An Optimization Method for Screw Compressor Unit Operation Based on a Hierarchical Predictive Control Model

Abstract

This invention belongs to the field of information technology and discloses a method for optimizing the operation of screw compressor units based on a hierarchical predictive control model. At the upper level, a least-squares support vector machine is used to predict the compressor unit load. Then, combined with the current system output, a rolling optimization method is applied to optimize the economic performance indicators of the compressor unit and generate setpoints for the lower-level control loop. At the lower level, for screw compressors of different sizes, a distributed model predictive controller and a proportional-integral-derivative controller are used in series to track the setpoints issued from the upper level, enabling the compressor unit to operate in an economically optimized state and achieving closed-loop optimization of the compressor unit. This invention can accurately and quickly track and optimize setpoints at the control layer, enabling joint control of compressor units of different sizes, and providing an effective solution for the economical operation of screw compressor units.

Description

Technical Field

[0001] This invention belongs to the field of information technology and involves modules such as data-driven air compressor modeling, distributed control of compressor units, and real-time optimization solutions. It is an optimization control method for screw compressor units based on two-level hierarchical model predictive control. This invention uses real-time load samples from the industrial field to update the predictive model training set, then uses least squares support vector machines to predict compressed air load online, and applies a two-stage optimization method to perform rolling optimization of the production plan for the next time step. Finally, it uses distributed model predictive control and proportional-integral-derivative (PID) controller cascade control to perform setpoint tracking control for each operating compressor with optimization as the goal, thereby realizing a hierarchical optimization control scheme from system-level real-time optimization to equipment-level dynamic tracking control. This method also has wide application value in various industrial fields. Background Technology

[0002] Compressed air is an expensive but clean energy source converted from electricity. Due to the relatively low cost of its main components and the ease of maintenance, compressed air systems are increasingly widely used in industrial production. However, compressed air systems are also major energy-consuming units in industries such as steel, pharmaceuticals, and chemicals. This is primarily due to factors such as improper compressor unit scheduling and untimely setpoint control, leading to high energy consumption and low efficiency. Therefore, it is crucial to optimize the scheduling of compressed air systems and implement dynamic tracking control for optimization purposes, while simultaneously improving the system's economy and control performance to ensure the feasibility of the optimized scheduling strategy.

[0003] Common approaches to improving the economic performance of compressed air systems can be broadly categorized into three approaches: First, from the perspective of optimizing the objective function, using mixed-integer linear programming (Adamson R, Hobbs M, Silcock A, et al. (2017) Steady-state optimization of a multiple cryogenic air separation unit and compressor plant. Applied Energy, Elsevier Ltd, 189: 221–232.) to solve the problem and reduce the operating cost of the compressor unit. Second, from the perspective of the prediction model designed in the optimization scheduling problem (Fu X, Kong D, Song R, et al. (2012) A dispatch method of air compressors based on forecasting consumption[J]. IEEE International Conference on Industrial Informatics (INDIN), IEEE, 218–222.) using neural networks to predict the load demand of the compressor unit, providing strong support for the optimal scheduling of the compressor unit. Third, from the perspective of the compressor's operating characteristics, using the modified adaptive method (Milosavljevic P, Marchetti P) for uncertain efficiency maps. AG, Cortinovis A, et al. (2020) Real-time optimization of load sharing for gas compressors in the presence of uncertainty. Applied Energy, 272.) This paper combines compressor optimization with compressor efficiency maps to achieve optimization of compressor load sharing.

