An online optimization method, device and medium for an industrial dynamic simulation model

Abstract

The application discloses an online optimization method of an industrial dynamic simulation model, and comprises the following steps: obtaining historical operation data of each device on site and n groups of simulation model operation data, and performing normalization processing on the data; the similarity of the normalized data is measured and matched through a weighted Euclidean distance method; and inherent parameters of each device in the simulation model are taken as optimization parameters, and an optimization objective function is constructed through a particle swarm optimization algorithm.The application combines big data with a mechanism model, can effectively improve the simulation precision and authenticity of the dynamic simulation model, and can better adapt to various different working conditions on site.The weighted Euclidean distance method is used to perform similarity matching on simulation data of the simulation model and historical operation data on site.The particle swarm optimization algorithm is used to optimize the device parameters of the simulation model, the mechanism model and the data are organically combined, and the simulation mechanism model is optimized and supplemented.

Description

Technical Field

[0001] This invention belongs to the field of process simulation, and in particular relates to an online optimization method, equipment and medium for industrial dynamic simulation models. Background Technology

[0002] Traditional operator training simulation systems (VxOTS) use process simulation software whose dynamic simulation models are based on rigorous mechanistic modeling. These models represent idealized simulations and therefore have limitations in application scenarios. However, actual industrial operating conditions are often influenced by various factors, such as ambient temperature, equipment performance degradation, and equipment malfunctions. These factors cause discrepancies between the actual operating conditions and the mechanistic simulation models, resulting in the mechanistic models failing to accurately reflect the actual operating conditions of the equipment in the field.

[0003] With technological advancements and the continuous improvement of factory automation and informatization, Distributed Control Systems (DCS) and System-on-Systems (SIS) are widely used in various factories, establishing real-time historical data centers across the entire plant. This allows for the collection, recording, and long-term storage of massive amounts of equipment operational data, forming a large-capacity historical database. This process data contains the most original operational information of the equipment, providing a solid data foundation for dynamic simulation model optimization. Simultaneously, big data mining and analysis technologies have also made significant progress with the development of computer and information technologies. Various machine learning algorithms, optimization algorithms, and clustering algorithms have become important means of extracting effective information from massive amounts of data, while also providing a rich methodological foundation for dynamic simulation model optimization. Summary of the Invention

[0004] The technical objective of this invention is to provide an online optimization method, device, and medium for industrial dynamic simulation models, addressing the technical problem of low accuracy in simulation results obtained in existing technologies.

[0005] To solve the above problems, the technical solution of the present invention is as follows:

[0006] An online optimization method for an industrial dynamic simulation model includes the following steps:

[0007] Acquire historical operating data for each piece of equipment on site, including temperature, pressure, and flow rate;

[0008] The simulation model is loaded with the steady-state operating conditions that need to be optimized. The simulation model is run n times under the corresponding operating conditions to obtain the simulation data of each parameter corresponding to the historical running data, and then n sets of simulation model running data are obtained.

[0009] Historical operating data is normalized according to its maximum and minimum values, and simulation model operating data is normalized according to the maximum and minimum values ​​of historical operating data.

[0010] By using the weighted Euclidean distance method, the similarity between the normalized simulation model running data and the historical running data is measured and matched to obtain the most similar historical time period and the corresponding historical running data between the two.

[0011] The particle swarm optimization algorithm is used to take the inherent parameters of each device in the simulation model as optimization parameters, and the objective function is to minimize the distance between the normalized simulation model running data and the historical running data.

[0012] Specifically, the normalization process is defined as follows:

[0013]

[0014] Where x and y are the data before and after normalization, respectively. mzx ,x min These represent the maximum and minimum values ​​of the data before normalization, y. max ,y min These are the maximum and minimum values ​​of the data after normalization, respectively.

[0015] Among them, when normalizing the simulation model running data, the maximum value x of the data before normalization is... max and minimum value x min Using the maximum and minimum values ​​of historical running data, and taking 1 and 0 as the maximum value y of the normalized data. max and minimum value y min .

