Fire hydrant pressure self-adaptive regulation and control system based on Internet of Things data

Abstract

The invention discloses a fire hydrant pressure adaptive regulation and control system based on Internet of Things data, and relates to the technical field of urban water supply pipe network fire hydrant pressure control, and the system comprises a multi-source data collection module which is used for collecting fire hydrant pressure time sequence data in real time from a plurality of Internet of Things nodes deployed in an urban water supply pipe network, and historical pressure data, pipeline topological structure data and user water consumption mode data of the water supply network are obtained from a historical database. Real-time and historical data are integrated through the multi-source data acquisition module, a comprehensive basis is provided for pressure regulation and control in combination with a pipe network topology and a user water consumption mode, and the integrity of data support is improved; the pipe network state prediction module is used for extracting multi-dimensional features and predicting pressure changes by means of an LSTM model, the pressure abnormal risk can be recognized in advance, and hysteresis of traditional regulation and control is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of fire hydrant pressure control in urban water supply pipe networks, and in particular to a fire hydrant pressure adaptive control system based on Internet of Things data. Background Art

[0002] In urban firefighting systems, fire hydrants are critical water supply equipment for firefighting, and their pressure stability is directly related to firefighting and rescue effectiveness. With the development of IoT technology, IoT-based fire hydrant monitoring systems have gradually replaced traditional manual inspections, enabling real-time monitoring of fire hydrant status parameters such as water pressure and water level.

[0003] However, the mainstream control scheme for existing fire hydrant pressure regulation systems primarily utilizes simple threshold control logic. When pressure is detected to exceed a preset range, it passively responds by starting and stopping pumps or regulating valves. This control approach exhibits numerous shortcomings in the complex environment of urban water supply networks. First, urban water supply networks are subject to the combined influence of factors such as multi-user water usage, pipeline aging, and terrain elevation differences. Pressure fluctuations are nonlinear and highly time-varying, making simple threshold control difficult to cope with these complex conditions. Second, existing systems lack the ability to predict pressure trends in the network, making it impossible to implement preventative control measures before pressure anomalies occur. Third, the fixed parameter control strategy employed by traditional systems cannot adapt to the dynamic changes in the network's operating status, easily leading to control lags or over-regulation. These issues can result in fire hydrants being unable to provide stable water supply pressure during sudden fires, seriously impacting firefighting and rescue efficiency. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a fire hydrant pressure adaptive control system based on Internet of Things data, with the aim of predicting the pressure change trend of the pipeline network in advance, dynamically generating precise control strategies, and realizing closed-loop adaptive control of fire hydrant pressure to improve pressure stability.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a fire hydrant pressure adaptive control system based on Internet of Things data, which is used for stable control of fire hydrant pressure in an urban water supply network environment. The system includes:

[0006] Multi-source data acquisition module: used to collect real-time fire hydrant pressure time series data from multiple IoT nodes deployed in the urban water supply network, and obtain historical pressure data, pipeline topology data and user water use pattern data of the water supply network from the historical database;

[0007] Pipeline network status prediction module: used to receive the fire hydrant pressure time series data, the historical pressure data, the pipeline topology data and the user water use pattern data output by the multi-source data acquisition module, and construct a pressure prediction model based on the LSTM neural network. Its training data is the historical pressure time series data. The network structure includes a 3-layer 128-unit LSTM layer. The module predicts the pipeline network pressure status within the next 24-hour time window with a prediction step of 5 minutes, and generates predictive status parameters including the predicted pressure value and pressure change trend;

[0008] Adaptive control strategy generation module: used to dynamically calculate and generate control instructions for adjusting the water pressure of fire hydrants based on the predictive state parameters generated by the pipe network state prediction module and in combination with a preset target pressure range, wherein the control instructions include the operation type and operation value of the pressure regulating actuator;

[0009] Pressure execution control module: used to receive the control instructions generated by the adaptive control strategy generation module, and parse the control instructions into specific drive signals to control the pressure regulating actuator connected to the fire hydrant to perform precise operation, thereby realizing closed-loop adaptive control of the fire hydrant pressure.

[0010] Furthermore, the multi-source data acquisition module includes:

[0011] Real-time data acquisition submodule: used to periodically obtain the real-time pressure readings measured by the pressure sensors installed on each fire hydrant through wireless communication protocols to form the original pressure time series data stream;

[0012] Historical data access submodule: used to access and read the structured database stored in the cloud or local server, and extract historical pressure fluctuation records, pipe network maintenance records, pipe material and aging information related to the current control scenario;

[0013] Correlation data fusion submodule: used to perform timestamp alignment and data format normalization on the original pressure time series data stream, the historical pressure fluctuation record and the pipeline information, and generate a multi-dimensional input data vector in a unified format for use by the pipeline network status prediction module.

[0014] Furthermore, the pipeline network status prediction module includes:

[0015] Feature engineering processing submodule: used to process the multi-dimensional input data vector and extract time domain features and frequency domain features for characterizing the dynamic characteristics of the pipe network pressure;

[0016] Prediction model calling submodule: used to load a pre-trained pressure prediction model based on a long short-term memory network (LSTM), and use the time domain features and the frequency domain features extracted by the feature engineering processing submodule as model input;

[0017] State parameter output submodule: used to run the pressure prediction model, output the prediction results of the pressure value sequence for multiple consecutive time steps in the future, and calculate the maximum predicted pressure value, minimum predicted pressure value and pressure change gradient based on the prediction results, which together constitute the predictive state parameters.

[0018] Furthermore, the feature engineering processing submodule includes:

[0019] A time domain feature extraction unit is used to calculate the mean, variance, peak value, valley value and zero-crossing rate of the pressure time series data within a sliding time window as the time domain feature;

[0020] A frequency domain feature extraction unit is used to apply a fast Fourier transform (FFT) to the pressure time series data to obtain its spectrum information, and extract the main frequency component, spectrum energy concentration and spectrum entropy as the frequency domain features;

[0021] A feature vector construction unit is used to combine the time domain features with the frequency domain features and then apply Z-score normalization: feature value = (original value - mean) / standard deviation to construct a normalized feature vector input to the pressure prediction model.