[0004] The above solutions for improving the economic performance of compressed air systems only achieve economic optimization of the compressor unit, but do not consider the feasibility of the optimization scheme at the equipment level. Therefore, it is necessary to establish a real-time optimization control method that can simultaneously achieve economic optimization and fast and accurate tracking control performance. Summary of the Invention

[0005] The problem this invention aims to solve is the real-time optimization control of screw compressed air systems. To address this issue, this invention first uses a least-squares support vector machine to predict the compressor unit load at the upper level. Then, combining this with the current system output, a rolling optimization method is applied to optimize the economic performance indicators of the compressor unit and generate setpoints for the lower-level control loop. At the lower level, for screw compressors of different sizes, a distributed model predictive controller and a proportional-integral-derivative controller are used in series to track the setpoints issued from the upper level, enabling the compressor unit to operate in an economically optimized state and achieving closed-loop optimization of the compressor unit. This invention can accurately and quickly track and optimize setpoints at the control layer, enabling joint control of compressor units of different sizes, and providing an effective solution for the economical operation of screw compressor units.

[0006] Technical solution of the present invention:

[0007] An optimized control method for screw compressor units based on two-level hierarchical model predictive control is shown in the appendix. Figure 1 As shown, the specific steps are as follows:

[0008] (1) Construction of compressor dynamic model: obtain historical operating data of each compressor in the air compressor station from the real-time database of the industrial site; use system identification method to identify the transfer function of each compressor;

[0009] (2) Construction of compressed air load prediction model: Least squares support vector machine is used as the prediction model for compressed air load to realize online prediction of compressed air load;

[0010] (3) Rolling optimization of the operating conditions of each compressor: The rolling optimization of the operating conditions of each compressor is divided into two stages. First, the operating conditions of N are calculated based on the predicted load. p1 The operating strategy that minimizes the number of start-ups and shutdowns of each unit within a given time step is then used to optimize the load allocation for each started unit under this strategy.

[0011] (4) Distributed controller model: Independent controllers are designed for compressor units of different types and specifications. The control scheme adopts a model predictive control and proportional integral derivative control cascade control structure. In the bottom secondary loop, a conventional proportional integral derivative controller is used to suppress minor disturbances in the system. The main loop adopts a model predictive controller to achieve good tracking and ensure good system robustness, thereby achieving control for optimization purposes.

[0012] (5) Gas tank model: The pressure fluctuation equation of the gas tank is derived by using the van der Waals equation, and then the pressure of the gas tank is obtained, and the current operating condition of the compressor is determined.

[0013] (6) Rolling optimization: After the compressor unit completes one optimization control action, it feeds back the current load and gas tank pressure to the real-time optimization layer. The real-time optimization layer adjusts the production plan for the next moment in a timely manner based on the predicted output and the feedback at the current moment.

[0014] The effects and benefits of this invention: This invention enables real-time optimized control of screw compressor units. While maintaining supply and demand balance at the system level based on user-side demand, it achieves steady-state economic optimization of the compressor unit and executes real-time tracking control of the optimized scheduling scheme at the equipment level. This allows the industrial system to handle constraints online, and finally returns the optimized control results to the optimization layer, realizing closed-loop optimized control. This solves the control problems caused by information mismatch between layers, fully reflects the plant's benefits in the control objectives, greatly improves the enterprise's economic efficiency, and reduces the energy consumption of the compressed air system. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a specific implementation of the present invention.

[0016] Figure 2 This describes the configuration of the screw-type compressed air system of the present invention.

[0017] Figure 3(a) shows the simulation results of distributed model predictive control for a 20 cubic meter compressor.

[0018] Figure 3(b) shows the control operation diagram of the 20 cubic meter compressor.

[0019] Figure 4(a) shows the simulation results of distributed model predictive control for a 40 cubic meter compressor.

[0020] Figure 4(b) shows the control operation diagram of the 40 cubic meter compressor.

[0021] Figure 5(a) shows the simulation results of distributed model predictive control for a 60 cubic meter compressor.

[0022] Figure 5(b) shows the control operation diagram of the 60 cubic meter compressor. Detailed Implementation

[0023] To better understand the technical solution of this invention, this invention takes a screw compressor system from a blower factory as an example, combined with the appendix... Figure 2 The embodiments of the present invention will be described in detail.

