Model-data hybrid driving-based water transfer pump station optimization scheduling method
By constructing a model-data hybrid driven optimization scheduling method for water transfer pumping stations, and combining the SE-Dueling DDQN algorithm and a two-level optimization scheduling model, the scheduling accuracy and energy consumption problems of traditional water transfer pumping stations under uncertain environments are solved, and precise control of reservoir capacity and reduction of energy consumption are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-29
- Publication Date
- 2026-04-03
AI Technical Summary
Traditional water diversion pump station optimization scheduling methods are difficult to guarantee the accuracy of water diversion plans and the safe and efficient operation of pump stations under uncertain environments. Existing algorithms have low computational efficiency and are prone to getting trapped in local optima, making it difficult to effectively reduce energy consumption.
A model-data hybrid-driven optimization scheduling method for water transfer pumping stations is adopted. By combining the SE-Dueling DDQN algorithm and a two-layer optimization scheduling model, a system operation model for water transfer pumping stations is constructed. The agent is trained through a Markov decision process to optimize the flow decision of the pumping stations.
It improved the accuracy of reservoir capacity control in regulating lakes, enhanced the precision of dispatching decisions, and reduced the energy consumption of pumping stations.
Smart Images
Figure CN121787835A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of water transfer pumping station optimization scheduling technology, and more specifically, relates to a model-data hybrid driven water transfer pumping station optimization scheduling method. Background Technology
[0002] In inter-basin water transfer projects, water transfer pumping stations are the core facilities for enabling water bodies to overcome topographical obstacles and raise the water conveyance elevation. Optimizing the scheduling of water transfer pumping stations to ensure their safe and efficient operation is of great significance for guaranteeing the long-term stable social benefits of inter-basin water transfer projects.
[0003] Predictions of lake runoff and surrounding water demand often contain errors. Traditional scheduling methods rely on deterministic scheduling based on these predictions, making it difficult to guarantee the precision of lake water level control. This impacts the formulation and implementation of subsequent water transfer plans for pumping stations, and even their safe operation. Furthermore, current mainstream algorithms for solving optimal scheduling models for pumping stations include dynamic programming and swarm intelligence algorithms such as genetic algorithms, particle swarm optimization, and sparrow search. However, dynamic programming often suffers from the "curse of dimensionality," leading to low computational efficiency and difficulty in guaranteeing a globally optimal solution. While swarm intelligence algorithms offer improvements in computational efficiency and flexibility, they are highly dependent on parameter tuning experience and prone to getting trapped in local optima. These limitations restrict their ability to reduce energy consumption at pumping stations under uncertain environments and hinder their ability to effectively improve the reservoir capacity control precision at the end of the scheduling period, ultimately failing to guarantee the safe and efficient operation of pumping stations. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention proposes a model-data hybrid-driven method for optimizing the scheduling of water transfer pumping stations. This method aims to obtain a robust scheduling strategy for water transfer pumping stations under uncertain scenarios, thereby improving the accuracy of reservoir capacity control at the end of the scheduling period and reducing the energy consumption of water transfer pumping stations.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: The present invention provides a model-data hybrid-driven method for optimizing the scheduling of water transfer pumping stations, characterized by the following steps: Step 1: Construct an operational model of the water transfer pumping station system, consisting of water transfer pumping stations, water conveyance channels, and regulating lakes; Step 2: Based on the operation model of the water transfer pumping station system, construct a two-layer optimized scheduling model for the water transfer pumping station; Step 3: Based on the two-layer optimization scheduling model of water transfer pumping stations, define the system state, actions, and reward function of water transfer pumping stations in the Markov decision process; Step 4: Based on the state, actions, and reward function of the water transfer pumping station system, the SE-Dueling DDQN algorithm is used to train the agent composed of a dual Q-value network. The trained agent is used to output the corresponding pumping station flow rate in response to the input state of the water transfer pumping station system.
[0006] The model-data hybrid-driven optimized scheduling method for water transfer pumping stations described in this invention is also characterized in that step 1 includes: Step 1.1: Construct the operation model of the water transfer pumping station using equations (1) and (2): (1) (2) In equations (1) and (2), For the first Energy consumption of time-segmented water diversion pumping stations; The density of water; It is the acceleration due to gravity; For the first Operating efficiency of time-segmented water diversion pumping stations; For the first Operating flow rate of the water diversion pumping station during specific time periods; For the first The head of the time-limited water diversion pumping station; For the first The first time period water diversion pumping station The operating flow rate of the unit; For the first The first time period water diversion pumping station Operating efficiency of the unit; Indicates the length of each time period. This indicates the total number of units within the water diversion pumping station. , ; Step 1.2: Construct an uncertainty model of the runoff inflow and surrounding water demand of the regulating lake using equations (3) and (4): (3) (4) In equations (3) and (4), It is a generalized logical distribution; For the first Error in predicting runoff inflow to time-limited reservoirs; and They are generalized logical distributions Position parameters, scale parameters, and shape parameters; It follows a normal distribution; For the first Errors in water demand forecasting around time-limited reservoirs; Normal distribution Expected value and standard deviation; It is a natural constant; Step 1.3: Based on the uncertainty model of runoff inflow and surrounding water demand of the regulating lake, calculate the first step using equations (5) and (6). The actual runoff of the time-limited lake and the The actual water demand of the surrounding area of the time-limited storage lake : (5) (6) In equations (5) and (6), For the first Predicted runoff inflow to time-limited reservoirs; For the first Predicted water demand in the surrounding area for time-limited water storage lakes.
