Energy-saving scheduling method, system, electronic device and storage medium for pumping station water supply system

Abstract

The present invention discloses an energy-saving scheduling method, system, electronic device and storage medium for a pumping station water supply system, which relates to the technical field of water supply. By using the trained and optimized DQN model, the optimal action combination corresponding to different water usage demands is determined, and the flow rates of each pump in the pumping station water supply system are regulated according to the optimal action combination, so as to ensure that each non-zero flow rate in the regulated flow rates of each pump is within the flow rate range corresponding to the high-efficiency section of the pump. When the flow rate of the pump is within the flow rate range corresponding to the high-efficiency section, the pump can operate efficiently. Since the flow rate regulation of the pump is achieved by adjusting the operating speed of the variable-frequency pump and controlling the start and stop of the power-frequency pump, the present invention can automatically adjust the operating speed of each variable-frequency pump and control the start and stop of each power-frequency pump according to the optimal action combination, so as to enable the pumping station to meet the water supply flow rate requirement and the head requirement, and at the same time effectively improve the overall operating efficiency of the water supply pumping station.

Description

Technical Field

[0001] The present invention relates to the technical field of water supply, and in particular to an energy-saving scheduling method, system, electronic device and storage medium for a pumping station water supply system. Background Art

[0002] With the continuous deepening of urbanization, the urban population in China has increased from 140 million in 1981 to 440 million today. To ensure the water demand of the urban population, the investment scale of the urban water supply system has been expanding. According to the data of the China Statistical Yearbook, the total national water supply reached 58.6 billion cubic meters at the end of 2020, and the water supply investment was 74.9 billion yuan. Water supply enterprises need to invest a large amount of energy costs every year to maintain a huge water supply system, and up to 90% of the power consumption is used to maintain the operation of water pumps.

[0003] The parallel water supply method is generally adopted in pumping station water supply, that is, the water pumps (including variable-frequency pumps and industrial-frequency pumps) in the pumping station work in parallel. The head of all water pumps is the same, but the flow rates of each water pump are different, and the sum of the flow rates of each water pump is the total flow rate of the pumping station. In the urban water supply system in China, most pumping stations adopt the manual regulation method, which often formulates the operating speeds of each variable-frequency pump and the start-stop schemes of each industrial-frequency pump according to historical experience, so that the pumping station meets the water supply flow rate requirements and head requirements. Although it can meet the current water use requirements of the user side (including water supply flow rate requirements and head requirements), it does not consider the overall operating efficiency of the water supply pumping station, and there are problems such as low operating efficiency and energy waste. Summary of the Invention

[0004] The purpose of the present invention is to provide an energy-saving scheduling method, system, electronic device and storage medium for a pumping station water supply system, which can automatically adjust the operating speeds of each variable-frequency pump and control the start-stop of each industrial-frequency pump, so that the pumping station meets the water supply flow rate requirements and head requirements, and at the same time effectively improves the overall operating efficiency of the water supply pumping station.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] An energy-saving scheduling method for a pumping station water supply system, the method includes:

[0007] Step S1: Obtain different water use requirements; the water use requirements include the total flow rate requirement and the head requirement;

[0008] Step S2: Construct a DQN model;

[0009] Step S3: For each of the water use requirements, input the water use requirements into the DQN model, and use the DQN model to randomly allocate the total flow rate requirement to each pump in the pumping station water supply system to obtain the state space corresponding to the water use requirements;

[0010] Step S4: Obtain the efficiency of multiple action combinations corresponding to the state space by using the DQN model; the action combinations include regulation actions for each pump in the pumping station water supply system.

[0011] Step S5: Regulate the flow rates of each pump in the pumping station water supply system according to the action combination with the highest efficiency corresponding to the state space, and obtain the regulated flow rates of each pump in the pumping station water supply system.

[0012] Step S6: Determine whether each non-zero flow rate among the regulated flow rates of each pump is within the set flow rate range corresponding to the pump with the non-zero flow rate; the set flow rate range is the flow rate range corresponding to the high-efficiency section of the pump.

[0013] If the output result of step S6 is no, then execute step S7: Update the flow rates of each pump in the state space with the regulated flow rates of each pump, and return to step S4 until the output result of step S6 is yes.

[0014] If the output result of step S6 is yes, then execute step S8: Record the action combination with the highest efficiency corresponding to the state space as the optimal action combination corresponding to the water usage demand.

[0015] Step S9: Construct a training data set by using different water usage demands and the optimal action combinations corresponding to the water usage demands.

[0016] Step S10: Train and optimize the DQN model by using the training data set to obtain an optimized DQN model.

[0017] Step S11: Input the actual water usage demand into the optimized DQN model to obtain the optimal action combination corresponding to the actual water usage demand.

[0018] Step S12: Perform energy-saving scheduling on the pumping station water supply system according to the optimal action combination corresponding to the actual water usage demand; the energy-saving scheduling includes regulating the flow rates of each pump in the pumping station water supply system according to the optimal action combination corresponding to the actual water usage demand.

[0019] Optionally, the state space includes the flow rates of each pump in the pumping station water supply system and the head demand; the sum of the flow rates of each pump in the pumping station water supply system is equal to the total flow rate demand.

[0020] Optionally, the pumping station water supply system includes multiple variable-frequency pumps and multiple power-frequency pumps; the regulation actions of the variable-frequency pumps include reducing the speed, keeping the speed unchanged, and increasing the speed; the regulation actions of the power-frequency pumps include starting and stopping.