[0024] An optimized control method for a screw compressor unit based on two-level hierarchical model predictive control is proposed, comprising the following steps:

[0025] Step 1: Construct a dynamic model of the compressor

[0026] Historical operating data of each compressor in the air compressor station is obtained from the industrial real-time database, including compressor discharge volume, discharge temperature, discharge pressure, speed and current data. Then, the transfer function of each compressor is obtained by using the system identification method. The transfer function takes speed as input and discharge volume and energy consumption as output.

[0027] Step 2: Construct a compressed air load prediction model

[0028] Historical load data of air compressor stations obtained from the industrial real-time database is used as the training set for a least squares support vector machine (LSVM). A LVM-based compressed air load prediction model is then trained. The latest load data samples from the industrial real-time database are added to the last item of the training set. The compressed air load prediction model is then used to predict the future N loads. p1 The load y(k+1:k+N) at each time step p1 |k), where k+1:k+N p1 |k represents the sequence of operations from k+1 to k+N at time k. p1 Predicting the timing;

[0029] Step 3: Continuously optimize the operating conditions of each compressor

[0030] Divided into two stages

[0031] 3.1 Based on the predicted load from step 2 and the type and specifications of each compressor, list all operating schemes that satisfy the predicted load, and use the grid search method to calculate the load at N. p1 The optimal operating strategy is to select the combination of all start-stop schemes with the fewest start-stop times within a given time step.

[0032]

[0033] Where m is the number of compressors; Let i represent the start / stop state of the i-th compressor at time k+1.

[0034] 3.2 Applying the rolling optimization approach, the optimal operating strategy is used to start the units at time k+1. The specific steps for optimizing load sharing are as follows:

[0035] a) Objective function

[0036]

[0037] Among them, speed i G represents the rotational speed of the i-th compressor; Power,i Let J represent the transfer function between the speed and energy consumption of the i-th compressor; J represents the operating energy consumption of the air compressor station.

[0038] b) Constraints

[0039] F imin <F i ≤F imax (i=1,2,3) (2)

[0040] Among them F i =G Flow,i (speed i ), G Flow,i F represents the transfer function between the speed and displacement of the i-th compressor; imin and F imax Let represent the upper and lower limits of the discharge capacity of the i-th compressor, respectively;

[0041] F imin =G Flow,i (G FC (f imin (3)

[0042] F imax =G Flow,i (G FC (f imax (4)

[0043] Among them, f imin and f imax G represents the lower and upper frequency limits that the inverter of the i-th compressor can reach due to physical time constraints, respectively; FC This is the transfer function between the frequency and speed of the frequency converter;

[0044] c) Optimization solution

[0045] The objective function is iteratively solved using the particle swarm optimization method. Iteration stops when the deviation between two adjacent generations is within a specified range, thus obtaining the optimal load allocation F for m compressors. i ref (i = 1, ..., m), which are used as the setpoints for the distributed controller model;

[0046] Step 4: Distributed Controller Model

[0047] The distributed controller model adopts a model predictive control and proportional-integral-derivative (PID) cascade control structure to track the setpoint in step 3 to achieve an economically optimized state; the PID controller is used in the bottom secondary loop of the distributed controller model, and the model predictive controller is used in the main loop of the distributed controller model.

[0048] Distributed control is performed within a 1-second sampling interval. The reference input optimal load distribution F of the model predictive controller for the main circuit of the i-th compressor is... i ref(i = 1, ..., m), the controlled object is a generalized controlled object composed of the secondary loop and the compressor unit, and the control scheme of the i-th compressor is implemented according to the following steps;

[0049] 4.1 Constructing the frequency converter model

[0050] Based on the working principle of the frequency converter, a frequency converter model with input frequency and output motor speed is established, and the transfer function is as follows:

[0051]

[0052] Among them, y speed (s) Laplace transform of the inverter output speed; u F (s) is the Laplace transform of the inverter input frequency;

[0053] 4.2 Constructing a Generalized Control Object Model

[0054] The proportional-integral-derivative (PID) frequency converter acts on the compressor to obtain the compressor's speed input and flow output data. The obtained data is then identified to obtain the transfer function of the generalized control object composed of the secondary loop and the compressor unit.