[0007] Step 1.4: Construct a regulating lake operation model using equation (7): (7) In equation (7), For the first The storage capacity of the regulating lake at the end of the period; For the first The storage capacity of the regulating lake at the end of the period; Step 1.5: Construct a one-dimensional unsteady flow model of the water conveyance channel using equations (8) and (9): (8) (9) In equations (8) and (9), This is the distance between the cross-section of the water conveyance channel and the starting point of the water conveyance channel; For a specific moment; The distance from the starting point of the water conveyance channel is The cross section at The water flow area at any given moment; The distance from the starting point of the water conveyance channel is The cross section at The water level at that moment; The distance from the starting point of the water conveyance channel is The cross section at Flow rate at any given moment; The distance from the starting point of the water conveyance channel is The cross section at The side stream of time; This is the momentum correction factor; The distance from the starting point of the water conveyance channel is The cross section at The hydraulic radius at any given time; The Manning coefficient for the water conveyance channel; Step 1.6: Calculate the distance to the starting point of the water conveyance channel using equation (10). The cross-section of the water conveyance channel at the first Water level during the period : (10) In equation (10), The symbol is for integrals.
[0008] Furthermore, the two-layer optimized scheduling model for the water transfer pumping station constructed in step 2 includes: an upper-level pumping station flow optimization model and a lower-level unit flow allocation model. Step 2.1: Construct the first objective function of the upper pump station flow optimization model using equation (11). : (11) In equation (11), Indicates the total number of time periods in the scheduling cycle; Step 2.2: Establish the second objective function of the upper pump station flow optimization model using equations (12) and (13). : (12) (13) In equations (12) and (13), For the first Penalty for deviation in storage capacity during a given period; For the first The target storage capacity of lakes at the end of the period; Step 2.3: Establish the operating flow constraints of the water transfer pumping station for the upper-level pumping station flow optimization model using equation (14): (14) In equation (14), and These are the minimum and maximum allowable flow rates for the water diversion pumping station, respectively. Step 2.4: Establish the water conveyance capacity constraints of the water conveyance channel for the upper pumping station flow optimization model using equation (15): (15) In equation (15), and These are the upper and lower limits of the permissible water conveyance flow rate for the water conveyance channel, respectively. Step 2.5: Establish the upstream water level constraint of the water transfer pumping station using equation (16) to optimize the flow rate of the upper pumping station: (16) In equation (16), For the first Water level in front of the water diversion pumping station during a specific time period; and These are the minimum and maximum water levels allowed for normal operation of the water diversion pumping station, respectively. Step 2.6: Establish the downstream water level constraint of the water transfer pumping station for the upper pumping station flow optimization model using equation (17): (17) In equation (17), For the first The water level after the pumping station during the designated water transfer period; and These are the lowest and highest post-station water levels allowed for normal operation of the water diversion pumping station, respectively. Step 2.7: Establish water level change safety constraints for the upper pump station flow optimization model using equation (18): (18) In equation (18), For the first The range of water level changes before and after the station during the specified time period; This refers to the maximum allowable water level fluctuation within a single time period. Step 2.8: Establish the head constraint of the water transfer pumping station in the flow optimization model of the upper pumping station using equation (19): (19) In equation (19), and These are the minimum and maximum head allowed for normal operation of the water diversion pumping station, respectively. Step 2.9: Using equation (20) to establish the flow optimization model of the upper pumping station, the storage capacity constraint of the storage lake is: (20) In equation (20), and These are the upper and lower limits allowed for the storage capacity of regulating lakes, respectively. Step 2.10: Construct the objective function of the lower-level unit flow allocation model using equation (21). : (twenty one) Step 2.11: Establish the total flow constraint of the water transfer pumping station for the flow distribution model of the lower-level units using equation (22): (twenty two) Step 2.12: Establish the flow capacity constraints of the units within the water transfer pumping station for the flow distribution model of the lower-level units using equation (23): (twenty three) In equation (23), and These are the minimum and maximum allowable flow rates for the units within the water diversion pumping station, respectively. Step 2.13: Use equation (24) to establish the minimum start-up and shutdown time constraints for the units within the water transfer pumping station in the lower-level unit flow distribution model: (twenty four) In equation (24): , and These refer to the start-up time, shutdown time, and minimum allowable start-up and shutdown time of the units within the water diversion pumping station.
