Regional water pump station intelligent gate management control method based on edge calculation
By using edge computing and the Attention-LSTM time-series prediction model, the accuracy and coordination issues of gate control in water pumping stations were solved, achieving efficient and low-latency intelligent gate management and optimizing the operating efficiency and energy consumption of water pumping stations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN INSTITUTE OF INFORMATION TECHNOLOGY
- Filing Date
- 2026-04-13
- Publication Date
- 2026-05-12
AI Technical Summary
Existing methods for controlling gates at pumping stations suffer from problems such as low control accuracy, high energy consumption, data transmission delays, control failures due to network congestion, and a lack of coordinated management and control of gates at multiple pumping stations within a region.
An edge computing-based intelligent gate management and control method for regional pumping stations is adopted. Data is collected through edge nodes, and a time-series prediction model Attention-LSTM is constructed to predict the gate opening. Collaborative management is carried out at the regional edge gateway to optimize the gate opening control.
It achieves high-precision gate opening control, reduces data processing and transmission delays, improves the coordination and rationality of gate control within the region, and reduces management difficulty.
Smart Images
Figure CN122018339A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent control of water pumping stations, and specifically to an intelligent gate management and control method for regional water pumping stations based on edge computing. Background Technology
[0002] As the core hub of a fluid transport system, the precise control of the gate opening of a pumping station directly affects water supply / drainage efficiency, energy costs, and pipeline safety. Existing gate control methods mainly fall into three categories: First, manual adjustment based on human experience, relying on operators' subjective judgment of water level and flow rate, resulting in low control accuracy, delayed response, and inability to adapt to complex operating conditions; second, local automatic control based on PLCs, often employing closed-loop regulation of a single parameter (such as water level or pressure), ignoring the coupled effects of multiple parameters such as flow rate, pipeline resistance, and medium viscosity, easily leading to control oscillations and high energy consumption; and third, centralized intelligent control based on the cloud, uploading sensor data to the cloud for modeling and decision-making. While this enables multi-parameter analysis, it suffers from data transmission delays (typically ≥100ms), control failure during network congestion, and poor adaptability between edge operating conditions and cloud models.
[0003] In recent years, edge computing technology has begun to be applied to the field of water pump station control due to its advantages of local data processing and low latency response. However, existing solutions still have obvious shortcomings: First, edge nodes only undertake data forwarding or simple threshold judgment functions, without realizing predictive control and deep optimization, and cannot avoid the risk of operating condition fluctuations in advance; Second, there is a lack of a collaborative management and control mechanism for the gates of multiple water pump stations in the region. The independent operation of each pump station is prone to causing pipeline pressure imbalance and uneven flow distribution. Summary of the Invention
[0004] To address the aforementioned shortcomings of existing technologies, this invention provides a method for managing and controlling intelligent gates in regional pumping stations based on edge computing, thereby achieving collaborative management and control of intelligent gates in regional pumping stations and optimizing intelligent control of gate opening.
[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for intelligent gate management and control of regional pumping stations based on edge computing is provided, comprising: Step S1: Set up edge nodes of the pumping station to collect historical operating data of the pumping station and calculate the similarity coefficient of operating status between pumping stations in the region based on the volatility of the operating data. Step S2: Calculate the operating environment similarity coefficient between regional pumping stations, and combine it with the operating status similarity coefficient to calculate the total operating similarity coefficient between regional pumping stations; Step S3: Based on the overall operational similarity coefficient between pumping stations, construct an objective function for allocating edge nodes to the regional edge gateway. Using the objective function Assign the required set of edge nodes to each regional edge gateway; Step S4: Build a temporal prediction model Attention-LSTM at the edge nodes to predict the opening of the smart gate, and preprocess the collected historical running data to obtain a temporal feature dataset, and then input it into the temporal prediction model Attention-LSTM to output the attention-weighted hidden state. Step S5: Output the hidden state through the fully connected layer to obtain the predicted value of the running data, and output the upper and lower boundaries of the predicted value of the running data through the Monte Carlo method to obtain the range of values for the smart gate opening in the future time step. Step S6: Each edge node inputs the predicted range of smart gate opening values into the regional edge gateway, retrieves the optimal range of smart gate opening values, and distributes it to each edge node; constructs an opening control objective function, and the edge nodes adjust the opening of the smart gate based on the optimal opening control objective function and within the optimal range of smart gate opening values to achieve control of the smart gate.