[0016] Specifically, the weighted Euclidean distance method is defined as follows:

[0017]

[0018]

[0019] w i =N0×e λi

[0020] Where D is the weighted Euclidean distance, the smaller the weighted Euclidean distance, the more similar the two sets of data sequences; A and B are data sequences; n is the length of the data sequence; m is the number of parameters in the data sequence; λ is the weighting exponent; w i The forgetting factor increases as it approaches the beginning of the data sequence; N0 is a constant coefficient.

[0021] The particle swarm optimization algorithm specifically includes the following steps:

[0022] Initialize the example group, which initializes the population and particles;

[0023] The fitness of each particle is calculated based on the fitness function, and the best fitness value of each particle and the best position of the population so far are obtained.

[0024] Update particle velocity and position according to velocity and position update formulas;

[0025] Determine if the stopping condition is met. If it is, output the optimal value; otherwise, continue updating the particle's velocity and position until the stopping condition is met.

[0026] Specifically, the initialization example group is as follows:

[0027] The population is initialized, including the upper and lower bounds of the search space, the learning factor, the maximum number of iterations, and the upper and lower bounds of the velocity of each particle.

[0028] Initialize the particles by randomly initializing the position and velocity of each particle;

[0029] The position of the i-th particle is represented by the following formula:

[0030] X i =(x i1 ,x i2 ,…,x iD1 ), i = 1, 2, ..., N

[0031] The velocity of the i-th particle moving to other positions is expressed by the following formula:

[0032] V i =(xv i1 ,v i2 ,…,v iD ), i = 1, 2, ..., N

[0033] Among them, X i V represents the position of the i-th particle. i Let be the velocity of the i-th particle, D1 be the dimension of the search space, and N be the total number of particles.

[0034] The specific formulas for updating velocity and position are as follows:

[0035] V id+1 =w×v id +c1r1(p id -x id )+c2r2(p gd -x id )

[0036] x id+1 =x id +vid

[0037] Among them, V id+1 Let x be the velocity of the i-th particle at step d+1. id+1 Let be the velocity of the i-th particle at step d+1, w be the inertia weight, c1 and c2 be learning factors, c1 = c2 ∈ [0, 4], r1 and r2 be random numbers in the range [0, 1], and p be the velocity of the i-th particle at step d+1. id p is the d-th dimension solution for the optimal fitness value of the i-th particle. gd The d-th dimension solution is given by the population at the known global optimal position.

[0038] The formula for calculating the objective function is as follows:

[0039]

[0040] Where, d 距离 The distance between the normalized simulation model running data and the historical running data is represented by A and B, where A and B are data sequences, n is the length of the data sequence, and m is the number of parameters in the data sequence.

[0041] An electronic device includes a processor and a memory, the memory storing computer instructions executable on the processor, the processor being used to invoke the computer instructions in the memory to execute an online optimization method for an industrial dynamic simulation model as described above.

[0042] A storage medium for storing computer instructions, wherein, when the computer instructions are executed, an online optimization method is provided to control the device on which the storage medium resides to execute an industrial dynamic simulation model as described above.

[0043] Because the present invention adopts the above technical solution, it has the following advantages and positive effects compared with the prior art:

[0044] This invention proposes an online optimization method for industrial dynamic simulation models, which combines big data with mechanistic models to effectively improve the simulation accuracy and realism of dynamic simulation models and can better adapt to various working conditions on site.

[0045] The simulation data of the simulation model and the historical operation data of the field were matched using the weighted Euclidean distance method.

[0046] The particle swarm optimization algorithm is used to optimize the equipment parameters of the simulation model by minimizing the distance between the normalized simulation model data and the historical operation data at the field.