[0009] Furthermore, step 3 includes: Step 3.1: Define the first using equation (25) Status of the water diversion pumping station system during the time period : (25) In equation (25), Indicates the sequence number of the time period; Step 3.2: Define the first Actions during a period of time For the first Pump station flow rate during the period ; Step 3.3: Construct the first equation using equation (26) Time-based rewards : (26) In equation (26), , , , and They are respectively , , , and Weighting coefficients; For the first The penalty for exceeding the storage capacity limit of the time-limited lake is obtained from equation (27). For the first The penalties for exceeding the water level limit at the time-limited water diversion pumping station before and after the station are obtained from equation (28). For the first The penalty for exceeding the limit of water level change range before and after the time period water diversion pumping station is obtained from equation (29). For the first The head over-limit penalty of the time-limited water diversion pumping station is obtained from equation (30): (27) (28) (29) (30) Furthermore, step 4 includes: Step 4.1: Construct a dual-Q value network consisting of an online Q-network and a target Q-network, and initialize the parameters of the online Q-network. and the parameters of the target Q network ; Step 4.2: Place the first Status of the water diversion pumping station system during the time period The input is processed in an online Q network to obtain the first... Actions during a period of time ; Will Substituting the values into the flow distribution model of the lower-level units and solving for the result, we obtain the... The operating flow rate of each unit within the time-limited water diversion pumping station is calculated and executed by the water diversion pumping station system to obtain the first... Time-based rewards and the Status of the water diversion pumping station system during the time period ; Will Each sample is stored in the experience pool until the number of samples in the experience pool reaches a certain threshold. up to the last one; Step 4.3: Randomly draw a sample from the experience pool. And input it into the online Q network for processing to obtain the first... Status of the water diversion pumping station system during the time period The following action and the Status of the water diversion pumping station system during the time period Next action Corresponding value ; The first Status of the water diversion pumping station system during the time period and its actions The input is processed in the target Q-network to obtain... Next action Corresponding value ; Step 4.4: Minimize the parameters of the online Q-network using equation (31) Update: (31) In equation (31), For the first The target value function for the time period is: (32) In equation (32), Discount factor; Step 4.5: Use equation (33) to establish the parameters of the target Q-network. Update formula: (33) In equation (33), This indicates assignment; Indicates the soft update coefficient; Step 4.6: Train the dual Q-value network according to the process of steps 4.3-4.5 until the maximum number of training iterations is reached, thereby obtaining the trained agent.
[0010] The present invention provides an electronic device, including a memory and a processor, characterized in that the memory is used to store a program supporting the processor in performing the method described therein, and the processor is configured to execute the program stored in the memory.
[0011] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program is executed by a processor to perform the steps of the method described thereon.
[0012] The advantages of this invention, which differ from existing technologies, are mainly reflected in the following aspects: 1. Traditional methods for optimizing the scheduling of water transfer pumping stations rely on deterministic scheduling based on runoff inflow and water demand forecasts, resulting in insufficient adaptability of the scheduling strategies to actual scenarios. This invention considers the impact of uncertainties in the runoff inflow of the regulating lake and the surrounding water demand on the optimized scheduling of water transfer pumping stations. It establishes an uncertainty model for the runoff inflow of the regulating lake and the surrounding water demand to achieve precise control of the lake's regulating capacity, providing a guarantee for the subsequent formulation and implementation of water transfer plans for the pumping stations, as well as the safe operation of the pumping stations.
[0013] 2. Traditional methods for optimizing the scheduling of water transfer pumping stations use a one-dimensional steady flow model to simulate the water transfer channel. However, this model neglects the assumption of dynamic transition processes in the water flow and fails to reflect the transient changes in river water level under short-term flow control at the pumping station, resulting in significant deviations between simulation results and actual operating conditions. The one-dimensional unsteady flow model for the water transfer channel constructed in this invention can accurately characterize the dynamic response of river water level during flow control, effectively overcoming the limitations of traditional models and improving the accuracy of scheduling decisions.