[0006] Further, step S1 includes: Step S11: Set the period for dynamically allocating edge nodes to be managed to the regional edge gateway. T Each edge node will record historical periods. T The pump station operation data collected by the inner edge terminal is sent to the pump station control center. Step S12: Obtain the timing operation data of each pumping station. ,in, M Historical cycle T The amount of runtime data collected internally. n For the types of running data, i Number the water pumping station. For the first i A water pumping station in historical cycles T The first internal collection M One running data n ; Step S13: Calculate the historical cycle of the water pumping station T Inner n Volatility of operating data ; Step S14: Obtain the historical cycle of the water pump station T Volatility data for all internal operating data , N For the number of types of data to be processed, For the water pump station in the historical cycle T Inner N Volatility of the data being processed; Step S15: Calculate historical cycles using volatility data T The similarity coefficient of operating status between internal water pumping stations.
[0007] Further, step S2 includes: Step S21: Based on the head and historical cycle of the intelligent gate of the pumping station between the upstream and downstream sides. T Instantaneous water quality data of water flowing through the intelligent gate are used to calculate the similarity coefficient of the operating environment between water pumping stations; Step S22: Calculate the total operational similarity coefficient between pumping stations based on the operational status similarity coefficient and the operational environment similarity coefficient.
[0008] Furthermore, in step S3, based on the number of regional edge gateways... W The objective function for building management capabilities to allocate edge nodes to regional edge gateways .
[0009] Further, step S4 includes: Step S41: Build a temporal prediction model, Attention-LSTM, within the edge nodes to predict the opening degree of the smart gate; collect historical operating data of the pumping station using the edge terminals associated with the edge nodes, normalize each historical operating data point, and map the historical operating data to... Within the interval, the time-series feature dataset is obtained. ; Step S42: Transfer the time series feature dataset The input feature sequence data is divided into continuous segments according to a set time window length. , b Assign numbers to the input feature sequence data segments, and then assign consecutive input feature sequence data segments. The input to the LSTM layer of the Attention-LSTM time-series prediction model extracts temporal features and outputs the hidden state. ; Step S43: Hidden state of output through attention mechanism Perform weighted analysis and output the attention-weighted hidden state. .
[0010] Further, step S5 includes: Step S51: The attention-weighted hidden state The predicted values of the running data are obtained by outputting the fully connected layer, and the uncertainty range of the 95% confidence interval is output by the Monte Carlo method, thus obtaining the upper and lower boundaries of the predicted values of the running data. Step S52: Obtain the upper boundary of the predicted value of the smart gate opening in the operating data. and lower boundary To form a future time step in the water pumping station Predicted range of smart gate opening values .
[0011] Further, step S6 includes: Step S61: Each edge node uploads the predicted range of smart gate opening values to the regional edge gateway that manages it. The regional edge gateway then obtains... E The predicted range of smart gate opening values is determined, and the values are taken as follows: E The overlapping region of the predicted smart gate opening range will be used as the future time step. Optimal range of smart gate opening values G ; Step S62: The regional edge gateway selects the optimal range of smart gate opening values. G The data is distributed to each edge node, which then constructs an objective function for opening control based on the optimal loss and energy consumption of the smart gate. ; Step S63: Future Time Step Internally, edge nodes are controlled based on the opening degree objective function. The goal is to minimize the value of the intelligent gate opening within the optimal range. G The opening degree of the intelligent gate can be adjusted internally to achieve control of the intelligent gate.
