Fluid pipe network intelligent management and control system and method based on multi-order spatial-temporal feature fusion

Abstract

The invention provides a fluid pipe network intelligent management and control system and method based on multi-order spatial-temporal feature fusion, and the system comprises an embedded Internet of Things sensing layer which comprises an electric valve control mainboard and / or a pneumatic valve control mainboard embedded with a remote information transmission module, the flow sensor, the heat sensor and the water invasion sensor are connected with the control mainboard; the multi-order spatio-temporal feature processing layer comprises a spatio-temporal feature extraction module configured to perform sliding window slicing processing on time sequence data of pressure, flow velocity and temperature parameters; and the feature fusion module is configured to perform feature alignment on the vibration atlas, the voiceprint signals and the infrared thermal imaging data through a multi-head attention mechanism. According to the fluid pipe network intelligent management and control system and method based on multi-order spatio-temporal feature fusion, the management and control efficiency and safety of the fluid pipe network are remarkably improved through embedded multi-source sensing, multi-order spatio-temporal feature fusion and an intelligent decision framework.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to an intelligent control system and method for fluid pipe networks based on multi-order spatio-temporal feature fusion. Background Art

[0002] Traditional fluid pipe network control systems mostly rely on single sensor data and static models, and it is difficult to effectively fuse multi-source heterogeneous spatio-temporal features, resulting in lagging fault warnings and insufficient robustness of control strategies.

[0003] In the prior art, the data acquisition frequency is low (usually lower than 1 kHz), the feature extraction dimension is single (such as only using time-domain statistics), and there is a lack of cross-platform real-time interaction ability. Especially when dealing with emergency working conditions such as sudden pressure changes, the dependence on manual decision-making is high, and misoperations are likely to occur.

[0004] In addition, the traditional database architecture is difficult to support high-frequency data concurrent processing, and the network security risks are prominent, unable to meet the real-time control requirements in the industrial Internet of Things scenario. Summary of the Invention

[0005] In view of this, the present invention aims to propose an intelligent control system and method for fluid pipe networks based on multi-order spatio-temporal feature fusion to solve at least one problem in the background art.

[0006] To achieve the above object, the technical solution of the present invention is realized as follows:

[0007] An intelligent control system for fluid pipe networks based on multi-order spatio-temporal feature fusion, comprising:

[0008] An embedded Internet of Things perception layer, including an electric valve control main board and / or a pneumatic valve control main board embedded with a remote information transmission module, and a flow sensor, a heat sensor, and a water intrusion sensor connected to the control main board;

[0009] A multi-order spatio-temporal feature processing layer, including:

[0010] A spatio-temporal feature extraction module configured to perform sliding window slicing processing on the time series data of pressure, flow rate, and temperature parameters, and extract first-order statistical features including mean and variance, second-order frequency domain features of Fourier transform coefficients, and third-order non-linear features of Lyapunov exponents;

[0011] A feature fusion module configured to perform feature alignment on vibration spectrograms, voiceprint signals, and infrared thermal imaging data through a multi-head attention mechanism;

[0012] An intelligent decision-making layer, the output end of the multi-order spatio-temporal feature processing layer is connected to the input end of the intelligent decision-making layer through a message queue middleware, including:

[0013] The digital twin module constructs a visualization model containing the three-dimensional topological structure of the pipeline and hydrodynamic parameters;

[0014] The fault prediction module includes an LSTM neural network with residual connections;

[0015] The linkage control module is configured to generate an instruction queue containing valve identifications and execution timings;

[0016] The cross-platform interaction layer includes a Linux server running a real-time in-memory database and a Windows terminal communicating with the server via the OPC UA protocol.

[0017] Furthermore, in the embedded IoT sensing layer:

[0018] The remote information transmission module includes a 4G / 5G dual-mode communication chip and an AES-256 encryption unit;

[0019] The vibration sensor array acquires vibration waveforms at a sampling rate of 50 kHz and is configured with a wavelet packet transform processor;

[0020] The water intrusion sensor includes an impedance spectrum analysis circuit with a working frequency range of 10 - 100 kHz. The water intrusion sensor is connected to the control main board via an RS-485 bus to transmit the change amount of the impedance spectrum phase angle.

