An optimization method for regulating valve dynamic simulation characteristic curve

Abstract

The application provides an optimization method for adjusting valve dynamic simulation characteristic curve, and the steps are as follows: obtaining historical data of a field adjusting valve, related parameters of a simulation adjusting valve and related physical parameters, and calculating corresponding simulation adjusting valve comprehensive opening degree; based on field adjusting valve real opening degree and simulation adjusting valve comprehensive opening degree data, pre-processing the data through an improved dynamic class center adjustment clustering method; dividing class center data obtained by clustering into n intervals according to a given opening degree of the field adjusting valve, and merging intervals without data in adjacent intervals to calculate initial characteristic curves of each interval; taking the distance sum of each data point to the initial characteristic curve as an optimization function, taking the initial characteristic curve slope k i of each interval and the interval number n as optimization parameters, and using an optimization algorithm to perform optimization to obtain the characteristic curve slope k i of each interval and the interval number n under the condition that the distance sum is minimum, and obtain a simulation adjusting valve characteristic curve with high accuracy and high authenticity.

Description

Technical Field

[0001] This invention relates to the field of dynamic simulation technology for control valves, and specifically to a method for optimizing the dynamic simulation characteristic curve of a control valve. Background Technology

[0002] Control valves are one of the most commonly used unit devices in chemical processes. They are the final control elements that change process parameters such as flow rate, pressure, temperature, and liquid level by receiving control signals from the control unit and using power operation. They are generally composed of an actuator and a valve.

[0003] To better simulate and accurately reproduce chemical production processes, dynamic simulation of control valves is often required. However, during dynamic simulation, the valve characteristics of the control valve often differ from those of the actual control valve in the field, leading to a discrepancy between the valve opening in the dynamic simulation and the actual opening in the field. Therefore, a special valve characteristic curve can be configured in the simulated control valve, reflecting the characteristic relationship between the actual opening (SetOP) of the field control valve and the combined opening (RealOP) of the simulated control valve.

[0004] By comparing the actual valve opening (SetOP) with this characteristic curve, the overall valve opening (RealOP) participating in the dynamic simulation calculation of the simulated control valve can be obtained. However, it is quite difficult to accurately obtain this curve.

[0005] With the advancement of technology, the characteristic curves of simulated control valves can be calculated from massive amounts of historical data of field control valves. However, the influence of noise in the field historical data can cause the characteristic curves of control valves obtained based on the historical data to be distorted. If the necessary processing is not performed, they may not be usable for simulating control valves.

[0006] The article "Optimization of Flow Characteristic Curves of Steam Turbine Valves in Thermal Power Plants" published in the journal Electric Power Safety Technology (09), 59-61, provides a method for adjusting and optimizing the characteristic curves of steam turbine valves on-site through experiments. This method is mainly for the optimization of actual valves on-site and requires obtaining the characteristic curves of valves through experiments. It is costly, inefficient, and has significant limitations in effectiveness. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide an optimization method for the dynamic simulation characteristic curve of a control valve, thereby improving the simulation accuracy and realism of the control valve in dynamic simulation and obtaining a high-precision and high-realism simulated control valve characteristic curve.

[0008] To address the aforementioned technical problems, the present invention provides the following technical solution.

[0009] This invention provides a method for optimizing the dynamic simulation characteristic curve of a control valve, comprising the following steps:

[0010] S1: Obtain historical data of the field control valve, relevant parameters of the simulation control valve, and relevant physical property parameters, and calculate the overall opening degree of the simulation control valve under the corresponding historical data;

[0011] S2: Based on the actual opening degree of the field control valve and the comprehensive opening degree data of the simulated control valve, the data is preprocessed using the improved dynamic core-based KMeans clustering method;

[0012] S3: Based on the given opening degree of the field control valve, the cluster core data obtained by clustering is divided into n intervals, and the intervals with no data in adjacent intervals are merged into a large empty interval. The number of merged intervals is n'. The initial characteristic curve is obtained based on the data of each merged interval.

[0013] S4: Use the sum of the distances from each data point to the initial characteristic curve as the optimization function, and calculate the slope k of the initial characteristic curve in each merged interval. i Where i = 1 to n', and the number of intervals n before merging are used as optimization parameters, the optimization algorithm is used to perform optimization operations to obtain the slope k of the characteristic curve of each interval in the case of minimizing the total distance. i And the number of intervals, n.

[0014] Furthermore, the historical data of the field control valve includes historical data related to the actual opening degree of the control valve, valve inlet pressure, valve outlet pressure, and valve flow rate; the relevant parameters of the simulation control valve are the Cv value of the simulation control valve; and the relevant physical property parameters are the mass flow density value.

[0015] Furthermore, the formula for calculating the overall opening degree of the simulated regulating valve is as follows:

[0016]

[0017] Where F is the valve flow rate; P in P represents the valve inlet pressure. out ρ is the valve outlet pressure; Cv is the simulated control valve Cv value; ρ is the mass flow density value.

[0018] Furthermore, in step S2, the steps for improving the dynamic core adjustment KMeans clustering method are as follows:

[0019] S201: Set the data deviation tolerance Δ and the data filtering tolerance (%) σ;

[0020] S202: Set the number of cluster centers K, and randomly select K data points as initial cluster centers, denoted as...