[0012] The beneficial effects of this invention are as follows: This invention constructs a hierarchical edge architecture in the intelligent gate control and management of pumping stations. Operational data acquisition and processing are achieved at edge nodes, enabling local prediction and control of the operation process, avoiding cloud dependency, and reducing data processing, transmission latency, and bandwidth consumption. At the edge nodes, multiple types of operational data are used to predict gate opening control, achieving high-precision control of gate opening in the future. In the control process of regional intelligent gates, it satisfies both the accuracy of individual intelligent gate opening control and the rationality and correlation of collaborative management and control of regional intelligent gates, reducing the difficulty of management and control. Attached Figure Description
[0013] Figure 1 This is a flowchart of a smart gate management and control method for regional pumping stations based on edge computing. Detailed Implementation
[0014] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0015] This embodiment constructs a three-tier edge computing architecture to support the implementation of an edge computing-based intelligent gate management and control method for regional pumping stations. The three-tier architecture includes edge terminals, edge nodes, and a regional edge gateway, with the regional edge gateway centrally managed by the pumping station control center. Edge terminals include sensor clusters deployed within individual pumping stations and control actuators for the intelligent gates. The sensor clusters include flow sensors, water level sensors, pressure sensors, temperature sensors, and gate opening encoders surrounding the pumping station gates. The control actuators control the opening degree of the intelligent gates. Edge nodes are paired with each pumping station, responsible for local data preprocessing, data prediction modeling, and optimization decision-making, and supporting iterative updates of the local model. Data from edge nodes can be transmitted to the regional edge gateway, which manages multiple similar edge nodes within the region. The regional edge gateway centrally manages and dynamically allocates the edge nodes to be managed, and is responsible for cross-pumping station data collaboration, global optimization coordination, fault diagnosis, and resource scheduling.
[0016] like Figure 1 As shown, a method for intelligent gate management and control of regional pumping stations based on edge computing includes: Step S1: Set up edge nodes to collect historical operating data of the pumping stations and analyze the volatility of the operating data to calculate the similarity coefficient of operating status between pumping stations in the region. Step S1 specifically includes: Step S11: Set the period for dynamically allocating edge nodes to be managed to the regional edge gateway. T Each edge node will record historical periods. T The pump station operation data collected by the inner edge terminal is sent to the pump station control center. The pump station operation data in this embodiment includes the upstream and downstream water levels of the smart gate, the instantaneous flow rate through the gate, the inlet and outlet pressures of the pump station pipeline network, the temperature of the medium passing through the pump station (mainly water temperature), the current opening degree of the smart gate, and the pump operating power.
[0017] Step S12: Obtain the timing operation data of each pumping station. ,in, M Historical cycle T The amount of runtime data collected internally. n For the types of running data, i Number the water pumping station. For the first i A water pumping station in historical cycles T The first internal collection M One running data n ; Step S13: Calculate the historical cycle of the water pumping station TInner n Volatility of operating data ; ; in, m For the number of the running data, For the first i A water pumping station in historical cycles T The first internal collection m One running data n , For the first water pumping station n Ideal values for the type of operational data; Step S14: Obtain the historical cycle of the water pump station T Volatility data for all internal operating data , N For the number of types of data to be processed, For the water pump station in the historical cycle T Inner N Volatility of the data being processed; Step S15: Calculate historical cycles using volatility data T The similarity coefficient of operating status between internal water pumping stations; ; in, For water pumping station j In historical cycles T Inner n The volatility of the operating data For the first n Volatility threshold of the type of operating data For the first n Weights of various types of operational data For water pumping station j Pumping stations i The similarity coefficient of their operating states.
[0018] In this embodiment, the similarity coefficient of operating status is used to characterize the similarity of operating status between pumping stations. The smaller the similarity coefficient, the more similar the operating status of the pumping stations is; the larger the similarity coefficient, the less similar the operating status is.