[0021] The present invention also provides the following solutions:

[0022] An energy-saving scheduling system for a pumping station water supply system, the system comprising:

[0023] A water demand acquisition module for acquiring different water demands; the water demands include the total flow demand and the head demand;

[0024] A model construction module for constructing a DQN model;

[0025] A state space obtaining module for, for each of the water demands, inputting the water demand into the DQN model, and using the DQN model to randomly allocate the total flow demand to each pump in the pumping station water supply system to obtain the state space corresponding to the water demand;

[0026] An efficiency obtaining module for various action combinations for using the DQN model to obtain the efficiencies of various action combinations corresponding to the state space; the action combinations include regulation actions for each pump in the pumping station water supply system;

[0027] A pump flow regulation module for regulating the flow rates of each pump in the pumping station water supply system according to the action combination with the highest efficiency corresponding to the state space to obtain the regulated flow rates of each pump in the pumping station water supply system;

[0028] A judgment module for judging whether each non-zero flow rate among the regulated flow rates of each pump is within the set flow rate range of the pump corresponding to the non-zero flow rate; the set flow rate range is the flow rate range corresponding to the high-efficiency section of the pump;

[0029] A flow rate update module for, when the output result of the judgment module is no, updating the flow rates of each pump in the state space with the regulated flow rates of each pump, and returning to the efficiency obtaining module for various action combinations until the output result of the judgment module is yes;

[0030] An optimal action combination recording module for, when the output result of the judgment module is yes, recording the action combination with the highest efficiency corresponding to the state space as the optimal action combination corresponding to the water demand;

[0031] A training data set construction module for constructing a training data set by using different water demands and the optimal action combinations corresponding to the water demands;

[0032] A model training and optimization module for training and optimizing the DQN model by using the training data set to obtain an optimized DQN model;

[0033] An optimal action combination obtaining module for inputting the actual water demand into the optimized DQN model to obtain the optimal action combination corresponding to the actual water demand;

[0034] The energy-saving scheduling module of the pumping station water supply system is used to perform energy-saving scheduling on the pumping station water supply system according to the optimal action combination corresponding to the actual water demand; the energy-saving scheduling includes regulating the flow rate of each pump in the pumping station water supply system according to the optimal action combination corresponding to the actual water demand.

[0035] Optionally, the state space includes the flow rate of each pump in the pumping station water supply system and the head demand; the sum of the flow rates of each pump in the pumping station water supply system is equal to the total flow demand.

[0036] Optionally, the pumping station water supply system includes a plurality of variable-frequency pumps and a plurality of industrial-frequency pumps; the regulation actions of the variable-frequency pumps include reducing the rotation speed, keeping the rotation speed unchanged, and increasing the rotation speed; the regulation actions of the industrial-frequency pumps include starting and stopping.

[0037] The present invention also provides the following solutions:

[0038] An electronic device includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the energy-saving scheduling method of the pumping station water supply system described above.

[0039] The present invention also provides the following solutions:

[0040] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the energy-saving scheduling method of the pumping station water supply system described above.

[0041] According to the specific embodiments provided by the present invention, the following technical effects are disclosed:

[0042] The energy-saving scheduling method, system, electronic device and storage medium of the pumping station water supply system disclosed by the present invention use the trained and optimized DQN model to determine the optimal action combination corresponding to different water demands, and regulate the flow rate of each pump in the pumping station water supply system according to the optimal action combination, so as to ensure that each non-zero flow rate in the regulated flow rates of each pump is within the flow rate range corresponding to the efficient section of the pump. When the flow rate of the pump is within the flow rate range corresponding to the efficient section, the pump can work efficiently (high operating efficiency). Since the flow rate regulation of the pump is achieved by adjusting the operating speed of the variable-frequency pump and controlling the start and stop of the industrial-frequency pump, the present invention can automatically adjust the operating speed of each variable-frequency pump and control the start and stop of each industrial-frequency pump according to the optimal action combination, so as to enable the pumping station to meet the water supply flow requirement (total flow demand) and head requirement (head demand), and at the same time effectively improve the overall operating efficiency of the water supply pumping station. Description of the Drawings

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0044] Figure 1 It is a flowchart of the first embodiment of the energy-saving scheduling method for the pump station water supply system of the present invention;

[0045] Figure 2 It is a schematic diagram of the high-efficiency area of the variable-frequency pump of the present invention;

[0046] Figure 3 It is a schematic diagram of the principle of the DQN algorithm of the present invention;

[0047] Figure 4 It is a schematic diagram of the training process of the pump unit operation state regulation based on the DQN algorithm of the present invention;

[0048] Figure 5 It is a schematic diagram of the working principle of the neural network of the present invention;

[0049] Figure 6 It is a schematic diagram of the fitting relationship between the pump efficiency η - flow rate Q and head H of the water pump of model 500*350CW11GM of the present invention;

[0050] Figure 7 It is a schematic diagram of the fitting relationship between the pump efficiency η - flow rate Q and head H of the water pump of model GSC400 / 500-500 / 7 of the present invention;

[0051] Figure 8 It is a schematic diagram of the fitting relationship between the pump efficiency η - flow rate Q and head H of the water pump of model GSC400 / 500-500 / 6 of the present invention;

[0052] Figure 9 It is a schematic diagram of the fitting relationship between the pump efficiency η - flow rate Q and head H of the water pump of model 600*400CW10GM of the present invention;

[0053] Figure 10 It is a comparison chart of the DQN regulation result and the manual regulation result of the present invention. Detailed implementation manners

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0055] The object of the present invention is to provide an energy-saving scheduling method, system, electronic device and storage medium for a pumping station water supply system, which can automatically adjust the operating speeds of each variable-frequency pump and control the start and stop of each industrial-frequency pump, so that the pumping station meets the water supply flow requirement and head requirement, and at the same time effectively improve the overall operating efficiency of the water supply pumping station.