[0014] 3. The SE-Dueling DDQN algorithm proposed in this invention integrates the powerful perception and decision-making capabilities of deep reinforcement learning in complex dynamic environments with the precise and efficient optimization characteristics of mathematical solvers by embedding Gurobi into the Dueling DDQN optimization framework of the deep reinforcement learning algorithm. This increases the training speed of the agent, overcomes the problem that current mainstream algorithms cannot cope with the uncertainties in the optimal scheduling of pumping stations, and helps to further reduce the energy consumption of pumping stations and improve the control accuracy of the storage capacity of the reservoir at the end of the scheduling period. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of a water diversion pumping station system; Figure 2 Framework diagram of a two-layer optimized scheduling model for water diversion pumping stations; Figure 3 Here is a framework diagram of the SE-Dueling DDQN algorithm; Figure 4 This is a schematic diagram of the sequential decision-making process of an intelligent agent. Detailed Implementation
[0016] In this embodiment, a model-data hybrid-driven water transfer pumping station optimization scheduling method is applied to, for example, Figure 1 The system comprises a water diversion pumping station, a water conveyance channel, and a regulating lake. The water conveyance channel originates at the pumping station and terminates at the regulating lake. The pumping station transports sufficient water from upstream to the regulating lake via the channel. The regulating lake, acting as a regulating node in the system, is responsible for receiving water from the pumping station and supplying water to the surrounding areas. The main steps of this method are as follows: Step 1: Construct an operational model of the water transfer pumping station system, consisting of water transfer pumping stations, water conveyance channels, and regulating lakes; Step 1.1: Construct the operation model of the water transfer pumping station using equations (1) and (2): (1) (2) Equations (1) and (2) are used to calculate the operating energy consumption and operating efficiency of the water diversion pumping station. In equations (1) and (2), For the first Energy consumption of time-segmented water diversion pumping stations; The density of water; It is the acceleration due to gravity; For the first Operating efficiency of time-segmented water diversion pumping stations; For the first Operating flow rate of the water diversion pumping station during specific time periods; For the first The head of the time-limited water diversion pumping station; For the first The first time period water diversion pumping station The operating flow rate of the unit; For the first The first time period water diversion pumping station Operating efficiency of the unit; Indicates the length of each time period. This indicates the total number of units within the water diversion pumping station. , .
[0017] Step 1.2: Construct an uncertainty model of the runoff inflow and surrounding water demand of the regulating lake using equations (3) and (4): (3) (4) In equations (3) and (4), It is a generalized logical distribution; For the first Error in predicting runoff inflow to time-limited reservoirs; and They are generalized logical distributions Position parameters, scale parameters, and shape parameters; It follows a normal distribution; For the first Errors in water demand forecasting around time-limited reservoirs; Normal distribution Expected value and standard deviation; It is a natural constant; By fitting statistical data on the prediction errors of runoff inflow to and water demand around the reservoirs, the generalized logistic distribution is determined. of and and normal distribution of Then, through Monte Carlo sampling, the first... The error in the prediction of runoff inflow to the time-limited lake and the first Errors in predicting water demand around time-limited reservoirs.
[0018] Step 1.3: Based on the uncertainty model of runoff inflow and surrounding water demand of the regulating lake, calculate the first step using equations (5) and (6). The actual runoff of the time-limited lake and the The actual water demand of the surrounding area of the time-limited storage lake : (5) (6) In equations (5) and (6), For the first Predicted runoff inflow to time-limited reservoirs; For the first Predicted water demand in the surrounding area for time-limited water storage lakes.
[0019] Step 1.4: Construct a water storage lake operation model based on the water balance principle using equation (7): (7) In equation (7), For the first The storage capacity of the regulating lake at the end of the period; For the first The storage capacity of the regulating lake at the end of the period.
[0020] Step 1.5: Construct a one-dimensional unsteady flow model of the water conveyance channel using equations (8) and (9): (8) (9) Equation (8) is the continuity equation in the Saint-Venant equation system, and equation (9) is the momentum equation in the Saint-Venant equation system. By solving the continuity equation and the momentum equation, we can obtain the equation at any time... The distance from the starting point of the water conveyance channel is The water level at the cross section, in equations (8) and (9), This is the distance between the cross-section of the water conveyance channel and the starting point of the water conveyance channel; For a specific moment; The distance from the starting point of the water conveyance channel is The cross section at The water flow area at any given moment; The distance from the starting point of the water conveyance channel is The cross section at The water level at that moment; The distance from the starting point of the water conveyance channel is The cross section at Flow rate at any given moment; The distance from the starting point of the water conveyance channel is The cross section at The side stream of time; This is the momentum correction factor; The distance from the starting point of the water conveyance channel is The cross section at The hydraulic radius at any given time; The Manning coefficient is the water conveyance channel.
[0021] Step 1.6: Calculate the first step using equation (10). Water level at the cross-section of the water conveyance channel during the period: (10) In equation (10), The distance from the starting point of the water conveyance channel is The cross section at the Water level during a given period; Let be the integral symbol. The distance to the starting point of the water conveyance channel is calculated using equation (10). The cross section at the Water level during a given period.