[0021] Furthermore, the digital twin modeling module:

[0022] Adopts the finite volume method to construct an unsteady fluid simulation model, and the solver is configured to execute the SIMPLE algorithm;

[0023] The dynamic boundary condition generation unit receives the monitoring data of the pressure sensor and the valve opening sensor in real time;

[0024] Includes a sealing ring friction coefficient correction database based on the number of valve openings and closings.

[0025] Furthermore, the fault prediction module:

[0026] The training dataset includes samples labeled with normal operating conditions and 5% - 15% fault injection operating conditions;

[0027] The anomaly detection unit is configured to calculate the Mahalanobis distance between the real-time feature vector and the cluster center;

[0028] The output layer includes an adaptive threshold adjustment unit based on the ambient temperature;

[0029] The residual connection is configured to span at least three time steps, and the skip link of the residual connection receives the frequency domain feature input from the spatio-temporal feature extraction module.

[0030] Furthermore, the linkage control module:

[0031] includes a valve priority sorting unit based on the pipe network topology;

[0032] The instruction queue is divided into a time-sensitive queue and a preventive maintenance queue;

[0033] configured with an instruction retransmission mechanism, the retransmission trigger time threshold is 2 seconds, and the 2-second threshold is determined according to a safety factor of 1.5 times the pipe network pressure wave propagation speed.

[0034] Furthermore, the cross-platform interaction layer:

[0035] The real-time in-memory database adopts a Redis cluster architecture, including a millisecond-level data cache, a feature intermediate result cache, and a diagnostic result cache;

[0036] The network isolation unit includes a hardware firewall and a whitelist access controller based on MAC addresses;

[0037] The visualization terminal is configured with a WebGL rendering engine and parallel pressure cloud map viewports and feature heat maps viewports.

[0038] Furthermore, the network isolation unit:

[0039] divides the IoT perception layer into independent security domains and configures one-way communication channels;

[0040] includes a Modbus TCP protocol parser, configured to filter write register instructions;

[0041] The control instruction transmission channel is configured with a quantum key distribution unit, and the quantum key distribution unit provides dynamic key updates for the AES-256 encryption unit.

[0042] Furthermore, this solution discloses an intelligent control method for fluid pipe networks based on multi-order spatio-temporal feature fusion, including:

[0043] Step S1: Collect valve vibration spectra, fluid pressure waveforms, and pipe wall temperature gradient data through embedded sensing nodes, with a sampling frequency not less than 10 kHz, and the valve vibration spectra are collected through the vibration sensor array;

[0044] Step S2: Use an improved empirical mode decomposition algorithm to decompose the original signal into intrinsic mode functions, and retain the 2-5th order IMF components to reconstruct the feature signal;

[0045] Step S3: Construct a spatio-temporal feature cube, divide the continuous 60-second data into 20 overlapping time windows, and extract 128-dimensional time-domain features and 64-dimensional frequency-domain features for each window;

[0046] Step S4: Train a multi-task learning model. The first task is to predict the pipeline leakage location, the second task is to estimate the remaining service life, and the third task is to generate an optimal maintenance strategy based on equipment health assessment and maintenance cost constraints;

[0047] Step S5: When a pressure mutation event is detected, start a fluid shock wave simulation, calculate the pressure peak value and arrival time of each node in the pipe network, and generate a valve action sequence;

[0048] Before implementing the control strategy, conduct a virtual rehearsal through the digital twin system, and issue an execution instruction after verifying that there is no excessive negative pressure fluctuation.

[0049] Furthermore, in step S4:

[0050] Adopt a hard parameter sharing architecture. The underlying convolutional layer extracts common features, and the upper-layer branch networks process different tasks respectively. The common features include first-order statistical features and second-order frequency domain features;

[0051] Introduce an uncertainty weighting mechanism, and dynamically adjust the weight coefficient according to the variance of each task loss function;

[0052] Set up a model interpretation module, generate a feature importance map through gradient backpropagation, and label the top 5 key influencing factors.