[0056] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Embodiment 1

[0058] Figure 1 It is a flowchart of Embodiment 1 of the energy-saving scheduling method for the pumping station water supply system of the present invention. As Figure 1 shown, this embodiment provides an energy-saving scheduling method for a pumping station water supply system, including the following steps:

[0059] Step S1: Obtain different water usage demands; the water usage demands include the total flow demand and the head demand.

[0060] Step S2: Construct a DQN model.

[0061] Step S3: For each water usage demand, input the water usage demand into the DQN model, and use the DQN model to randomly distribute the total flow demand to each pump in the pumping station water supply system to obtain the state space corresponding to the water usage demand.

[0062] Step S4: Use the DQN model to obtain the efficiency of multiple action combinations corresponding to the state space; the action combinations include the regulation actions for each pump in the pumping station water supply system.

[0063] Step S5: Regulate the flow of each pump in the pumping station water supply system according to the action combination with the highest efficiency corresponding to the state space to obtain the regulated flow of each pump in the pumping station water supply system.

[0064] Step S6: Judge whether each non-zero flow in the regulated flow of each pump is within the set flow range of the pump corresponding to the non-zero flow; the set flow range is the flow range corresponding to the high-efficiency section of the pump.

[0065] If the output result of Step S6 is no, then execute Step S7: Update the flow of each pump in the state space with the regulated flow of each pump, and return to Step S4 until the output result of Step S6 is yes.

[0066] If the output result of Step S6 is yes, then execute Step S8: Record the action combination with the highest efficiency corresponding to the state space as the best action combination corresponding to the water usage demand.

[0067] Step S9: Construct a training dataset using different water usage demands and the corresponding optimal action combinations for the water usage demands.

[0068] Step S10: Train and optimize the DQN model using the training dataset to obtain an optimized DQN model.

[0069] Step S11: Input the actual water usage demand into the optimized DQN model to obtain the optimal action combination corresponding to the actual water usage demand.

[0070] Step S12: Perform energy-saving scheduling on the pump station water supply system according to the optimal action combination corresponding to the actual water usage demand; the energy-saving scheduling includes regulating the flow rates of each pump in the pump station water supply system according to the optimal action combination corresponding to the actual water usage demand.

[0071] Specifically, the state space includes the flow rate and head demands of each pump in the pump station water supply system; the sum of the flow rates of each pump in the pump station water supply system is equal to the total flow rate demand.

[0072] The pump station water supply system includes multiple variable-frequency pumps and multiple industrial-frequency pumps; the regulation actions of the variable-frequency pumps include reducing the rotational speed, keeping the rotational speed unchanged, and increasing the rotational speed; the regulation actions of the industrial-frequency pumps include starting and stopping.

[0073] The following uses a specific embodiment to illustrate the technical solution of the present invention:

[0074] The energy-saving scheduling method for a pump station water supply system provided by the present invention is an energy-saving scheduling method for a pump station water supply system based on the DQN algorithm. It mathematically describes and Markov decision process describes the problem of optimizing the state of the pump group, and at the same time defines the state space, action space, and immediate reward value during the operation of the pump group, constructs a DQN network, and improves the overall efficiency of the pump group.

[0075] An energy-saving scheduling method for a pump station water supply system based on the DQN algorithm provided by the present invention specifically includes the following steps:

[0076] Step 1: Convert the operating state of the pump into a mathematical description.

[0077] This step 1 includes establishing an objective function and setting constraint conditions, specifically as follows:

[0078] ① Define the high-efficiency area of the pump

[0079] Determine the maximum efficiency of the pump (usually taking η>90%), and then define the high-efficiency area of the pump according to the characteristic curve of the pump, such as Figure 2 shown.

[0080] ② Establish the objective function:

[0081] Suppose there are x variable-frequency pumps and y constant-frequency pumps in the pump group. The pump types of each pump are determined, the pipeline characteristic curve is determined, and the parallel water supply mode is adopted. Given the required water supply flow rate Q sum and the required head H e , the operating combination of the water pumps and the speed regulation ratio k i of each water pump are obtained. On the premise that the pumping station meets the water supply requirements and each water pump is in an efficient operating state, the highest total efficiency of the pumping station is achieved. The operating combination of the water pumps and the speed regulation ratio k i of each water pump are determined by drawing the characteristic curve after series connection according to the characteristic curves of each water pump (using the method of adding the flow rates at the same head of each water pump). The speed regulation ratio of the water pump is obtained by dividing the adjusted water pump speed by the designed speed.

[0082] Based on the above problem description, when the pumping station supplies water, the objective function of system optimization is as follows:

[0083]

[0084] This objective function is determined for the head and flow rate requirements. This objective function is determined based on the DQN algorithm and can ensure that each pump works in its efficient section as much as possible. Among them, η is the total efficiency of the pump group; γ is the unit weight of water; Q sum is the total required water supply flow rate (water supply flow rate requirement), that is, the total flow rate provided by the pumping station; H e is the required water supply head, that is, the head provided by the pumping station; N i is the power of the i-th pump.

[0085] ③ Set the constraint conditions: The operating point of the water pump must be within its efficient area. Taking this as the constraint condition, the mathematical description is as follows:

[0086]

[0087] The constraint condition is that under the required Q and H, it is ensured that the water pump N (efficiency) should be in its efficient section. Among them, H x is the virtual total head; S x is the virtual resistance loss coefficient in the pump body; x represents x variable-frequency pumps; k min represents the minimum speed regulation ratio of the pump; Q i is the flow rate at any speed; Q min , Q max represent the efficient area limits, and the expressions are as follows:

[0088]

[0089]

[0090] Among them, Q A and H Aare the minimum flow rate corresponding to the efficient section of the water pump operating at the rated speed and the head at the minimum flow rate; H C is the minimum flow rate corresponding to the efficient section of the water pump operating at the lowest speed; Q B and H B are the maximum flow rate corresponding to the efficient section of the water pump operating at the rated speed and the head at the maximum flow rate respectively.

