Valve system control parameter optimization method based on differential evolution algorithm

Abstract

The invention discloses a valve system control parameter optimization method based on a differential evolution algorithm, and mainly solves the problem that related parameters cannot be quickly obtained to enable the control performance of a valve system to be optimal in the prior art under the complex conditions of multiple working conditions. Comprising the following steps: 1) collecting and preprocessing data of a valve system under different working conditions, and constructing a working condition data set of the valve system; 2) determining initial parameters, and generating an initial population by using Latin hypercube sampling; 3) considering a plurality of optimization targets and constraint conditions of the valve control system, establishing a control parameter evaluation model, and adopting a random evaluation strategy to ensure the randomness and richness of evaluation working condition combination; and 4) optimizing the control parameters of the valve system by using a differential evolution algorithm, and outputting the control parameters of the valve system represented by the optimal individual when a termination condition is reached. According to the method, valve system control parameters which are high in stability and have a certain anti-interference capability on disturbance can be obtained, and the optimization time of the parameters is effectively shortened.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and further relates to valve system control technology. Specifically, it is an optimization method for control parameters of a valve system based on a differential evolution algorithm, which can be used for optimizing control parameters of a valve system. Background Art

[0002] In the optimization problems of industrial automation and control systems, valve control, as a key link, directly affects the stability and efficiency of the production process. Valve systems usually need to operate under changing operating conditions and environments, switch between different working conditions, and adjust control parameters. Factors such as measurement errors, equipment aging, and sensor data loss cause changes in the dynamic characteristics of the system, and the optimization algorithm needs to be able to handle these uncertainties. At the same time, the adjustment of the valve system has hysteresis, which means that in the process of valve control, the response of the controlled variable to the control signal has a delay or does not fully follow. During this adjustment process, the system is unstable and is likely to have an adverse impact on production. Considering the working scenario and hysteresis of the valve control system, in the face of working condition switching and multiple working condition combinations, it is required that the adjustment of control parameters be as small as possible to avoid the impact of large changes on the system, and at the same time have a certain anti-interference ability against small disturbances.

[0003] In current enterprise practices, the traditional PID controller is a commonly used control method in valve control systems. This method is simple and easy to implement, but the parameter adjustment process depends on experience and trial and error, lacking a general optimization method. For complex systems, parameter adjustment is particularly difficult. In the face of complex multi-objective and multi-constraint optimization problem scenarios, it cannot provide the optimal control performance, and the control parameter setting method only considers the optimum for a single working condition and does not consider the stability under working condition switching and multiple working condition combinations.

[0004] In the published literature with the patent number CN202010828608.9, a parameter tuning method, device, storage medium, and parameter tuning unit are provided, including creating a model library for the controlled object under different working conditions, where each model in the model library corresponds to one working condition; selecting the optimal model from the model library that matches the current data of the coordinated control system; generating the optimal control parameters of the predictive controller according to historical data, the optimal model, and the improved differential evolution algorithm. However, this solution is for the matching of a single working condition and does not consider the randomness and richness of the working condition combinations of the valve system. Summary of the Invention

[0005] The object of the present invention is to propose an optimization method for control parameters of a valve system based on the differential evolution algorithm in view of the deficiencies of the above-mentioned prior art, so as to solve the problem that the valve control parameters cannot provide optimal control performance in the face of working condition switching and multi-working condition combinations. The present invention fully considers the working scenario and hysteresis of the valve control system, collects and processes data through the method for establishing a valve working condition data set to obtain a valve system working condition data set; at the same time, based on multiple optimization objectives and constraint conditions of the valve control system, a control parameter evaluation model is established, and the differential evolution algorithm is used to complete the optimization of control parameters, and a random evaluation strategy is adopted to effectively reduce the parameter optimization time and obtain control parameters with high stability. The present invention can significantly reduce the number and amplitude of control parameter adjustments and effectively improve the anti-interference ability against disturbances under the condition of working condition transformation.