[0019] Step S2: Calculate the operating environment similarity coefficient among the regional pumping stations, and combine it with the operating status similarity coefficient to calculate the total operating similarity coefficient among the regional pumping stations. Step S2 specifically includes: Step S21: Based on the head and historical cycle of the intelligent gate of the pumping station between the upstream and downstream sides. T Instantaneous water quality data of water flowing through the intelligent gate are used to calculate the similarity coefficient of the operating environment between water pumping stations; ; in, They are water pumping stations i Pumping stations j The lifting height between the upstream and downstream of the intelligent gate, For the ideal head between upstream and downstream, u Number the types of water quality data. U The types and quantities of water quality data. They are water pumping stations j Pumping stations i In historical cycles T The first flow through the smart gate m Instantaneous water quality data u , Water quality data during pumping station operation u The ideal value, The weights of the impact of head and water quality data on the operation of the water pumping station are respectively. For water pumping station j Pumping stations i The similarity coefficient of the operating environments between them Historical cycle T The number of instantaneous water quality data collected internally; In this embodiment, the operating environment similarity coefficient is used to represent the similarity of the operating environment between two pumping stations based on head and water quality data. The smaller the operating environment similarity coefficient, the more similar the operating environments of the two pumping stations are; the larger the operating environment similarity coefficient, the less similar the operating environments are.
[0020] Step S22: Calculate the total operational similarity coefficient between pumping stations based on the operational status similarity coefficient and the operational environment similarity coefficient; ; in, The weighting coefficients are for the operating status and operating environment. For water pumping station j Pumping stations i The overall operational similarity coefficient between them.
[0021] Weighting coefficients in this implementation The value is determined based on the type of water network in which the pumping station is located. For pumping stations within a tap water network, where the water quality is relatively good and the head fluctuation is not significant, the value is [value to be filled in]. ,For example For pumping stations in sewage or flood control / drainage networks, where water quality is poor, head fluctuations are large, and the operating environment has a greater impact on operation, then [the following is taken]: ,For example Among the other balanced water network types, the following can be selected: .
[0022] The overall operational similarity coefficient represents the similarity of the operational processes between two pumping stations. A smaller overall operational similarity coefficient indicates greater similarity in the operational processes between the two pumping stations, and vice versa. This invention uses the overall operational similarity coefficient to group pumping stations with similar operational processes into the same regional edge gateway for management. This reduces the difficulty and complexity of data processing at the regional edge gateway, ensuring that the operational data collected by pumping stations corresponding to edge nodes managed by the same regional edge gateway are highly similar, thus reducing subsequent management difficulties.
[0023] Step S3: Based on the overall operational similarity coefficient between pumping stations, construct an objective function for allocating edge nodes to the regional edge gateway. Using the objective function Assign the required set of edge nodes to each regional edge gateway. Specifically: Based on the number of regional edge gateways W The objective function for building management capabilities to allocate edge nodes to regional edge gateways ; ; in, A set of edge nodes managed by a regional edge gateway. Represents the set of edge nodes The summation of the total operational similarity coefficients among the pumping stations corresponding to the inner edge nodes. This represents the maximum number of edge nodes that can be managed, reflecting the management capabilities of the regional edge gateway. To meet the minimum management revenue requirements of the regional edge gateway, the number of edge nodes to be managed is required. To allocate the number of managed edge nodes to the regional edge gateway based on the objective function, For the set of edge nodes The maximum geographical distance between mid-edge nodes. Manage the allowed geographical distance thresholds between edge nodes for the regional edge gateway; Step S4: Build a temporal prediction model, Attention-LSTM, at the edge nodes to predict the opening degree of the smart gate. Preprocess the collected historical running data to obtain a temporal feature dataset, then input it into the Attention-LSTM model to output the attention-weighted hidden states. Step S4 specifically includes: Step S41: Build a temporal prediction model, Attention-LSTM, within the edge nodes to predict the opening degree of the smart gate; collect historical operating data of the pumping station using the edge terminals associated with the edge nodes, normalize each historical operating data point, and map the historical operating data to... Within the interval, the time-series feature dataset is obtained. ; ; in, For the first n Normalized values of the running data t For the data collection time during operation, They are first-order difference and second-order difference, respectively. For collection time t -1 is the normalized value of the running data. For collection time t -1 first difference This is the normalized value of the pipeline pressure difference. These are the normalized values of the inlet and outlet pressures of the water pumping station's pipeline network, respectively. This is the normalized value of the upstream water level difference. These are the normalized values of the upstream and downstream water levels of the intelligent gate, respectively. Step S42: Transfer the time series feature dataset The input feature sequence data is divided into continuous segments according to a set time window length. , b Assign numbers to the input feature sequence data segments, and then assign consecutive input feature sequence data segments. The input to the LSTM layer of the Attention-LSTM time-series prediction model extracts temporal features and outputs the hidden state. ; Step S43: Hidden state of output through attention mechanism Perform weighted analysis and output the attention-weighted hidden state. ; ; in, To score attention, k The dimension of attention coefficient. K The number of dimensions for the attention coefficient. Attention coefficient For the first k Dimensional attention coefficient For attention weights, This is for attentional bias.