[0091] Step 2: Express the water pump operation problem as a Markov decision process, and define the parameters of the Markov decision model (construct the Markov decision model according to the state space S, action space A, state transition function, and immediate reward).

[0092] The Markov decision process (MDP) provides a framework for reinforcement learning algorithms, and its model can be represented by the tuple (S, A, T, R, γ). Among them, S represents the state space of the system; A represents the action space; T represents the state transition matrix, that is, the probability of changing from one state to another after executing a specific action; R is the immediate reward obtained after executing a specific action; γ is the discount factor, that is, the importance of future rewards.

[0093] In the water supply system, due to the fact that the water supply flow rate requirement Q sum and the head requirement H e are random variables, the problem of optimizing the operation state of the water pump group can be regarded as a sequential decision-making problem under uncertain conditions. The present invention expresses it as a Markov decision process and defines the parameters in the model, including: state space S, action space A, state transition function, and immediate reward.

[0094] ① State space S (flow rate and head under the operation of each water pump)

[0095] At a certain time step, the intelligent agent will observe the state of the environment, which includes the operation state of each water pump (which describes the operation state of each variable-frequency pump), including parameters such as flow rate and head. Its expression is as follows (since the water supply method of the pumping station generally adopts parallel water supply, the heads of each water pump are regarded as the same):

[0096] S = [Q 1 , Q 2 , …, Q x+y , H]

[0097] Among them, Q 1 , Q 2 , …, Q x+y represent the water supply flow rates of each pump, and H represents the water supply head of the pumping station.

[0098] ② Action space A (speed regulation and opening / closing of each variable-frequency pump)

[0099] After the agent obtains the environmental state S, it will give an action A, which includes regulating the rotational speeds of the variable-frequency pumps and the opening and closing states of the power-frequency pumps. The goal of the water pump group operation state regulation problem is to optimally determine the rotational speeds of the variable-frequency pumps and the opening and closing states of the power-frequency pumps. The present invention divides the regulation of the variable-frequency pumps and the power-frequency pumps into action spaces A v and A g respectively, and the expressions are as follows:

[0100]

[0101]

[0102] Then the action of the agent can be expressed as:

[0103]

[0104] where the meanings of the parameters are as follows: a ∈ A, i ∈ [1, x], j ∈ [1, y]. A v represents the regulation of the variable-frequency pumps by the agent, represents the regulation of the x-th variable-frequency pump by the agent. Specifically, "0" means to lower the rotational speed, "1" means to keep the rotational speed unchanged, and "2" means to increase the rotational speed. A g represents the opening and closing state of the power-frequency pumps by the agent, represents the opening and closing state of the y-th power-frequency pump by the agent. Specifically, "0" means to open, and "1" means to close.

[0105] ③ State transition function (the pump changes from one working state to another)

[0106] The traditional Markov model predicts future states through a state transition matrix. However, when the state space is high-dimensional and continuous, a state transition function can be used for prediction, that is:

[0107]

[0108] State transition function can be obtained through neural network learning (obtained through a DQN network), where f represents a function.

[0109] ④ Immediate reward (that is, the reward obtained when the pump operates in the high-efficiency section is recorded)

[0110] When the agent executes an action a in state s, it will obtain an immediate reward r. Since the goal of optimizing the water pump group operation state problem is to maximize the total efficiency of the pumping station, assuming the motor efficiency remains unchanged, the immediate reward r can be set as:

[0111]

[0112] Step 3: Construct the DQN algorithm process. The overall process is as follows: Based on the rewards feedback from the environment, the agent learns the optimal action for each state.

[0113] The DQN algorithm is one of the algorithms for solving reinforcement learning problems under the framework provided by the Markov decision process and needs to satisfy the assumption conditions of the Markov decision process. Therefore, based on the above description of the Markov decision process, the DQN algorithm process as shown Figure 3 can be constructed. The specific steps are as follows: In the pumping station, the pumps work in parallel (the head of all pumps is the same, but the flow rates are different, and the sum of the flow rates is the total flow rate). Thus, DQN is used for regulation. The regulation process is to determine the flow rate distribution of each pump under the given head. The sum of the flow rates of each pump is equal to the total flow rate demand. If the total flow rate demand changes, first randomly distribute the total flow rate, and then the agent will decide how to adjust each pump according to the current head and the flow rate of each pump, and finally adjust it until the agent thinks the most efficient scheme is reached. The adjustment method for each pump is to increase or decrease the flow rate step by step while ensuring that their flow rates add up to a constant value. If the efficiency of a certain pump is too low during the process, the pump will be directly turned off, that is, the flow rate becomes 0. Only steps 6, 7, and 8 are the regulation processes. Before this is the initialization, that is, the preparation stage, and after this is to optimize the parameters of the network. Obtain the results step by step according to the above steps. The agent automatically adjusts the rotational speeds of each variable-frequency pump after the demand changes, and on the basis of meeting the water supply demand, makes the total efficiency of the pump group as high as possible.

