Intelligent water tank water inlet valve control system

Abstract

The invention discloses an intelligent water tank water inlet valve control system comprising a data acquisition and preprocessing module used for acquiring multi-source time sequence data through a multi-source water affair sensor unit and preprocessing the multi-source time sequence data; the water level prediction modeling module is used for constructing a water level prediction neural network model; the control objective function construction module is used for setting an optimization objective function containing a valve control performance index according to the liquid level prediction result and the operation state; the valve control optimization module is used for generating an optimal valve control strategy sequence by adopting an improved cat swarm algorithm; the electric valve execution control module is used for sequentially inputting the target opening degree instructions into the electric control valves, sequentially executing the instructions according to a control cycle and adjusting the opening degree of the electric control valve group in real time; and the Internet of Things visual management module is used for carrying out visual management through the Internet of Things monitoring operation and maintenance platform. Accurate intelligent water tank water inlet valve control is provided for a user.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent water control, and in particular to an intelligent water tank inlet valve control system. Background Art

[0002] In modern urban water supply systems and building water management, water tanks serve as crucial water supply buffers and pressure regulators. Their inlet control strategies directly impact water resource utilization efficiency, pipe network load balancing, and terminal water quality safety. Currently, water tank inlet valve control systems often rely on liquid level sensors as their primary input, employing timed opening and closing or simple upper and lower limit triggering mechanisms to control valve status. While these traditional methods offer low cost and ease of implementation, they suffer from widespread regulation lag, rigid response, and an inability to dynamically adapt to water usage fluctuations. These methods struggle to meet the high-precision, highly adaptable control requirements of current smart water systems.

[0003] In terms of control methods, existing water inlet valve control systems mostly use rule-based control logic and fail to effectively introduce predictive modeling and optimization algorithms. Because future water consumption trends, user load characteristics, or dynamic changes in pipe network pressure are not taken into account, start-stop control is performed solely based on the current liquid level value, which can easily lead to frequent opening and closing, control oscillation, and even water inlet delays or insufficient water supply during peak water consumption periods. In addition, frequent valve execution will also increase the mechanical wear of the electric valve, reduce the system life, and increase maintenance costs. In complex scenarios, this single logic control cannot effectively coordinate the contradictions between different control objectives, especially it cannot simultaneously guarantee the multi-objective regulation requirements of liquid level stability, peak avoidance control, and pipe network pressure stability.

[0004] In terms of data processing and decision-making mechanisms, existing systems lack a complete "prediction-optimization-execution-feedback" closed-loop control structure and generally lack the ability to deeply utilize historical operating data and conduct online learning. Even some high-end systems have introduced IoT platforms for remote data monitoring, but these systems are mostly limited to display functions and lack intelligent prediction capabilities based on historical data modeling. Furthermore, existing solutions lack high-frequency, high-precision mechanisms for monitoring water inflow execution behavior, making it difficult to capture the nonlinear dynamic processes between valve execution and liquid level response. This results in the control system being unable to promptly perceive execution deviations, causing prediction model errors to accumulate, further impacting system stability.

[0005] Some studies have attempted to incorporate machine learning methods into water level prediction models, such as using support vector machines and shallow neural networks to perform regression modeling of water level trends. However, these models have simple structures and weak time series modeling capabilities, making it difficult to fully extract the time-dependent characteristics of liquid level changes. Regarding optimization scheduling algorithms, traditional genetic algorithms and particle swarm algorithms suffer from weak local search capabilities and premature convergence. This is particularly true when solving multi-objective optimization problems with high dimensions and nonlinear constraints. Numerous iterations are often required to approach the optimal solution, resulting in poor real-time performance and unsuitability for valve control tasks in dynamic environments.

[0006] Furthermore, current mainstream systems lack a well-designed joint feedback mechanism for control and model errors. Even with a prediction module, real-time errors during control execution cannot be effectively fed back to the prediction model itself, resulting in a disconnect between the prediction model and control behavior, making it impossible to achieve self-evolution and optimization. The lack of error-driven dynamic correction capabilities leads to a significant decrease in model accuracy after long-term operation, seriously affecting the effectiveness and robustness of control strategy generation. Especially in complex and changing water management scenarios, if the prediction model cannot adapt to changing boundary conditions in a timely manner, it is very likely to suffer from serious deviations.

[0007] On the other hand, while some systems have IoT access, their primary functions focus on remote liquid level reading and manual control, lacking visual analysis of valve control processes and system-level O&M capabilities. Advanced functions such as historical data query, trend analysis, water usage modeling, remote policy adjustments, and anomaly alerts have not yet been systematically integrated, limiting the IoT platform's role in supporting decision-making in smart water management. Furthermore, the monitoring platform lacks a closed-loop interconnection with optimization algorithms and predictive models, preventing the coordinated linkage of platform-level policy intervention and adaptive control.

[0008] Therefore, how to provide an intelligent water tank inlet valve control system is a problem that those skilled in the art urgently need to solve. Summary of the Invention

[0009] One purpose of the present invention is to propose an intelligent water tank inlet valve control system. The present invention provides users with accurate intelligent water tank inlet valve control by combining liquid level prediction, optimization control, closed-loop feedback and Internet of Things operation and maintenance.

[0010] An intelligent water tank water inlet valve control system according to an embodiment of the present invention includes:

[0011] The data acquisition and preprocessing module is used to collect multi-source time series data through multi-source water affairs sensor units and preprocess the multi-source time series data;

[0012] Water level prediction modeling module, used to build a water level prediction neural network model;

[0013] A control objective function construction module is used to set an optimization objective function including valve control performance indicators based on the liquid level prediction results and operating status;

[0014] Valve control optimization module, used to generate the optimal valve control strategy sequence using improved cat swarm algorithm;

[0015] The electric valve execution control module is used to input the target opening instructions to the electric control valve in sequence, execute the instructions in accordance with the control cycle sequence, and adjust the opening of the electric control valve group in real time;

[0016] The IoT visualization management module is used for visual management through the IoT monitoring and operation platform.

[0017] Optionally, modules can be connected using the following methods:

[0018] S1. Collect multi-source time series data through multi-source water affairs sensor units and perform pre-processing;

[0019] S2. Build a water level prediction neural network model based on multi-source time series data. Use a gated recurrent unit network to model the historical liquid level time series data. Through the gated recurrent layer, calculate the future water tank level change trend based on the current input and historical status, and output the liquid level prediction sequence.