[0053] Furthermore, in step S5:

[0054] Use the method of characteristics to solve the water hammer equation, calculate the pressure wave propagation path and reflection superposition effect. The fluid parameters of the water hammer equation are derived from the unsteady fluid simulation model;

[0055] Establish a valve response delay model, and set an action delay time of 0.5 - 2 seconds according to the actuator model;

[0056] Generate an instruction set containing N + 1 backup strategies, and automatically switch to the backup plan when the main strategy fails.

[0057] Compared with the prior art, the intelligent control system and method for fluid pipe networks based on multi-order spatio-temporal feature fusion of the present invention have the following advantages:

[0058] (1) The intelligent control system and method for fluid pipe networks based on multi-order spatio-temporal feature fusion of the present invention significantly improve the control efficiency and safety of fluid pipe networks through an embedded multi-source perception, multi-order spatio-temporal feature fusion and intelligent decision-making architecture:

[0059] (2) The intelligent control system and method for fluid pipe networks based on multi-order spatio-temporal feature fusion of the present invention, based on the 50kHz high-frequency sampling and encrypted transmission technology, combined with time-domain statistics, frequency-domain analysis and non-linear feature extraction, improve the accuracy of leakage detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The drawings forming a part of this invention are used to provide a further understanding of the invention. The schematic embodiments and descriptions thereof of the invention are used to explain the invention and shall not unduly limit the invention. In the drawings:

[0061] Figure 1 FIG. is a schematic diagram of an intelligent control and management system for a fluid pipe network based on multi - order spatio - temporal feature fusion according to an embodiment of the present invention;

[0062] Figure 2 FIG. is a schematic diagram of an intelligent control and management method for a fluid pipe network based on multi - order spatio - temporal feature fusion according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0064] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first", "second", etc. may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise stated, the meaning of "a plurality" is two or more.

[0065] In the description of the present invention, it should be noted that, unless otherwise clearly defined and limited, the terms "installed", "connected", "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood through specific situations.

[0066] The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0067] This embodiment provides an intelligent control and management system for a fluid pipe network based on multi - order spatio - temporal feature fusion, and the specific implementation is as follows:

[0068] Implementation of the embedded IoT sensing layer: This layer is deployed at key nodes of the pipe network. The core components include the STM32H743 control mainboard integrated with a 4G / 5G dual-mode communication chip (model Quectel EC25), and the mainboard is built-in with an AES-256 encryption unit to ensure the security of data transmission.

[0069] The sensing unit consists of a triaxial vibration sensor array 7, a Coriolis flowmeter 7, and a customized water intrusion sensor. Among them, the vibration sensor collects signals at a sampling rate of 50kHz and performs seven-layer wavelet packet decomposition in real time through an ADSP-21489 chip. The water intrusion sensor obtains phase angle data by scanning in the frequency band of 10 - 100kHz through an AD5933 impedance analyzer. All sensors are connected to the control mainboard through an RS-485 bus 7.

[0070] The actuator adopts the Bernard AQ series electric valve, which receives the 4 - 20mA adjustment signal output by the mainboard to achieve precise control of the opening degree.

[0071] Implementation of the multi-order spatio-temporal feature processing layer: This layer is deployed at the edge computing node (NVIDIA Jetson AGX Xavier). The spatio-temporal feature extraction module performs sliding window processing 7 on the pressure and flow rate data, extracts the mean, variance, and approximate entropy 7 within the window to form the first-order statistical features, calculates the first 32 harmonic amplitudes through a 1024-point FFT as the second-order frequency domain features, and uses the Rosenstein algorithm to calculate the Lyapunov exponent 7 to form the third-order non-linear features.

[0072] The feature fusion module constructs an 8-head Transformer model, where the MFCC features of the vibration signal are used as the key vectors, the gray-level co-occurrence matrix of the infrared thermal image is used as the value vectors, and the mel-frequency cepstral coefficients are used as the query vectors. Cross-modal feature dynamic alignment is achieved through the multi-head attention mechanism, and the attention weight matrix is updated in real time at a frequency of 100Hz.