[0114] The principle of the DQN algorithm is as shown Figure 3 below. The overall process is as follows: Based on the rewards feedback from the environment, the agent learns the optimal behavior for each state (that is, under the given demand working state, each pump will wait for its own state with high efficiency under mutual combination and then be memorized in the system). Compared with the traditional reinforcement learning method Q-Learning, the DQN algorithm uses a neural network to replace the Q-table and introduces an experience replay pool (that is, the efficiencies of each pump under different working states are memorized and put into the recycling pool) to solve the problems of continuous high-dimensional state space and catastrophic forgetting.

[0115] As shown Figure 4 below, based on the DQN algorithm, the training process of the pump group operation state regulation is as follows (that is, for the flow rate demand under parallel connection, first set an initial Q for each pump, and then start iterative calculation until the flow rate reaches the target Q value, and then each pump will have the corresponding flow rate, and at this flow rate, it is just in the high-efficiency section). In this process, first set a flow rate:

[0116] (1) Initialize the evaluation network (Q) and initialize the target network (targetQ).

[0117] (2) Initialize the experience replay pool (replay_memory).

[0118] (3) Initialize the environment state (environment).

[0119] (4) Start the first episode.

[0120] (5) Start iteration from step = 1.

[0121] (6) The agent observes the environment and obtains the state S t (Flow rate and head of each pump under the total flow rate of demand).

[0122] (7) Take the state S t as the input of the evaluation network (Q), output the action values of all actions in the current state, and then select and execute the action A using the ε-greedy strategy t (i.e., whether to adjust the speed or open / close the pump at this flow rate and head).

[0123] (8) The agent obtains the new state S t+1 and the immediate reward r t (Obtain the new flow rate and head, and the reward is whether it is in the high-efficiency area, and there is a large reward if it is in the high-efficiency area).

[0124] (9) Store (S t , A t , S t+1 , r t ) in the experience replay pool (store the obtained values in the system).

[0125] (10) Randomly sample a batch of samples from the experience replay pool as the input of the target network and the evaluation network, and then output their respective Q values (reset the new flow rate demand and the target flow rate).

[0126] (11) Calculate the loss function based on the difference between the two Q values and update the weights of the evaluation network through backpropagation (i.e., the degree of closeness).

[0127] (12) Judge whether N steps have passed. If so, assign the weights of the evaluation network to the target network.

[0128] (13) Judge whether M steps have passed. If so, increment the episode count by 1 and return to step (3); if not, increment the step count by 1 and return to step (5).

[0129] (14) Judge whether the episode count is equal to max_episode. If so, end the training.

[0130] Step 4: Build a simulation environment;

[0131] ① Set the water supply scenario

[0132] Set the total water supply flow rate and head, as well as parameters of the pump group such as rated power, rated voltage, rated flow rate, rated head, speed range, efficiency, etc. (These parameters are required according to the water supply demand. With these parameters, the model can determine how to combine the pumps to operate in the high-efficiency section as much as possible for energy conservation).

[0133] ② Determine the power N and efficiency η of each pump during the water supply process (The purpose of calculating the power N and efficiency η of each pump is to determine whether each pump is operating in the high-efficiency section)

[0134] The basis for evaluating the advantages and disadvantages of the DQN algorithm in this problem is whether the simulation environment can accurately simulate the water supply demand and the operating state of the pump group. If simulating the operating state of a variable-frequency pump, a quantitative relationship of the power N of the variable-frequency pump with respect to the flow rate Q and head H at any speed is required (i.e., for the simulation of the variable-frequency pump, a relationship of its power N with respect to the flow rate Q and head H at any speed is needed). For medium and large-sized pumps that have been put into operation, it is difficult to obtain the relationship between various parameters through on-site testing. Therefore, this relationship must be obtained from the pump performance curve diagram.

[0135] Simply put, first obtain the quantitative relationship of the power N of the variable-frequency pump with respect to the flow rate Q and head H at any speed by analyzing the pump performance curve diagram, that is, at the rated speed, a series of (m) efficiencies η 0 , flow rate Q 0 , head H 0 and the calculated power N 0 of discrete values, as shown in the following formula:

[0136] η 0 =[η 0 1 ,η 0 2 ,…,η 0 m

[0137] Q 0 =[Q 0 1 ,Q 0 2 ,…,Q 0 m

[0138] N 0 =[N 0 1 ,N 0 2 ,…,N​​0 m

[0139] H 0 = [H 0 1 , H 0 2 , …, H 0 m

[0140] According to the similarity law, any speed ratio k can be obtained i Under this condition, the power N of each water pump i , flow rate Q i , head H i , that is:

[0141]

[0142]

[0143]

[0144] Then build a supervised learning neural network: Under the PyTorch framework, build 2 fully connected layers, with 128 neurons in each layer, and the activation function between each layer is the ReLU function. Its working principle is as Figure 5 shown.

[0145] Thus, a non-linear relationship model of power with respect to flow rate and head can be obtained, that is:

[0146] N = f(Q, H)

[0147] ③ Set DQN parameters (DQN parameters are used to simulate the actual situation)

[0148] Adopt the operation data of the actual water pump station. Using PyTorch as the deep learning tool, the set state space S includes the flow rate Q of each water pump and the head H provided by the water pump group, and the action space A is to adjust the speed of each water pump. Adopt the structure of a deep neural network (DNN), with three hidden layers and 128 neurons in each layer. The activation function between layers uses the ReLU function, and set hyperparameters.

[0149] Step 5: Obtain the actual operation data of the water pump, including the number of water pumps started and the flow rate, head, power, efficiency, etc. of each water pump as the time and water consumption change.