[0022] Furthermore, the adaptive control strategy generation module includes:

[0023] Target interval definition submodule: used to set a daily maintenance target interval of 0.4-0.6MPa, including the target pressure upper limit and target pressure lower limit, according to the fire incident level or daily maintenance requirements; the target interval for level 1 fire alarm is 0.6-0.8MPa;

[0024] Control decision logic submodule: used to compare the predicted pressure value in the predictive state parameter with the daily maintenance target range of 0.4-0.6MPa and the first level fire alarm target range of 0.6-0.8MPa to determine whether there is a potential overpressure or underpressure risk;

[0025] Control parameter calculation submodule: used to calculate the optimal valve opening adjustment amount or water pump start-stop frequency through a preset nonlinear mapping function based on the deviation between the predicted pressure value and the target pressure range and the severity of the pressure change trend when judging the existence of potential risks.

[0026] Furthermore, the control parameter calculation submodule includes:

[0027] Deviation quantization unit: used to calculate the difference ΔP between the predicted pressure value and the center value of the target pressure interval, and the pressure change gradient G;

[0028] Weight coefficient query unit: used to maintain a preset weight coefficient lookup table, which establishes a mapping relationship between different deviation quantization values ​​ΔP and pressure change gradient G combinations and corresponding control weight coefficients K_p and K_d;

[0029] Control quantity calculation unit: used to calculate the final operation quantity value in the control instruction according to the queried control weight coefficients K_p and K_d through the formula Control_Output=K_p*ΔP+K_d*G, where Control_Output is the final operation quantity value.

[0030] Furthermore, the pressure execution control module includes:

[0031] Instruction decoding submodule: used to receive and parse the control instructions, identify the target actuator address, operation type (such as increase, decrease, hold) and specific operation value;

[0032] A drive signal generation submodule is used to convert the parsed operation value into an electric control signal compatible with the target pressure regulating actuator, wherein the electric control signal is a pulse width modulation (PWM) signal or a serial communication data frame;

[0033] State feedback interface submodule: used to receive execution status receipt information from the pressure regulating actuator, and feed back the receipt information to the adaptive control strategy generation module to realize verification and correction of the control closed loop.

[0034] Furthermore, the pipeline network status prediction module also includes:

[0035] The operating mode recognition submodule is used to classify the current pipe network operating status into one of stable operating conditions, peak water consumption conditions, or sudden abnormal conditions based on the statistical distribution characteristics of the input data before pressure prediction.

[0036] Model selection decision submodule: used to dynamically select a specific pressure prediction model that best matches the current working condition from the model library based on the working condition category output by the working condition mode recognition submodule; the model library pre-stores multiple pressure prediction models trained and optimized for different working conditions.

[0037] Furthermore, the operating mode recognition submodule includes:

[0038] A data segmentation processing unit: configured to divide the continuous multi-dimensional input data vector into data segments of fixed length;

[0039] Feature extraction unit: used for calculating the mean vector, covariance matrix and kurtosis coefficient of the internal data of each data segment;

[0040] Pattern classification unit: used to input the mean vector, covariance matrix and kurtosis coefficient into the pre-trained Gaussian mixture model, and determine the operating condition category to which the current data segment belongs based on the maximum a posteriori probability output by the model, wherein each Gaussian component of the Gaussian mixture model corresponds to a preset pipeline network operating condition.

[0041] Furthermore, the adaptive control strategy generation module also includes:

[0042] Control cost evaluation submodule: used to pre-establish the cost model formula between control operations and energy consumption and equipment wear: Cost = 0.8 × valve opening change + 0.2 × pump start and stop times. The coefficient is calibrated through equipment wear experiments. The candidate operation values ​​output by the control parameter calculation submodule are used to calculate the corresponding predicted control costs.

[0043] Multi-objective optimization submodule: It is used to fine-tune the operation value output by the control parameter calculation submodule on the premise of meeting the pressure control requirements and taking minimizing the predicted control cost as the optimization goal, so as to generate the final control instruction that takes into account both the control effect and economy.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention integrates real-time and historical data through a multi-source data acquisition module, combines the pipe network topology and user water usage patterns, provides a comprehensive basis for pressure control, and improves the integrity of data support; uses the pipe network status prediction module to extract multidimensional features and uses the LSTM model to predict pressure changes, which can identify pressure abnormality risks in advance and avoid the lag of traditional control; the adaptive control strategy generation module dynamically sets the target interval, combines the pressure deviation and change trend to generate control instructions, and through cost evaluation and optimization, reduces energy consumption and equipment wear while ensuring pressure stability; the closed-loop feedback mechanism of the pressure execution control module ensures accurate control and reduces pressure fluctuations, so that the fire hydrant can provide stable water pressure in the event of a fire, improves fire extinguishing efficiency, and at the same time extends equipment life and reduces maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0047] The technical solutions of this application will be described clearly and completely below, in conjunction with the accompanying drawings. It should be understood that the described embodiments represent only a portion of the embodiments of this application, and not all of them. The components of this application, generally described and illustrated in the drawings herein, may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but rather merely represents selected embodiments of this application. All other embodiments derived by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. Furthermore, in the description of this application, the terms "first," "second," etc., are used solely to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0048] In traditional fire hydrant pressure control systems, pressure sensors and controllers trigger pump or valve adjustments based on fixed thresholds. However, urban water supply networks experience dynamic pressure fluctuations due to multiple users, aging pipes, and terrain variations. This results in a lack of predictive control instructions for dynamic changes in network pressure, making it impossible to proactively intervene before pressure spikes or troughs occur. This results in delayed and inaccurate control. For example, in urban areas with multiple high-rise buildings and aging pipelines, network pressure during peak hours is affected by sudden changes in user water usage patterns, resulting in periodic pressure fluctuations. Conventional systems can only trigger control based on current pressure readings and are unable to analyze pressure trends and future fluctuations. This results in frequent actuator starts and stops, but pressure still deviates from the target range. When a sudden pipeline leak causes a sudden drop in pressure, traditional systems, without integrating network topology and historical maintenance data, are unable to quickly identify abnormal operating conditions and generate a matching control strategy. Delayed actuator response further exacerbates pressure imbalances. If the above problems are not solved, the continuous deviation of pipeline network pressure from the safety threshold will lead to unstable water supply pressure of fire hydrants. When a fire occurs, it will be impossible to provide water pressure that meets the fire extinguishing flow requirements, which will prolong the fire control time; frequent operation of actuators will also accelerate the wear of mechanical components, increase maintenance costs and system failure rate; long-term fluctuations in pipeline network pressure may induce pipeline water hammer effect, causing pipeline rupture and secondary disasters.