[0073] Implementation of the intelligent decision-making layer: The digital twin module constructs a three-dimensional pipe network model based on ANSYS Fluent, uses the finite volume method to solve the transient N-S equations, the dynamic boundary condition interface accesses the data of 300 pressure sensors at a frequency of 10Hz, and the friction coefficient of the sealing ring is dynamically corrected according to the number of valve actions according to the formula μ = 0.15 + 0.02log(N).

[0074] The fault prediction module adopts a residual LSTM network. The input layer with 128 nodes receives spatio-temporal features. The residual connection spans three time steps and introduces a frequency domain feature branch. The training data contains 12% fault injection samples, and the anomaly detection unit calculates the Mahalanobis distance between the real-time feature vector and the cluster center.

[0075] The linkage control module divides the valve priorities according to the node betweenness centrality calculated from the pipe network topology. Time-sensitive instructions are transmitted through the RabbitMQ priority channel. The instruction retransmission mechanism triggers quantum key update and instruction retransmission when no feedback is received within 2 seconds.

[0076] Implementation of the cross-platform interaction layer: The real-time in-memory database adopts a Redis cluster architecture, which is divided into a millisecond-level buffer, a feature intermediate buffer, and a diagnostic result persistent storage area.

[0077] The network isolation unit uses the Huawei USG6350 firewall to divide the Internet of Things perception layer into an independent VLAN (ID 100), sets a one-way communication policy to only allow the OPC UA port to go out, the Modbus TCP parser filters write register instructions, and the MAC white list binds 6 authorized device addresses. The visualization terminal renders the three-dimensional pipe network model based on the WebGL engine. The pressure cloud map uses HSL color mapping, and the feature heat viewport synchronously displays the activation state of the LSTM hidden layer. The abnormal area is warned with red light flashing.

[0078] This embodiment provides an intelligent control method for fluid pipe networks based on multi-order spatio-temporal feature fusion, and the specific implementation is as follows:

[0079] Step S1 Data acquisition: Embedded sensing terminals are deployed at key nodes of the pipe network, integrating triaxial vibration sensors, broadband pressure transmitters, and distributed temperature sensing arrays. Among them, the vibration monitoring covers the mechanical vibration characteristic frequency band, the pressure acquisition module is set with an adaptive range switching function, the temperature sensing network adopts a bus-type topology to achieve continuous spatial monitoring, and multi-source data is collected synchronously in time through an industrial communication protocol. The original signal is transmitted to the edge computing node after anti-aliasing filtering and format standardization processing, and a packet verification retransmission mechanism is adopted during the transmission process to ensure integrity.

[0080] Step S2 Signal reconstruction: Aiming at the signal aliasing problem under complex working conditions of the pipe network, an improved empirical mode decomposition algorithm is used to perform intrinsic mode separation on the original signal. Effective components are screened through band energy analysis and time-frequency distribution correlation evaluation. The reconstructed feature signal retains the core information of fluid pulsation, mechanical vibration, and heat conduction, while suppressing background noise interference. A sliding window mechanism is introduced during the reconstruction process to achieve dynamic adaptation of signal features, providing a high-quality input source for subsequent spatio-temporal feature extraction.

[0081] Step S3 Feature Cube Construction: Extract multi-dimensional operation features based on a sliding time window. The time-domain analysis module calculates waveform statistics and non-linear dynamics indicators. The frequency-domain processing unit obtains energy distribution features through time-frequency joint analysis. The spatial correlation module calculates the parameter gradient change rate in combination with the pipe network topology. Finally, organize the time-series evolution features, frequency-domain energy features, and spatial distribution features into a three-dimensional data cube. This data structure fully characterizes the spatio-temporal evolution law of the pipe network system and meets the input dimension requirements of the deep learning model.

[0082] Step S4 Multi-task Learning: Construct a multi-task model with a parameter sharing mechanism. The underlying feature extraction network uses time convolutional layers to capture cross-modal spatio-temporal correlations. The upper task branches respectively implement functions of leakage location, life prediction, and maintenance decision-making. During the training process, balance the learning progress of each task by dynamically adjusting the loss weights. The model interpretation module uses gradient backpropagation technology to identify key feature parameters, providing a traceable technical basis for operation and maintenance decisions. Set an online update mechanism in the model deployment stage to achieve continuous optimization of the knowledge base.