[0022] Step 2: Based on the aforementioned water transfer pumping station system operation model, construct a two-layer optimized scheduling model for the water transfer pumping station. For example... Figure 2 As shown, the two-layer optimal scheduling model for water transfer pumping stations includes an upper-level pumping station flow optimization model and a lower-level unit flow allocation model. The upper-level model is based on the first... The flow rate of the water diversion pumping station during each time period is used as the decision variable and is passed to the lower-level model. The lower-level model then optimizes the flow rate of the units within the water diversion pumping station based on the flow rate of the water diversion pumping station given by the upper-level model for each time period to obtain the result. The system calculates the operating efficiency of the water diversion pumping stations during different time periods and feeds this efficiency back to the upper level. This process is repeated continuously to achieve coordination between the upper and lower level models.
[0023] Step 2.1: Construct the first objective function of the upper pump station flow optimization model using equation (11). : (11) In equation (11), This indicates the total number of time periods in the scheduling cycle.
[0024] Step 2.2: Establish the second objective function of the upper pump station flow optimization model using equations (12) and (13). : (12) (13) In equations (12) and (13), For the first Penalty for deviation in storage capacity during a given period; For the first The target storage capacity of the lake at the end of the period.
[0025] Step 2.3: Establish the operating flow constraints of the water transfer pumping station for the upper-level pumping station flow optimization model using equation (14): (14) In equation (14), and These are the minimum and maximum allowable flow rates for the water diversion pumping station, respectively.
[0026] Step 2.4: Establish the water conveyance capacity constraints of the water conveyance channel for the upper pumping station flow optimization model using equation (15): (15) In equation (15), and These are the upper and lower limits of the permissible water conveyance flow rate for the water conveyance channel, respectively.
[0027] Step 2.5: Establish the upstream water level constraint of the water transfer pumping station using equation (16) to optimize the flow rate of the upper pumping station: (16) In equation (16), For the first Water level in front of the water diversion pumping station during a specific time period; and These represent the minimum and maximum water levels upstream of the water diversion pumping station, respectively, allowed for normal operation. Due to sufficient upstream water flow, the water level upstream of the pumping station remains constant throughout the various time periods, i.e., the [missing information - likely a specific timeframe]. Water level in front of the time-limited water diversion pumping station The water level is always equal to the initial time period. .
[0028] Step 2.6: Establish the downstream water level constraint of the water transfer pumping station for the upper pumping station flow optimization model using equation (17): (17) In equation (17), For the first The water level after the pumping station during the designated water transfer period; and These are the lowest and highest post-station water levels allowed for the normal operation of the water diversion pumping station, respectively; the water diversion pumping station is the starting point of the water conveyance channel, therefore the first... Post-station water level of time-limited water diversion pumping station equal to the The distance between the time period and the water diversion pumping station is Water level at the cross section of the water conveyance channel .
[0029] Step 2.7: Establish water level change safety constraints for the upper pump station flow optimization model using equation (18): (18) In equation (18), For the first The range of water level changes before and after the station during the specified time period; This refers to the maximum allowable water level fluctuation within a single time period. Step 2.8: Establish the head constraint of the water transfer pumping station in the flow optimization model of the upper pumping station using equation (19): (19) In equation (19), and These are the minimum and maximum head allowed for normal operation of the water diversion pumping station, respectively.
[0030] Step 2.9: Using equation (20) to establish the flow optimization model of the upper pumping station, the storage capacity constraint of the storage lake is: (20) In equation (20), and These represent the upper and lower limits allowed for the storage capacity of regulating lakes, respectively.
[0031] Step 2.10: Construct the objective function of the lower-level unit flow allocation model using equation (21). : (twenty one) Step 2.11: Establish the total flow constraint of the water transfer pumping station for the flow distribution model of the lower-level units using equation (22): (twenty two) Step 2.12: Establish the flow capacity constraints of the units within the water transfer pumping station for the flow distribution model of the lower-level units using equation (23): (twenty three) In equation (23), and These are the minimum and maximum flow rates allowed for the units within the water diversion pumping station, respectively.
[0032] Step 2.13: Use equation (24) to establish the minimum start-up and shutdown time constraints for the units within the water transfer pumping station in the lower-level unit flow distribution model: (twenty four) In equation (24): , and These refer to the start-up time, shutdown time, and minimum allowable start-up and shutdown time of the units within the water diversion pumping station.
[0033] Step 3: Based on the two-layer optimization scheduling model of water transfer pumping stations, define the system state, actions, and reward function of water transfer pumping stations in the Markov decision process; Step 3.1: Define the first using equation (25) Status of the water diversion pumping station system during the time period : (25) In equation (25), Indicates the sequence number of the time period.