[0049] When faced with the above problems, this application first analyzes that the fundamental reason for the lag in traditional system regulation is the lack of predictive ability for dynamic changes in the pipeline network. Conventional methods only rely on current pressure readings to trigger regulation and cannot predict future pressure fluctuation trends. In this regard, this application considers introducing a time series data acquisition and prediction model, and constructs a pipeline network pressure dynamic model by fusing multi-source historical data to predict pressure changes in future time windows in advance. At the same time, in response to the problem of frequent actuator action, this application explores dynamically matching the prediction results with the target pressure range, and generates control quantities based on the pressure change trend to avoid mechanical wear caused by a single threshold trigger. In addition, this application further studies how to convert predictive control instructions into closed-loop control signals, and correct execution errors through real-time feedback to ensure the accuracy of pressure regulation. Through the above analysis, this application determines that it is necessary to establish a complete closed-loop system including data acquisition, state prediction, strategy generation and execution control to achieve full process optimization from data perception to precise regulation.

[0050] In this regard, this application proposes a fire hydrant pressure adaptive control system based on IoT data, as follows:

[0051] Please refer to Figure 1 A fire hydrant pressure adaptive control system based on Internet of Things data is used to stabilize fire hydrant pressure in urban water supply network environments. The system includes:

[0052] Multi-source data acquisition module: used to collect real-time fire hydrant pressure time series data from multiple IoT nodes deployed in the urban water supply network, and obtain historical pressure data, pipeline topology data and user water use pattern data of the water supply network from the historical database;

[0053] The pipeline network status prediction module receives the fire hydrant pressure time series data, historical pressure data, pipeline topology data, and user water usage pattern data output by the multi-source data acquisition module, and constructs a pressure prediction model based on the LSTM neural network. Its training data is the historical pressure time series data. The network structure consists of a three-layer 128-unit LSTM layer. It predicts the pipeline network pressure status within a 5-minute prediction step within the next 24-hour time window, and generates predictive status parameters including the predicted pressure value and pressure change trend.

[0054] Adaptive control strategy generation module: This module is used to dynamically calculate and generate control instructions for adjusting the water pressure of fire hydrants based on the predictive state parameters generated by the pipe network state prediction module and the preset target pressure range. The control instructions include the operation type and operation value of the pressure regulating actuator;

[0055] Pressure execution control module: used to receive the control instructions generated by the adaptive control strategy generation module, and parse the control instructions into specific drive signals to control the pressure regulating actuator connected to the fire hydrant for precise operation, thereby realizing closed-loop adaptive control of the fire hydrant pressure.

[0056] Among them, the Internet of Things node refers to a hardware device deployed in the urban water supply network with data collection and communication functions. Specifically, it can be implemented using an embedded device that integrates a pressure sensor and a wireless transmission module. It is used to obtain fire hydrant pressure data in real time and transmit it to the central processing system, solving the problem of the traditional system's single data source and lack of real-time performance.

[0057] Among them, historical pressure data refers to the set of pressure values ​​recorded during the past operation of the water supply network. This can be achieved by regularly retrieving historical records stored in structured form through the database query interface, which can be used to analyze the long-term trend of network pressure changes and enhance the predictability of control strategies.

[0058] Among them, pipeline topology data refers to the spatial information that describes the connection relationship between each pipeline in the water supply network. Specifically, it can be implemented using the node connection matrix derived from the GIS geographic information system to establish a physical model of pipeline network pressure conduction and accurately simulate the propagation path of pressure fluctuations.

[0059] Among them, user water use pattern data refers to the statistical characteristics of water consumption of end users in different time periods. It can be achieved through cluster analysis of periodic water use records collected by smart water meters, and is used to predict the impact of future changes in water demand on pipeline network pressure.

[0060] Among them, the pipeline network pressure dynamic model refers to the mathematical expression that describes the change of pressure over time. It can be implemented by a hybrid modeling method based on the combination of fluid mechanics equations and machine learning, which is used to convert multi-source input data into a computational framework that can predict pressure states.

[0061] Among them, the predictive state parameter refers to a quantitative indicator that includes future pressure values ​​and their rate of change. Specifically, it can be achieved by outputting a pressure value sequence through a time series prediction algorithm and calculating its first-order derivative, which is used to identify the risk of pressure deviating from the target range in advance.

[0062] The target pressure range refers to the pre-set allowable pressure fluctuation range. This can be achieved by setting upper and lower thresholds based on the type of fire hydrant and fire regulations, providing a benchmark reference for generating control instructions. The pressure regulating actuator is a mechanical device that can change the pressure of the pipeline fluid. This can be achieved using an electric valve or variable frequency water pump. It is used to adjust the water pressure in the pipeline network to the target range according to the control instructions.

[0063] Among them, closed-loop adaptive control refers to a technical means of continuously optimizing the control effect based on real-time feedback. Specifically, it can be achieved by transmitting the actuator status information back to the strategy generation module to form a control loop, which is used to eliminate the response lag problem existing in traditional open-loop control.

[0064] The core innovation of this application lies in the construction of an adaptive control system that integrates multi-source data acquisition, dynamic pressure prediction and closed-loop feedback control. Fire hydrant pressure data is obtained in real time through IoT nodes, and a prediction model is established by combining historical data with pipe network topology to generate control instructions that take into account future pressure changes. The control parameters are continuously optimized based on actuator feedback to achieve precise and stable control of fire hydrant pressure in complex water supply environments.