[0150] Step 6: Train the DNN network;

[0151] Use the obtained past operation data (the data of the number of water pumps started and the flow rate, head, power, efficiency, etc. of each water pump as the time and water consumption change obtained in Step 5) to train the network.​​

[0152] Step 7: Input the water supply demand, intelligently adjust the rotational speeds of each water pump, and achieve energy-saving dispatching of the pumping station water supply system (when the water supply demand is input, the water pumps will automatically adjust their rotational speeds to operate in the high-efficiency section).

[0153] The following takes the application of the energy-saving dispatching method for the pumping station water supply system of the present invention to a certain waterworks for water supply to its responsible area as an example to further illustrate the energy-saving dispatching method for the pumping station water supply system of the present invention:

[0154] The selected scenario is that the waterworks supplies water to its responsible area, and the water supply situation of the waterworks and its responsible area is used as the training scenario. The total flow rate and head data are sourced from the weekly water supply data provided by the waterworks, among which the data of the first 5 days are used for training, and the data of the remaining days are used for result evaluation. Among the pump sets used for water supply, 5 variable-frequency pumps are set, and the parallel water supply method is adopted. The specific parameters are shown in Table 1:

[0155] Table 1 Performance Parameter Table of Water Pumps

[0156]

[0157] Determine the power N and efficiency η of each water pump during the water supply process through the method described in the second part of Step 4.

[0158] To verify the model effect, the present invention calculates the efficiency η of each water pump at different flow rates and heads through the following formula:

[0159]

[0160] Among them, g represents the acceleration due to gravity, ρ represents the density of water, N represents the power of the water pump, Q represents the flow rate of the water pump, and H represents the head provided by the pump set; and a surface plot is drawn in Matlab software, and the results are as shown in Figure 6 、 Figure 7 、 Figure 8 and Figure 9 shown.

[0161] From Figure 6 、 Figure 7 、 Figure 8 and Figure 9 it can be seen that the efficiency of the water pumps is all between 40% and 90%, and the results are in line with the actual situation; while the high-efficiency area of the water pumps requires an efficiency of more than 70%, which provides an operating space for the program to adjust the water pumps to the high-efficiency operation area.

[0162] Using PyTorch as the deep learning tool, the set state space includes the flow rates of 5 pumps and the head H provided by the pump group. The action space is to adjust the rotational speeds of each pump. Both the evaluation network and the target network adopt the deep neural network (DNN) structure, with three hidden layers, 128 neurons in each layer, and the ReLU function is used as the activation function between layers. The hyperparameter settings are shown in Table 2:

[0163] Table 2 Hyperparameter Settings Table

[0164]

[0165] Using the weekly water supply data obtained by the water plant, the data of the first 5 days are used for training, and the data of the last 2 days are used for result evaluation. On the premise of the same water supply demand, the power consumption of the pump group under two methods of manual regulation and DQN regulation at different time periods is compared, as Figure 10 shown.

[0166] According to the calculation, the average loss power of the pump group using the current manual regulation method is 143.67 kW, while the average loss power of the pump group regulated by the DQN algorithm is 130.96 W, with an 8.84% reduction in power loss compared to manual regulation. It can be seen that the DQN algorithm regulation has good economic efficiency in optimizing the operating state of the pump group and can effectively improve the operating efficiency of the pump station.

[0167] Based on the premise of meeting the water use demand of urban residents, the energy-saving benefits brought by improving the overall efficiency of the pump group in the water supply pump station are very significant. Studying a practical system optimization algorithm is one of the important directions to solve the problem of energy waste in the water supply system. This invention proposes to introduce a deep reinforcement learning algorithm - the DQN algorithm, taking the operating state of the pump group as the input of the evaluation network. When the user demand changes, it automatically adjusts the rotational speed of the pump to enable the pump to work efficiently under the required working conditions, thereby achieving the goal of efficient energy utilization and cost reduction. Since the energy-saving optimization problem of the pump group operating state in the pump station can be abstracted as a dynamic programming problem with the total efficiency of the pump group as the optimization goal and the high-efficiency area of the pump as the constraint condition under the changing water supply demand. Deep reinforcement learning not only has good perception ability but also has good decision-making ability, and is very suitable for dynamic programming problems. It has been widely applied in fields such as industrial manufacturing, optimization scheduling, and game playing, and has not been applied in the Chinese water supply field yet. Among the current mature deep reinforcement learning algorithms, the DQN algorithm is more suitable for continuous state spaces and has good research prospects. Therefore, this invention realizes the energy-saving scheduling of the pump station water supply system based on the DQN algorithm. The energy-saving scheduling method of the pump station water supply system of this invention can automatically adjust the operating rotational speed of each variable-frequency pump and control the start and stop of each power-frequency pump, so as to enable the pump station to meet the water supply flow requirement and head requirement, and at the same time effectively improve the overall operating efficiency of the water supply pump station.

[0168] Compared with the prior art, the advantages of the present invention are as follows:

[0169] (1) Better energy conservation and emission reduction effects

[0170] The DQN algorithm is applied to the water supply field for the first time, and the energy conservation effect is good. It not only meets the water use requirements of the user side, but also makes each pump and pumping station operate at high efficiency, reducing the energy consumption of the pumping station operation.

[0171] (2) Realize automatic regulation of the state of the pump group

[0172] According to the real-time water use requirements of users, the optimal pump combination and their respective speeds are quickly solved, saving human resources, reducing the risk of non-instantaneous regulation by operators at the same time, and improving the water supply service level.

[0173] (3) Adaptive update algorithm

[0174] This algorithm can update the network weights in an online learning manner. Even if there are abnormal water supply requirements or the characteristic curves of the pump group change, it can adaptively change.

