Intelligent energy-saving method for lifting pump house

Abstract

The invention belongs to the technical field of industrial automation and intelligent control, and particularly relates to an intelligent energy-saving method for a lifting pump room. According to the method, pressure, water flow, current and other sensors are deployed in a lifting pump room to collect data and the start-stop state of a pump set; based on a machine learning algorithm, an improved LSTM network is adopted to construct a water flow prediction model, and data in the past three years are obtained to predict the water flow in the next 24 hours. Secondly, aiming at mixed operation of a variable frequency pump and a power frequency pump, constructing a multi-target optimization scheduling model, considering energy consumption, start-stop cost and equipment balance, taking water flow and the minimum operation time of a single pump as constraints, and solving by using reinforcement learning to obtain the operation frequency and the number of pump sets; and finally, starting and stopping and frequency adjustment of the pump set are automatically controlled through the programmable logic controller. The method can solve the problems of adjustment lag, incapability of multi-pump collaborative optimization and the like in the traditional technology, and realizes the effects of saving energy, reducing consumption and prolonging the service life of equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial automation and intelligent control, and in particular relates to an intelligent energy-saving method for lifting a pump room. Background Art

[0002] The booster pump room is a core energy-consuming unit in a water plant, typically consuming 60% to 80% of the plant's total energy consumption. It is a key component in water plant energy conservation and optimization. The main reasons for this high energy consumption are: first, reliance on manual observation of pressure data and adjustment of pump frequency; second, the mixed operation of variable-frequency pumps and power-frequency pumps, without intelligent scheduling strategies, prevents the coordinated optimization of multiple pumps and the energy savings inherent in the booster pump room.

[0003] Traditional solutions to this problem rely on combining flow and pressure data to dynamically adjust control parameters. A pressure reducing valve is used to reduce the pressure in the high-pressure zone and supply it to the corresponding pressurized zone. The pressurized zone's pressure-maintaining shutdown function then shuts down the pump, saving energy.

[0004] However, current technologies all rely on real-time pressure feedback and fail to combine historical data for optimization, resulting in lag in regulation, inability to achieve multi-pump coordinated optimization, and low energy efficiency. Some control systems even cause water pumps to start and stop frequently when the water volume fluctuates greatly, affecting the life of the equipment. Summary of the Invention

[0005] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes an intelligent energy-saving method for lifting pump rooms.

[0006] In order to achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0007] S1. Data collection and transmission: Various sensors are deployed in the lift pump room to collect data and the start and stop status of the pump group;

[0008] S2. Water flow prediction model construction: Water flow prediction model based on machine learning algorithm;

[0009] S3. Intelligent Pump Group Scheduling: Develop a scheduling strategy for the mixed operation of variable-frequency pumps and power-frequency pumps, taking into account energy costs, start-stop costs, and equipment balance. Using water flow and the minimum operating time of a single pump as constraints, this strategy achieves multi-pump collaborative optimization and calculates the operating frequency and number of pump groups. Specific operations include building a multi-objective optimization scheduling model, establishing constraints, and solving strategies.

[0010] Multi-objective optimization scheduling model construction: Build a scheduling model with energy consumption cost, start-stop cost, and equipment balanced operation as optimization objectives;

[0011] Constraint construction: water flow and minimum single pump operating time are used as constraints to constrain the optimization objective;

[0012] Strategy solution: Reinforcement learning is used to solve the multi-objective optimization scheduling model in combination with constraints to obtain the final pump group operating frequency and number;

[0013] S4. Automatic control execution: Based on the scheduling results, the programmable logic controller automatically controls the start and stop of the pump group and the frequency adjustment.

[0014] Preferably, the sensors deployed in the lift pump room include pressure sensors, water flow sensors, and current sensors, which collect pipeline pressure data, water pump operating flow, and power consumption parameters in real time.

[0015] Preferably, the water flow prediction model in step S2 includes: data acquisition, model construction, model input, and model output;

[0016] Data acquisition: Obtain hourly water flow data and precipitation data for the past three years;

[0017] Model construction: Use improved LSTM network to build prediction model;

[0018] Model input: input the acquired water flow data and precipitation data;

[0019] Model output: predict the water flow in the next 24 hours and obtain the predicted value.