[0024] Step S5: The hidden state is output through a fully connected layer to obtain the predicted value of the running data. The upper and lower boundaries of the predicted running data value are then output using the Monte Carlo method to obtain the range of the smart gate opening value predicted for future time steps. Step S5 specifically includes: Step S51: The attention-weighted hidden state The predicted values of the running data are obtained by outputting the fully connected layer, and the uncertainty range of the 95% confidence interval is output by the Monte Carlo method, thus obtaining the upper and lower boundaries of the predicted values of the running data. ; in, These represent the weights and biases of the fully connected layer, Stepping into the future The predicted vector of the running data, Stepping into the future Predicted values of the running data, To obtain the set of predicted values by setting M Monte Carlo sampling times, This represents the quantile function. These are the lower and upper boundaries of the predicted values for the running data, respectively. Step S52: Obtain the upper boundary of the predicted value of the smart gate opening in the operating data. and lower boundary To form a future time step in the water pumping station Predicted range of smart gate opening values .
[0025] Step S6: Each edge node inputs the predicted smart gate opening range into the regional edge gateway, retrieves the optimal smart gate opening range, and distributes it to each edge node; an opening control objective function is constructed, and the edge nodes adjust the smart gate opening based on the optimal opening control objective function and within the optimal smart gate opening range to achieve control of the smart gate. Step S6 specifically includes: Step S61: Each edge node uploads the predicted range of smart gate opening values to the regional edge gateway that manages it. The regional edge gateway then obtains... E The predicted range of smart gate opening values is determined, and the values are taken as follows: E The overlapping region of the predicted smart gate opening range will be used as the future time step. Optimal range of smart gate opening values G ; Step S62: The regional edge gateway selects the optimal range of smart gate opening values. G The data is distributed to each edge node, which then constructs an objective function for opening control based on the optimal loss and energy consumption of the smart gate. ; ; in, The energy loss coefficient and energy consumption coefficient of the intelligent gate opening are taken in this embodiment. , This represents the maximum allowable opening adjustment per unit time step. The opening degree of the smart gate at two adjacent time steps; the control objective function. The smaller the value, the lower the loss and energy consumption of the intelligent gate control.
[0026] Step S63: Future Time Step Internally, edge nodes are controlled based on the opening degree objective function. The goal is to minimize the value of the intelligent gate opening within the optimal range. G The opening degree of the intelligent gate can be adjusted internally to achieve control of the intelligent gate.
[0027] This invention constructs a hierarchical edge architecture for the intelligent gate control and management of pumping stations. Operational data acquisition and processing are achieved at edge nodes, enabling local prediction and control of the operation process. This avoids cloud dependency and reduces data processing, transmission latency, and bandwidth consumption. At the edge nodes, multiple types of operational data are used to predict gate opening control, achieving high-precision control of gate opening in the future. In the control of regional intelligent gates, this invention satisfies both the accuracy of individual intelligent gate opening control and the rationality and correlation of collaborative management and control of regional intelligent gates, reducing the difficulty of management and control.