[0065] The working process and principle of this application are as follows: the system includes a multi-source data acquisition module, a pipe network state prediction module, an adaptive control strategy generation module and a pressure execution control module. The multi-source data acquisition module collects fire hydrant pressure time series data from the Internet of Things node in real time, and obtains historical pressure data, pipeline topology data and user water use pattern data. The pipe network state prediction module receives these data, and the pressure prediction model constructed based on the LSTM neural network has its training data as historical pressure time series data. The network structure contains a 3-layer 128-unit LSTM layer, which predicts the pipe network pressure state within the next 24-hour time window with a prediction step of 5 minutes, and generates predictive state parameters. The adaptive control strategy generation module dynamically calculates and generates control instructions based on the predictive state parameters and the preset target pressure range. The pressure execution control module receives the control instructions and parses them into drive signals to control the pressure regulating actuator for precise operation.

[0066] These modules work together to achieve closed-loop adaptive control, from data acquisition and state prediction to strategy generation and execution control. Multi-source data acquisition provides comprehensive input for prediction. Pipeline network state prediction enables anticipation of future pressure changes. Adaptive control strategy generation dynamically generates control instructions based on the prediction results, and pressure execution control translates these instructions into specific actions. Through predictive control and closed-loop control, the system can promptly respond to changes in pipeline network pressure and achieve precise control.

[0067] Key technical features were chosen for their effectiveness: multi-source data fusion for improved prediction accuracy, predictive models for early intervention, dynamic control strategy generation to adapt to complex operating conditions, and closed-loop control to ensure precise execution. These features effectively address the control lag and inaccuracy inherent in traditional systems.

[0068] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0069] The multi-source data acquisition module uses a wireless sensor network to collect real-time pressure data from pressure sensors distributed throughout the city's water supply network, sampling at a frequency of 1 Hz. Simultaneously, the module connects to the city's water supply management system database to obtain historical pressure data from the past 30 days, along with a network topology map and user water consumption statistics.

[0070] The pipeline network status prediction module uses a long short-term memory (LSTM) network to build a pressure prediction model. The input layer receives time-series pressure data, historical data, and topology information from the multi-source data acquisition module. The hidden layer contains 128 LSTM units to capture the long-term and short-term dependencies between pressure changes. The output layer predicts a series of pressure values ​​at 5-minute intervals for the next hour.

[0071] The adaptive control strategy generation module is designed based on a fuzzy control algorithm. First, the target pressure range is defined as 0.4-0.6 MPa. The predicted pressure value is then compared with the target range, and the deviation and rate of change are calculated. Based on the deviation and trend, the module queries a pre-set fuzzy rule library to infer the corresponding control variable.

[0072] The pressure actuator control module uses a PID controller for closed-loop control. The controller receives the target pressure value generated by the adaptive control strategy and calculates the valve opening adjustment based on real-time pressure feedback. A 4-20mA analog signal drives the actuator to precisely adjust the valve opening, achieving dynamic regulation of the fire hydrant's outlet pressure.

[0073] Through the above scheme, this application realizes predictive and adaptive control of fire hydrant pressure. The system can predict the trend of changes in pipeline network pressure in advance and generate matching control strategies in time, avoiding the problem of lag in traditional system control. The accuracy of pressure prediction is improved by integrating multi-source data and adopting advanced prediction algorithms. The dynamically generated control strategy can adapt to complex and changeable pipeline network conditions and is more flexible and effective than the fixed threshold trigger method. The closed-loop control mechanism ensures execution accuracy and reduces pressure fluctuations. In actual application, the system can stably maintain the fire hydrant pressure within the target range, providing reliable protection for fire water supply. At the same time, due to the reduction of frequent start and stop of actuators, equipment wear is reduced and the service life of the system is extended.

[0074] The present application further proposes that the multi-source data acquisition module includes a real-time data acquisition submodule, a historical data access submodule, and a related data fusion submodule.

[0075] Among them, the real-time data acquisition submodule periodically acquires pressure sensor data through wireless communication protocol to form the original pressure time series data stream. The data acquisition period can be set to 1-5 seconds;

[0076] The historical data access submodule accesses the structured database on the cloud or local server through the API interface to extract historical pressure fluctuation records, pipeline maintenance records, pipeline material and aging information. The database access frequency is synchronized with the real-time data collection cycle;

[0077] The associated data fusion submodule uses a time series alignment algorithm to match the timestamps of real-time data streams with historical data, and uses a data format converter to convert data of different units and precisions into a standard floating-point format to construct a multi-dimensional input data vector.

[0078] Specifically, the real-time data acquisition submodule collects raw pressure values ​​from the fire hydrant pressure sensor at fixed time intervals through the ZigBee or LoRa wireless communication protocol, and generates a raw data stream containing timestamps, device IDs, and pressure values.

[0079] The historical data access submodule reads historical pressure fluctuation records from the relational database, including hourly average pressure values, peak pressures, and durations, and extracts material types, installation years, and corrosion level indicators from the pipeline topology.

[0080] The associated data fusion submodule uses a sliding time window mechanism to partition the real-time data stream into minute-level windows. A sliding average of the pressure data within each window is calculated and spatially correlated with the historical data within the same time window. During data normalization, pressure units are uniformly converted to megapascals, timestamp accuracy is adjusted to milliseconds, and missing data is filled using linear interpolation. The multidimensional input data vector ultimately contains dimensions such as the real-time mean pressure, historical peak pressure values, and pipeline aging coefficients, which are then processed by the prediction module.

[0081] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0082] The multi-source data acquisition module consists of a real-time data acquisition submodule, a historical data access submodule, and a correlated data fusion submodule. The real-time data acquisition submodule uses the LoRaWAN wireless communication protocol to acquire pressure readings from pressure sensors installed on fire hydrants every five minutes, generating a raw pressure time-series data stream. The historical data access submodule uses SQL queries to extract pressure fluctuation records from the past 30 days, recent pipeline maintenance records, and information about pipeline material and age from a cloud-based MySQL database. The correlated data fusion submodule aligns the timestamps of the raw pressure time-series data stream with the historical data and converts them into a unified JSON format, generating a multidimensional input data vector containing fields such as pressure value, timestamp, and pipeline ID.

[0083] By the technical solution, efficient collection and fusion processing of multi-source heterogeneous data are realized. Thus, the completeness and timeliness of the data are improved, comprehensive and real-time input information is provided for subsequent pipe network state prediction, and the accuracy and adaptability of pressure regulation are enhanced. Specifically, high-frequency collection of real-time data enables the system to quickly respond to sudden pressure changes, introduction of historical data helps to capture long-term pressure change trends, and fusion processing of associated data ensures consistency and availability of data from different sources.