[0034] Step 3.2: Define the first Actions during a period of time For the first Pump station flow rate during the period ; Step 3.3: Construct the first equation using equation (26) Time-based rewards : (26) In equation (26), , , , and They are respectively , , , and Weighting coefficients; For the first The penalty for exceeding the storage capacity limit of a time-limited regulating lake is used to prevent violations of the storage capacity constraints of the regulating lake, and is obtained from equation (27). For the first The penalties for exceeding the water level limits before and after the water transfer pumping station during the time period are used to prevent violations of the water level constraints before and after the water transfer pumping station, and are obtained from equation (28). For the first The penalty for exceeding the limit of water level change range before and after the time-limited water diversion pumping station is used to prevent violations of water level change safety constraints, and is obtained from equation (29). For the first The head over-limit penalty for time-limited water transfer pumping stations is used to prevent violations of the head constraints of water transfer pumping stations, and is obtained from equation (30): (27) (28) (29) (30).
[0035] Step 4: Based on the state, actions, and reward function of the water transfer pumping station system, the SE-Dueling DDQN algorithm is used to train the agent composed of a dual Q-value network. The trained agent is used to output the corresponding pumping station flow rate in response to the input state of the water transfer pumping station system.
[0036] like Figure 3 As shown, the SE-Dueling DDQN algorithm is a model-data hybrid driven algorithm that embeds a Gurobi solver into the deep reinforcement learning Dueling DDQN optimization framework. The SE-Dueling DDQN agent consists of an online Q-network and a target Q-network. The action selected by the agent in each time period must first be solved by the Gurobi solver to obtain the executable unit flow rate of the water transfer pumping station system. After the water transfer pumping station system executes the action, a state transition occurs, and a reward signal is fed back to the agent. The agent updates its strategy based on the samples obtained from interacting with the water transfer pumping station system, and through continuous strategy iteration, finally obtains the optimal scheduling strategy that maximizes the expected cumulative reward. The specific steps of the algorithm are as follows: Step 4.1: Construct a dual-Q value network consisting of an online Q-network and a target Q-network, and initialize the parameters of the online Q-network. and the parameters of the target Q network .
[0037] Step 4.2: Place the first Status of the water diversion pumping station system during the time period The input is processed in an online Q network to obtain the first... Actions during a period of time ; Will Substituting the values into the flow distribution model of the lower-level units and solving for the result, we obtain the... The operating flow rate of each unit within the time-limited water diversion pumping station is calculated and executed by the water diversion pumping station system to obtain the first... Time-based rewards and the Status of the water diversion pumping station system during the time period ; Will Each sample is stored in the experience pool until the number of samples in the experience pool reaches a certain threshold. That's all.
[0038] Step 4.3: Randomly draw a sample from the experience pool. And input it into the online Q network for processing to obtain the first... Status of the water diversion pumping station system during the time period The following action and the Status of the water diversion pumping station system during the time period Next action Corresponding value ; The first Status of the water diversion pumping station system during the time period and its actions The input is processed in the target Q-network to obtain... Next action Corresponding value .
[0039] Step 4.4: Minimize the parameters of the online Q-network using equation (31) Update: (31) In equation (31), For the first The target value function for the time period is: (32) In equation (32), This is the discount factor.
[0040] Step 4.5: Use equation (33) to establish the parameters of the target Q-network. Update formula: (33) In equation (33), This indicates assignment; This represents the soft update coefficient.
[0041] Step 4.6: Train the dual Q-value network according to the process of steps 4.3-4.5 until the maximum number of training iterations is reached, thereby obtaining the trained agent.
[0042] See Figure 4 The application of intelligent agents in water pumping station scheduling decisions is explained: In the first... When the time period arrives, the status of the water diversion pumping station system should be observed and obtained first. The system then inputs the trained agent into the data. The agent then outputs action decisions based on the state information. That is, the first Time-based water pump station flow rate and will The input is fed into the Gurobi solver. The Gurobi solver solves the lower-level unit flow distribution model and outputs the first... Operating strategy of units within time-segmented water diversion pumping stations, namely , After the strategy is executed, the water diversion pumping station system undergoes a state transition, changing from the current state... Enter the next state And at the same time generate an instant reward signal. Feedback is then given to the agent. Subsequently, the agent uses the new input state information... Output the action again This process continues in a loop until the scheduling task is completed.
[0043] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.
[0044] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.
