. During the training process, the Adam optimizer is used to optimize the model parameters, and the gradient of each layer is calculated to update the weight values, thereby improving the accuracy of the model.

[0050] In one possible implementation, the loss function of the LSTM model can be defined as the mean square error (MSE): where L is the loss function, N is the number of samples, y t is the actual value, is the predicted value of the model.

[0051] Once the model is trained, the state recognition module can make real-time predictions for new input data. By receiving and processing data from sensors, the LSTM model can predict the future state of the pipeline network. For example, it can predict the changing trends of parameters such as pressure and flow in a specific section of the pipeline network over a period of time. This provides basic data for the anomaly detection module, enabling it to promptly identify potential pipeline network failures or anomalies.

[0052] The LSTM state recognition model described in this example can be used not only to predict the state of a single pipeline network, but also to combine data from multiple monitoring points to perform global state prediction. For example, by analyzing real-time data from multiple pipeline segments, the system can predict load changes across the entire pipeline network and determine in advance whether there is a risk of water shortage or pressure anomalies.

[0053] Alternatively, the LSTM model can be combined with other time series data analysis methods (such as ARIMA and XGBoost) to further improve forecast accuracy and model robustness. Specifically, ensemble learning can be used to combine multiple forecasting methods, leveraging the strengths of each model to provide more stable and accurate forecast results.

[0054] In practical applications, the performance of LSTM models may be affected by multiple factors, including data quality, training data volume, and computing resources. Therefore, in some embodiments, prediction accuracy can be further improved by optimizing the model structure and hyperparameters (such as the number of network layers and learning rate). Furthermore, distributed computing methods can be used to train LSTM networks in parallel on multiple machines to accelerate the training process and meet the needs of large-scale municipal pipe networks.

[0055] As demonstrated in this example, the state recognition module, based on the LSTM model employed in deep learning, effectively processes time-series data from municipal water distribution networks and accurately identifies their operational status. Through its gating mechanism and long-term memory, the module successfully captures long-term dependencies in network data, providing reliable predictive support for subsequent anomaly detection and intelligent scheduling. Furthermore, through flexible model training and optimization, the module can be widely applied to municipal water management systems of varying scales, providing highly accurate network status analysis.

[0056] The anomaly detection module receives the network status prediction results and real-time observation data, compares the differences between the two, and generates an anomaly alarm; In the municipal water distribution network data analysis system, the core task of the anomaly detection module is to monitor real-time network status and promptly identify and report potential anomalies. Based on the difference between predicted results and actual monitored data, this module calculates and detects residual values ​​to determine whether there are any anomalies in network operation. This process not only ensures early warning of potential problems but also provides timely feedback for scheduling optimization and troubleshooting.

[0057] Typically, the anomaly detection module relies on the previous state recognition module, which predicts the network's status. By comparing the predicted values ​​with the observed values, the anomaly detection module determines whether anomalies exist. To achieve efficient anomaly detection, this module employs a residual analysis-based detection method. This method uses threshold settings and residual value comparisons to accurately identify potential faults or anomalies in the network.

[0058] In this embodiment, the anomaly detection module's basic process includes residual calculation and threshold determination. First, the module receives predicted network status data from the state identification module and compares it with the actual sensor observations. The residual value represents the difference between the predicted value and the actual value. By calculating this residual, the system can promptly determine whether an anomaly exists.

[0059] In some embodiments, the basic formula for residual calculation is: Among them, r t represents the residual at time t, y t is the actual observed value, is the model's predicted value.

[0060] Residual values ​​represent the deviation between the system's predictions and actual data and are the basis for anomaly detection. Excessive residual values ​​indicate a significant gap between the system's predictions and actual conditions, which typically indicates an anomaly in some part of the network (e.g., excessive pressure or abnormal flow).

[0061] The determination of residual values ​​is the core of the anomaly detection module. To effectively identify anomalies, a dynamic threshold needs to be set. This threshold is adaptively adjusted based on the statistical characteristics of historical data. Specifically, the threshold can be calculated based on the mean and standard deviation of the residuals: threshold=μ r +α·σ r ; Among them, μ r and σ r The thresholds are dynamically adjusted to ensure the system can adapt to different operating conditions and changes in the network status.

[0062] When the residual r at a certain moment t When the dynamic threshold is exceeded, the anomaly detection module triggers an alarm mechanism. This threshold-based detection method ensures that the system can respond quickly when an anomaly occurs.

[0063] For example, if the flow rate in a particular section of pipeline deviates significantly from the predicted value, the system will issue an alarm, notifying operations and maintenance personnel to inspect that section of the pipeline network for faults, leaks, or other issues. The alarm typically includes the type of anomaly, its location, and the time of occurrence, allowing operations and maintenance personnel to promptly locate and address the problem.