[0175] Embodiment 2

[0176] In order to execute the method corresponding to the above Embodiment 1 to achieve the corresponding functions and technical effects, an energy-saving dispatching system for a pumping station water supply system is provided below. The system includes:

[0177] A water use demand acquisition module for acquiring different water use demands; the water use demands include the total flow demand and the head demand.

[0178] A model construction module for constructing a DQN model.

[0179] A state space obtaining module for inputting the water use demand into the DQN model for each water use demand, and randomly distributing the total flow demand to each pump in the pumping station water supply system by using the DQN model to obtain the state space corresponding to the water use demand.

[0180] An efficiency obtaining module for various action combinations for obtaining the efficiencies of various action combinations corresponding to the state space by using the DQN model; the action combinations include regulation actions for each pump in the pumping station water supply system.

[0181] A flow regulation module for pumps for regulating the flow of each pump in the pumping station water supply system according to the action combination with the highest efficiency corresponding to the state space to obtain the regulated flow of each pump in the pumping station water supply system.

[0182] A judgment module, configured to judge whether each non-zero flow rate among the flow rates after regulating each pump is within the set flow rate range of the pump corresponding to the non-zero flow rate; the set flow rate range is the flow rate range corresponding to the high-efficiency section of the pump.

[0183] A flow rate update module, configured to, when the output result of the judgment module is negative, update the flow rates of each pump in the state space with the flow rates after regulating each pump, and return to the efficiency obtaining module of multiple action combinations until the output result of the judgment module is positive.

[0184] An optimal action combination recording module, configured to, when the output result of the judgment module is positive, record the action combination with the highest efficiency corresponding to the state space as the optimal action combination corresponding to the water use demand.

[0185] A training data set construction module, configured to construct a training data set by using different water use demands and the optimal action combinations corresponding to the water use demands.

[0186] A model training and optimization module, configured to train and optimize a DQN model by using the training data set to obtain an optimized DQN model.

[0187] An optimal action combination obtaining module, configured to input the actual water use demand into the optimized DQN model to obtain the optimal action combination corresponding to the actual water use demand.

[0188] A pump station water supply system energy-saving scheduling module, configured to perform energy-saving scheduling on the pump station water supply system according to the optimal action combination corresponding to the actual water use demand; the energy-saving scheduling includes regulating the flow rates of each pump in the pump station water supply system according to the optimal action combination corresponding to the actual water use demand.

[0189] Wherein, the state space includes the flow rate and head demand of each pump in the pump station water supply system; the sum of the flow rates of each pump in the pump station water supply system is equal to the total flow rate demand.

[0190] Specifically, the pump station water supply system includes a plurality of variable-frequency pumps and a plurality of industrial-frequency pumps; the regulation actions of the variable-frequency pumps include reducing the rotation speed, keeping the rotation speed unchanged, and increasing the rotation speed; the regulation actions of the industrial-frequency pumps include starting and stopping.

[0191] Embodiment III

[0192] Embodiment III of the present invention provides a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable an electronic device to execute the energy-saving scheduling method for the pump station water supply system in Embodiment I.

[0193] The above-mentioned electronic device may be a server.

[0194] Embodiment IV

[0195] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the energy-saving scheduling method for the pumping station water supply system in Embodiment 1.

[0196] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0197] Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method of the present invention and its core idea. At the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. An energy-saving scheduling method for a pumping station water supply system, characterized in that, the method includes: Step S1: Obtain different water usage demands; the water usage demands include the total flow demand and the head demand; Step S2: Construct a DQN model; Step S3: For each of the water usage demands, input the water usage demand into the DQN model, and use the DQN model to randomly allocate the total flow demand to each pump in the pumping station water supply system to obtain the state space corresponding to the water usage demand; Step S4: Use the DQN model to obtain the efficiencies of multiple action combinations corresponding to the state space; the action combinations include regulation actions for each pump in the pumping station water supply system; Step S5: Regulate the flow rates of each pump in the pumping station water supply system according to the action combination with the highest efficiency corresponding to the state space to obtain the regulated flow rates of each pump in the pumping station water supply system; Step S6: Determine whether each non-zero flow rate among the regulated flow rates of each pump is within the set flow rate range of the pump corresponding to the non-zero flow rate; the set flow rate range is the flow rate range corresponding to the high-efficiency section of the pump; If the output result of Step S6 is no, then execute Step S7: Update the flow rates of each pump in the state space with the regulated flow rates of each pump, and return to Step S4 until the output result of Step S6 is yes; If the output result of Step S6 is yes, then execute Step S8: Record the action combination with the highest efficiency corresponding to the state space as the best action combination corresponding to the water usage demand; Step S9: Use different water usage demands and the best action combinations corresponding to the water usage demands to construct a training data set; Step S10: Use the training data set to train and optimize the DQN model to obtain an optimized DQN model; the training process is as follows: (1) Initialize the evaluation network and initialize the target network; (2) Initialize the experience replay pool; (3) Initialize the environmental state; (4) Start the first episode; (5) Start iteration from step = 1; step is the time step; (6) The agent observes the environment and obtains the state ; the state i.e., the flow rate and head of each pump under the total flow rate of the demand (7) Take the state as the input of the evaluation network, output the action values of all actions in the current state, and then select and execute an action using the ε-greedy strategy ; The action is whether speed regulation or pump opening / closing is required at this flow head; The agent obtains a new state and an immediate reward ; the new state is the new flow head; the immediate reward i.e., whether it is in the high-efficiency zone, and if it is in the high-efficiency zone, the reward is large; (9) Store into the experience replay pool; (10) Randomly extract a batch of samples from the experience replay pool as the input of the target network and the evaluation network, and then output their Q values, that is, reset the new flow demand and the target flow; (11) Calculate the loss function based on the difference between the two Q values, and update the weights of the evaluation network through backpropagation; (12) Determine whether N steps have passed. If so, assign the weights of the evaluation network to the target network; (13) Determine whether M steps have passed. If so, increment the episode number by 1 and return to Step (3); if not, increment the step number by 1 and return to Step (5); (14) Determine whether the episode number is equal to max_episode. If so, end the training; Step S11: Input the actual water usage demand into the optimized DQN model to obtain the best action combination corresponding to the actual water usage demand; Step S12: Perform energy-saving scheduling on the pumping station water supply system according to the optimal action combination corresponding to the actual water demand; the energy-saving scheduling includes regulating the flow rates of the pumps in the pumping station water supply system according to the optimal action combination corresponding to the actual water demand.