[0047] It makes full use of historical operational data from the site and the mechanism model of VxOTS simulation software, and organically combines the mechanism model and data to optimize and supplement the simulation mechanism model. Attached Figure Description

[0048] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0049] Figure 1 This is a flowchart of an online optimization method for an industrial dynamic simulation model according to the present invention;

[0050] Figure 2 This is a flowchart of the particle swarm optimization algorithm of the present invention;

[0051] Figure 3 To utilize the dynamic simulation model of the primary air system in a power plant according to the present invention;

[0052] Figure 4 The diagram shows the optimization results of the dynamic simulation model of the primary wind turbine using the present invention. Detailed Implementation

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0054] To keep the drawings concise, only the parts relevant to the invention are shown schematically in each figure, and they do not represent the actual structure of the product. Furthermore, for ease of understanding, in some figures, only one of components with the same structure or function is shown schematically, or only one is labeled. In this document, "one" can mean not only "only one" but also "more than one".

[0055] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a method, apparatus, and medium for online optimization of industrial dynamic simulation models proposed in this invention. The advantages and features of this invention will become clearer from the following description and claims.

[0056] Example

[0057] See Figures 1 to 4This embodiment provides an online optimization method for industrial dynamic simulation models. The operator first loads the corresponding factory's dynamic simulation model configuration into the SimuBuilder dynamic simulation model configuration software within the training simulation system (VxOTS), and then loads the desired steady-state operating condition into the configuration. Historical operating data from the on-site DCS and simulation data from each device (such as temperature, pressure, and flow rate) from n rounds of model operation starting from the corresponding operating condition are acquired in the SimuBuilder dynamic simulation model configuration software. Since the data volumes of different parameters are different, normalization processing of the two sets of data is required. Using the weighted Euclidean distance method, the simulation data under the corresponding operating condition obtained from the n rounds of model configuration is similarly matched with the historical operating data in the DCS to obtain the historical time period most similar to the simulation model's operating data under that operating condition. The objective function is to minimize the sum of the absolute values ​​of the errors between the n rounds of simulation data corresponding to the simulation model and the field operation data of the most similar historical time period. The Particle Swarm Optimization (PSO) algorithm is used to optimize the model and adjust the inherent parameters of each device in the simulation model to minimize the objective function. This makes the simulation data of the simulation model closer to the historical operation data in the DCS, thereby effectively improving the accuracy and realism of the dynamic simulation model. It can simulate the field operation conditions well under various conditions and ensure the accuracy of the simulation results.

[0058] Specifically, it can be divided into the following steps:

[0059] by Figure 3 As shown, taking the dynamic simulation model of the primary air system of a power plant as an example, its actual operating conditions in the industrial field are affected by many factors, such as ambient temperature, equipment performance degradation, and device failure, which causes the dynamic simulation model to deviate from the actual operating conditions on site to a certain extent. This makes it difficult for the dynamic simulation model to accurately simulate the actual operating conditions of the on-site devices. Therefore, it is necessary to optimize the relevant equipment parameters in the model to improve the accuracy and realism of the simulation model.

[0060] See Figure 1 First, it is necessary to acquire historical operating data of each piece of equipment on site. The data collected under different operating conditions will vary, such as temperature, pressure, and flow rate. Next, the steady-state operating condition that needs to be optimized in the simulation model is loaded. The simulation model is then run n times under the corresponding operating condition to obtain simulation data of each parameter in the simulation model that corresponds to the historical operating data of each piece of equipment on site, thereby obtaining n sets of simulation model operating data.

[0061] Then, the historical operating data at the site is normalized according to its maximum and minimum values, and the operating data of the simulation model is normalized according to the maximum and minimum values ​​of the historical operating data.

[0062] The formula for normalization calculation is shown below:

[0063]

[0064] Where x and y are the data before and after normalization, respectively. max ,x min These represent the maximum and minimum values ​​of the data before normalization, y. max ,y min These are the maximum and minimum values ​​of the data after normalization, respectively.