Claims
1. A method for intelligent gate management and control of regional water pumping stations based on edge computing, characterized in that, include: Step S1: Set up edge nodes of the pumping station to collect historical operating data of the pumping station and calculate the similarity coefficient of operating status between pumping stations in the region based on the volatility of the operating data. Step S2: Calculate the operating environment similarity coefficient between regional pumping stations, and combine it with the operating status similarity coefficient to calculate the total operating similarity coefficient between regional pumping stations; Step S3: Based on the overall operational similarity coefficient between pumping stations, construct an objective function for allocating edge nodes to the regional edge gateway. Using the objective function Assign the required set of edge nodes to each regional edge gateway; Step S4: Build a temporal prediction model Attention-LSTM at the edge nodes to predict the opening of the smart gate, and preprocess the collected historical running data to obtain a temporal feature dataset, and then input it into the temporal prediction model Attention-LSTM to output the attention-weighted hidden state. Step S5: Output the hidden state through the fully connected layer to obtain the predicted value of the running data, and output the upper and lower boundaries of the predicted value of the running data through the Monte Carlo method to obtain the range of values for the smart gate opening in the future time step. Step S6: Each edge node inputs the predicted range of smart gate opening values into the regional edge gateway, retrieves the optimal range of smart gate opening values, and distributes it to each edge node. An opening control objective function is constructed. The edge nodes adjust the opening of the smart gate based on the optimal opening value of the opening control objective function and within the optimal range of smart gate opening values, thereby realizing the control of the smart gate.
2. The intelligent gate management and control method for regional pumping stations based on edge computing according to claim 1, characterized in that, Step S1 includes: Step S11: Set the period for dynamically allocating edge nodes to be managed to the regional edge gateway. T Each edge node will record historical periods. T The pump station operation data collected by the inner edge terminal is sent to the pump station control center. Step S12: Obtain the timing operation data of each pumping station. ,in, M Historical cycle T The amount of runtime data collected internally. n For the types of running data, i Number the water pumping station. For the first i A water pumping station in historical cycles T The first internal collection M One running data n ; Step S13: Calculate the historical cycle of the water pumping station T Inner n Volatility of operating data ; Step S14: Obtain the historical cycle of the water pump station T Volatility data for all internal operating data , N For the number of types of data to be processed, For the water pump station in the historical cycle T Inner N Volatility of the data being processed; Step S15: Calculate historical cycles using volatility data T The similarity coefficient of operating status between internal water pumping stations; ； in, For water pumping station j In historical cycles T Inner n The volatility of the operating data For the first n Volatility threshold of the type of operating data For the first n Weights of various types of operational data For water pumping station j Pumping stations i The similarity coefficient of their operating states.
3. The intelligent gate management and control method for regional pumping stations based on edge computing according to claim 2, characterized in that, Step S2 includes: Step S21: Based on the head and historical cycle of the intelligent gate of the pumping station between the upstream and downstream sides. T Instantaneous water quality data of water flowing through the intelligent gate are used to calculate the similarity coefficient of the operating environment between water pumping stations; ； in, They are water pumping stations i Pumping stations j The lifting height between the upstream and downstream of the intelligent gate, For the ideal head between upstream and downstream, u Number the types of water quality data. U The types and quantities of water quality data. They are water pumping stations j Pumping stations i In historical cycles T The first flow through the smart gate m Instantaneous water quality data u , Water quality data during pumping station operation u The ideal value, The weights of the impact of head and water quality data on the operation of the water pumping station are respectively. For water pumping station j Pumping stations i The similarity coefficient of the operating environments between them Historical cycle T The number of instantaneous water quality data collected internally; Step S22: Calculate the total operational similarity coefficient between pumping stations based on the operational status similarity coefficient and the operational environment similarity coefficient; ； in, The weighting coefficients are for the operating status and operating environment. For water pumping station j Pumping stations i The overall operational similarity coefficient between them.