[0020] S3. According to the liquid level prediction sequence, an optimization objective function including valve control performance indicators is set;

[0021] S4. Using the improved cat swarm algorithm, each cat swarm individual is equipped with a dynamic observation function and an inertial guidance factor. In the optimization state, the cat swarm individual's detection and search behavior for different opening intervals is simulated. In the tracking state, the imitation and approach path behavior of the current optimal individual is simulated. Based on the optimization objective function, the optimal valve control strategy sequence is generated;

[0022] S5. Using the optimal valve control strategy sequence as a control instruction, adjust the opening of the electric regulating valve group in real time to control the water inlet flow of the water tank;

[0023] S6. Input the valve opening execution result and the actual water tank liquid level feedback data into the control system to form a closed-loop feedback mechanism and dynamically correct the water level prediction neural network model parameters;

[0024] S7. Visual management is carried out through the IoT monitoring and operation and maintenance platform, which also provides remote operation, historical data query, water consumption trend analysis and abnormal alarm functions.

[0025] Optionally, the multi-source time series data includes water tank liquid level, water inlet network pressure, instantaneous flow, water tank water output and water load.

[0026] Optionally, the valve control performance indicators include a water level stability indicator, a water inlet peak avoidance rate indicator, a pipe network pressure fluctuation constraint indicator, and a valve adjustment frequency constraint indicator.

[0027] Optionally, S2 includes the following specific steps:

[0028] S21. Construct a water level prediction neural network model, which consists of an input layer, multiple gated recurrent unit layers, and an output layer;

[0029] S22. Receive multi-source time series data at the input layer of the water level prediction neural network model, input the multi-source time series data of each time step into the gated recurrent unit layer, process the data of each time step in sequence and transfer the hidden state;

[0030] S23. Within each gated recurrent unit, a state update result of the current time step is generated based on the current input and the state of the previous time step, capturing the time-dependent characteristics of the water level change;

[0031] S24, after processing all time steps, extracting hidden state information of the water level prediction neural network model;

[0032] S25. Generate the water tank level prediction results for multiple future time steps through the output layer and output the level prediction sequence.

[0033] Optionally, S3 includes the following specific steps:

[0034] S31. Setting an optimization objective function of the valve control strategy based on the liquid level prediction sequence and data related to the water tank operation status;

[0035] S32. The optimization objective function is composed of multiple valve control performance indicators, and the optimization objective function adopts a normalized multi-objective weighted minimization structure:

[0036]

[0037] Among them, J is the optimization objective function, ω i is the weighted coefficient of the i-th indicator, N is the number of indicators involved in the calculation, M i,t is the measured value of the i-th indicator at time step t, μ i is the mean value of the i-th indicator in the historical window, σ i is the standard deviation of the i-th indicator in the historical window, T is the total number of steps in the forecast time series, ΔU t is the change in valve opening between the tth and t-1th time steps, λ is the smoothness adjustment parameter of the control strategy, i is the i-th indicator, t is the time step, and max is the maximum value operator;

[0038] S33, setting a liquid level stability index, and calculating a mean square error based on the difference between the predicted value of the water tank liquid level and the target liquid level;

[0039] S34. Set a peak avoidance efficiency index, which is calculated based on the proportion of water flow during peak hours to the total water flow;

[0040] S35. Set a pressure stability index, which is calculated based on the range and standard deviation of pressure measurements within a continuous time period;

[0041] S36, setting the regulation smoothness index, which is calculated based on the maximum change range of the valve opening in consecutive time steps;

[0042] S37. Incorporate various indicators into the optimization objective function.

[0043] Optionally, S4 includes the following specific steps:

[0044] S41. Construct a cat swarm optimization population, set a fixed number of individuals, each cat swarm individual represents a valve opening control strategy sequence within a complete time period, initialize the strategy position and velocity vector of each cat swarm individual, and simultaneously construct a historical trajectory record for each cat swarm individual, dynamically storing the state change information of the cat swarm individual in multiple consecutive iterations;

[0045] S42. Compared with the cat swarm algorithm that uses a fixed state transition probability method, the improved cat swarm algorithm introduces a dynamic state switching mechanism, setting three behavior modes: optimization state, tracking state, and mixed state. The individual state is determined by the adaptive state function:

[0046]

[0047] Among them, P s (k) is the behavior state selection result of the kth cat group individual, k is the cat group individual number, j is the behavior state index, arg max is the index operator corresponding to the maximum value, T s is the sliding time window length for state evaluation, t is the time step, is the fitness change of the kth cat group individual in the tth cycle and the behavior state is j, is the reference fitness value of the kth individual in the tth cycle, ε is a very small positive number, W j is the weight coefficient corresponding to behavior state j;

[0048] S43. In the optimization state, the cat group individuals perform non-uniform directional search based on their historical trajectories and current policy positions by introducing a weighted perturbation function. The perturbation amplitude is determined by the individual's current convergence trend and the policy gradient direction. In the tracking state, the cat group individuals use the cat group individual control strategy with the best fitness value in the current cat group as the target to guide the update of the current position. The optimal cat group individual control strategy is defined as the cat group individual control strategy sequence with the smallest fitness value in this round of iteration.

[0049] S44. In the tracking state, the individual cats in the group refer to the optimal individual cat control strategy and the historical optimal individual cat control strategy at the same time, and perform bias-following updates through a dual-guidance mechanism;

[0050] S45. In each iteration, the valve control strategy sequence represented by the current position of each cat individual is substituted into the optimization objective function, the corresponding fitness value is calculated, and all cat individuals are ranked by performance according to the fitness value. The strategy sequence corresponding to the individual with the smallest fitness value is recorded as the current optimal solution, and the global optimal solution is continuously updated;

[0051] S46. When the maximum number of iterations reaches 1000, the optimal valve control strategy sequence corresponding to the cat group individual with the best current comprehensive fitness value is output.

[0052] Optionally, S5 includes the following specific steps:

[0053] S51, receiving an optimal valve control strategy sequence, where the optimal valve control strategy sequence includes a target valve opening value corresponding to each time step in a future prediction time period;

[0054] S52, inputting the target opening instructions into the electric control valves in sequence, executing the instructions in accordance with the control cycle sequence, and adjusting the opening of the electric control valve group in real time;

[0055] S53. In each control cycle, the electric actuator adjusts the valve opening angle according to the current target opening instruction. The valve opening is continuously adjustable within the range of 0 to 90 degrees.