[0020] Preferably, in the model construction, the specific improvements of using the improved LSTM network to construct the prediction model are as follows:

[0021] Based on the input gate and forget gate of LSTM, the feature interaction gate G is introduced f , calculated as: G f =σ(W f ·[h t-1 ,x t ]+b f ), where σ is the sigmoid activation function, W f is the weight matrix, h t-1 Represents the hidden state of the previous moment, x t is the feature of the current input, b f represents the bias term;

[0022] Perform weighted fusion on the input features to obtain Where ⊙ represents the element-by-element product, and FFN is the feedforward network;

[0023] The enhanced features The improved LSTM unit mentioned above is input to replace the initial input for prediction to obtain the improved LSTM network.

[0024] Preferably, the specific operations of constructing the multi-objective optimization scheduling model in step S3 are:

[0025] First determine the frequency of M variable frequency pumps and N power frequency pumps. For the variable frequency pump, [f min ,f max ], the power frequency pump has a fixed frequency f fix , the operating state of each pump is S i ∈{0,1}, frequency is f i =[f min ,f max ]∪f fix ;

[0026] Introducing the start-stop cost matrix in represents the starting energy consumption and mechanical loss cost of the i-th pump, Cost of downtime;

[0027] Construct a three-dimensional optimization objective including energy consumption, start-stop cost and equipment balance: min(λ1L energy +λ2L switch +λ3L balance ), where L energy , L switch , L balance are energy consumption cost, start-stop cost and balanced operation respectively, λ1, λ2 and λ3 are the weight coefficients of the three;

[0028] Among them, energy consumption cost Among them, P i (f i (t)) is the frequency power function;

[0029] Start-Stop Cost Identify start and stop actions through 0-1 indication function;

[0030] Balanced operation Where T i (t) is the cumulative operating time of the i-th pump, is the average running time.

[0031] Preferably, the specific operation of constructing the constraint conditions in step S3 is:

[0032] Establish single pump flow formula q i =η i (f i )·q rated , where η i is the frequency efficiency function, q rated is the rated flow rate, and the constraint on the total flow rate is: where Q pred(t) is the water flow rate at the future time t predicted in step S2, and ε represents the allowable flow rate deviation threshold;

[0033] Constraint minimum running time constraint Through the state transfer matrix S(t) = [s1(t),...,s M+N (t)], if S i (t) = 0, and S i (t-1)=1, then the subsequent within a time step

[0034] Preferably, the specific steps of the strategy solution in step S3 include:

[0035] State space definition: Construct the state space based on the current pump group operation state vector, the accumulated operation time of each pump, and the future predicted water flow sequence;

[0036] Action space definition: The frequency adjustment action of each variable frequency pump and the start and stop action of the working pump constitute a joint action vector;

[0037] Reward function construction: Use the three-dimensional optimization objective as the reward function and embed the constraints into the policy space;

[0038] Reinforcement Learning Solution: Using deep deterministic policy gradients for model training and policy generation.

[0039] Preferably, the reinforcement learning solution process includes offline training and online fine-tuning:

[0040] Offline phase: Using three years of historical water flow and pump group operation data, the reinforcement learning model is trained by simulating scheduling scenarios under different working conditions until the reward function converges;

[0041] Online stage: Real-time sensor data is collected to update the state space, and a strategy solution is triggered every ten minutes to output the pump group operation frequency and number scheduling sequence for the next hour, and the weight coefficient is dynamically adjusted according to the actual operation feedback error.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are:

[0043] 1. Existing technologies rely on real-time pressure feedback, fail to incorporate historical data for optimization, and result in delayed regulation. This invention builds a water flow prediction model and employs an improved LSTM network to predict future hourly water flow. This model can predict water flow trends in advance, providing a forward-looking basis for decision-making in pump scheduling. This allows scheduling strategies to proactively respond to load changes and avoid energy waste caused by delayed regulation.

[0044] 2. Develop a scheduling strategy for the mixed operation of variable frequency pumps and industrial frequency pumps, build a multi-objective optimization scheduling model, consider energy consumption costs, start-stop costs, and equipment balance, use water flow and the minimum operating time of a single pump as constraints, achieve multi-pump collaborative optimization, and calculate the operating frequency and number of pump groups.