[0064] In practical applications, a single threshold detection may be affected by multiple factors, such as sensor accuracy or changes in environmental conditions. To this end, in some embodiments, the anomaly detection module can also be optimized by combining multiple technical means.

[0065] Alternatively, the anomaly detection module can combine data from multiple sensors for more detailed anomaly analysis. For example, it can compare residual differences between different monitoring points to determine whether cross-regional anomalies exist. This approach, through multi-channel data fusion, enables more comprehensive and accurate anomaly detection.

[0066] Specifically, assuming there are multiple monitoring points i, the corresponding prediction values ​​are The actual value is y i,t , then the residual of each monitoring point is: Among them, r i,t Represents the residual value of the i-th node or variable at time t, which represents the absolute error between the true value and the predicted value; y i,t represents the actual observation value of the i-th node at time t; Represents the predicted value of the i-th node at time t.

[0067] The overall residual error of the entire system can be calculated by weighted averaging, which comprehensively considers the data from different monitoring points. This multi-channel detection method can better detect some localized and difficult-to-detect anomalies.

[0068] Specifically, in some embodiments, the calculation of dynamic thresholds relies not only on the mean and standard deviation of the residuals but can also be adjusted based on the system's historical operating data and external environmental factors (such as weather and holidays). For example, the pipe network system may experience significant flow fluctuations during certain seasons (such as the summer peak season), and the system can appropriately amplify the threshold based on this characteristic. This can avoid overly sensitive alarms and reduce the occurrence of false alarms.

[0069] In some embodiments, the anomaly detection module can further incorporate pattern recognition algorithms, combining anomaly patterns in historical data to automatically identify possible fault types within the pipeline network. By combining deep learning with clustering algorithms, the system can identify different types of anomalies (such as pipeline ruptures and pump station failures) and generate corresponding alarm messages based on the identification results. This not only improves the accuracy of anomaly detection but also provides more detailed troubleshooting information for operations and maintenance personnel.

[0070] The anomaly detection module works closely with other modules to support the intelligent operation of the entire municipal water distribution network data analysis system. Specifically, when the anomaly detection module triggers an alarm, the intelligent scheduling module adjusts the water resource scheduling strategy based on the alarm information to avoid water waste or the expansion of pipeline network failures.

[0071] As a possible implementation method, the collaboration between the anomaly detection module, the state recognition module and the scheduling module can be achieved through a centralized control platform, which monitors the operation of each module in real time and provides decision support when anomalies occur.

[0072] The description of this embodiment demonstrates the crucial role of the anomaly detection module in municipal water distribution network data analysis systems. Through residual calculation and dynamic threshold determination, the system can detect anomalies in the network in real time, providing a crucial basis for subsequent troubleshooting and scheduling optimization. Furthermore, the anomaly detection module's flexibility and scalability enable it to adapt to water distribution network systems of varying sizes in different cities, providing accurate anomaly identification and feedback.

[0073] Please see the attached Figure 4 ,The intelligent scheduling module receives the abnormal information and pipe network status prediction results output by the anomaly detection module, and generates a water resource scheduling optimization plan; The deep learning algorithm is a long short-term memory (LSTM) network, which is used to analyze pipeline network time series data and predict future pipeline network status.

[0074] The intelligent scheduling module is a core component of the municipal water distribution network data analysis system, responsible for optimizing water resource allocation based on real-time network status and operational needs. By combining the outputs of the status recognition module and the anomaly detection module, this module automatically adjusts the water resource scheduling strategy to ensure efficient network operation and avoid energy waste and leakage. The intelligent scheduling module's goal is to minimize energy consumption and maximize water resource utilization.

[0075] Generally, the intelligent scheduling module relies on the results of the previous anomaly detection and state recognition modules. The anomaly detection module provides the scheduling module with real-time pipeline network fault alarm information, while the state recognition module provides predicted data on future pipeline network conditions. Based on this information, the intelligent scheduling module calculates the optimal scheduling plan and adjusts operations such as pump station startup and shutdown and valve opening and closing, thereby improving overall operational efficiency.

[0076] In this embodiment, the intelligent scheduling module mainly optimizes the water resource scheduling of the pipe network through the particle swarm optimization (PSO) algorithm. The PSO algorithm is an optimization algorithm based on swarm intelligence, which simulates the information exchange and collaboration between individuals in a swarm to find the global optimal solution.

[0077] The goal of scheduling is to minimize the energy consumption and leakage of the entire system. Specifically, the objective function can be expressed as: Among them, J is the optimization target, c i represents the energy consumption cost of the i-th pumping station, x i is the scheduling decision variable (such as start and stop time), λ is the penalty factor, φ j Represents the constraints (e.g., flow rate, pressure limit, etc.). Minimizing this objective function is the core task of scheduling optimization.