[0006] The specific steps for the present invention to achieve the above object are as follows:

[0007] (1) Through the method for establishing a valve working condition data set, collect and process the data of multiple valves under different working conditions, obtain effective working condition data, and construct a valve system working condition data sample set;

[0008] (2) Use Latin hypercube sampling to generate an initial population in the search space, and let each individual in the population represent a set of control parameters, that is, a D-dimensional vector, where D represents the number of control parameters; (3) Establish a control parameter fitness evaluation model:

[0009]

[0010] Among them, Fitness represents the fitness evaluation value of the control parameters, n represents the number of working conditions to be evaluated, m represents the number of constraints, and C ij represents the violation of the j-th constraint of the control parameters of this individual under the i-th working condition;

[0011] (4) Set the crossover probability, mutation scaling factor, and maximum number of iterations, and use the differential evolution algorithm to obtain a new generation of population from the initial population through mutation, crossover, and selection operations. In the selection operation of each generation of evolution, use the control parameter fitness evaluation model to evaluate the fitness of the current population, and adopt a random evaluation strategy until the optimal individual is output when the iteration termination condition is reached, that is, obtain the optimal valve system control parameters.

[0012] Compared with the prior art, the present invention has the following advantages:

[0013] First, considering the particularity and hysteresis of the working scenario of the valve control system, the present invention obtains control parameters with high stability through the valve working condition data set establishment method and the improved differential evolution evaluation method (the valve working condition collection method is used in conjunction with the improved differential evolution evaluation method), so as to significantly reduce the number and amplitude of control parameter adjustments under the condition of working condition transformation, and effectively improve the anti-interference ability of the system.

[0014] Second, since the traditional control parameter adjustment process of the valve system relies on experience and trial and error, lacks a general optimization method, and is difficult to handle complex multi-objective and multi-constraint scenarios, the present invention proposes a differential evolution algorithm to specifically handle multi-objective and multi-constraint scenarios, complete the optimization of control parameters in complex scenarios, and is not limited to the optimal control parameter acquisition method under a single valve working condition, with high reliability and easy implementation.

[0015] Third, since the present invention adopts a random evaluation strategy in the optimization process, the optimization time of the control parameters is effectively reduced, the optimization efficiency is improved, and control parameters with high stability can be obtained in a shorter time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the implementation flowchart of the method of the present invention;

[0017] Figure 2 is the comparison chart of the time-consuming results of obtaining optimization parameters using the traditional differential evolution algorithm and the random evaluation strategy adopted by the present invention;

[0018] Figure 3 is the schematic diagram of the algorithm convergence curve for optimizing the control parameters of the valve system using the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In order to enable those of ordinary skill in the art to better understand the technical solution of the present invention, the present invention will be further described below with reference to the accompanying drawings.

[0020] Example 1. Referring to the attached Figure 1 , a method for optimizing the control parameters of a valve system based on a differential evolution algorithm proposed by the present invention includes the following steps:

[0021] Step 1. Through the method for establishing the valve operating condition data set, collect and process the data of multiple valves under different operating conditions, obtain the effective operating condition data, and construct the valve system operating condition data sample set. For the above method for establishing the valve operating condition data set, in this embodiment, specifically, first fix the time interval segment, and then collect the real data of each valve under different operating conditions within each time period through sensors, that is, the key monitoring indicators, and perform preprocessing operations including outlier detection, elimination, and normalization on them. Finally, use the preprocessed data to construct the operating condition sample set. The above key monitoring indicators at least include the temperature, flow rate, valve position, pressure before the valve, and pressure after the valve of each valve, etc.

[0022] Step 2. Use Latin hypercube sampling in the search space to generate the initial population, and let each individual in the population represent a set of control parameters, that is, a D-dimensional vector, where D represents the number of control parameters;

[0023] Step 3. Establish a control parameter fitness evaluation model:

[0024]

[0025] Among them, Fitness represents the fitness evaluation value of the control parameters. The obtained fitness evaluation value here is used to reflect the control quantity relationship between valves. The smaller the fitness value, the smaller the constraint violation of the set of valve system control parameters in the operating condition evaluation, that is, the better the valve system control parameters. n represents the number of operating conditions to be evaluated, m represents the number of constraints, and C ij represents the jth constraint violation of the control parameters of this individual under the ith operating condition, introduced according to the actual requirements of the key monitoring indicators of the valve system, and the specific expression is as follows:

[0026] C ij = ∑d i * f(x j ),

[0027] Among them, f(x) represents the non-linear relationship determined according to the valve system, and d is the coefficient determined according to the actual valve system.