[0056] S54. In each control cycle, the electric actuator adjusts the valve opening angle of the electric regulating valve group in real time according to the target opening instruction. At the same time, the actual opening in the cycle is collected and compared with the target opening to obtain the valve opening error:

[0057]

[0058] Among them, E adaptive is the adaptive control error feedback value constructed based on the current behavior state selection result, P s(k) is the behavior state selection result of the k-th cat group individual, k is the cat group individual number, is the deviation amplification factor corresponding to the current behavior state, T w N is the total number of cycles of the control feedback sliding time window. w is the number of subsampling in each control cycle, is the target valve opening at the nth sampling point in the tth cycle, is the actual valve opening of the nth sampling point in the tth cycle, t is the time step, and n is the sub-sampling point number;

[0059] S55. The control system synchronously collects the current target opening, the actual execution opening and the water tank liquid level change value, calculates the execution deviation and records the actual response behavior of the water inlet flow in the current cycle.

[0060] Optionally, S6 includes the following specific steps:

[0061] S61. In each control cycle, the actual execution opening data of the electric regulating valve group and the actual water tank liquid level feedback data of the corresponding time step are collected to form a multi-source feedback data pair, and the data is stored in the control system in a time series;

[0062] S62. Construct an actual control execution sequence and an actual liquid level observation sequence. The actual control execution sequence is a sequence consisting of the actual valve openings at each time step recorded within the control cycle. The actual liquid level observation sequence is a sequence consisting of the actual values of the water tank liquid level obtained by the liquid level sensor at each time step. The actual control execution sequence and the actual liquid level observation sequence have the same time step length.

[0063] S63. At the same time step, obtain a water tank level prediction sequence output by the neural network prediction model, where the water tank level prediction sequence corresponds one-to-one to the actual liquid level observation sequence;

[0064] S64. Construct a joint loss function to fuse the tank level prediction sequence and valve control execution deviation for dynamic feedback learning:

[0065]

[0066] Among them, L joint is the value of the joint loss function, α is the weighted coefficient of the liquid level prediction error, T' is the total number of time steps, t is the time step, is the actual water tank level value at time step t, is the predicted water tank level value at time step t, is the square of the liquid level prediction error, β is the weighted coefficient of the control execution deviation, E adaptive is an adaptive control error feedback value constructed based on the current behavior state selection result;

[0067] S65. Based on the joint loss function value, a gradient descent optimization method is used to dynamically adjust the weight parameters in the water level prediction neural network model, and incremental training is performed on the collected feedback data;

[0068] S66. After each round of closed-loop training is completed, the water level prediction neural network model parameters are updated and the current version is saved.

[0069] Optionally, S7 includes the following specific steps:

[0070] S71. Upload the water tank level data, valve control strategy, actual valve opening, water inlet flow, pipe network pressure, and water load parameters in the control system to the monitoring and operation and maintenance platform to achieve real-time access to multi-source operation data;

[0071] S72. Construct a visualization interface in the monitoring and operation and maintenance platform to display sensor and control execution data in the form of curve graphs, bar graphs, and color block graphs, including the trend of water tank level changes, comparison of valve control and execution status, and operation status distribution;

[0072] S73. Establish a remote operation interface on the platform side, receive the control adjustment instructions issued by the user side, convert the control adjustment instructions into standard control commands, and perform control according to the standard control commands;

[0073] S74. Configure historical data query function to support periodic indexing and display of water tank level, water inlet flow, valve control, and execution error, and support data screening and retrospective analysis by day, week, and month;

[0074] S75. Set up a water consumption trend analysis function, perform periodic modeling and trend extraction on historical water consumption data, and generate a trend change report;

[0075] S76. Build anomaly detection and alarm functions to perform status assessment based on control execution deviation, water tank level fluctuation, valve response delay, and flow anomalies. When an abnormal event is detected, abnormal alarm information is automatically generated and pushed to the remote user terminal and operation and maintenance background.

[0076] The beneficial effects of the present invention are:

[0077] The intelligent water tank inlet valve control system provided by the present invention, based on the deep integration of artificial intelligence algorithms and water engineering scenario requirements, realizes intelligent, refined and visual management of the entire process from water level prediction to water inlet control, and has achieved remarkable beneficial effects in improving system operation performance, improving water scheduling efficiency, and optimizing water quality management.

[0078] By constructing a water level prediction neural network model based on a gated cyclic unit network, the system can accurately predict the liquid level change trend for multiple time steps in the future based on multi-source time series data such as water tank liquid level, pipe network pressure, water inlet flow, and water load. This makes valve control no longer rely on simple upper and lower limit logic, but instead has forward-looking and dynamic characteristics. On this basis, the system further introduces a control strategy optimization mechanism based on an improved cat swarm algorithm. Based on multi-objective optimization functions including liquid level stability, water inlet peak avoidance rate, pressure fluctuation constraints, and regulation smoothness, it generates the optimal valve control sequence to achieve intelligent regulation of water inlet flow. Through this mechanism, water tanks can store water in advance during non-peak hours and reduce water withdrawal from the municipal pipe network during peak hours, thereby effectively improving the storage capacity of water pools / tanks and alleviating the operating pressure of the municipal pipe network during peak water use periods.

[0079] Furthermore, the system incorporates a sliding window feedback mechanism and a joint loss function during control execution. By comparing actual execution results with predicted values in real time, the neural network model parameters are dynamically modified, achieving the co-evolution of control behavior and the predictive model. This closed-loop feedback mechanism not only improves the model's prediction accuracy and robustness, but also provides real-time perception of control errors and equipment response deviations, effectively reducing fluctuations in the pipeline network caused by frequent pressure regulation, improving system operation stability, and providing a basis for scientific scheduling of water plants, ultimately reducing energy consumption and operating costs.