[0078] Alternatively, the intelligent scheduling module uses a particle swarm optimization algorithm to solve the aforementioned objective function. The PSO algorithm simulates a swarm of particles searching for an optimal solution in the solution space, gradually converging to the global optimal solution. At each time step, particles search the solution space by updating their velocity and position, adjusting their positions based on the current particle's position, velocity, and the particle's historical optimal position.

[0079] The particle update formula of PSO is as follows: in, is the velocity vector of the i-th particle in the k-th generation; is the velocity vector of the i-th particle in the k+1th generation; is the position vector of the i-th particle in the k-th generation; is the position vector of the k+1th particle in the kth generation; is the best historical position of the particle itself; g * is the global optimal position; c1 and c2 are acceleration constants; r1 and r2 are random vectors between [0,1] (usually each dimension is independently and uniformly distributed) to increase the randomness of the search; w is the inertia weight, which controls the influence of the current velocity of the particle on the velocity of the next step.

[0080] Through iterative updates, the particle swarm can continuously adjust its position in the search space and eventually converge to the optimal solution. In this way, the particle swarm can effectively search for a scheduling solution that minimizes energy consumption and meets the network operation requirements.

[0081] In pipeline network scheduling, in addition to minimizing energy consumption and leakage, various constraints must be considered, such as the flow rate and pressure of each pumping station. To ensure the feasibility of the scheduling plan, this embodiment adopts a soft constraint approach, which adjusts the plan by weighting the penalty terms for constraint violations.

[0082] For example, assuming the flow constraint is φ1 and the pressure constraint is φ2, the corresponding penalty factors λ1 and λ2 are added to the objective function so that when the constraints are violated, the value of the objective function increases, thus forcing the algorithm to consider the constraints during the optimization process: Where J is the objective function value, which represents the quantitative result of the comprehensive optimization goal and needs to be minimized or maximized; n is the number of variables, that is, the number of decision variables involved in the optimization problem; x i : The i-th decision variable represents the object to be optimized (such as resource allocation, flow, time, etc.); c i : The unit cost or weight coefficient of the i-th variable, which measures the direct contribution of the variable to the target; λ1, λ2: weight coefficients used to balance the importance between the main objective and the constraint penalty term; φ1, φ2: Constraint penalty functions or additional indicators (such as deviation constraints, systematic risk, volatility, etc.) used to guide the solution to meet additional requirements.

[0083] In practice, the intelligent dispatching module not only relies on historical and real-time monitoring data but also dynamically adjusts to meet varying operational needs. Specifically, when dealing with large-scale pipe networks, the intelligent dispatching system may automatically adjust its dispatching strategy based on factors such as seasonal changes and fluctuations in water demand during holidays. By incorporating external environmental data, such as weather changes and water demand forecasts, the intelligent dispatching system can adapt to varying loads in real time, enabling more flexible dispatching.

[0084] For example, in some embodiments, the scheduling system automatically increases the frequency of water pump operation during peak summer demand, or reduces the energy consumption of pump stations when water demand is low. By combining weather data and water use forecast models, the system can achieve more precise scheduling strategies.

[0085] The intelligent dispatch module not only provides optimized dispatch solutions during network operation but also further optimizes dispatch strategies based on real-time feedback. In one possible implementation, the dispatch module regularly checks the discrepancies between the system's actual operating results and predicted results, and uses this discrepancy to adjust future dispatch. For example, if leakage increases or energy consumption is too high during dispatch, the system can fine-tune dispatch parameters to minimize these adverse effects.

[0086] This feedback mechanism enables the intelligent scheduling system to self-optimize and adapt to the ever-changing pipeline network environment, thereby ensuring the operating efficiency of the entire system and maximizing resource utilization.

[0087] The intelligent scheduling module is closely linked to the state recognition module and the anomaly detection module. When the anomaly detection module detects an anomaly in the pipeline network, the scheduling module responds to the alarm and makes appropriate scheduling adjustments. For example, if a pipeline leaks or overpressure occurs, the scheduling module can temporarily adjust valve openings or shut down pump stations to prevent the problem from escalating.

[0088] At the same time, the intelligent scheduling module can also carry out long-term scheduling planning based on the future pipeline load changes predicted by the state recognition module to avoid resource waste or system overload due to load fluctuations.

[0089] As demonstrated in this example, the intelligent scheduling module, through its particle swarm optimization algorithm, constraint processing, and dynamic feedback mechanism, enables intelligent, optimized scheduling of the municipal water distribution network. Through this module's real-time scheduling, the system not only reduces energy consumption and water loss but also automatically adjusts scheduling strategies based on varying network conditions and external environments. The efficiency and flexibility of the intelligent scheduling module enable the entire municipal water distribution network to maintain efficient and stable operation under a variety of operating conditions.