[0065] When normalizing the simulation model's running data, the maximum value x of the data before normalization is... max and minimum value x min The maximum and minimum values ​​from historical data are used. In this embodiment, 1 and 0 are used as the maximum value y of the normalized data. max and minimum value y min , that is, y max =1, y min =0.

[0066] Next, the weighted Euclidean distance method is used to measure and match the similarity between the normalized simulation model running data and the historical running data, so as to obtain the most similar historical time period and its historical running data.

[0067] The weighted Euclidean distance method is defined as follows:

[0068]

[0069]

[0070] w i =N0×e λi

[0071] Where D is the weighted Euclidean distance, the smaller the weighted Euclidean distance, the more similar the two sets of data sequences; A and B are data sequences; n is the length of the data sequence; m is the number of parameters in the data sequence; λ is the weighting exponent; w i The forgetting factor increases as it approaches the beginning of the data sequence; N0 is a constant coefficient.

[0072] Furthermore, using the particle swarm optimization algorithm, the inherent parameters of each device in the simulation model are used as optimization parameters, such as the heat transfer coefficient U of the heat exchanger, the Cv value of the valve, and the characteristic curve coefficient of the compressor. The objective function is then set as minimizing the distance between the normalized simulation model operating data and the historical operating data from the field.

[0073] Particle Swarm Optimization (PSO) is a stochastic search algorithm based on group cooperation, developed by simulating the foraging behavior of bird flocks. Its core idea is to find the optimal solution through cooperation and information sharing among individuals within the group. PSO's advantages include simplicity, ease of implementation, no need for gradient information, few parameters, and its inherent real-number encoding characteristics, making it particularly suitable for handling real optimization problems. Furthermore, it has a profound intelligent background, making it suitable for both scientific research and engineering applications.

[0074] PSO is initialized with a swarm of random particles (random solutions), and then finds the optimal solution through iteration. In each iteration, the particles update themselves by tracking two "extremes". The first is the optimal solution found by the particle itself, called the individual extreme value pBest; the other extreme value is the optimal solution found by the entire swarm, called the global extreme value gBest. Alternatively, instead of the entire swarm, only a subset of the best particles' neighbors can be used; in this case, the extreme value among all neighbors is the local extreme value.

[0075] See Figure 1 and Figure 2 The specific implementation of the particle swarm optimization algorithm in this embodiment will now be explained step by step:

[0076] First, in step S1, the example population is initialized, including initializing both the population and the particles. Initializing the population includes setting the upper and lower bounds of the search space, learning factors c1 and c2, the maximum number of iterations T, and the upper and lower bounds of the velocity for each particle. Initializing the particles involves randomly initializing the position and velocity of each particle.

[0077] Suppose that in a D1-dimensional search space, there are N particles forming a community, where the position of the i-th particle is represented as a D1-dimensional vector;

[0078] The position of the i-th particle is represented by the following formula:

[0079] X i =(x i1 ,x i2 ,…,x iD1 ), i = 1, 2, ..., N

[0080] The velocity of the i-th particle moving to other positions is expressed by the following formula:

[0081] V i =(xv i1 ,v i2 ,…,v iD1 ), i = 1, 2, ..., N

[0082] Among them, Xi V represents the position of the i-th particle. i Let be the velocity of the i-th particle.

[0083] Next, in step S2, the fitness of each particle is calculated based on the fitness function, yielding the optimal fitness value for each particle and the best position of the population so far. The optimal fitness value is the historically best position found for each particle. A global optimal solution is then found from these historical optimal solutions and compared with the historical optimal solutions, selecting the best one as the current historical optimal solution. The optimal solution pbest (optimal fitness value) found for each individual and the optimal solution gbest (optimal position) found for the entire population are saved. By substituting the position information of each particle into the objective function, the value of the objective function for each particle can be obtained. The position of the particle with the minimum objective function value is the optimal solution for this round of optimization.