[0084] The application further proposes that the pipe network state prediction module comprises a feature engineering processing submodule, a prediction model calling submodule, and a state parameter output submodule. The feature engineering processing submodule is used to process the multi-dimensional input data vector and extract time domain features and frequency domain features representing the dynamic characteristics of pipe network pressure. The prediction model calling submodule is used to load a pre-trained pressure prediction model based on a long short-term memory network and input the time domain features and frequency domain features as model inputs. The state parameter output submodule is used to run the pressure prediction model, output a pressure value sequence prediction result for a future continuous number of time steps, and calculate a maximum predicted pressure value, a minimum predicted pressure value, and a pressure change gradient based on the prediction result, which together constitute predictive state parameters.

[0085] The feature engineering processing submodule calculates the mean, variance, peak value, valley value, and zero-crossing rate in the sliding time window through a time domain feature extraction unit, a frequency domain feature extraction unit applies fast Fourier transform to obtain frequency spectrum information and extract the main frequency component, frequency spectrum energy concentration degree, and spectrum entropy, and a feature vector construction unit combines and normalizes the time domain and frequency domain features to form an input vector. The prediction model calling submodule uses a long short-term memory network to process time series features, and the state parameter output submodule calculates the gradient change trend through the pressure value sequence.

[0086] Specifically, the multidimensional input data vector first enters the feature engineering processing submodule. Within a sliding time window, the time-domain feature extraction unit calculates statistical indicators of the pressure time series data. For example, for a 5-second window length, the mean reflects the average pressure level, the variance characterizes the fluctuation amplitude, the peak and valley values ​​record extreme values, and the zero-crossing rate reflects the fluctuation frequency. The frequency-domain feature extraction unit performs a fast Fourier transform on the same window data. For example, for a sampling frequency of 10 Hz, spectrum analysis determines whether the main frequency component is within the normal fluctuation range of 0.5-2 Hz. A spectral energy concentration exceeding 70% indicates periodic fluctuation, and a spectral entropy below 1.5 bits indicates a concentrated spectral distribution. The feature vector construction unit normalizes the time-domain and frequency-domain features, for example, using the Z-score method to eliminate dimensionality differences, to form an 8-dimensional feature vector for input into the prediction model. The long-short-term memory network uses memory cells to store historical states, handle temporal dependencies, and output a sequence of predicted pressure values ​​for each second over the next 30 seconds. The state parameter output submodule extracts the maximum and minimum values ​​from the predicted sequence and calculates the average of the pressure differences between adjacent time steps as the change gradient. For example, when the predicted pressure value rises from 0.5 MPa to 0.6 MPa within 10 seconds, the gradient is calculated to be 0.01 MPa / s. This technical solution significantly improves the accuracy and timeliness of pressure state prediction by combining multidimensional feature extraction with a deep learning model, providing a reliable basis for adaptive regulation.

[0087] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0088] The pipeline network status prediction module includes a feature engineering processing submodule, a prediction model calling submodule and a status parameter output submodule.

[0089] The feature engineering submodule receives multidimensional input data vectors and extracts time-domain and frequency-domain features that characterize the dynamic characteristics of pipeline network pressure. Specifically, this submodule applies sliding window processing to the input pressure time series data, with a window size of 1 hour and a sliding step of 5 minutes. Within each window, the mean, variance, peak value, valley value, and zero-crossing rate are calculated as time-domain features. Simultaneously, a fast Fourier transform is performed on the data within the window to extract the main frequency component, spectral energy concentration, and spectral entropy as frequency-domain features.

[0090] The prediction model call submodule loads a pre-trained stress prediction model based on a long short-term memory network. This model uses a three-layer LSTM structure, with each layer containing 128 neurons. The input layer corresponds to the features extracted by the feature engineering submodule, and the output layer predicts stress values ​​every 5 minutes for the next 24 hours.

[0091] The state parameter output submodule runs the pressure prediction model and outputs a 24-hour pressure forecast sequence. Based on this sequence, it calculates the maximum and minimum predicted pressure values ​​within the 24-hour period, as well as the pressure gradient between adjacent time steps. These parameters together constitute the predictive state parameters used in subsequent pressure control decisions.

[0092] Through the above technical solution, this application realizes the accurate prediction of the future pressure state of the urban water supply network. As a result, the system can identify potential pressure anomalies in advance, provide a decision-making basis for the adaptive regulation of fire hydrant pressure, and effectively avoid the regulation lag problem caused by the lack of foresight in traditional methods. Furthermore, by combining time domain and frequency domain features, this solution can capture the complex dynamic characteristics of pressure changes, and the prediction accuracy is significantly improved compared to methods that rely only on a single type of feature. In addition, the LSTM network is used as the core prediction model, which fully utilizes the time series correlation of pressure data and overcomes the limitations of traditional linear prediction methods in dealing with nonlinear and non-stationary time series.

[0093] The present application further proposes that the feature engineering processing submodule includes a time domain feature extraction unit, a frequency domain feature extraction unit and a feature vector construction unit.

[0094] Among them, the time domain feature extraction unit calculates the mean, variance, peak value, valley value and zero-crossing rate of the pressure time series data within the sliding time window. These statistics are used to quantify the fluctuation amplitude and change frequency of the pressure data;

[0095] The frequency domain feature extraction unit applies fast Fourier transform to the pressure time series data to extract the main frequency component, spectrum energy concentration and spectrum entropy to characterize the periodic characteristics and energy distribution characteristics of pressure fluctuations;

[0096] The feature vector construction unit combines the time domain features with the frequency domain features and performs standardization processing to eliminate the impact of the differences in different feature dimensions on model training and form a standardized feature vector.