[0090] Example 2: A method for analyzing municipal water distribution network data, see the attached Figure 2 ,include: S1. Collect data from multiple sensors and clean, standardize, and fuse the data; the cleaning step includes using interpolation methods to fill missing data, the standardization step includes using Z-score standardization method to convert data, and the fusion step includes fusing the data through the Kalman filter algorithm.

[0091] In the municipal water distribution network data analysis method provided by this invention, data collection and preprocessing, as the initial step in the process, have a direct impact on the accuracy and effectiveness of subsequent core modules such as data modeling, state recognition, anomaly detection, and scheduling optimization. Therefore, to ensure data quality and processing efficiency, a systematic technical design is implemented in step S1 for the collection, cleaning, standardization, and fusion processing of sensor data. This step not only marks the starting point of the entire method's logical chain but also lays the foundation for the stable operation of subsequent modules.

[0092] Typically, municipal water distribution networks generate data from a wide variety of sensors, including pressure sensors, flow sensors, and water quality testing equipment. These sensors vary in sampling frequency, unit system, and data format. Directly using these sensors without standardized processing can easily lead to data error propagation, impacting subsequent prediction and optimization processes.

[0093] In this embodiment, step S1 includes the following processing flow: multi-source data collection, noise removal, unit and format unification, missing value filling, standardization conversion and feature fusion.

[0094] Data collection: Specifically, the system first collects pressure P periodically by accessing the remote transmission interface of various sensors. t , flow Q t , turbidity T t , residual chlorine concentration C t etc. original data.

[0095] Where t represents the sampling time step, P t Unit is MPa, Q t Unit is m 3 / h,T t The unit is NTU, C t The unit is mg / L.

[0096] The sampling frequency can be set to every 5 minutes or every 15 minutes, and the specific frequency is configured by the system operation strategy.

[0097] As an option, in this embodiment, a local outlier factor (LOF) algorithm is used to detect outliers in response to possible abnormal values ​​and missing values ​​in the collected data.

[0098] If a data point x i If the local density of a point deviates greatly from its neighboring points, its LOF value is much higher than 1, and the system marks the point as an outlier.

[0099] At the same time, linear interpolation is used to fill in short-term missing data: Among them, x tIndicates missing points, x t-1 、x t+1 They are the valid data at the adjacent moments before and after respectively.

[0100] In order to eliminate the influence of the order of magnitude differences between different physical quantities, the Z-score normalization method is introduced in this embodiment to uniformly process all variables.

[0101] The normalization formula is as follows: Among them, z i,t represents the normalized value of the i-th sensor at time t, x i,t is the original sampling value, μ i is the historical mean of the variable, σ i is its standard deviation.

[0102] In some embodiments, if the variable has obvious periodicity, the system will update the mean and variance based on the sliding window to adapt to the dynamic changes in the system state.

[0103] Generally speaking, in order to enhance the feature representation capability of subsequent models, it is necessary to fuse multiple types of sensor data.

[0104] In one possible implementation, the system horizontally concatenates multiple variables at the same geographic location to form a multidimensional feature vector at time t: X t =[z P,t ,z Q,t ,z T,t ,z C,t ]; Among them, X t ∈R 1×d , d represents the feature dimension, which is equal to the number of variables.

[0105] If there are multiple monitoring points, the following feature matrix is ​​constructed: in, Represents the feature vector of the i-th monitoring point, and n is the total number of monitoring points.

[0106] This feature matrix is ​​then provided as input to the state recognition and anomaly detection module to achieve cross-space and cross-variable information collaboration.

[0107] In some embodiments, considering that different sensors may have inconsistent sampling times, this embodiment adopts a time alignment strategy based on a master clock signal to uniformly map all data to a standard time axis.

[0108] If there is a timing offset in the sensor data, the system can achieve data synchronization through time interpolation or sliding window resampling methods.

[0109] In another implementation, the data fusion process is not limited to numerical concatenation, but can also extract the main eigenvectors to reduce redundancy through dimensionality reduction methods such as principal component analysis (PCA).

[0110] For example, the standardized feature matrix X∈R T×d Input to the PCA transformation module to obtain the principal component representation: Z = X·W; Among them, Z is the feature matrix after dimensionality reduction, W∈R d×k is the eigenvector matrix of the first k principal component directions.

[0111] This method can effectively improve the stability and computational efficiency of subsequent models when the dimension is too high or there is multicollinearity.

[0112] Through the above technical solutions, this embodiment provides a complete and highly scalable implementation path for data collection, cleaning, standardization, and fusion processing. All involved mathematical processing procedures and parameter definitions are clearly listed to avoid technical omissions. This step serves as the prerequisite for the entire municipal water distribution network analysis method and plays a key supporting role in ensuring data quality, standardizing input formats, and improving downstream model effectiveness.