[0084] Next, in step S3, the particle velocity and position are updated according to the velocity and position update formulas. Specifically, the i-th particle updates its velocity and position according to the following formulas:

[0085] V id+1 =w×v id +c1r1(p id -x id )+c2r2(p gd -x id )

[0086] x id+1 =x id +v id

[0087] In the formula, V id+1 Let x be the velocity of the i-th particle at step d+1. id+1 Let be the velocity of the i-th particle at step d+1, w be the inertia weight, c1 and c2 be learning factors, c1 = c2 ∈ [0, 4], r1 and r2 be random numbers in the range [0, 1], and p be the velocity of the i-th particle at step d+1. id p is the d-th dimension solution for the optimal fitness value of the i-th particle. gd Given the d-th dimension solution of the population with the known global optimal position.

[0088] Furthermore, the velocity update formula consists of three parts, from left to right: the inertial or momentum part, the self-awareness part, and the social part. The inertial or momentum part reflects the particle's motion habits and is an inheritance of its previous velocity. The self-awareness part represents the influence of the particle's previous flight experience on its subsequent flight direction; the particle has an advantage in approaching its historical best position. The social part represents the influence of the flight experience of all particles in the population on the subsequent flight direction of each particle; particles tend to approach the historical best position of the group or domain.

[0089] Therefore, the velocity update formula can be interpreted as: the velocity of the i-th particle at step d+1 = its own velocity inertia from the previous step + self-awareness component + social awareness component. The position update formula can be interpreted as: the position of the i-th particle at step d+1 = position at step d + velocity at step d * motion time, where the motion time for each step is generally taken as 1.

[0090] Next, proceed to steps S4 and S5 sequentially to calculate the updated fitness value of each particle. Compare the best fitness value of each particle with its fitness value at its historical best position. If it is better, then its current position is taken as the best position for that particle. For each particle, compare the fitness value corresponding to its best position with the best fitness value of the population. If it is better, then update the best position and best fitness value of the population.

[0091] Finally, in step S6, it is determined whether the search result meets the stopping condition (reaching the maximum number of iterations or meeting the accuracy requirement). If it does, the optimal value is output; otherwise, the velocity and position of the particles are updated until the stopping condition is met.

[0092] Specifically, the formula for calculating the objective function is as follows:

[0093]

[0094] Where, d 距离 The distance between the normalized simulation model running data and the historical running data is represented by A and B, where A and B are data sequences, n is the length of the data sequence, and m is the number of parameters in the data sequence.

[0095] See Figure 4 The figure shows the optimized results of the dynamic simulation model in this embodiment. Four reference digits are selected as the benchmark reference digits for the dynamic simulation model and the field equipment data. To make the simulation data of these four reference digits in the dynamic simulation model closer to the operational data of the corresponding reference digits in the field, the method proposed in this embodiment is used... Figure 3The Cv values ​​of the four valves in the dynamic simulation model, as well as the coefficients of the flow / pressure rise characteristic curve equation and the fan efficiency equation of the primary air fan A, were used as optimization parameters to optimize the dynamic model. The flow / pressure rise characteristic curve equation and the fan efficiency equation are shown below:

[0096] Ax 3 +Bx 2 +Cx+D=0

[0097] from Figure 4 It can be clearly seen that after optimization, the simulation data of the four reference numbers obtained by the dynamic simulation model are basically close to the actual operating data of the corresponding reference numbers on site, indicating that this embodiment has significantly improved the accuracy of the dynamic simulation model.

[0098] Furthermore, this embodiment also provides an electronic device, which includes a processor and a memory. The memory stores computer instructions that can be executed on the processor, and the processor is used to call the computer instructions in the memory to execute an online optimization method for an industrial dynamic simulation model as described above.

[0099] Furthermore, this embodiment also provides a storage medium for storing computer instructions, wherein, when the computer instructions are executed, the device where the storage medium is located is controlled to execute an online optimization method for an industrial dynamic simulation model as described above.

[0100] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalents, they shall still fall within the protection scope of the present invention.