[0080] To ensure water quality, the system integrates liquid level control strategies with water age management strategies. Combined with the water usage trend analysis module and the monitoring capabilities of the IoT platform, it dynamically identifies the risk of long-term water retention in the tank. By rationally adjusting the water inflow frequency and water exchange cycle, it significantly reduces the probability of dead and stale water, maintains the freshness of water quality in the tank, and further improves the sanitation and safety of water users. Furthermore, the system's built-in anomaly detection module provides real-time alerts for abnormal water level fluctuations, valve actuation anomalies, and abnormal water inflow conditions, further enhancing the system's intelligent diagnostic capabilities and operational safety.

[0081] In summary, this invention not only achieves an organic combination of deep prediction and intelligent optimization in terms of technical means, but also achieves a systematic innovation in traditional water inlet control methods in terms of application effects. Specific benefits include: improving water tank storage and pipe network regulation capabilities, alleviating peak water supply pressure; reducing pipe network pressure fluctuations, ensuring stable operation and energy conservation; and reducing water age, preventing stagnant water, and ensuring water quality safety and user health. It possesses a high degree of intelligence, scalability, and engineering feasibility, with significant social and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0083] Figure 1 This is a flow chart of the intelligent water tank inlet valve control system proposed by the present invention;

[0084] Figure 2 This is a schematic diagram of the intelligent water tank inlet valve control system proposed by the present invention;

[0085] Figure 3 This is the data flow diagram of the intelligent water tank inlet valve control system proposed by the present invention. DETAILED DESCRIPTION

[0086] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0087] refer to Figure 1-3 , intelligent water tank inlet valve control system, including:

[0088] The data acquisition and preprocessing module is used to collect multi-source time series data through multi-source water affairs sensor units and preprocess the multi-source time series data;

[0089] Water level prediction modeling module, used to build a water level prediction neural network model;

[0090] A control objective function construction module is used to set an optimization objective function including valve control performance indicators based on the liquid level prediction results and operating status;

[0091] Valve control optimization module, used to generate the optimal valve control strategy sequence using improved cat swarm algorithm;

[0092] The electric valve execution control module is used to input the target opening instructions to the electric control valve in sequence, execute the instructions in accordance with the control cycle sequence, and adjust the opening of the electric control valve group in real time;

[0093] The IoT visualization management module is used for visual management through the IoT monitoring and operation platform.

[0094] This invention forms an end-to-end closed-loop control system by integrating a data acquisition and preprocessing module, a water level prediction modeling module, a control objective function construction module, a valve control optimization module, an electric valve execution control module, and an Internet of Things (IoT) visualization management module. It uses a gated recurrent neural network to predict liquid level trends and optimizes multi-objective valve control strategies using an improved cat swarm algorithm to ensure accurate and efficient control responses. The system supports remote visualization management and abnormality alarms, improving the water tank storage capacity and pipe network operational stability.

[0095] In this embodiment, the modules are connected through the following methods:

[0096] S1. Collect multi-source time series data through multi-source water affairs sensor units and perform pre-processing;

[0097] S2. Build a water level prediction neural network model based on multi-source time series data. Use a gated recurrent unit network to model the historical liquid level time series data. Through the gated recurrent layer, calculate the future water tank level change trend based on the current input and historical status, and output the liquid level prediction sequence.

[0098] S3. According to the liquid level prediction sequence, an optimization objective function including valve control performance indicators is set;

[0099] S4. Using the improved cat swarm algorithm, each cat swarm individual is equipped with a dynamic observation function and an inertial guidance factor. In the optimization state, the cat swarm individual's detection and search behavior for different opening intervals is simulated. In the tracking state, the imitation and approach path behavior of the current optimal individual is simulated. Based on the optimization objective function, the optimal valve control strategy sequence is generated;

[0100] S5. Using the optimal valve control strategy sequence as a control instruction, adjust the opening of the electric regulating valve group in real time to control the water inlet flow of the water tank;

[0101] S6. Input the valve opening execution result and the actual water tank liquid level feedback data into the control system to form a closed-loop feedback mechanism and dynamically correct the water level prediction neural network model parameters;

[0102] S7. Visual management is carried out through the IoT monitoring and operation and maintenance platform, which also provides remote operation, historical data query, water consumption trend analysis and abnormal alarm functions.

[0103] This invention uses a gated recurrent neural network driven by multi-source time series data to predict liquid level trends. Combined with an improved cat swarm algorithm to simulate optimization and tracking behavior, it dynamically generates the optimal valve control strategy. A closed feedback loop is established to correct the prediction model in real time based on execution errors, improving control accuracy and stability. Leveraging an Internet of Things platform, remote monitoring, trend analysis, and anomaly warnings are implemented, ensuring intelligent system operation and efficient management.

[0104] In this embodiment, the multi-source time series data includes the water tank liquid level, the water inlet network pressure, the instantaneous flow rate, the water tank water output and the water load.

[0105] This invention builds a dynamic data system covering both supply and demand by collecting data on water tank levels, inlet pipe pressure, instantaneous flow rate, water tank output, and water load, enabling comprehensive modeling of water service status. Multidimensional data input enhances the level prediction model's ability to detect complex operating conditions, improving prediction accuracy and control strategy adaptability. The system can adjust control parameters based on real-world operating conditions, making valve adjustments more proactive and stable, effectively improving regulation and storage efficiency and pipe network operational safety.

[0106] In this embodiment, the valve control performance indicators include a water level stability indicator, a water inlet peak avoidance rate indicator, a pipe network pressure fluctuation constraint indicator, and a valve adjustment frequency constraint indicator.

[0107] This method constructs a multi-objective optimization function by setting water level stability, water inflow peak avoidance ratio, pipe network pressure fluctuation constraints, and valve adjustment frequency constraints. This function comprehensively balances water tank level accuracy, peak-shifting capability, pipe network pressure stability, and equipment lifespan. Each indicator is normalized and used in control strategy evaluation, making the optimization process more targeted and robust. This allows for refined and adaptive adjustment of valve control strategies, effectively improving system operational stability and energy efficiency.

[0108] In this embodiment, S2 includes the following specific steps:

[0109] S21. Construct a water level prediction neural network model, which consists of an input layer, multiple gated recurrent unit layers, and an output layer;

[0110] S22. Receive multi-source time series data at the input layer of the water level prediction neural network model, input the multi-source time series data of each time step into the gated recurrent unit layer, process the data of each time step in sequence and transfer the hidden state;

[0111] S23. Within each gated recurrent unit, a state update result of the current time step is generated based on the current input and the state of the previous time step, capturing the time-dependent characteristics of the water level change;

[0112] S24, after processing all time steps, extracting hidden state information of the water level prediction neural network model;

[0113] S25. Generate the water tank level prediction results for multiple future time steps through the output layer and output the level prediction sequence.