[0028] Step 4. Set the crossover probability, mutation scaling factor, and maximum number of iterations. Use the differential evolution algorithm to obtain a new generation of population from the initial population through mutation, crossover, and selection operations. In the selection operation of each generation of evolution, use the control parameter fitness evaluation model to evaluate the fitness of the contemporary population, and adopt a random evaluation strategy until the optimal individual is output when the iteration termination condition is reached, that is, obtain the optimal valve system control parameters.

[0029] The above mutation operation is implemented using any one of the mutation strategies of DE / rand / 1, DE / best / 1, DE / current-to-best / 1, DE / rand / 2, and DE / best / 2. In this embodiment, DE / rand / 1 is preferably used for the mutation operation. The above random evaluation strategy is to randomly select 1 / 10 to 1 / 2 of the working condition data from all the working condition data for evaluation during each generation of evaluation, which is used to reduce the number and amplitude of control parameter adjustments under the condition of working condition transformation and improve the anti-interference ability.

[0030] Embodiment 2. The overall implementation steps of the optimization method proposed in this embodiment are the same as those in Embodiment 1. The following gives the specific parameter settings to further describe the implementation process of the present invention in detail:

[0031] Step S1. To prove the effectiveness of the method, real data collected by sensors from the valve system under different working conditions is used, including the working condition characteristics of 41 dimensions of key monitoring indicators such as the temperature, humidity, and pressure of each valve. After outlier detection and elimination, a total of 2000 working condition samples are obtained.

[0032] The method for establishing the working condition dataset of the valve system is a fixed time interval segment. The sensors collect the temperature, flow rate, valve position, pressure before the valve, and pressure after the valve of each valve in the valve system. A plurality of working condition data are collected within one interval. The data within the interval only has small-range changes and essentially represents the same working condition data. The purpose is to improve the anti-interference of control parameters during evaluation and eliminate the error data caused by sensor failures and input failures. The sensor data is uniformly retained within the numerical range, ensuring the representativeness of the working condition data of the valve system. Since the pressure before the valve may be a relatively large quantity (such as several hundred Pa), while the valve position is usually within a relatively small range (such as 0-100%), data normalization operation is performed to avoid the unbalanced influence on model training caused by the difference in the numerical range of features. This method for collecting and processing the valve working condition data ensures the richness and representativeness of the working condition data of the valve system.

[0033] Step S2. In the search space, each individual in the population represents a set of control parameters, that is, a D-dimensional vector, where D represents the number of control parameters. In this embodiment, the number of control parameters D = 20, the population size NP = 160, the differential evolution iteration times Gmax = 30, the upper limit of the decision variable X_max = 100, and the lower limit X_min = 0. The initial population is generated using Latin hypercube sampling, and the individual is represented as follows: x_i = (d_1, d_2, d_3, …, d_20).

[0034] Step S3. The evolutionary process of each generation in the differential evolution algorithm specifically performs the following steps: Establish a control parameter fitness evaluation model. The evaluation strategy used is a random evaluation strategy. Compared with the full-scale evaluation that uses all working condition data for evaluation, using partial working conditions for evaluation ensures the diversity of the valve working condition scenario combinations. Along with the increase in the number of iteration rounds, it ensures the comprehensiveness of the use of all valve working condition scenarios, takes less time, and increases the stability of the parameter results. The implementation steps include:

[0035] S31) In this embodiment, the DE / rand / 1 / mutation strategy is adopted, and the formula is as follows:

[0036] v i,G+1 = x r1,G + F * (x r2,G - x r3,G ),

[0037] where x r1 , x r2 , x r3 represent individuals randomly selected from the current population, v i represents the mutant individual, and G represents the current iterative population;

[0038] In this embodiment, the crossover is implemented using the following formula:

[0039]

[0040] where x i represents the parent individual, v i represents the mutant individual, u i represents the trial vector, and u i,j is the value of the trial vector in the j dimension. CR is the crossover factor, and its usual value range is [0, 1]. In this embodiment, CR is preferably set to 0.3.