[0055]

[0056] Among them, y flow (s) is the Laplace transform of the compressor output flow rate; u speed (s) is the Laplace transform of the compressor input speed;

[0057] 4.3 Predictive Model of Model Predictive Controller

[0058] The model predictive controller models in the N² time domain and predicts in the N time domain. p2 The control time domain is N c2 According to the proportional superposition property of linear time-invariant systems, the output of the generalized controlled object is:

[0059]

[0060] Where, This represents a known quantity determined by past control actions; ΔU(t) = [Δu i (t|t)…Δu i (t+N c2 -1|t)] T Represents the future N c2 The control increment of each demand solution; Represents the future N p2 The output of the prediction model at each step;

[0061] It is a dynamic matrix composed of the step response coefficients of the generalized controlled object;

[0062] 4.4 Rolling Optimization of Model Predictive Controller

[0063] The rolling optimization objective function for the i-th compressor is:

[0064]

[0065] The constraints are

[0066]

[0067] Among them, u i (t+q) represents the actual control quantity from time t to time q; Δu i (t+q) is the control increment from time t to time q; u i (t+q)=u i (t-1+q)+Δu i (t+q), u i (t-1+q) is the actual control quantity from time t-1 to time q; u imin and u imax These are the minimum and maximum speeds of the compressor motor, respectively; Δu imin and Δu imax These are the minimum and maximum adjustable step sizes of the compressor motor, respectively. It is the load forecast from time t to time q; y imin and y imax These are the lower and upper limits of the compressor's displacement, respectively; q ip and r iq It is the weight matrix; F iref (t+p) is the p-th given expected value starting from time t; J represents the predicted flow rate at time t for time t+p; i (t) represents the i-th compressor's effect on the future N at time t. p2 The deviation between the predicted output and the given expected value at each time step;

[0068] 4.5 Feedback Correction of Model Predictive Controller

[0069] There is a discrepancy between the output of the predictive model of the model predictive controller and the output of the generalized controlled object;

[0070]

[0071] Where e(t+1) is the output error at time t+1; The exhaust output of the generalized controlled object at time t+1; The output of the model predictor controller at time t+1;

[0072] The obtained output error is used to correct the prediction of future output.

[0073]

[0074] in, From time t+1 to time t+N p2 Correction values ​​for load forecasts at any given time; For time t, the time from time t+1 to time t+N p2 Load forecast at time t; h is the error correction matrix;

[0075] 4.6 Solving for the optimal control quantity

[0076] Iteratively solve for the optimal control quantity Δu of the i-th compressor at the current moment. i and control quantity u i (t)=u i (t-1)+Δu i It applies to objects under generalized control.

[0077] Step 5: Gas storage tank model

[0078] According to the van der Waals equation, which is an improvement on the ideal gas law.

[0079]

[0080] In the formula, p is the pressure of the gas storage tank, R is the gas constant; T is the absolute temperature of the gas; a and b are van der Waals constants; V m It is the molar volume of the gas;

[0081] Through derivation, the relationship between the pressure change inside the gas storage tank and the amount of compressed air generated and consumed can be written as follows:

[0082]

[0083] In the formula, Δp is the pressure change; m i It is the amount of compressed air generated; m o V is the amount of compressed air consumed; V is the effective volume of the air tank; M is the molar mass of the gas. It is the mass flow rate of the gas inside the gas storage tank;

[0084] Step 6: Scrolling Optimization

[0085] After the compressor unit completes one optimized control action, it will display the current load y. real (k+1) and the pressure p of the gas storage tank tank(k+1) are fed back to the prediction model in step 2 and the optimization layer in step 3, respectively. The production plan for the next moment is adjusted in combination with the prediction output to realize the real-time optimization of the compressor unit and the tracking control of the set value.