[0114] The present invention constructs a neural network structure within the water level prediction model, consisting of an input layer, multiple gated recurrent unit layers, and an output layer. The model receives multi-source time series data and feeds it into the gated recurrent units time-step by time. Each layer dynamically captures the relationship between the current input and the historical state, forming a time-dependent signature of water level changes. After time series processing, the final hidden state information is extracted, and the output layer generates liquid level predictions for multiple future time steps, improving the consistency and accuracy of the predictions and providing dynamic support for optimizing valve control strategies.

[0115] In this embodiment, S3 includes the following specific steps:

[0116] S31. Setting an optimization objective function of the valve control strategy based on the liquid level prediction sequence and data related to the water tank operation status;

[0117] S32. The optimization objective function is composed of multiple valve control performance indicators, and the optimization objective function adopts a normalized multi-objective weighted minimization structure:

[0118]

[0119] Among them, J is the optimization objective function, ω i is the weighted coefficient of the i-th indicator, N is the number of indicators involved in the calculation, M i,t is the measured value of the i-th indicator at time step t, μ i is the mean value of the i-th indicator in the historical window, σ i is the standard deviation of the i-th indicator in the historical window, T is the total number of steps in the forecast time series, ΔU t is the change in valve opening between the tth and t-1th time steps, λ is the smoothness adjustment parameter of the control strategy, i is the i-th indicator, t is the time step, and max is the maximum value operator;

[0120] S33, setting a liquid level stability index, and calculating a mean square error based on the difference between the predicted value of the water tank liquid level and the target liquid level;

[0121] S34. Set a peak avoidance efficiency index, which is calculated based on the proportion of water flow during peak hours to the total water flow;

[0122] S35. Set a pressure stability index, which is calculated based on the range and standard deviation of pressure measurements within a continuous time period;

[0123] S36, setting the regulation smoothness index, which is calculated based on the maximum change range of the valve opening in consecutive time steps;

[0124] S37. Incorporate various indicators into the optimization objective function.

[0125] During the valve control strategy optimization process, this invention constructs a multi-objective optimization function based on tank level predictions and current operating status data to comprehensively evaluate the pros and cons of different control strategies. This optimization objective function integrates four key indicators: level stability, peak avoidance efficiency, pressure stability, and regulation smoothness. It is designed using a normalized multi-objective weighted minimization structure to balance control accuracy, operating efficiency, and system stability.

[0126] In this embodiment, S4 includes the following specific steps:

[0127] S41. Construct a cat swarm optimization population, set a fixed number of individuals, each cat swarm individual represents a valve opening control strategy sequence within a complete time period, initialize the strategy position and velocity vector of each cat swarm individual, and simultaneously construct a historical trajectory record for each cat swarm individual, dynamically storing the state change information of the cat swarm individual in multiple consecutive iterations;

[0128] S42. Compared with the cat swarm algorithm that uses a fixed state transition probability method, the improved cat swarm algorithm introduces a dynamic state switching mechanism, setting three behavior modes: optimization state, tracking state, and mixed state. The individual state is determined by the adaptive state function:

[0129]

[0130] Among them, P s (k) is the behavior state selection result of the kth cat group individual, k is the cat group individual number, j is the behavior state index, arg max is the index operator corresponding to the maximum value, T s is the sliding time window length for state evaluation, t is the time step, is the fitness change of the kth cat group individual in the tth cycle and the behavior state is j, is the reference fitness value of the kth individual in the tth cycle, ε is a very small positive number, W j is the weight coefficient corresponding to behavior state j;

[0131] S43. In the optimization state, the cat group individuals perform non-uniform directional search based on their historical trajectories and current policy positions by introducing a weighted perturbation function. The perturbation amplitude is determined by the individual's current convergence trend and the policy gradient direction. In the tracking state, the cat group individuals use the cat group individual control strategy with the best fitness value in the current cat group as the target to guide the update of the current position. The optimal cat group individual control strategy is defined as the cat group individual control strategy sequence with the smallest fitness value in this round of iteration.

[0132] S44. In the tracking state, the individual cats in the group refer to the optimal individual cat control strategy and the historical optimal individual cat control strategy at the same time, and perform bias-following updates through a dual-guidance mechanism;

[0133] S45. In each iteration, the valve control strategy sequence represented by the current position of each cat individual is substituted into the optimization objective function, the corresponding fitness value is calculated, and all cat individuals are ranked by performance according to the fitness value. The strategy sequence corresponding to the individual with the smallest fitness value is recorded as the current optimal solution, and the global optimal solution is continuously updated;

[0134] S46. When the maximum number of iterations reaches 1000, the optimal valve control strategy sequence corresponding to the cat group individual with the best current comprehensive fitness value is output.

[0135] This invention utilizes an improved cat swarm optimization algorithm, introduces a dynamic behavior state switching mechanism and an adaptive state selection function, and constructs weighted perturbation search and dual-guided update strategies in the optimization and tracking phases, respectively, to achieve global exploration and local refinement of the valve control strategy. By evaluating individual fitness and continuously updating the global optimal solution in each iteration, the dynamic optimality of the control strategy under multi-objective constraints is guaranteed. A maximum number of iterations is also set to improve the algorithm's convergence efficiency and operational stability, ensuring accurate and efficient control optimization under variable water service conditions.

[0136] In this embodiment, S5 includes the following specific steps:

[0137] S51, receiving an optimal valve control strategy sequence, where the optimal valve control strategy sequence includes a target valve opening value corresponding to each time step in a future prediction time period;

[0138] S52, inputting the target opening instructions into the electric control valves in sequence, executing the instructions in accordance with the control cycle sequence, and adjusting the opening of the electric control valve group in real time;

[0139] S53. In each control cycle, the electric actuator adjusts the valve opening angle according to the current target opening instruction. The valve opening is continuously adjustable within the range of 0 to 90 degrees.