Claims
1. A model-data hybrid-driven optimized scheduling method for water transfer pumping stations, characterized in that, Includes the following steps: Step 1: Construct an operational model of the water transfer pumping station system, consisting of water transfer pumping stations, water conveyance channels, and regulating lakes; Step 2: Based on the operation model of the water transfer pumping station system, construct a two-layer optimized scheduling model for the water transfer pumping station; Step 3: Based on the two-layer optimization scheduling model of water transfer pumping stations, define the system state, actions, and reward function of water transfer pumping stations in the Markov decision process; Step 4: Based on the state, actions, and reward function of the water transfer pumping station system, the SE-Dueling DDQN algorithm is used to train the agent composed of a dual Q-value network. The trained agent is used to output the corresponding pumping station flow rate in response to the input state of the water transfer pumping station system.
2. The model-data hybrid-driven optimized scheduling method for water transfer pumping stations as described in claim 1, characterized in that, Step 1 includes: Step 1.1: Construct the operation model of the water transfer pumping station using equations (1) and (2): (1) (2) In equations (1) and (2), For the first Energy consumption of time-segmented water diversion pumping stations; The density of water; It is the acceleration due to gravity; For the first Operating efficiency of time-segmented water diversion pumping stations; For the first Operating flow rate of the water diversion pumping station during specific time periods; For the first The head of the time-limited water diversion pumping station; For the first The first time period water diversion pumping station The operating flow rate of the unit; For the first The first time period water diversion pumping station Operating efficiency of the unit; Indicates the length of each time period. This indicates the total number of units within the water diversion pumping station. , ; Step 1.2: Construct an uncertainty model of the runoff inflow and surrounding water demand of the regulating lake using equations (3) and (4): (3) (4) In equations (3) and (4), It is a generalized logical distribution; For the first Error in predicting runoff inflow to time-limited reservoirs; and They are generalized logical distributions Position parameters, scale parameters, and shape parameters; It follows a normal distribution; For the first Errors in water demand forecasting around time-limited reservoirs; Normal distribution Expected value and standard deviation; It is a natural constant; Step 1.3: Based on the uncertainty model of runoff inflow and surrounding water demand of the regulating lake, calculate the first step using equations (5) and (6). The actual runoff of the time-limited lake and the The actual water demand of the surrounding area of the time-limited storage lake : (5) (6) In equations (5) and (6), For the first Predicted runoff inflow to time-limited reservoirs; For the first Predicted water demand in the surrounding areas of time-limited water storage lakes; Step 1.4: Construct a regulating lake operation model using equation (7): (7) In equation (7), For the first The storage capacity of the regulating lake at the end of the period; For the first The storage capacity of the regulating lake at the end of the period; Step 1.5: Construct a one-dimensional unsteady flow model of the water conveyance channel using equations (8) and (9): (8) (9) In equations (8) and (9), This is the distance between the cross-section of the water conveyance channel and the starting point of the water conveyance channel; For a specific moment; The distance from the starting point of the water conveyance channel is The cross section at The water flow area at any given moment; The distance from the starting point of the water conveyance channel is The cross section at The water level at that moment; The distance from the starting point of the water conveyance channel is The cross section at Flow rate at any given moment; The distance from the starting point of the water conveyance channel is The cross section at The side stream of time; This is the momentum correction factor; The distance from the starting point of the water conveyance channel is The cross section at The hydraulic radius at any given time; The Manning coefficient for the water conveyance channel; Step 1.6: Calculate the distance to the starting point of the water conveyance channel using equation (10). The cross-section of the water conveyance channel at the first Water level during the period : (10) In equation (10), The symbol is for integrals.