[0045] 3. By constructing constraints, the minimum operating time of a single pump is limited to reduce the frequent start and stop of the pump; at the same time, the balanced operation of the equipment is taken into consideration. By setting the balanced operation target, the cumulative operating time of each pump is averaged, which reduces equipment loss and extends equipment service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0047] Figure 1 A structural flow chart of an intelligent energy-saving method for lifting pump rooms; DETAILED DESCRIPTION

[0048] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] In industrial and urban water supply systems, the pump room is the core energy consumption unit, and its electricity consumption usually accounts for 60% to 80% of the total energy consumption of the water plant. The traditional operation mode relies on manual experience to adjust the frequency of the pump group, and there is a lack of intelligent coordination strategy when the variable frequency pump and the industrial frequency pump are mixed in operation, resulting in serious energy waste, frequent equipment start and stop, shortened life and other problems. The present invention proposes an intelligent energy-saving method for the pump room. The specific implementation process is as follows: Figure 1 shown.

[0051] First, a variety of sensors are deployed in the booster pump room to collect data and the start and stop status of the pump group. The sensors deployed in the booster pump room include pressure sensors, water flow sensors, and current sensors, which collect pipeline pressure data, water pump operating flow, and power consumption parameters in real time.

[0052] Taking into account the fact that the existing technology does not combine historical data for flow prediction, resulting in the pump group adjustment lagging behind actual demand and large water volume fluctuations, the present invention constructs a water flow prediction model based on a machine learning algorithm. The water flow prediction model includes: data acquisition, model construction, model input, and model output. Data acquisition: acquiring hourly water flow data and precipitation data for the past three years; model construction: using an improved LSTM network to build a prediction model; model input: inputting the acquired water flow data and precipitation data; model output: predicting the water flow for the next 24 hours to obtain a predicted value. Specifically, the specific improvement of using the improved LSTM network to construct the prediction model is to introduce a feature interaction gate G on the basis of the input gate and forget gate of the LSTM. f , calculated as: G f =σ(W f ·[h t-1 ,x t ]+b f ), where σ is the sigmoid activation function, W f is the weight matrix, h t-1 Represents the hidden state of the previous moment, x t is the feature of the current input, b f Represents the bias term; perform weighted fusion on the input features to obtain Where ⊙ represents the element-by-element product, FFN is the feedforward network; the enhanced features The improved LSTM unit described above is inputted instead of the initial input for prediction, resulting in an improved LSTM network. This predicts water flow trends in advance, providing a forward-looking decision-making basis for pump group scheduling. This enables the scheduling strategy to respond to load changes in advance, avoiding energy waste caused by delayed regulation.

[0053] Next, considering the lack of coordinated optimization of energy consumption, start-stop costs, and equipment balance when variable frequency pumps and industrial frequency pumps are mixed in operation, it leads to high energy consumption and short equipment life. This step performs intelligent scheduling of pump groups: formulates a scheduling strategy for the mixed operation of variable frequency pumps and industrial frequency pumps, considers energy consumption costs, start-stop costs, and equipment balance, and uses water flow and the minimum operating time of a single pump as constraints to achieve multi-pump coordinated optimization and calculate the operating frequency and number of pump groups; the specific operations include the construction of a multi-objective optimization scheduling model, the construction of constraint conditions, and the solution of strategies;

[0054] Multi-objective optimization scheduling model construction: Taking energy consumption cost, start-stop cost, and equipment balanced operation as optimization objectives, a scheduling model is constructed. Specifically, the frequencies of M variable frequency pumps and N power frequency pumps are first determined. For the variable frequency pump, [f min ,f max ], the power frequency pump has a fixed frequency f fix , the operating state of each pump is S i∈{0,1}, frequency is f i =[f min ,f max ]∪f fix ;Introduce the start-stop cost matrix in represents the starting energy consumption and mechanical loss cost of the i-th pump, is the downtime cost; construct a three-dimensional optimization objective including energy consumption, start-stop cost and equipment balance: min(λ1L energy +λ2L switch +λ3L balance ), where L energy , L switch , L balance are energy consumption cost, start-stop cost and balanced operation, λ1, λ2, λ3 are the weight coefficients of the three; Among them, P i (f i (t)) is the frequency power function; the start-stop cost Identify start and stop actions through 0-1 indication function; balanced operation Where T i (t) is the cumulative operating time of the i-th pump, is the average running time.