[0140] S54. In each control cycle, the electric actuator adjusts the valve opening angle of the electric regulating valve group in real time according to the target opening instruction. At the same time, the actual opening in the cycle is collected and compared with the target opening to obtain the valve opening error:

[0141]

[0142] Among them, E adaptive is the adaptive control error feedback value constructed based on the current behavior state selection result, P s (k) is the behavior state selection result of the k-th cat group individual, k is the cat group individual number, is the deviation amplification factor corresponding to the current behavior state, T w N is the total number of cycles of the control feedback sliding time window. w is the number of subsampling in each control cycle, is the target valve opening at the nth sampling point in the tth cycle, is the actual valve opening of the nth sampling point in the tth cycle, t is the time step, and n is the sub-sampling point number;

[0143] S55. The control system synchronously collects the current target opening, the actual execution opening and the water tank liquid level change value, calculates the execution deviation and records the actual response behavior of the water inlet flow in the current cycle.

[0144] This invention converts the optimal valve control strategy sequence into cycle-by-cycle execution instructions, driving the electric control valve to achieve continuous adjustment from 0 to 90 degrees. It also introduces a multi-point sampling mechanism within a sliding time window to collect real-time deviations between the actual and target openings, constructing an adaptive control error feedback function associated with the behavior state. The system simultaneously collects liquid level changes and flow rate responses, dynamically assessing execution accuracy and hydraulic performance, achieving precise, stable, and closed-loop water inlet control, improving overall operational safety and intelligent regulation.

[0145] In this embodiment, S6 includes the following specific steps:

[0146] S61. In each control cycle, the actual execution opening data of the electric regulating valve group and the actual water tank liquid level feedback data of the corresponding time step are collected to form a multi-source feedback data pair, and the data is stored in the control system in a time series;

[0147] S62. Construct an actual control execution sequence and an actual liquid level observation sequence. The actual control execution sequence is a sequence consisting of the actual valve openings at each time step recorded within the control cycle. The actual liquid level observation sequence is a sequence consisting of the actual values of the water tank liquid level obtained by the liquid level sensor at each time step. The actual control execution sequence and the actual liquid level observation sequence have the same time step length.

[0148] S63. At the same time step, obtain a water tank level prediction sequence output by the neural network prediction model, where the water tank level prediction sequence corresponds one-to-one to the actual liquid level observation sequence;

[0149] S64. Construct a joint loss function to fuse the tank level prediction sequence and valve control execution deviation for dynamic feedback learning:

[0150]

[0151] Among them, L joint is the value of the joint loss function, α is the weighted coefficient of the liquid level prediction error, T' is the total number of time steps, t is the time step, is the actual water tank level value at time step t, is the predicted water tank level value at time step t, is the square of the liquid level prediction error, β is the weighted coefficient of the control execution deviation, E adaptive is an adaptive control error feedback value constructed based on the current behavior state selection result;

[0152] S65. Based on the joint loss function value, a gradient descent optimization method is used to dynamically adjust the weight parameters in the water level prediction neural network model, and incremental training is performed on the collected feedback data;

[0153] S66. After each round of closed-loop training is completed, the water level prediction neural network model parameters are updated and the current version is saved.

[0154] This method establishes a one-to-one correspondence between the liquid level prediction sequence, the actual executed opening, and the water tank feedback liquid level. It then designs a joint loss function that integrates prediction error and control error, guiding the incremental training and dynamic optimization of the water level prediction neural network. The system uses gradient descent to adjust network parameters, enabling adaptive model correction and ensuring that prediction results consistently match actual operating conditions. The model version is automatically updated after each round of closed-loop training, effectively improving long-term operational accuracy and stability, and implementing an efficient closed-loop learning mechanism for water inlet control prediction, execution, and correction.

[0155] In this embodiment, S7 includes the following specific steps:

[0156] S71. Upload the water tank level data, valve control strategy, actual valve opening, water inlet flow, pipe network pressure, and water load parameters in the control system to the monitoring and operation and maintenance platform to achieve real-time access to multi-source operation data;

[0157] S72. Construct a visualization interface in the monitoring and operation and maintenance platform to display sensor and control execution data in the form of curve graphs, bar graphs, and color block graphs, including the trend of water tank level changes, comparison of valve control and execution status, and operation status distribution;

[0158] S73. Establish a remote operation interface on the platform side, receive the control adjustment instructions issued by the user side, convert the control adjustment instructions into standard control commands, and perform control according to the standard control commands;

[0159] S74. Configure historical data query function to support periodic indexing and display of water tank level, water inlet flow, valve control, and execution error, and support data screening and retrospective analysis by day, week, and month;

[0160] S75. Set up a water consumption trend analysis function, perform periodic modeling and trend extraction on historical water consumption data, and generate a trend change report;

[0161] S76. Build anomaly detection and alarm functions to perform status assessment based on control execution deviation, water tank level fluctuation, valve response delay, and flow anomalies. When an abnormal event is detected, abnormal alarm information is automatically generated and pushed to the remote user terminal and operation and maintenance background.

[0162] This system achieves comprehensive visual management of operational status by uploading multi-source data, including tank level, valve control, flow, and pressure, to a monitoring and maintenance platform in real time. Trend analysis and historical query capabilities are built in to assist users in understanding water usage patterns. A remote operation interface is configured to enable online adjustment of control strategies. Combined with a multi-factor anomaly detection mechanism, the system automatically issues alerts and pushes notifications when control deviations, tank fluctuations, or response delays occur, effectively improving the operational intelligence, safety, and response efficiency of the water system.

[0163] Example 1:

[0164] To verify the feasibility of the present invention, it was applied to the central water supply tank system of a municipal government service center. The park has a total construction area of over 90,000 square meters, serves over 3,000 people, and provides an average daily water supply of approximately 350 to 500 cubic meters. The original system employed a water inlet strategy that combined timed control with upper and lower liquid level limits. This resulted in issues such as untimely control, low tank storage capacity, overlapping water inlet and water consumption peaks, and large pressure fluctuations at the water end. Long-term operation also presented safety risks such as localized stagnant water in the tank and unstable water quality.

[0165] To address these issues, this embodiment deploys the intelligent water tank inlet valve control system described in this invention on top of the existing water tank system. The system incorporates multiple water sensor units, including liquid level sensors, inlet pipe pressure sensors, flow meters, and outlet load measurement modules. It also constructs a water level prediction neural network model and deploys edge computing devices to access the control logic. Through an IoT communication module, it connects to the park's smart management and control platform, achieving intelligent upgrades without disrupting the original water supply.