[0113] S2. Analyze the processed data based on a long short-term memory (LSTM) network to predict the future status of the pipeline network. The LSTM network is used to analyze pipeline network time series data. The network training process optimizes network parameters through a backpropagation algorithm, and the network uses a gating mechanism to control the transmission and update of information.

[0114] After cleaning, standardizing, and fusing data from multiple sensors, the system now has a unified input format for modeling. To further predict pipeline network operating trends and perceive their status, the analysis method proposed in this invention introduces time series modeling technology based on deep learning. In this process, a long short-term memory network (LSTM) is used to model and extract features from historical data sequences to achieve high-precision predictions of future pipeline network status. Step S2 is developed based on this context and constitutes the core prediction unit of this method, providing a dynamic reference for subsequent anomaly detection and intelligent scheduling.

[0115] Typically, the operational data of municipal water distribution networks exhibits distinct time series characteristics, with strong periodicity, trends, and short-term disturbances. Traditional linear models struggle to accurately capture these nonlinear dynamics. However, LSTM networks, an improved form of recurrent neural networks (RNNs), possess the ability to memorize long-term dependencies and effectively capture the complex temporal relationships between multidimensional variables such as water pressure and flow.

[0116] In this embodiment, the state prediction module in step S2 uses a multi-layer LSTM network as a modeling basis to perform in-depth analysis on the fused data formed in the above steps.

[0117] Specifically, the feature matrix obtained after step S1 is processed Window construction is performed on the time dimension to form a time series input tensor: X [t-w+1:t] ={X t-w+1 ,X t-w+2 ,…,X t}; Among them, w is the time window length, x t ∈R n×d , n is the number of monitoring points, and d is the feature dimension of each point.

[0118] In some embodiments, the input data is normalized and expanded into a three-dimensional tensor to meet the LSTM input format: X input ∈R w×n×d ; In one possible implementation, the LSTM network consists of three stacked layers. Each layer of units performs information transfer between time steps according to the following formula: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) o t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t *tanh(c t ) Where: xt ∈R n×d is the current input data, h t ∈R n×h is the current hidden state, h is the number of hidden layer units, c t ∈R n×h is the current unit state, f t ,i t , o t ∈R n×h They are forget gate, input gate and output gate respectively, W f , W i , W o , W c is the network weight matrix, b f , b i , b o , b c is the bias term, σ(·) represents the Sigmoid function, and tanh(·) is the hyperbolic tangent function.

[0119] The final hidden state vector h output by LSTM t As the high-order time series feature expression of the current time step, it is subsequently used to predict the status of the pipeline network.

[0120] Alternatively, the system uses a fully connected regression layer to map the output of the LSTM to future state predictions: in: represents the system state at the prediction time t+Δ, d′ is the prediction target dimension (such as future water pressure, flow, etc.), W out ∈R h×d′ is the output layer weight matrix, b out ∈R d′ is the bias term.

[0121] To improve the accuracy of multi-step predictions, in some embodiments, the system adopts a sequence-to-sequence (Seq2Seq) structure, which outputs the prediction results of a time sequence through a decoder structure: The decoder initialization state is passed from the encoder final hidden state to maintain temporal context continuity.

[0122] In this embodiment, the LSTM network uses mean square error (MSE) as the loss function for supervised learning training: Where: N is the number of samples, Δ is the prediction time span, y i,t+δ is the true value of the i-th sample at time t+δ, is the corresponding predicted value.

[0123] In some implementations, to prevent overfitting, the system introduces the Dropout mechanism to randomly mask the hidden state, and uses the Adam optimizer to automatically adjust the learning rate to speed up convergence.

[0124] Specifically, in scenarios that support multi-region and multi-pump station forecasting needs, the LSTM structure can be configured as a multi-input-multi-output form according to the network topology. Each node is modeled and predicted independently, and then summarized into a complete forecast matrix.

[0125] In some embodiments, an attention mechanism may be introduced to assign dynamic weights to different time steps: This forms a weighted representation Used to enhance the role of key time information in prediction.

[0126] Through the technical solutions of this embodiment, the LSTM network provides reliable support for modeling complex temporal relationships and capturing nonlinear evolutionary characteristics. As the core component of the municipal water distribution network analysis method, the prediction module not only improves the accuracy of future state perception but also provides a solid data foundation for anomaly detection and real-time optimization of scheduling strategies. The formula definitions and structural parameters are clearly disclosed, meeting the requirements of sufficient technical disclosure under patent law.

[0127] S3. Compare the differences between the pipeline network status prediction results and the actual observation data, and determine whether there is an anomaly by calculating the residual; In the municipal water distribution network data analysis method proposed in this invention, after completing the modeling of historical operating data and the prediction of future states, it is necessary to promptly determine whether the current system operation deviates from expectations, thereby accurately identifying potential abnormal states. To this end, in step S3, an anomaly detection mechanism based on residual analysis is introduced to identify abnormal events by comparing the deviation between the predicted results and the real-time observed data. This step serves as a key bridge between the prediction module and the scheduling module, providing a basis for subsequent system response and resource scheduling.