4. The intelligent gate management and control method for regional pumping stations based on edge computing according to claim 3, characterized in that, In step S3, the number of regional edge gateways is used as a basis. W The objective function for building management capabilities to allocate edge nodes to regional edge gateways ; ； in, A set of edge nodes managed by a regional edge gateway. Represents the set of edge nodes The summation of the total operational similarity coefficients among the pumping stations corresponding to the inner edge nodes. This represents the maximum number of edge nodes that can be managed, reflecting the management capabilities of the regional edge gateway. To meet the minimum management revenue requirements of the regional edge gateway, the number of edge nodes to be managed is required. To allocate the number of managed edge nodes to the regional edge gateway based on the objective function, For the set of edge nodes The maximum geographical distance between mid-edge nodes. Manage the allowed geographical distance thresholds between edge nodes for the regional edge gateway.
5. The intelligent gate management and control method for regional pumping stations based on edge computing according to claim 4, characterized in that, Step S4 includes: Step S41: Build a temporal prediction model, Attention-LSTM, within the edge nodes to predict the opening degree of the smart gate; collect historical operating data of the pumping station using the edge terminals associated with the edge nodes, normalize each historical operating data point, and map the historical operating data to... Within the interval, the time-series feature dataset is obtained. ; ； in, For the first n Normalized values of the running data t For the data collection time during operation, They are first-order difference and second-order difference, respectively. For collection time t -1 is the normalized value of the running data. For collection time t -1 first difference This is the normalized value of the pipeline pressure difference. These are the normalized values of the inlet and outlet pressures of the water pumping station's pipeline network, respectively. This is the normalized value of the upstream water level difference. These are the normalized values of the upstream and downstream water levels of the intelligent gate, respectively. Step S42: Transfer the time series feature dataset The input feature sequence data is divided into continuous segments according to a set time window length. , b Assign numbers to the input feature sequence data segments, and then assign consecutive input feature sequence data segments. The input to the LSTM layer of the Attention-LSTM time-series prediction model extracts temporal features and outputs the hidden state. ; Step S43: Hidden state of output through attention mechanism Perform weighted analysis and output the attention-weighted hidden state. ; ； in, To score attention, k The dimension of attention coefficient. K The number of dimensions for the attention coefficient. Attention coefficient For the first k Dimensional attention coefficient For attention weights, This is for attentional bias.
6. The intelligent gate management and control method for regional pumping stations based on edge computing according to claim 5, characterized in that, Step S5 includes: Step S51: The attention-weighted hidden state The predicted values of the running data are obtained by outputting the fully connected layer, and the uncertainty range of the 95% confidence interval is output by the Monte Carlo method, thus obtaining the upper and lower boundaries of the predicted values of the running data. ； in, These represent the weights and biases of the fully connected layer, Stepping into the future The predicted vector of the running data, Stepping into the future Predicted values of the running data, To obtain the set of predicted values by setting M Monte Carlo sampling times, This represents the quantile function. These are the lower and upper boundaries of the predicted values for the running data, respectively. Step S52: Obtain the upper boundary of the predicted value of the smart gate opening in the operating data. and lower boundary To form a future time step in the water pumping station Predicted range of smart gate opening values .
7. The intelligent gate management and control method for regional pumping stations based on edge computing according to claim 6, characterized in that, Step S6 includes: Step S61: Each edge node uploads the predicted range of smart gate opening values to the regional edge gateway that manages it. The regional edge gateway then obtains... E The predicted range of smart gate opening values is determined, and the values are taken as follows: E The overlapping region of the predicted smart gate opening range will be used as the future time step. Optimal range of smart gate opening values G ; Step S62: The regional edge gateway selects the optimal range of smart gate opening values. G The data is distributed to each edge node, which then constructs an objective function for opening control based on the optimal loss and energy consumption of the smart gate. ; ； in, The energy loss coefficient and energy consumption coefficient of the intelligent gate opening. This represents the maximum allowable opening adjustment per unit time step. These represent the opening degree of the smart gate at two adjacent time steps; Step S63: Future Time Step Internally, edge nodes are controlled based on the opening degree objective function. The goal is to minimize the value of the intelligent gate opening within the optimal range. G The opening degree of the intelligent gate can be adjusted internally to achieve control of the intelligent gate.