[0055] Constraint construction: The water flow rate and the minimum operating time of a single pump are used as constraints to constrain the optimization objective. Specifically, the single pump flow formula q is established. i =η i (f i )·q rated , where η i is the frequency efficiency function, q rated is the rated flow rate, and the constraint on the total flow rate is: where Q pred (t) is the water flow at time t predicted in step S2, ε represents the allowable flow deviation threshold; the minimum running time constraint Through the state transfer matrix S(t) = [s1(t),...,s M+N (t)], if S i (t) = 0, and S i (t-1)=1, then the subsequent within a time step

[0056] Policy Solution: Reinforcement learning is used to solve the multi-objective optimization scheduling model, combined with constraints, to determine the final pump operating frequency and number. The steps are: state space definition: constructing a state space based on the current pump operating state vector, the accumulated operating time of each pump, and the future predicted water flow sequence; action space definition: constructing a joint action vector for each variable frequency pump's frequency adjustment action and the start / stop action of the working pump; reward function construction: using the three-dimensional optimization objective as the reward function and embedding the constraints into the policy space; and reinforcement learning solution: employing deep deterministic policy gradient descent. Specifically, in the policy solution of the multi-objective optimization scheduling model, reinforcement learning is used in combination with constraints to determine the final pump operating frequency and number. The state space definition constructs a space that comprehensively reflects the system's operating status. The current pump operating state vector specifies whether each pump is on or off, the accumulated operating time of each pump is used to assess equipment utilization, and the future predicted water flow sequence is derived from the water flow prediction model, providing forward-looking information for pump scheduling. These three elements together constitute the state space, providing the basis for intelligent decision-making.

[0057] The action space is defined as a set of all possible actions. These actions are combined into a joint action vector, encompassing all possible scheduling operations. The reward function is constructed to transform the three-dimensional optimization objective into a quantitative assessment of the policy's performance. Energy cost, start / stop costs, and equipment balance are the core elements of the reward function, enabling the system to minimize pump starts and stops and balance equipment usage while pursuing low energy consumption. Furthermore, flow matching constraints and minimum run time constraints are embedded in the reward function. Negative rewards are applied if these constraints are violated, guiding the agent to learn a legal and optimal policy. Reinforcement learning solution: Deep deterministic policy gradients are used for model training and policy generation. This involves both offline training and online fine-tuning. In the offline phase, the reinforcement learning model is trained using three years of historical water flow and pump operation data until the reward function converges. This convergence is achieved by achieving a maximum number of iterations. In the online phase, sensor data is collected in real time to update the state space. A policy solution is triggered every ten minutes, outputting a schedule for pump operation frequency and number for the next hour. Weight coefficients are dynamically adjusted based on actual operational feedback errors.

[0058] Finally, to address the issues of traditional manual operation, which often result in delayed adjustments and inaccurate execution of optimization strategies, automated control was implemented. Based on the scheduling results, a programmable logic controller (PLC) automatically controls the pump start / stop and frequency adjustment. Specifically, a Siemens S7-1200 PLC was selected as the core controller, communicating with the inverter and soft starter via the Modbus protocol, supporting real-time data acquisition and command issuance.

[0059] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any person skilled in the art may utilize the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes for application in other fields. However, any simple modification, equivalent change, and modification of the above embodiments made in accordance with the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An intelligent energy-saving method for lifting pump rooms, characterized in that: The following steps are involved: S1. Data collection and transmission: Various sensors are deployed in the lift pump room to collect data and the start and stop status of the pump group; S2. Water flow prediction model construction: Water flow prediction model based on machine learning algorithm; S3. Intelligent Pump Group Scheduling: Develop a scheduling strategy for the mixed operation of variable-frequency pumps and power-frequency pumps, taking into account energy costs, start-stop costs, and equipment balance. Using water flow and the minimum operating time of a single pump as constraints, this strategy achieves multi-pump collaborative optimization and calculates the operating frequency and number of pump groups. Specific operations include building a multi-objective optimization scheduling model, establishing constraints, and solving strategies. Multi-objective optimization scheduling model construction: Build a scheduling model with energy consumption cost, start-stop cost, and equipment balanced operation as optimization objectives; Constraint construction: water flow and minimum single pump operating time are used as constraints to constrain the optimization objective; Strategy solution: Reinforcement learning is used to solve the multi-objective optimization scheduling model in combination with constraints to obtain the final pump group operating frequency and number; S4. Automatic control execution: Based on the scheduling results, the programmable logic controller automatically controls the start and stop of the pump group and the frequency adjustment.