Claims

1. An online optimization method for an industrial dynamic simulation model, characterized in that, Includes the following steps: Acquire historical operating data for each piece of equipment on site, including temperature, pressure, and flow rate; Load the steady-state operating conditions that need to be optimized into the simulation model, and then run the simulation model under the corresponding operating conditions. The process involves retrieving simulation data for various parameters corresponding to historical operational data, and then obtaining... Group simulation model running data; Historical operating data is normalized according to its maximum and minimum values, and simulation model operating data is normalized according to the maximum and minimum values ​​of historical operating data. By using the weighted Euclidean distance method, the similarity between the normalized simulation model running data and the historical running data is measured and matched to obtain the most similar historical time period and the corresponding historical running data between the two. The particle swarm optimization algorithm is used to take the inherent parameters of each device in the simulation model as optimization parameters, and the objective function is to minimize the distance between the normalized simulation model running data and the historical running data. The weighted Euclidean distance method is defined as follows: in, The weighted Euclidean distance is used; the smaller the weighted Euclidean distance, the more similar the two sets of data sequences are. For data sequences; The length of the data sequence; λ is the number of parameters in the data sequence; λ is the weighting exponent. The forgetting factor increases as the data sequence approaches its starting point. These are constant coefficients.

2. The online optimization method for the industrial dynamic simulation model according to claim 1, characterized in that, The normalization process is defined as follows: in, These are the data before and after normalization, respectively. These are the maximum and minimum values ​​of the data before normalization. These are the maximum and minimum values ​​of the data after normalization, respectively.

3. The online optimization method for the industrial dynamic simulation model according to claim 2, characterized in that, When normalizing simulation model running data, the maximum value of the data before normalization. and minimum value Using the maximum and minimum values ​​of historical running data, and taking 1 and 0 as the maximum values ​​of the normalized data. and minimum value .

4. The online optimization method for the industrial dynamic simulation model according to claim 1, characterized in that, The particle swarm optimization algorithm specifically includes the following steps: Initialize the example group, which initializes the population and particles; The fitness of each particle is calculated based on the fitness function, and the best fitness value of each particle and the best position of the population so far are obtained. Update particle velocity and position according to velocity and position update formulas; Determine if the stopping condition is met. If it is, output the optimal value; otherwise, continue updating the particle's velocity and position until the stopping condition is met.

5. The online optimization method for the industrial dynamic simulation model according to claim 4, characterized in that, The initialization example group is as follows: The population is initialized, including the upper and lower bounds of the search space, the learning factor, the maximum number of iterations, and the upper and lower bounds of the velocity of each particle. Initialize the particles by randomly initializing the position and velocity of each particle; No. The position of each particle is represented by the following formula: No. The velocity of a particle moving to other positions is expressed by the following formula: in, For the first The position of each particle. For the first The speed of each particle For the dimensions of the search space, This represents the total number of particles.

6. The online optimization method for the industrial dynamic simulation model according to claim 4, characterized in that, The specific formulas for updating velocity and position are as follows: in, For the first The particle in the first The speed of the step, For the first The particle in the first The speed of the step, For inertial weights, As a learning factor, , For range random numbers, For the first The optimal fitness value of the particle Solution, For a population whose global optimal position is known, the first... Solution.

7. The online optimization method for an industrial dynamic simulation model according to claim 1, characterized in that, The formula for calculating the objective function is as follows: in, The distance between the normalized simulation model running data and the historical running data, For data sequences, The length of the data sequence; This represents the number of parameters within the data sequence.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing computer instructions executable on the processor, and the processor being used to invoke the computer instructions in the memory to execute the online optimization method of the industrial dynamic simulation model as described in any one of claims 1-7.

9. A storage medium, characterized in that, The storage medium is used to store computer instructions, wherein, when the computer instructions are executed, the device where the storage medium is located is controlled to perform an online optimization method for the industrial dynamic simulation model as described in any one of claims 1-7.

Images