[0097] Specifically, the time domain feature extraction unit continuously calculates statistical indicators of the pressure data through a sliding window mechanism, such as calculating the mean value within a 30-second window to reflect the pressure baseline level, the variance to reflect the fluctuation intensity, and the zero-crossing rate to reflect the frequency of switching the direction of pressure change; the frequency domain feature extraction unit converts the pressure signal to the frequency domain using fast Fourier transform, identifies the periodic fluctuation mode by extracting the main frequency component, measures the energy proportion of the main frequency band by the spectral energy concentration, and evaluates the spectral complexity by the spectral entropy; the feature vector construction unit normalizes the combined multi-dimensional features using the z-score standardization method, so that each dimension of the features has a zero mean and a unit variance. Through the above processing, the prediction model can obtain high-quality input data containing both time domain dynamic characteristics and frequency domain periodic characteristics, improve the prediction accuracy of future stress states, and thus provide a reliable basis for generating accurate control instructions.

[0098] As a preferred embodiment, the scheme of the application is implemented as follows:

[0099] The feature engineering processing submodule includes a time domain feature extraction unit, a frequency domain feature extraction unit, and a feature vector construction unit. The time domain feature extraction unit calculates the mean value, variance, peak value, valley value, and zero-crossing rate of the pressure time series data within a sliding time window as time domain features. The frequency domain feature extraction unit applies fast Fourier transform to the pressure time series data to obtain its frequency spectrum information and extracts the main frequency component, spectral energy concentration, and spectral entropy as frequency domain features. The feature vector construction unit combines the time domain features and the frequency domain features and applies Z-score standardization: feature value = (original value - mean value) / standard deviation, to construct a standardized feature vector input to the pressure prediction model.

[0100] In specific implementation, the time domain feature extraction unit uses a sliding time window with a length of 60 seconds, sliding every 10 seconds. Within each time window, the arithmetic mean, sample variance, maximum value, minimum value, and the number of times the signal crosses zero of the pressure data are calculated. The frequency domain feature extraction unit performs 512-point FFT transformation on the data of each 60-second window, extracts the top three frequency components with the largest amplitude in the frequency spectrum, calculates the proportion of spectral energy in the 0-1 Hz, 1-5 Hz, and 5-10 Hz frequency bands, and calculates the entropy value of the frequency spectrum using the Shannon entropy formula. The feature vector construction unit combines the above 13 features (5 time domain features and 8 frequency domain features) into a feature vector and normalizes each feature using the Z-score method.

[0101] Through the above technical solution, this application achieves comprehensive feature extraction of pressure time series data. Time domain features reflect the statistical characteristics of pressure fluctuations, and frequency domain features capture the periodic patterns of pressure signals. The combination of the two provides rich input information. Standardization ensures the comparability of features of different dimensions. This multi-dimensional feature engineering provides high-quality input for subsequent pressure prediction models, improves the accuracy and robustness of predictions, and thus supports more precise pressure control decisions.

[0102] The deviation quantification unit converts pressure deviation and dynamic trends into quantitative parameters by calculating the difference ΔP between the predicted pressure value and the center value of the target interval, as well as the pressure change gradient G. The weight coefficient query unit includes a preset lookup table that stores the proportional coefficient K_p and differential coefficient K_d corresponding to different ΔP and G combinations. The control quantity calculation unit uses the formula Control_Output = K_pΔP + K_dG to generate the final operation value, where K_p and K_d are dynamically adjusted based on the current pressure deviation level and rate of change.

[0103] Specifically, when the predicted pressure value deviates from the center of the target interval, the deviation quantization unit calculates the absolute value of ΔP and simultaneously determines the direction of pressure change based on the pressure gradient G. The weight coefficient query unit matches the corresponding K_p and K_d value combinations in the lookup table based on the absolute value of ΔP and the sign of G. For example, when ΔP = 0.3 MPa and G = 0.05 MPa / s, the query results in K_p = 2.5 and K_d = 1.8. Through linear combination calculation, the valve opening adjustment amount is 2.5 × 0.3 + 1.8 × 0.05 = 0.84 units. This calculation method ensures that the control amount reflects the current deviation while taking into account the inertia of the pressure change trend, avoiding overshoot caused by single proportional control. By dynamically adjusting the weight coefficient, the system can increase the control force in advance during the rapid pressure rise phase and automatically reduce the adjustment range when the pressure stabilizes, achieving precise and energy-saving pressure control.

[0104] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0105] The adaptive control strategy generation module includes a target interval definition submodule, a control decision logic submodule and a control parameter calculation submodule.

[0106] The target range definition submodule sets the daily maintenance target range of 0.4-0.6MPa and the level 1 fire alarm target range of 0.6-0.8MPa based on the fire incident level or daily maintenance requirements. For example, when a level 1 fire alarm signal is received from the fire command center, the target pressure range is set to 0.6-0.8MPa; for a level 2 fire alarm, it is set to 0.5-0.7MPa; and in the daily maintenance state, it is set to 0.4-0.6MPa.

[0107] The control decision logic submodule compares the predicted pressure value in the predictive state parameter with the daily maintenance target range of 0.4-0.6 MPa and the first-level fire alarm target range of 0.6-0.8 MPa. Specifically, if the predicted pressure value is lower than the target pressure lower limit, an underpressure risk is determined; if it is higher than the target pressure upper limit, an overpressure risk is determined.

[0108] When a potential risk is identified, the control parameter calculation submodule uses a pre-set nonlinear mapping function to calculate the optimal valve opening adjustment or pump start / stop frequency based on the deviation between the predicted pressure value and the target pressure range, as well as the severity of the pressure fluctuation trend. Furthermore, a fuzzy control algorithm is used to construct the nonlinear mapping function, taking the pressure deviation and pressure fluctuation trend as input variables and outputting the corresponding control variable. This enables differentiated handling of varying pressure fluctuations, improving both the accuracy and flexibility of control.

[0109] Through the above technical solution, the present application realizes dynamic adaptive regulation of fire hydrant pressure. Due to the introduction of a dynamic definition mechanism for the target pressure range, the system can flexibly adjust the pressure control target according to different fire event levels or maintenance needs. At the same time, by comparing the predicted pressure value with the target range and making decisions based on the pressure change trend, the system's ability to predict potential pressure anomalies is improved. In addition, the use of nonlinear mapping functions to calculate control parameters makes the control strategy more refined, and corresponding control measures can be taken for different degrees of pressure deviation, thereby effectively improving the accuracy and response speed of fire hydrant pressure control, providing more reliable water pressure protection for fire rescue.