[0166] In actual application, the system trained 45 consecutive days of historical data through a gated cyclic unit network, including factors such as liquid level change curves, water load changes, and pipe network pressure changes, to form a water level prediction model. In the first week after deployment, the system began to adjust the opening of the electric control valve according to the prediction results. The control strategy is generated by an improved cat swarm algorithm, and the optimization objective function comprehensively considers water level stability, peak avoidance rate, pressure stability, and valve adjustment smoothness. The platform sets the maximum number of iterations to 1000 rounds, and the actual convergence is between 280 and 320 rounds. The generated control strategy covers the water-sensitive periods of the morning peak, midday fluctuations, and evening peak.

[0167] Table 1 Comparison of optimization effects of intelligent customer service system based on reinforcement learning

[0168]

[0169]

[0170] Table 1 demonstrates the dynamic regulation of water storage, achieving "peak shaving and valley filling," by predicting and preserving water storage and controlling peak inflows. Compared with traditional timed control methods, the fluctuation range of water tank levels has been reduced from ±20 cm to ±8 cm, and the adjustment frequency has been reduced from 15 times per day to 5-7 times, significantly reducing the number of valve openings and closings. Secondly, the instantaneous impact of inlet pressure on the pipe network has been reduced by approximately 40%, maintaining a more stable water pressure at the end of the pipe network and significantly improving user satisfaction. Thirdly, the system automatically identifies water sections with a water age of more than 48 hours and initiates water exchange operations, keeping the average daily dead water ratio below 5%, compared to nearly 15% in the previous system. Furthermore, through the platform's visual interface, on-duty personnel can remotely view the current control strategy and execution results, real-time liquid level curves, and early warning information. Control strategy parameters can also be manually adjusted to meet specific scheduling needs.

[0171] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. Intelligent water tank inlet valve control system, characterized by: include: The data acquisition and preprocessing module is used to collect multi-source time series data through multi-source water affairs sensor units and preprocess the multi-source time series data; Water level prediction modeling module, used to build a water level prediction neural network model; A control objective function construction module is used to set an optimization objective function including valve control performance indicators based on the liquid level prediction results and operating status; Valve control optimization module, used to generate the optimal valve control strategy sequence using improved cat swarm algorithm; The electric valve execution control module is used to input the target opening instructions to the electric control valve in sequence, execute the instructions in accordance with the control cycle sequence, and adjust the opening of the electric control valve group in real time; The IoT visualization management module is used for visual management through the IoT monitoring and operation platform.

2. The intelligent water tank inlet valve control system according to claim 1, characterized in that: The modules are implemented as follows: S1. Collect multi-source time series data through multi-source water affairs sensor units and perform pre-processing; S2. Build a water level prediction neural network model based on multi-source time series data. Use a gated recurrent unit network to model the historical liquid level time series data. Through the gated recurrent layer, calculate the future water tank level change trend based on the current input and historical status, and output the liquid level prediction sequence. S3. According to the liquid level prediction sequence, an optimization objective function including valve control performance indicators is set; S4. Using the improved cat swarm algorithm, each cat swarm individual is equipped with a dynamic observation function and an inertial guidance factor. In the optimization state, the cat swarm individual's detection and search behavior for different opening intervals is simulated. In the tracking state, the imitation and approach path behavior of the current optimal individual is simulated. Based on the optimization objective function, the optimal valve control strategy sequence is generated; S5. Using the optimal valve control strategy sequence as a control instruction, adjust the opening of the electric regulating valve group in real time to control the water inlet flow of the water tank; S6. Input the valve opening execution result and the actual water tank liquid level feedback data into the control system to form a closed-loop feedback mechanism and dynamically correct the water level prediction neural network model parameters; S7. Visual management is carried out through the IoT monitoring and operation and maintenance platform, which also provides remote operation, historical data query, water consumption trend analysis and abnormal alarm functions.

3. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The multi-source time series data includes water tank liquid level, water inlet network pressure, instantaneous flow rate, water tank water output and water load.

4. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The valve control performance indicators include a water level stability index, a water inlet peak avoidance rate index, a pipe network pressure fluctuation constraint index, and a valve adjustment frequency constraint index.

5. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The S2 includes the following specific steps: S21. Construct a water level prediction neural network model, which consists of an input layer, multiple gated recurrent unit layers, and an output layer; S22. Receive multi-source time series data at the input layer of the water level prediction neural network model, input the multi-source time series data of each time step into the gated recurrent unit layer, process the data of each time step in sequence and transfer the hidden state; S23. Within each gated recurrent unit, a state update result of the current time step is generated based on the current input and the state of the previous time step, capturing the time-dependent characteristics of the water level change; S24, after processing all time steps, extracting hidden state information of the water level prediction neural network model; S25. Generate the water tank level prediction results for multiple future time steps through the output layer and output the level prediction sequence.

6. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The S3 includes the following specific steps: S31. Setting an optimization objective function of the valve control strategy based on the liquid level prediction sequence and data related to the water tank operation status; S32. The optimization objective function is composed of multiple valve control performance indicators, and the optimization objective function adopts a normalized multi-objective weighted minimization structure: Among them, J is the optimization objective function, ω i is the weighted coefficient of the i-th indicator, N is the number of indicators involved in the calculation, M i,t is the measured value of the i-th indicator at time step t, μ i is the mean value of the i-th indicator in the historical window, σ i is the standard deviation of the i-th indicator in the historical window, T is the total number of steps in the forecast time series, ΔU t is the change in valve opening between the tth and t-1th time steps, λ is the smoothness adjustment parameter of the control strategy, i is the i-th indicator, t is the time step, and max is the maximum value operator; S33, setting a liquid level stability index, and calculating a mean square error based on the difference between the predicted value of the water tank liquid level and the target liquid level; S34. Set a peak avoidance efficiency index, which is calculated based on the proportion of water flow during peak hours to the total water flow; S35. Set a pressure stability index, which is calculated based on the range and standard deviation of pressure measurements within a continuous time period; S36, setting the regulation smoothness index, which is calculated based on the maximum change range of the valve opening in consecutive time steps; S37. Incorporate various indicators into the optimization objective function.

7. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The S4 includes the following specific steps: S41. Construct a cat swarm optimization population, set a fixed number of individuals, each cat swarm individual represents a valve opening control strategy sequence within a complete time period, initialize the strategy position and velocity vector of each cat swarm individual, and simultaneously construct a historical trajectory record for each cat swarm individual, dynamically storing the state change information of the cat swarm individual in multiple consecutive iterations; S42. Compared with the cat swarm algorithm that uses a fixed state transition probability method, the improved cat swarm algorithm introduces a dynamic state switching mechanism, setting three behavior modes: optimization state, tracking state, and mixed state. The individual state is determined by the adaptive state function: Among them, P s (k) is the behavior state selection result of the kth cat group individual, k is the cat group individual number, j is the behavior state index, arg max is the index operator corresponding to the maximum value, T s is the sliding time window length for state evaluation, t is the time step, is the fitness change of the kth cat group individual in the tth cycle and the behavior state is j, is the reference fitness value of the kth individual in the tth cycle, ε is a very small positive number, W j is the weight coefficient corresponding to behavior state j; S43. In the optimization state, the cat group individuals perform non-uniform directional search based on their historical trajectories and current policy positions by introducing a weighted perturbation function. The perturbation amplitude is determined by the individual's current convergence trend and the policy gradient direction. In the tracking state, the cat group individuals use the cat group individual control strategy with the best fitness value in the current cat group as the target to guide the update of the current position. The optimal cat group individual control strategy is defined as the cat group individual control strategy sequence with the smallest fitness value in this round of iteration. S44. In the tracking state, the individual cats in the group refer to the optimal individual cat control strategy and the historical optimal individual cat control strategy at the same time, and perform bias-following updates through a dual-guidance mechanism; S45. In each iteration, the valve control strategy sequence represented by the current position of each cat individual is substituted into the optimization objective function, the corresponding fitness value is calculated, and all cat individuals are ranked by performance according to the fitness value. The strategy sequence corresponding to the individual with the smallest fitness value is recorded as the current optimal solution, and the global optimal solution is continuously updated; S46. When the maximum number of iterations reaches 1000, the optimal valve control strategy sequence corresponding to the cat group individual with the best current comprehensive fitness value is output.

8. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The S5 includes the following specific steps: S51, receiving an optimal valve control strategy sequence, where the optimal valve control strategy sequence includes a target valve opening value corresponding to each time step in a future prediction time period; S52, inputting the target opening instructions into the electric control valves in sequence, executing the instructions in accordance with the control cycle sequence, and adjusting the opening of the electric control valve group in real time; S53. In each control cycle, the electric actuator adjusts the valve opening angle according to the current target opening instruction. The valve opening is continuously adjustable within the range of 0 to 90 degrees. S54. In each control cycle, the electric actuator adjusts the valve opening angle of the electric regulating valve group in real time according to the target opening instruction. At the same time, the actual opening in the cycle is collected and compared with the target opening to obtain the valve opening error: Among them, E adaptive is the adaptive control error feedback value constructed based on the current behavior state selection result, P s (k) is the behavior state selection result of the k-th cat group individual, k is the cat group individual number, is the deviation amplification factor corresponding to the current behavior state, T w N is the total number of cycles of the control feedback sliding time window. w is the number of subsampling in each control cycle, is the target valve opening at the nth sampling point in the tth cycle, is the actual valve opening of the nth sampling point in the tth cycle, t is the time step, and n is the sub-sampling point number; S55. The control system synchronously collects the current target opening, the actual execution opening and the water tank liquid level change value, calculates the execution deviation and records the actual response behavior of the water inlet flow in the current cycle.

9. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The S6 includes the following specific steps: S61. In each control cycle, the actual execution opening data of the electric regulating valve group and the actual water tank liquid level feedback data of the corresponding time step are collected to form a multi-source feedback data pair, and the data is stored in the control system in a time series; S62. Construct an actual control execution sequence and an actual liquid level observation sequence. The actual control execution sequence is a sequence consisting of the actual valve openings at each time step recorded within the control cycle. The actual liquid level observation sequence is a sequence consisting of the actual values of the water tank liquid level obtained by the liquid level sensor at each time step. The actual control execution sequence and the actual liquid level observation sequence have the same time step length. S63. At the same time step, obtain a water tank level prediction sequence output by the neural network prediction model, where the water tank level prediction sequence corresponds one-to-one to the actual liquid level observation sequence; S64. Construct a joint loss function to fuse the tank level prediction sequence and valve control execution deviation for dynamic feedback learning: Among them, L joint is the value of the joint loss function, α is the weighted coefficient of the liquid level prediction error, T' is the total number of time steps, t is the time step, is the actual water tank level value at time step t, is the predicted water tank level value at time step t, is the square of the liquid level prediction error, β is the weighted coefficient of the control execution deviation, E adaptive is an adaptive control error feedback value constructed based on the current behavior state selection result; S65. Based on the joint loss function value, a gradient descent optimization method is used to dynamically adjust the weight parameters in the water level prediction neural network model, and incremental training is performed on the collected feedback data; S66. After each round of closed-loop training is completed, the water level prediction neural network model parameters are updated and the current version is saved.

10. The intelligent water tank inlet valve control system according to claim 2, characterized in that: The S7 includes the following specific steps: S71. Upload the water tank level data, valve control strategy, actual valve opening, water inlet flow, pipe network pressure, and water load parameters in the control system to the monitoring and operation and maintenance platform to achieve real-time access to multi-source operation data; S72. Construct a visualization interface in the monitoring and operation and maintenance platform to display sensor and control execution data in the form of curve graphs, bar graphs, and color block graphs, including the trend of water tank level changes, comparison of valve control and execution status, and operation status distribution; S73. Establish a remote operation interface on the platform side, receive the control adjustment instructions issued by the user side, convert the control adjustment instructions into standard control commands, and perform control according to the standard control commands; S74. Configure historical data query function to support periodic indexing and display of water tank level, water inlet flow, valve control, and execution error, and support data screening and retrospective analysis by day, week, and month; S75. Set up a water consumption trend analysis function, perform periodic modeling and trend extraction on historical water consumption data, and generate a trend change report; S76. Build anomaly detection and alarm functions to perform status assessment based on control execution deviation, water tank level fluctuation, valve response delay, and flow anomalies. When an abnormal event is detected, abnormal alarm information is automatically generated and pushed to the remote user terminal and operation and maintenance background.

Images