[0086] Taking the compressed air system of a blower factory as an example, assuming ideal transmission of compressed air in the pipeline, and that the load of the compressed air system changes continuously according to user demand, without considering differences in electricity prices at different times (i.e., the electricity price is calculated at 0.458 yuan / kWh), Table 1 shows a comparison of the results of the method of the present invention and the manual dispatching method. Figures 3, 4, and 5 respectively show the decentralized control results of the present invention for 20 cubic meter, 40 cubic meter, and 60 cubic meter compressors.

[0087] Table 1 shows a comparison of the effectiveness of the method of the present invention and the manual scheduling method.

[0088]

[0089]

[0090]

Claims

1. An optimized control method for a screw compressor unit based on two-level hierarchical model predictive control, characterized in that, The specific steps are as follows: Step 1: Construct a dynamic model of the compressor Historical operating data of each compressor in the air compressor station is obtained from the industrial real-time database, including compressor discharge volume, discharge temperature, discharge pressure, speed and current data. Then, the transfer function of each compressor is obtained by using the system identification method. The transfer function takes speed as input and discharge volume and energy consumption as output. Step 2: Construct a compressed air load prediction model Historical load data of air compressor stations obtained from the industrial real-time database is used as the training set for a least squares support vector machine (LSVM). A LVM-based compressed air load prediction model is then trained. The latest load data samples from the industrial real-time database are added to the last item of the training set. The compressed air load prediction model is then used to predict the future N loads. p1 The load y(k+1:k+N) at each time step p1 |k), where k+1:k+N p1 |k represents the sequence of operations from k+1 to k+N at time k. p1 Predicting the timing; Step 3: Continuously optimize the operating conditions of each compressor Divided into two stages 3.1 Based on the predicted load from step 2 and the type and specifications of each compressor, list all operating schemes that satisfy the predicted load, and use the grid search method to calculate the load at N. p1 The optimal operating strategy is to select the combination of all start-stop schemes with the fewest start-stop times within a given time step. Where m is the number of compressors; Let i represent the start / stop state of the i-th compressor at time k+1. 3.2 Applying the rolling optimization approach, the optimal operating strategy is used to start the units at time k+1. The specific steps for optimizing load sharing are as follows: a) Objective function Among them, speed i G represents the rotational speed of the i-th compressor; Power,i Let J represent the transfer function between the speed and energy consumption of the i-th compressor; J represents the operating energy consumption of the air compressor station. b) Constraints F imin <F i ≤F imax (i=1,2,3) (2) Among them F i =G Flow,i (speed i ), G Flow,i F represents the transfer function between the speed and displacement of the i-th compressor; imin and F imax Let represent the upper and lower limits of the discharge capacity of the i-th compressor, respectively; F imin =G Flow,i (G FC (f imin )) (3) F imax =G Flow,i (G FC (f imax )) (4) Among them, f imin and f imax G represents the lower and upper frequency limits that the inverter of the i-th compressor can reach due to physical time constraints, respectively; FC This is the transfer function between the frequency and speed of the frequency converter; c) Optimization solution The objective function is iteratively solved using the particle swarm optimization method. Iteration stops when the deviation between two adjacent generations is within a specified range, thus obtaining the optimal load allocation F for m compressors. i ref (i = 1, ..., m), which are used as the setpoints for the distributed controller model; Step 4: Distributed Controller Model The distributed controller model adopts a model predictive control and proportional-integral-derivative (PID) cascade control structure to track the setpoint in step 3 to achieve an economically optimized state; the PID controller is used in the bottom secondary loop of the distributed controller model, and the model predictive controller is used in the main loop of the distributed controller model. Distributed control is performed within a 1-second sampling interval. The reference input optimal load distribution F of the model predictive controller for the main circuit of the i-th compressor is... i ref (i = 1, ..., m), the controlled object is a generalized controlled object composed of the secondary loop and the