[0110] The present application further proposes that the control parameter calculation submodule includes a deviation quantization unit, a weight coefficient query unit and a control quantity calculation unit.

[0111] Among them, the deviation quantization unit calculates the difference ΔP between the predicted pressure value and the center value of the target pressure interval and the pressure change gradient G; the weight coefficient query unit maintains a preset weight coefficient lookup table, which establishes a mapping relationship between the control weight coefficients K_p and K_d corresponding to different ΔP and G combinations; the control quantity calculation unit calculates the operation quantity value based on the queried K_p and K_d through the formula Control_Output=K_pΔP+K_dG.

[0112] Specifically, the deviation quantization unit quantifies the difference ΔP between the predicted pressure value and the target center value into a specific value, such as ΔP=0.3MPa, and extracts the pressure change gradient G=0.05MPa / s. The weight coefficient query unit maps the combination of ΔP=0.3MPa and G=0.05MPa / s to K_p=2 and K_d=0.5 based on the preset lookup table. The control quantity calculation unit substitutes the value into the formula and obtains Control_Output=20.3+0.50.05=0.625, which corresponds to the adjustment amount of the valve opening. Through the pre-established weight coefficient mapping relationship, the system can adaptively adjust the control weight according to the dynamic combination of deviation and gradient, avoid the over-adjustment or under-adjustment problem caused by a single weight coefficient under complex working conditions, and improve the control accuracy and response speed.

[0113] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0114] The control parameter calculation submodule includes a deviation quantization unit, a weight coefficient query unit, and a control quantity calculation unit. The deviation quantization unit calculates the difference ΔP between the predicted pressure value and the center value of the target pressure range, as well as the pressure change gradient G. The weight coefficient query unit maintains a preset weight coefficient lookup table, which establishes a mapping relationship between different combinations of deviation quantization values ​​ΔP and pressure change gradient G and corresponding control weight coefficients K_p and K_d. Based on the queried control weight coefficients K_p and K_d, the control quantity calculation unit calculates the final control instruction operation value using the formula Control_Output = K_pΔP + K_dG, where Control_Output is the final operation value.

[0115] During specific implementation, the deviation quantization unit receives the predicted pressure value and pressure change gradient from the pipeline state prediction module, as well as the target pressure interval information from the target interval definition submodule. The deviation quantization unit calculates the difference ΔP between the predicted pressure value and the center value of the target pressure interval, and directly obtains the pressure change gradient G. The weight coefficient query unit pre-stores a weight coefficient lookup table, which contains multiple sets of ΔP and G combinations and their corresponding K_p and K_d values. For example, when |ΔP|_0.1MPa and |G|_0.01MPa / min, K_p=0.5 and K_d=0.2 correspond to it; when 0.1MPa≤|ΔP|_0.3MPa and 0.01MPa / min≤|G|_0.05MPa / min, K_p=0.8 and K_d=0.4 correspond to it. The control quantity calculation unit substitutes the queried K_p and K_d into the formula to calculate Control_Output. If Control_Output is positive, a control command is generated to increase pressure; if it is negative, a control command is generated to decrease pressure. Ultimately, the absolute value of Control_Output is converted into a specific valve opening adjustment amount or pump start and stop frequency.

[0116] Through the above technical solution, the present application realizes the generation of adaptive control strategies based on predicted pressure values ​​and pressure change trends. As a result, the system can quickly calculate the optimal control parameters according to the dynamic changes of the current pipeline network status, avoiding the lag and inflexibility of the fixed threshold control method. Furthermore, by introducing a weight coefficient lookup table, the system can take control measures of corresponding intensity for different degrees of pressure deviation and change rate, thereby improving the accuracy and response speed of the control. Specifically, when faced with sudden pressure fluctuations, the system can adjust the fire hydrant pressure in a timely manner to ensure that it always remains within a safe and effective pressure range, thereby improving the reliability and effectiveness of fire hydrants in emergency situations.

[0117] In summary, the present invention uses a multi-source data acquisition module to acquire real-time pressure data from IoT nodes at a 1Hz sampling frequency. It also simultaneously retrieves 30 days of historical pressure, network topology, and user water usage data, which are then fused to form a multidimensional vector. The network state prediction module extracts time-domain (mean, variance, etc.) and frequency-domain (dominant frequency, spectral entropy, etc.) features from this data, inputting them into a three-layer LSTM model to predict the pressure series for the next 24 hours, generating maximum / minimum pressures and gradients. The adaptive control strategy generation module sets a dynamic target range of 0.4-0.8 MPa based on the fire rating. Based on the predicted deviation ΔP and the gradient G, it matches K_p and K_d using a weighting table, calculating the control variable as Control_Output = K_p・ΔP + K_d・G. Energy consumption and wear are also optimized using a cost model. The pressure execution control module interprets the instructions as PWM signals or serial data to drive the actuator, forming a closed loop through state feedback to achieve adaptive and stable pressure control.

[0118] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. Various changes and improvements are possible without departing from the spirit and scope of the present invention, and such changes and improvements fall within the scope of the invention as claimed.

Claims

1. The fire hydrant pressure adaptive control system based on Internet of Things data is used for stable control of fire hydrant pressure in urban water supply network environment, which is characterized by: The system includes: Multi-source data acquisition module: used to collect real-time fire hydrant pressure time series data from multiple IoT nodes deployed in the urban water supply network, and obtain historical pressure data, pipeline topology data and user water use pattern data of the water supply network from the historical database; Pipeline network status prediction module: used to receive the fire hydrant pressure time series data, the historical pressure data, the pipeline topology data and the user water use pattern data output by the multi-source data acquisition module, and construct a pressure prediction model based on the LSTM neural network. Its training data is the historical pressure time series data. The network structure includes a 3-layer 128-unit LSTM layer. The module predicts the pipeline network pressure status within the next 24-hour time window with a prediction step of 5 minutes, and generates predictive status parameters including the predicted pressure value and pressure change trend; Adaptive control strategy generation module: used to dynamically calculate and generate control instructions for adjusting the water pressure of fire hydrants based on the predictive state parameters generated by the pipe network state prediction module and in combination with a preset target pressure range, wherein the control instructions include the operation type and operation value of the pressure regulating actuator; Pressure execution control module: used to receive the control instructions generated by the adaptive control strategy generation module, and parse the control instructions into specific drive signals to control the pressure regulating actuator connected to the fire hydrant to perform precise operation, thereby realizing closed-loop adaptive control of the fire hydrant pressure.

2. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 1 is characterized in that: The multi-source data acquisition module includes: Real-time data acquisition submodule: used to periodically acquire the real-time pressure readings measured by the pressure sensors installed on each fire hydrant through wireless communication protocols to form the original pressure time series data stream; Historical data access submodule: used to access and read the structured database stored in the cloud or local server, and extract historical pressure fluctuation records, pipe network maintenance records, pipe material and aging information related to the current control scenario; Correlation data fusion submodule: used to perform timestamp alignment and data format normalization on the original pressure time series data stream, the historical pressure fluctuation record and the pipeline information, and generate a multi-dimensional input data vector in a unified format for use by the pipeline network status prediction module.

3. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 1 is characterized in that: The pipeline network status prediction module includes: Feature engineering processing submodule: used to process the multi-dimensional input data vector and extract time domain features and frequency domain features for characterizing the dynamic characteristics of the pipe network pressure; Prediction model calling submodule: used to load a pre-trained pressure prediction model based on a long short-term memory network (LSTM), and use the time domain features and the frequency domain features extracted by the feature engineering processing submodule as model input; State parameter output submodule: used to run the pressure prediction model, output the prediction results of the pressure value sequence for multiple consecutive time steps in the future, and calculate the maximum predicted pressure value, minimum predicted pressure value and pressure change gradient based on the prediction results, which together constitute the predictive state parameters.

4. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 3 is characterized in that: The feature engineering processing submodule includes: A time domain feature extraction unit is used to calculate the mean, variance, peak value, valley value and zero-crossing rate of the pressure time series data within a sliding time window as the time domain feature; A frequency domain feature extraction unit is used to apply a fast Fourier transform (FFT) to the pressure time series data to obtain its spectrum information, and extract the main frequency component, spectrum energy concentration and spectrum entropy as the frequency domain features; A feature vector construction unit is used to combine the time domain features with the frequency domain features and then apply Z-score normalization: feature value = (original value - mean) / standard deviation to construct a normalized feature vector input to the pressure prediction model.

5. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 1 is characterized in that: The adaptive control strategy generation module includes: Target interval definition submodule: used to set a daily maintenance target interval of 0.4-0.6MPa, including the target pressure upper limit and target pressure lower limit, according to the fire incident level or daily maintenance requirements; the target interval for level 1 fire alarm is 0.6-0.8MPa; Control decision logic submodule: used to compare the predicted pressure value in the predictive state parameter with the daily maintenance target range of 0.4-0.6MPa and the first level fire alarm target range of 0.6-0.8MPa to determine whether there is a potential overpressure or underpressure risk; Control parameter calculation submodule: used to calculate the optimal valve opening adjustment amount or water pump start-stop frequency through a preset nonlinear mapping function based on the deviation between the predicted pressure value and the target pressure range and the severity of the pressure change trend when judging the existence of potential risks.

6. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 5 is characterized in that: The control parameter calculation submodule includes: Deviation quantization unit: used to calculate the difference ΔP between the predicted pressure value and the center value of the target pressure interval, and the pressure change gradient G; Weight coefficient query unit: used to maintain a preset weight coefficient lookup table, which establishes a mapping relationship between different deviation quantization values ​​ΔP and pressure change gradient G combinations and corresponding control weight coefficients K_p and K_d; Control quantity calculation unit: used to calculate the final operation quantity value in the control instruction according to the queried control weight coefficients K_p and K_d through the formula Control_Output=K_p*ΔP+K_d*G, where Control_Output is the final operation quantity value.

7. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 1 is characterized in that: The pressure execution control module includes: Instruction decoding submodule: used to receive and parse the control instruction, identify the target actuator address, operation type (such as increase, decrease, hold) and specific operation value; A drive signal generation submodule is used to convert the parsed operation value into an electric control signal compatible with the target pressure regulating actuator, wherein the electric control signal is a pulse width modulation (PWM) signal or a serial communication data frame; State feedback interface submodule: used to receive execution status receipt information from the pressure regulating actuator, and feed back the receipt information to the adaptive control strategy generation module to realize verification and correction of the control closed loop.

8. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 3 is characterized in that: The pipeline network status prediction module also includes: The operating mode recognition submodule is used to classify the current pipe network operating status into one of stable operating conditions, peak water consumption conditions, or sudden abnormal conditions based on the statistical distribution characteristics of the input data before pressure prediction. Model selection decision submodule: used to dynamically select a specific pressure prediction model that best matches the current working condition from the model library based on the working condition category output by the working condition mode recognition submodule; the model library pre-stores multiple pressure prediction models trained and optimized for different working conditions.

9. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 8 is characterized in that: The operating mode recognition submodule includes: A data segmentation processing unit: configured to divide the continuous multi-dimensional input data vector into data segments of fixed length; Feature extraction unit: used for calculating the mean vector, covariance matrix and kurtosis coefficient of the internal data of each data segment; Pattern classification unit: used to input the mean vector, covariance matrix and kurtosis coefficient into the pre-trained Gaussian mixture model, and determine the operating condition category to which the current data segment belongs based on the maximum a posteriori probability output by the model, wherein each Gaussian component of the Gaussian mixture model corresponds to a preset pipeline network operating condition.

10. The fire hydrant pressure adaptive control system based on Internet of Things data according to claim 5 is characterized in that: The adaptive control strategy generation module also includes: Control cost evaluation submodule: used to pre-establish the cost model formula between control operations and energy consumption and equipment wear: Cost = 0.8 × valve opening change + 0.2 × pump start and stop times. The coefficient is calibrated through equipment wear experiments. The candidate operation values ​​output by the control parameter calculation submodule are used to calculate the corresponding predicted control costs. Multi-objective optimization submodule: It is used to fine-tune the operation value output by the control parameter calculation submodule on the premise of meeting the pressure control requirements and taking minimizing the predicted control cost as the optimization goal, so as to generate the final control instruction that takes into account both the control effect and economy.

Images