[0128] Generally speaking, the prediction results represent the ideal state of the system in the absence of abnormal disturbances, while the actual observations reflect the actual operating conditions affected by external environmental and internal disturbances. The difference between the two can indicate whether the system is operating abnormally.

[0129] In this embodiment, step S3 includes the following core processes: state difference calculation, residual extraction, threshold setting and abnormality determination.

[0130] Specifically, the system compares the predicted state at each prediction time t and real-time observation state y t , calculate the residual value: Where: r t ∈R d′ Represents the residual vector at time t, y t ∈R d′ is the actual observed value, is the network status predicted based on LSTM, and d′ is the state variable dimension, which usually includes indicators such as water pressure and flow.

[0131] In some embodiments, if the number of monitoring points is n, the residual r is calculated for each monitoring point i. i,t : The system can construct the complete residual matrix: In order to effectively determine whether there is an anomaly, this embodiment introduces a statistical dynamic threshold setting mechanism. Specifically, the system calculates the historical residual sequence Perform statistical analysis and calculate the mean and standard deviation: Then set the motion detection threshold: threshold=μ r +α·σ r ; Where: μ r ∈R d′ is the residual mean vector, σ r ∈R d′ is the residual standard deviation vector, α∈R + It is the sensitivity adjustment factor, and its value is generally between [2,3], which can be flexibly set according to the system tolerance.

[0132] In one possible implementation, the system sets a threshold for each variable dimension independently to prevent high-volatility variables from interfering with the judgment results of low-volatility variables.

[0133] As an option, if the residual value of a certain dimension of any monitoring point Exceeds the corresponding threshold (j) , the system determines that there is an abnormality at this point at this moment: In some embodiments, to avoid false alarms caused by short-term fluctuations, the system can introduce a sliding window mechanism for cumulative abnormality judgment. That is, an alarm is finally triggered only when an abnormality is detected for κ consecutive sampling periods.

[0134] The alarm information may include: abnormal time, abnormal location (sensor number or area number), abnormal type (high pressure, sudden drop in flow, etc.) and abnormal value.

[0135] In more complex scenarios, this embodiment introduces a multi-dimensional residual fusion index to comprehensively evaluate the degree of deviation of the overall system operation: Where: j ∈R + is the weighting coefficient of each dimension, is the scalar indicator of the fusion residual.

[0136] The system sets the fusion threshold uniformly based on this Used to comprehensively determine whether the entire network is in an abnormal state.

[0137] In some embodiments, the anomaly detection module can combine the pipeline network topology to cluster and analyze multiple spatially close anomalies to determine whether there is a regional systemic fault. For example, if multiple adjacent nodes experience pressure anomalies simultaneously, the system can determine that the pipeline segment has a leak or scheduling imbalance.

[0138] In addition, in another implementation method, the anomaly detection results can be used to reversely adjust the LSTM model training samples, eliminate obvious outliers, improve the robustness of the prediction model, and achieve model-detection linkage optimization.

[0139] In summary, step S3 achieves real-time anomaly identification of the network's operational status through residual calculation and a statistical threshold determination mechanism. All formula variable definitions are clearly disclosed, including vector dimensions, statistical parameters, and judgment logic, ensuring completeness and adequacy. The anomaly detection module not only serves as a system early warning but also provides important decision-making basis for subsequent scheduling strategy adjustments, forming a key link in the core process chain of this invention.

[0140] S4. Generate a water resource scheduling optimization plan based on the anomaly detection results and the pipeline network status prediction results.

[0141] After completing the prediction of the pipeline network status (step S2) and the determination of anomalies (step S3), the system has obtained the operating status information and potential anomaly risks for the current and future time periods. In order to optimize resource allocation and maximize operational efficiency, the present invention further proposes a mechanism for generating a scheduling optimization plan as a key step in the final step of the entire data analysis process. Step S4 is to dynamically formulate a reasonable water resource scheduling strategy based on the aforementioned detection and prediction results, taking into account the balance of supply and demand, operational safety, and energy economy, to ensure the continued stability of the municipal water distribution system under different operating scenarios.

[0142] Generally, intelligent scheduling strategies must comprehensively consider multiple factors, including but not limited to water pressure, water volume, energy consumption, supply-demand matching, avoidance of abnormal pipe sections, and pump station start-up and shutdown strategies. In this embodiment, the scheduling scheme is constructed based on an optimization model, using the particle swarm optimization algorithm (PSO) to search for the optimal set of control parameters.