[0083] Step S5 Pressure Sudden Change Response: When a pressure fluctuation exceeding the threshold event is detected, the fluid simulation engine solves the water hammer equation based on real-time data-driven, predicts the propagation path and intensity distribution of the pressure wave, generates a hierarchical control strategy in combination with the pipe network topology and valve response characteristics. The main strategy preferentially adjusts the proximal valves to suppress the wave propagation, and the backup strategy activates the compensation device to cope with extreme conditions. The strategy generation module synchronously calculates the expected effect and evaluates the secondary risk, forming an optimized instruction set that takes into account both safety and economy.

[0084] Step S6 Control Strategy Verification: Build a multi-physical field coupling verification platform in the digital twin environment. The fluid dynamics module checks the pressure fluctuation suppression effect, the mechanical strength module evaluates the stress change of the pipe material, and the thermal coupling module analyzes the stability of the temperature field distribution. Exclude defective strategies that may cause system oscillation or equipment overload through virtual rehearsal. The verified instruction set is sent to the on-site execution unit through a double-encrypted channel. Set a status feedback closed-loop verification mechanism during the execution process to ensure the timing accuracy and reliability of the control actions.

[0085] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent control system for fluid pipe networks based on multi-order spatio-temporal feature fusion, characterized in that, Comprising: An embedded Internet of Things perception layer, including an electric valve control main board and / or a pneumatic valve control main board embedded with a remote information transmission module, as well as a flow sensor, a heat sensor, and a water intrusion sensor connected to the control main board; A multi-order spatio-temporal feature processing layer, including: A spatio-temporal feature extraction module configured to perform sliding window slicing processing on the time series data of pressure, flow rate, and temperature parameters, and extract first-order statistical features including mean and variance, second-order frequency domain features of Fourier transform coefficients, and third-order non-linear features of Lyapunov exponents; A feature fusion module configured to perform feature alignment on vibration spectrograms, voiceprint signals, and infrared thermal imaging data through a multi-head attention mechanism; An intelligent decision-making layer, the output end of the multi-order spatio-temporal feature processing layer is connected to the input end of the intelligent decision-making layer through a message queue middleware, including: A digital twin module that constructs a visualization model including a three-dimensional pipeline topological structure and hydrodynamic parameters; A fault prediction module, including an LSTM neural network with residual connections; A linkage control module configured to generate an instruction queue including valve identification and execution time series; A cross-platform interaction layer, including a Linux server running a real-time in-memory database, and a Windows terminal communicating with the server through the OPCUA protocol.

2. The intelligent control system for fluid pipe networks based on multi-order spatio-temporal feature fusion according to claim 1, wherein In the embedded Internet of Things perception layer: The remote information transmission module includes a 4G / 5G dual-mode communication chip and an AES-256 encryption unit; The vibration sensor array collects vibration waveforms at a sampling rate of 50 kHz and is configured with a wavelet packet transform processor; The water intrusion sensor includes an impedance spectrum analysis circuit with a working frequency range of 10 - 100 kHz, and the water intrusion sensor is connected to the control main board through an RS-485 bus to transmit the change amount of the impedance spectrum phase angle.

3. The intelligent control system for fluid pipe networks based on multi-order spatio-temporal feature fusion according to claim 1, wherein The digital twin modeling module: Adopts the finite volume method to construct an unsteady fluid simulation model, and the solver is configured to execute the SIMPLE algorithm; The dynamic boundary condition generation unit receives the monitoring data of the pressure sensor and the valve opening sensor in real time; Includes a sealing ring friction coefficient correction database based on the number of valve openings and closings.