compressor unit, and the control scheme of the i-th compressor is implemented according to the following steps; 4.1 Constructing the frequency converter model Based on the working principle of the frequency converter, a frequency converter model with input frequency and output motor speed is established, and the transfer function is as follows: Among them, y speed (s) Laplace transform of the inverter output speed; u F (s) is the Laplace transform of the inverter input frequency; 4.2 Constructing a Generalized Control Object Model The proportional-integral-derivative (PID) frequency converter acts on the compressor to obtain the compressor's speed input and flow output data. The obtained data is then identified to obtain the transfer function of the generalized control object composed of the secondary loop and the compressor unit. Among them, y flow (s) is the Laplace transform of the compressor output flow rate; u speed (s) is the Laplace transform of the compressor input speed; 4.3 Predictive Model of Model Predictive Controller The model predictive controller models in the N² time domain and predicts in the N time domain. p2 The control time domain is N c2 According to the proportional superposition property of linear time-invariant systems, the output of the generalized controlled object is: In the formula, This represents a known quantity determined by past control actions; ΔU(t) = [Δu i (t|t)…Δu i (t+N c2 -1|t)] T Represents the future N c2 The control increment of each demand solution; Represents the future N p2 The output of the prediction model at each step; It is a dynamic matrix composed of the step response coefficients of the generalized controlled object; 4.4 Rolling Optimization of Model Predictive Controller The rolling optimization objective function for the i-th compressor is: The constraints are Among them, u i (t+q) represents the actual control quantity from time t to time q; Δu i (t+q) is the control increment from time t to time q; u i (t+q)=u i (t-1+q)+Δu i (t+q), u i (t-1+q) is the actual control quantity from time t-1 to time q; u imin and u imax These are the minimum and maximum speeds of the compressor motor, respectively; Δu imin and Δu imax These are the minimum and maximum adjustable step sizes of the compressor motor, respectively. It is the load forecast from time t to time q; y imin and y imax These are the lower and upper limits of the compressor's displacement, respectively; q ip and r iq It is the weight matrix; F iref (t+p) is the p-th given expected value starting from time t; J represents the predicted flow rate at time t for time t+p; i (t) represents the i-th compressor's effect on the future N at time t. p2 The deviation between the predicted output and the given expected value at each time step; 4.5 Feedback Correction of Model Predictive Controller There is a discrepancy between the output of the predictive model of the model predictive controller and the output of the generalized controlled object; Where e(t+1) is the output error at time t+1; The exhaust output of the generalized controlled object at time t+1; The output of the model predictor controller at time t+1; The obtained output error is used to correct the prediction of future output. in, From time t+1 to time t+N p2 Correction values ​​for load forecasts at any given time; For time t, the time from time t+1 to time t+N p2 Load forecast at time t; h is the error correction matrix; 4.6 Solving for the optimal control quantity Iteratively solve for the optimal control quantity Δu of the i-th compressor at the current moment. i and control quantity u i (t)=u i (t-1)+Δu i It applies to objects under generalized control. Step 5: Gas storage tank model According to the van der Waals equation, which is an improvement on the ideal gas law. In the formula, p is the pressure of the gas storage tank, R is the gas constant; T is the absolute temperature of the gas; a and b are van der Waals constants; V m It is the molar volume of the gas; Through derivation, the relationship between the pressure change inside the gas storage tank and the amount of compressed air generated and consumed can be written as follows: In the formula, Δp is the pressure change; m i It is the amount of compressed air generated; m o V is the amount of compressed air consumed; V is the effective volume of the air tank; M is the molar mass of the gas. It is the mass flow rate of the gas inside the gas storage tank; Step 6: Scrolling Optimization After the compressor unit completes one optimized control action, it will display the current load y. real (k+1) and the pressure p of the gas storage tank tank (k+1) are fed back to the prediction model in step 2 and the optimization layer in step 3, respectively. The production plan for the next moment is adjusted in combination with the prediction output to realize the real-time optimization of the compressor unit and the tracking control of the set value.

Images