3. The model-data hybrid driven optimized scheduling method for water transfer pumping stations as described in claim 2, characterized in that, The two-layer optimized scheduling model for water transfer pumping stations constructed in step 2 includes: an upper-level pumping station flow optimization model and a lower-level unit flow allocation model. Step 2.1: Construct the first objective function of the upper pump station flow optimization model using equation (11). : (11) In equation (11), Indicates the total number of time periods in the scheduling cycle; Step 2.2: Establish the second objective function of the upper pump station flow optimization model using equations (12) and (13). : (12) (13) In equations (12) and (13), For the first Penalty for deviation in storage capacity during a given period; For the first The target storage capacity of lakes at the end of the period; Step 2.3: Establish the operating flow constraints of the water transfer pumping station for the upper-level pumping station flow optimization model using equation (14): (14) In equation (14), and These are the minimum and maximum allowable flow rates for the water diversion pumping station, respectively. Step 2.4: Establish the water conveyance capacity constraints of the water conveyance channel for the upper pumping station flow optimization model using equation (15): (15) In equation (15), and These are the upper and lower limits of the permissible water conveyance flow rate for the water conveyance channel, respectively. Step 2.5: Establish the upstream water level constraint of the water transfer pumping station using equation (16) to optimize the flow rate of the upper pumping station: (16) In equation (16), For the first Water level in front of the water diversion pumping station during a specific time period; and These are the minimum and maximum water levels allowed for normal operation of the water diversion pumping station, respectively. Step 2.6: Establish the downstream water level constraint of the water transfer pumping station for the upper pumping station flow optimization model using equation (17): (17) In equation (17), For the first The water level after the pumping station during the designated water transfer period; and These are the lowest and highest post-station water levels allowed for normal operation of the water diversion pumping station, respectively. Step 2.7: Establish water level change safety constraints for the upper pump station flow optimization model using equation (18): (18) In equation (18), For the first The range of water level changes before and after the station during the specified time period; This refers to the maximum allowable water level fluctuation within a single time period. Step 2.8: Establish the head constraint of the water transfer pumping station in the flow optimization model of the upper pumping station using equation (19): (19) In equation (19), and These are the minimum and maximum head allowed for normal operation of the water diversion pumping station, respectively. Step 2.9: Using equation (20) to establish the flow optimization model of the upper pumping station, the storage capacity constraint of the storage lake is: (20) In equation (20), and These are the upper and lower limits allowed for the storage capacity of regulating lakes, respectively. Step 2.10: Construct the objective function of the lower-level unit flow allocation model using equation (21). : (21) Step 2.11: Establish the total flow constraint of the water transfer pumping station for the flow distribution model of the lower-level units using equation (22): (22) Step 2.12: Establish the flow capacity constraints of the units within the water transfer pumping station for the flow distribution model of the lower-level units using equation (23): (23) In equation (23), and These are the minimum and maximum flow rates allowed for the units within the water diversion pumping station, respectively. Step 2.13: Use equation (24) to establish the minimum start-up and shutdown time constraints for the units within the water transfer pumping station in the lower-level unit flow distribution model: (24) In equation (24): , and These refer to the start-up time, shutdown time, and minimum allowable start-up and shutdown time of the units within the water diversion pumping station.
4. The model-data hybrid-driven optimized scheduling method for water transfer pumping stations as described in claim 3, characterized in that, Step 3 includes: Step 3.1: Define the first using equation (25) Status of the water diversion pumping station system during the time period : (25) In equation (25), Indicates the sequence number of the time period; Step 3.2: Define the first Actions during a period of time For the first Pump station flow rate during the period ; Step 3.3: Construct the first equation using equation (26) Time-based rewards : (26) In equation (26), , , , and They are respectively , , , and Weighting coefficients; For the first The penalty for exceeding the storage capacity limit of the time-limited lake is obtained from equation (27). For the first The penalties for exceeding the water level limit at the time-limited water diversion pumping station before and after the station are obtained from equation (28). For the first The penalty for exceeding the limit of water level change range before and after the time period water diversion pumping station is obtained from equation (29). For the first The head over-limit penalty of the time-limited water diversion pumping station is obtained from equation (30): (27) (28) (29) (30)。 5. The model-data hybrid-driven optimized scheduling method for water transfer pumping stations as described in claim 4, characterized in that, Step 4 includes: Step 4.1: Construct a dual-Q value network consisting of an online Q-network and a target Q-network, and initialize the parameters of the online Q-network. and the parameters of the target Q network ; Step 4.2: Place the first Status of the water diversion pumping station system during the time period The input is processed in an online Q network to obtain the first... Actions during a period of time ; Will Substituting the values into the flow distribution model of the lower-level units and solving for the result, we obtain the... The operating flow rate of each unit within the time-limited water diversion pumping station is calculated and executed by the water diversion pumping station system to obtain the first... Time-based rewards and the Status of the water diversion pumping station system during the time period ; Will Each sample is stored in the experience pool until the number of samples in the experience pool reaches a certain threshold. up to the last one; Step 4.3: Randomly draw a sample from the experience pool. And input it into the online Q network for processing to obtain the first... Status of the water diversion pumping station system during the time period The following action and the Status of the water diversion pumping station system during the time period Next action Corresponding value ; The first Status of the water diversion pumping station system during the time period and its actions The input is processed in the target Q-network to obtain... Next action Corresponding value ; Step 4.4: Minimize the parameters of the online Q-network using equation (31) Update: (31) In equation (31), For the first The target value function for the time period is: (32) In equation (32), Discount factor; Step 4.5: Use equation (33) to establish the parameters of the target Q-network. Update formula: (33) In equation (33), This indicates assignment; Indicates the soft update coefficient; Step 4.6: Train the dual Q-value network according to the process of Steps 4.3-4.5 until the maximum number of training iterations is reached, thereby obtaining the trained agent.
6. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports a processor in executing the method of any one of claims 1-5, the processor being configured to execute the program stored in the memory.
7. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the steps of the method according to any one of claims 1-5.