2. The method for energy-saving scheduling of a pumping station water supply system according to claim 1, wherein, the state space includes the flow rates of the pumps in the pumping station water supply system and the head demand; the sum of the flow rates of the pumps in the pumping station water supply system is equal to the total flow demand.

3. The method for energy-saving scheduling of a pumping station water supply system according to claim 1, wherein, the pumping station water supply system includes a plurality of variable-frequency pumps and a plurality of industrial-frequency pumps; the control actions of the variable-frequency pumps include reducing the speed, keeping the speed unchanged, and increasing the speed; the control actions of the industrial-frequency pumps include starting and stopping.

4. An energy-saving scheduling system for a pumping station water supply system, wherein, the system includes: a water demand acquisition module for acquiring different water demands; the water demands include the total flow demand and the head demand; a model construction module for constructing a DQN model; a state space obtaining module for inputting the water demand into the DQN model for each water demand, and using the DQN model to randomly distribute the total flow demand to the pumps in the pumping station water supply system to obtain the state space corresponding to the water demand; an efficiency obtaining module for various action combinations for using the DQN model to obtain the efficiencies of various action combinations corresponding to the state space; the action combinations include control actions for the pumps in the pumping station water supply system; a pump flow regulation module for regulating the flow rates of the pumps in the pumping station water supply system according to the action combination with the highest efficiency corresponding to the state space to obtain the regulated flow rates of the pumps in the pumping station water supply system; a judgment module for judging whether each non-zero flow rate among the regulated flow rates of the pumps is within the set flow rate range of the pump corresponding to the non-zero flow rate; the set flow rate range is the flow rate range corresponding to the efficient section of the pump; a flow rate update module for, when the output result of the judgment module is no, updating the flow rates of the pumps in the state space with the regulated flow rates of the pumps, and returning to the efficiency obtaining module for various action combinations until the output result of the judgment module is yes; an optimal action combination recording module for, when the output result of the judgment module is yes, recording the action combination with the highest efficiency corresponding to the state space as the optimal action combination corresponding to the water demand; a training data set construction module for constructing a training data set by using different water demands and the optimal action combinations corresponding to the water demands; a model training and optimization module for training and optimizing the DQN model by using the training data set to obtain an optimized DQN model; the training process is as follows: (1) Initialize the evaluation network and initialize the target network; (2) Initialize the experience replay pool; (3) Initialize the environmental state; (4) Start the first round; (5) Start iteration from step = 1; step is the time step; (6) The agent observes the environment and obtains the state ; State i.e., the flow rate and head of each pump under the total flow rate of the demand; (7) Take the state as the input of the value network, output the action values of all actions in the current state, and then select and execute an action using the ε-greedy strategy ; The action is whether speed regulation or pump opening / closing is required at this flow head; The agent obtains a new state and an immediate reward ; the new state is the new flow head; the immediate reward i.e., whether it is in the high-efficiency zone, and there is a large reward if it is in the high-efficiency zone; (9) Store in the experience replay pool; (10) Randomly extract a batch of samples from the experience replay pool as the input of the target network and the evaluation network, and then output their respective Q values, that is, reset the new traffic demand and the target traffic; (11) Calculate the loss function based on the difference between the two Q values, and update the weights of the evaluation network through backpropagation; (12) Determine whether N steps have passed. If so, assign the weights of the evaluation network to the target network; (13) Determine whether M steps have passed. If so, increment the number of rounds by 1 and return to step (3); if not, increment the step number by 1 and return to step (5); (14) Determine whether the number of rounds is equal to max_episode. If so, end the training; The optimal action combination obtaining module is used to input the actual water demand into the optimized DQN model to obtain the optimal action combination corresponding to the actual water demand; The energy-saving scheduling module of the pumping station water supply system is used to perform energy-saving scheduling on the pumping station water supply system according to the optimal action combination corresponding to the actual water demand; the energy-saving scheduling includes regulating the flow rates of each pump in the pumping station water supply system according to the optimal action combination corresponding to the actual water demand.

5. The energy-saving scheduling system of the pumping station water supply system according to claim 4, characterized in that, The state space includes the flow rates of each pump in the pumping station water supply system and the head demand; the sum of the flow rates of each pump in the pumping station water supply system is equal to the total flow demand.

6. The energy-saving scheduling system of the pumping station water supply system according to claim 4, characterized in that, The pumping station water supply system includes a plurality of variable-frequency pumps and a plurality of industrial-frequency pumps; the control actions of the variable-frequency pumps include reducing the speed, keeping the speed unchanged, and increasing the speed; the control actions of the industrial-frequency pumps include starting and stopping.

7. An electronic device, characterized in that, It includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the energy-saving scheduling method of the pumping station water supply system according to any one of claims 1-3.

8. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by a processor, it implements the energy-saving scheduling method of the pumping station water supply system according to any one of claims 1-3.

Images