2. The intelligent energy-saving method for lifting pump room according to claim 1, characterized in that: The sensors deployed in the lift pump room include pressure sensors, water flow sensors, and current sensors, which collect pipeline pressure data, water pump operating flow, and power consumption parameters in real time.

3. The intelligent energy-saving method for lifting pump room according to claim 1, characterized in that: The water flow prediction model in step S2 includes: data acquisition, model construction, model input, and model output; Data acquisition: Obtain hourly water flow data and precipitation data for the past three years; Model construction: Use improved LSTM network to build prediction model; Model input: input the acquired water flow data and precipitation data; Model output: predict the water flow in the next 24 hours and obtain the predicted value.

4. The intelligent energy-saving method for lifting pump room according to claim 3, characterized in that: In the model construction, the specific improvements of using the improved LSTM network to build the prediction model are as follows: Based on the input gate and forget gate of LSTM, the feature interaction gate G is introduced f , calculated as: G f =σ(W f ·[h t-1 ,x t ]+b f ), where σ is the sigmoid activation function, W f is the weight matrix, h t-1 Represents the hidden state of the previous moment, x t is the feature of the current input, b f represents the bias term; Perform weighted fusion on the input features to obtain Where ⊙ represents the element-by-element product, and FFN is the feedforward network; The enhanced features The improved LSTM unit mentioned above is input to replace the initial input for prediction to obtain the improved LSTM network.

5. The intelligent energy-saving method for lifting pump room according to claim 1, characterized in that: The specific operations of constructing the multi-objective optimization scheduling model in step S3 are: First determine the frequency of M variable frequency pumps and N power frequency pumps. For the variable frequency pump, [f min ,f max ], the power frequency pump has a fixed frequency f fix , the operating state of each pump is S i ∈{0,1}, frequency is f i =[f min ,f max ]∪f fix ; Introducing the start-stop cost matrix in represents the starting energy consumption and mechanical loss cost of the i-th pump, Cost of downtime; Construct a three-dimensional optimization objective including energy consumption, start-stop cost and equipment balance: min(λ1L energy +λ2L switch +λ3L balance ), where L energy , L switch , L balance are energy consumption cost, start-stop cost and balanced operation respectively, λ1, λ2 and λ3 are the weight coefficients of the three; Among them, energy consumption cost Among them, P i (f i (t)) is the frequency power function; Start-Stop Cost Identify start and stop actions through 0-1 indication function; Balanced operation Where T i (t) is the cumulative operating time of the i-th pump, is the average running time.

6. The intelligent energy-saving method for lifting pump room according to claim 1, characterized in that: The specific operations of constructing the constraint conditions in step S3 are: Establish single pump flow formula q i =η i (f i )·q rated , where η i is the frequency efficiency function, q rated is the rated flow rate, and the constraint on the total flow rate is: where Q pred (t) is the water flow rate at the future time t predicted in step S2, and ε represents the allowable flow rate deviation threshold; Constraint minimum running time constraint Through the state transfer matrix S(t) = [s1(t),...,s M+N (t)], if S i (t) = 0, and S i (t-1)=1, then the subsequent within a time step 7. The intelligent energy-saving method for lifting pump room according to claim 1, characterized in that: The specific steps of the S3 step strategy solution include: State space definition: Construct the state space based on the current pump group operation state vector, the accumulated operation time of each pump, and the future predicted water flow sequence; Action space definition: The frequency adjustment action of each variable frequency pump and the start and stop action of the working pump constitute a joint action vector; Reward function construction: Use the three-dimensional optimization objective as the reward function and embed the constraints into the policy space; Reinforcement Learning Solution: Using deep deterministic policy gradients for model training and policy generation.

8. The intelligent energy-saving method for lifting pump room according to claim 7, characterized in that: The reinforcement learning solution process includes offline training and online fine-tuning: Offline phase: Using three years of historical water flow and pump group operation data, the reinforcement learning model is trained by simulating scheduling scenarios under different working conditions until the reward function converges; Online stage: Real-time sensor data is collected to update the state space, and a strategy solution is triggered every ten minutes to output the pump group operation frequency and number scheduling sequence for the next hour, and the weight coefficient is dynamically adjusted according to the actual operation feedback error.

Images