[0041] S32) Establish a control parameter fitness evaluation model. According to the requirements, introduce 4 valve system constraints, namely C1, C2, C3, and C4, and calculate the 41-dimensional features of the data set in step S1 and the 20-dimensional control parameters of the population individuals in step S2. The constraint is defined as C k = ∑d i * f(x j ) where f(x) represents the non-linear relationship determined according to the valve system, and the fitness of the member x in the current generation population P is evaluated. The model inputs through sensor data, calculates the constraint violation, and accumulates it as the fitness value of the parameter. The smaller the fitness value, the smaller the constraint violation of the parameter in the working condition evaluation, indicating that the solution is better.

[0042] S33) Adopt a random evaluation strategy. Randomly select 200 working conditions in each generation of evolution to evaluate 4 constraints. The fitness value Fitness calculation formula of its individual control parameters is as follows The time-consuming comparison between the original differential evolution algorithm and the adopted random evaluation strategy is as Figure 2 shown. It can be seen that when 200 working conditions are selected for evaluation, the time consumption is significantly reduced.

[0043] S34) After the differential evolution process is completed, the optimization convergence curve of the valve control parameters is as Figure 3 shown. After reaching the maximum number of iterations, output the individual with the optimal fitness, that is, the optimization result of the valve control parameters;

[0044] The parts not detailed in the present invention belong to the common general knowledge of those skilled in the art. The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Obviously, for professionals in the field, after understanding the content and principle of the present invention, various modifications and changes in form and details may be made without departing from the principle and structure of the present invention. However, these modifications and changes based on the idea of the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A valve system control parameter optimization method based on differential evolution algorithm, characterized in that: The steps include: (1) Through the valve operating condition data set establishment method, the data of multiple valves under different operating conditions are collected and processed to obtain effective operating condition data and construct a valve system operating condition data sample set; (2) Using Latin hypercube sampling in the search space to generate an initial population, each individual in the population is represented by a set of control parameters, that is, a D-dimensional vector, where D represents the number of control parameters; (3) Establishing a control parameter fitness evaluation model: Among them, Fitness represents the fitness evaluation value of the control parameter, n represents the number of working conditions to be evaluated, m represents the number of constraints, C ij represents the jth constraint violation of the individual control parameter under the i-th operating condition; (4) The crossover probability, mutation scaling factor, and maximum number of iterations are set, and a differential evolution algorithm is used to obtain a new generation of population from the initial population through mutation, crossover, and selection operations. In the selection operation of each generation of evolution, the control parameter fitness evaluation model is used to evaluate the fitness of the contemporary population, and a random evaluation strategy is used until the optimal individual is output when the iteration termination condition is reached, that is, the optimal valve system control parameters are obtained.

2. The method according to claim 1, characterized in that: The method for establishing the valve operating condition data set described in step (1) is as follows: at fixed time intervals, sensors are used to collect real data of each valve under different operating conditions in each time period, i.e., key monitoring indicators, and the data are preprocessed, and a working condition sample set is constructed using the preprocessed data.

3. The method according to claim 2, characterized in that: The key monitoring indicators include at least the temperature, flow rate, valve position, valve front pressure and valve back pressure of each valve.

4. The method according to claim 2, characterized in that: The preprocessing includes outlier detection, elimination and normalization operations.

5. The method according to claim 1, characterized in that: The fitness evaluation value in step (3) is used to reflect the control quantity relationship between valves. The smaller the fitness value, the smaller the constraint violation of the control parameters of the valve system group in the working condition evaluation, that is, the better the valve system control parameters.

6. The method according to claim 1, characterized in that: The constraint violation in step (3) is introduced according to the actual requirements of the key monitoring indicators of the valve system. The specific expression is as follows: C ij =∑d i *f(x j ), Among them, f(x) represents the nonlinear relationship determined according to the valve system, and d is the coefficient determined according to the actual valve system.

7. The method according to claim 1, characterized in that: The mutation in step (4) is implemented using any one of the mutation strategies DE / rand / 1, DE / best / 1, DE / current-to-best / 1, DE / rand / 2 and DE / best / 2.

8. The method according to claim 1, characterized in that: The random evaluation strategy described in step (4) is to randomly select 1 / 10 to 1 / 2 of the operating condition data from all the operating condition data for evaluation in each generation, which is used to reduce the number and amplitude of control parameter adjustments when the operating conditions change, and improve the anti-interference ability.

Images