4. The intelligent control system for fluid pipe networks based on multi-order spatio-temporal feature fusion according to claim 1, wherein, The fault prediction module: The training data set includes samples labeled with normal working conditions and 5% - 15% fault injection working conditions; The anomaly detection unit is configured to calculate the Mahalanobis distance between the real-time feature vector and the cluster center; The output layer includes an adaptive threshold adjustment unit based on the ambient temperature; The residual connection is configured to span at least three time steps, and the skip link of the residual connection receives the frequency domain feature input from the spatio-temporal feature extraction module.

5. The intelligent control system for fluid pipe networks based on multi-order spatio-temporal feature fusion according to claim 1, characterized in that The linkage control module: Includes a valve priority sorting unit based on the pipe network topological structure; The instruction queue is divided into a time-sensitive queue and a preventive maintenance queue; Configured with an instruction retransmission mechanism, the retransmission trigger time threshold is 2 seconds, and the 2-second threshold is determined according to a safety factor of 1.5 times the pipe network pressure wave propagation speed.

6. The intelligent control system for fluid pipe networks based on multi-order spatio-temporal feature fusion according to claim 1, characterized in that The cross-platform interaction layer: The real-time in-memory database adopts a Redis cluster architecture, including a millisecond-level data cache, a feature intermediate result cache, and a diagnostic result cache; The network isolation unit includes a hardware firewall and a MAC address-based whitelist access controller; The visualization terminal is configured with a WebGL rendering engine and parallel pressure cloud map viewports and feature heat maps viewports.

7. The intelligent control system for fluid pipe networks based on multi-level spatio-temporal feature fusion according to claim 1, wherein The network isolation unit: Divides the Internet of Things perception layer into independent security domains and configures one-way communication channels; Includes a Modbus TCP protocol parser configured to filter write register instructions; The control instruction transmission channel is configured with a quantum key distribution unit, and the quantum key distribution unit provides dynamic key updates for the AES-256 encryption unit.

8. An intelligent control method for fluid pipe networks based on multi-order spatio-temporal feature fusion, based on the system according to any one of claims 1-7, characterized in that Includes: Step S1: Collect valve vibration spectra, fluid pressure waveforms, and pipe wall temperature gradient data through embedded sensing nodes, with a sampling frequency not lower than 10 kHz. The valve vibration spectra are collected through the vibration sensor array; Step S2: Use an improved empirical mode decomposition algorithm to decompose the original signal into intrinsic mode functions, and retain the 2-5th order IMF components to reconstruct the feature signal; Step S3: Construct a spatio-temporal feature cube, divide the continuous 60-second data into 20 overlapping time windows, and extract 128-dimensional time-domain features and 64-dimensional frequency-domain features for each window; Step S4: Train a multi-task learning model. The first task predicts the pipeline leakage location, the second task estimates the remaining service life, and the third task generates an optimal maintenance strategy based on equipment health assessment and maintenance cost constraints; Step S5: When a pressure mutation event is detected, start a fluid shock wave simulation, calculate the pressure peak value and arrival time of each node in the pipe network, and generate a valve action sequence; Step S6: Before implementing the control strategy, perform a virtual rehearsal through the digital twin system, and issue an execution instruction after verifying that there is no over-standard negative pressure fluctuation.

9. The intelligent control method for fluid pipe networks based on multi-order spatio-temporal feature fusion according to claim 8, wherein In step S4: Adopt a hard parameter sharing architecture. The underlying convolutional layer extracts common features, and the upper-level branch networks process different tasks respectively. The common features include first-order statistical features and second-order frequency-domain features; Introduce an uncertainty weighting mechanism to dynamically adjust the weight coefficient according to the variance of each task loss function; Set up a model interpretation module to generate a feature importance map through gradient backpropagation and label the top 5 key influencing factors.

10. The intelligent control method for fluid pipe networks based on multi - order spatio - temporal feature fusion according to claim 8, characterized in that, In step S5: Use the method of characteristics to solve the water hammer equation, calculate the pressure wave propagation path and reflection superposition effect, and the fluid parameters of the water hammer equation are derived from the unsteady fluid simulation model; Establish a valve response delay model and set an action delay time of 0.5-2 seconds according to the actuator model; Generate an instruction set including N+1 backup strategies, and automatically switch to the backup plan when the main strategy fails.

Images