[0143] In this embodiment, the scheduling optimization aims to minimize the system operation cost, and the following objective function is constructed: Where: J∈R represents the total scheduling cost of the system, n p ∈N is the number of pumps involved in the scheduling, C i ∈R + is the unit energy consumption cost of the i-th pumping station, x i ∈{0,1} indicates whether the i-th pump station is enabled (0 means disabled, 1 means enabled), n s ∈N is the number of scheduling constraints that the system needs to satisfy, Φ j ∈R + Indicates the degree of violation of the j-th constraint, λ∈R + is the constraint penalty factor, which is used to balance the weight between energy consumption cost and safety constraints.

[0144] As an option, the system introduces the following typical scheduling constraints: Water pressure constraint: Ensure that the pressure of all water supply nodes is within a safe range: Supply and demand balance constraint: The total water supply at each time step should meet the predicted demand D t : Abnormal avoidance constraints: Avoid faulty pipe sections or faulty nodes identified by the anomaly detection module and adjust the flow direction and pump station operation strategy.

[0145] In some embodiments, the scheduling constraints are incorporated into the optimization objective in the form of a penalty function to avoid the complexity of directly solving the constrained optimization problem.

[0146] In one possible implementation, the intelligent scheduling module uses a particle swarm optimization algorithm to solve the above scheduling objectives. The particle update process is shown in the particle update formula of the PSO.

[0147] The above process is iterated continuously until the objective function converges or reaches the set number of rounds, and finally the optimal pump station operation status and water resource allocation plan within the current scheduling cycle are output.

[0148] The scheduling results are output in table form, including the pump station start-stop state, water distribution, flow direction adjustment, and other parameters in each time period. For example: Time period Pumping station 1 status Pumping station 2 status Water supply in area A Water supply in area B t+1 start up stop <![CDATA[180m 3 ]]> <![CDATA[220m 3 ]]> t+2 start up start up 200 m 3 ]] <![CDATA[240m 3 ]]> The system supports rolling scheduling, and the scheduling strategy is regenerated based on the latest prediction results and abnormal feedback in each cycle to ensure flexible response.

[0149] In some embodiments, the intelligent scheduling module also has the following enhanced functions: Supports running costs, carbon emissions, and water hammer as multi-objective optimization content, and introduces Pareto frontier to determine the optimal scheduling solution; Combines GIS system to realize spatial visualization scheduling feedback and assist manual intervention; Supports emergency scheduling mode, such as forced switching to backup water supply path or prioritizing maintaining high-priority area water supply safety when abnormal high frequency occurs.

[0150] In this embodiment, the scheduling optimization scheme generated based on prediction and abnormal results has operability and dynamic adaptability, and can respond to complex changes in the operation state of the urban pipe network in real time. The optimization target, variable, and constraint are completely disclosed, and the formula parameter definition is clear and explicit, meeting the requirements of the Patent Law for full technical disclosure. As a process terminal, scheduling optimization not only completes the data analysis closed loop, but also provides key support for urban water supply safety and efficiency.

[0151] Embodiment 3: Municipal water distribution pipe network data analysis electronic device, please refer to the attached Figure 3 , including a data interface module for accessing data from sensors or other protocol devices, supporting access to multiple communication protocols; this module is responsible for obtaining real-time data of the water pipe network from sensor devices, including flow, water pressure, valve state, and other key indicators. By supporting multiple communication protocols (such as TCP / IP, Modbus, OPC, etc.), this module can seamlessly interface with devices from different manufacturers, ensuring the accuracy and real-time nature of data collection.

[0152] The cache and processing unit is used for cleaning and synchronous processing of collected data; this unit performs necessary preprocessing on received data, such as removing noise, synchronizing timing, and filling missing values. Through data cleaning and synchronization processing, the data quality and consistency in the subsequent analysis process are ensured.

[0153] The edge computing module is used to perform model inference and prediction based on input data, outputting predictions about the pipeline network's status. This module, the core of the device, leverages advanced computing capabilities for real-time data processing. The model employed can make predictions based on historical data and real-time input data. For example, an LSTM (Long Short-Term Memory) model can be used for status prediction, accurately predicting the future operational status of the pipeline network and providing data support for subsequent anomaly analysis.

[0154] The anomaly analysis and optimization control module is used to identify anomalies by calculating the residual between the predicted and actual values ​​and generate an optimized scheduling strategy using a particle swarm optimization algorithm. After state prediction, the system calculates the residual between the predicted and observed values ​​and uses a set dynamic threshold to determine whether the system is anomaly. If a system anomaly occurs (such as low water pressure or excessive flow), the module uses a particle swarm optimization (PSO) algorithm to optimize the scheduling strategy. Optimization objectives include minimizing energy consumption and maintaining stable water pressure.

[0155] The communication and control output module transmits optimized dispatch commands to the execution terminal for control. The final optimized dispatch plan is transmitted through this module to the control terminal (such as pump stations, valves, and other execution equipment), enabling the automated issuance of dispatch instructions. This module's efficient communication capabilities ensure the real-time and accuracy of dispatch instructions, guaranteeing the reliability and efficiency of water network operations.

[0156] Through the synergistic effect of the above modules, the electronic device of the present invention can realize intelligent analysis, anomaly detection and optimized scheduling of the water distribution network in the urban water supply system, thereby improving the automation and intelligent management level of the urban water supply system.

[0157] The examples of this specific embodiment are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Identical components are represented by the same reference numerals. Therefore, any equivalent changes made based on the structure, shape, and principle of this application should be included in the scope of protection of this application.

Claims

1. Municipal water distribution network data analysis system, characterized by: include, The data acquisition and preprocessing module is used to collect data from various sensors, and clean, standardize and fuse the data to generate standardized data; A state recognition module, based on a deep learning algorithm, receives the standardized data and outputs a state prediction result of the pipeline network; an anomaly detection module receives the state prediction result of the pipeline network and the real-time observation data, compares the difference between the two, and generates an anomaly alarm; An intelligent scheduling module receives the abnormal information and pipe network status prediction results output by the abnormality detection module and generates a water resource scheduling optimization plan; The deep learning algorithm is a long short-term memory network, which is used to analyze pipeline network time series data and predict future pipeline network status.

2. The municipal water distribution network data analysis system according to claim 1, characterized in that: The data acquisition and preprocessing module includes: The data cleaning unit is used to perform interpolation processing on the collected data and fill in the missing values ​​in the data; Data standardization unit, used to standardize data and convert data into a unified scale using the Z-score standardization method; The data fusion unit is used to fuse data from different sensors through the Kalman filter algorithm.

3. The municipal water distribution network data analysis system according to claim 1, characterized in that: The state recognition module is modeled by a long short-term memory network, which includes: Forget gate, used to control the degree of forgetting of input information; Input gate, used to control the extent to which new information is written; Output gate, used to control the hidden state output at the current moment; The long short-term memory network is trained by a back-propagation algorithm to optimize the parameters of the network.

4. The municipal water distribution network data analysis system according to claim 1, characterized in that: The anomaly detection module is based on the residual calculation method. It determines whether there is an anomaly by calculating the residual between the pipeline network status prediction result and the actual observation data. Specifically: Calculating the residual, and determining that it is an abnormal state if the residual exceeds a preset threshold; The threshold is dynamically calculated based on the mean and standard deviation of the residuals.

5. The municipal water distribution network data analysis system according to claim 1, characterized in that: The intelligent scheduling module schedules water resources using a particle swarm optimization algorithm. The update formula of the particle swarm optimization algorithm is: in, is the velocity of the i-th particle in the k-th generation, is the current position of the i-th particle, is the particle’s own best historical position, g * is the global optimal position, c1 and c2 are acceleration constants, r1 and r2 are random numbers, and w is the inertia weight.

6. A method for analyzing municipal water distribution network data, characterized in that: The municipal water distribution network data analysis system according to any one of claims 1 to 5 comprises: S1. Collect data from multiple sensors and clean, standardize and fuse the data; S2. Analyze the processed data based on the long short-term memory network to predict the future status of the pipeline network; S3. Compare the difference between the pipeline network status prediction result and the actual observation data, and determine whether there is an abnormality by calculating the residual; S4. Generate a water resource scheduling optimization plan based on the anomaly detection results and the pipeline network status prediction results.

7. A municipal water distribution network data analysis method according to claim 6, characterized in that: The cleaning step includes filling missing data using an interpolation method, the standardization step includes converting data using a Z-score standardization method, and the fusion step includes fusing data using a Kalman filter algorithm.

8. A municipal water distribution network data analysis method according to claim 6, characterized in that: The long short-term memory network is used to analyze pipeline network time series data. The network training process optimizes network parameters through a back-propagation algorithm, and the network controls the transmission and update of information through a gating mechanism.

9. Municipal water distribution network data analysis electronic equipment, characterized in that: A municipal water distribution network data analysis method as described in any one of claims 6 to 8, comprising: Data interface module, used to access data from sensors or other protocol devices, supporting access to multiple communication protocols; Cache and processing unit, used to clean and synchronize the collected data; The edge computing module is used to perform model inference and prediction based on the input data and output the prediction results of the pipeline network status; The anomaly analysis and optimization control module is used to identify anomalies by calculating the residual between the predicted value and the actual value and generate an optimized scheduling strategy in combination with the particle swarm optimization algorithm; The communication and control output module is used to transmit the optimized scheduling commands to the execution terminal for control.

10. The municipal water distribution network data analysis electronic device according to claim 9, characterized in that: The data interface module supports real-time access and protocol conversion of sensor data, supports compatible access of multiple industrial communication protocols, and ensures seamless data interaction between devices.

Images