[0021] S203: Define the loss function:

[0022] S204: Let t = 0, 1, 2, ... be the number of iterations. For each sample x... i Assign it to the nearest class center. The i-th class cluster at step t is:

[0023]

[0024] S205: For each cluster, recalculate the cluster center. The i-th cluster center at iteration t+1 is:

[0025]

[0026] S206: Repeat steps 4 and 5 until J converges;

[0027] S207: Determine the value of each sample x i Check if the distance to its class core is less than the data deviation tolerance Δ, record the number N of class cores that do not meet the condition, and reset the number of class cores K = K + N;

[0028] S208: Repeat steps 2-7 until all samples x are obtained. i The distance to the center of its class is less than the data deviation tolerance Δ;

[0029] S209: Calculate the percentage of samples within each class out of the total number of samples. If the percentage of samples within a class is less than the data filtering tolerance (%) σ, then remove that class.

[0030] Furthermore, in step S3, obtaining the initial characteristic curve specifically involves:

[0031] Using the origin (0,0) as the starting point of the first interval's initial characteristic curve, and passing through the pre-set initial slope k of each interval's initial characteristic curve... i Obtain the initial slope k i The initial characteristic curves within each interval are used to obtain the linear initial characteristic curve y = k within the range of 0-100% opening of the field control valve. i *x+b i (i = 1 to n'); where the initial slope k is set. i The methods include: presetting an initial k based on empirical data. i The value is a fixed value, or the center of each cluster obtained through clustering is calculated by the least squares method.

[0032] Furthermore, in step S4, the distance from the data points to the initial characteristic curve is calculated as follows: the distances from each data point (x, y, y) obtained through clustering are... i yi x in ) i Substituting into the initial characteristic curve, we can obtain the ordinate y corresponding to the initial characteristic curve. i '=k i *x i +b i The ordinate y under the initial characteristic curve i 'and the actual ordinate y i Subtract and take the absolute value |y i -y i By substituting '|' into the initial characteristic curve, we can obtain the distance from the data point to the initial characteristic curve. Substituting all data points into the initial characteristic curve, we can obtain the sum of the distances from all data points to the initial characteristic curve, sum|y. i -y i '|.

[0033] Furthermore, in step S4, the optimization algorithm includes one or more of Nelder-Mead Simplex, Principal AXIS, Constrained Optimization BY Linear Approximations, BOBYQA, NEWUOA, and based on Subplex.

[0034] Compared with existing technologies, the optimization method for the dynamic simulation characteristic curve of the control valve provided by this invention has the following beneficial effects:

[0035] This invention proposes a method for optimizing the dynamic simulation characteristic curve of a control valve. First, by using historical data and relevant parameters and physical properties of the control valve in the simulation model, the comprehensive opening of the simulated control valve under field data conditions is obtained. Then, an improved KMeans clustering method with a dynamic core number is used to preprocess the calculated comprehensive opening of the simulated control valve and the actual opening data of the field control valve, effectively removing abnormal noise data and reducing data duplication. Specifically, the improved KMeans clustering algorithm dynamically adjusts the number of cores by setting core filtering tolerance and maximum core deviation distance, making the obtained cores more representative of the original data distribution characteristics. The opening range of the field control valve from 0% to 100% is divided into n equal intervals, and intervals with no data in adjacent intervals are merged. The slope k of the initial characteristic curve in each merged interval is then calculated. i Using the number of intervals n as optimization parameters, the minimum sum of distances from each data point to the initial characteristic curve is taken as the optimization function. Through the corresponding optimization algorithm, the optimized characteristic curve of the control valve can be obtained, thereby improving the simulation accuracy and simulation realism of the control valve in dynamic simulation and obtaining a high-precision and high-realism simulated control valve characteristic curve. Attached Figure Description

[0036] Figure 1(a) is a schematic diagram of the configuration before optimization of the simulated control valve characteristic curve in an embodiment of the present invention;

[0037] Figure 1(b) is a schematic diagram of the optimized configuration of the simulated control valve characteristic curve in an embodiment of the present invention;

[0038] Figure 2 This is a flowchart illustrating the optimization steps for the dynamic simulation characteristic curve of the regulating valve of the present invention.

[0039] Figure 3 This is a flowchart of the improved dynamic core adjustment KMeans clustering method in an embodiment of the present invention;

[0040] Figure 4 This is a comparison chart of the clustering results of KMeans and the improved KMeans algorithm;

[0041] Figure 5 This is a schematic diagram of the data after merging the data division intervals according to an embodiment of the present invention;

[0042] Figure 6 This is a comparison chart of three data points: the actual flow rate of the control valve in the field, the flow rate obtained from the dynamic simulation of the control valve before the optimization of the control valve characteristic curve, and the flow rate obtained from the dynamic simulation of the control valve after the optimization of the control valve characteristic curve. Detailed Implementation

[0043] To clearly illustrate the objectives, technical solutions, and advantages of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0044] Figure 1(a) shows a schematic diagram of the characteristic curve configuration of the control valve. The horizontal axis represents the given opening degree (%), which is the actual opening degree of the control valve in the field (SetOP). The vertical axis represents the comprehensive opening degree (%), which is the simulated control valve opening degree (RealOP) participating in the dynamic simulation calculation. The characteristic curve can be configured by inputting the given opening degree and the comprehensive opening degree. By comparing the given opening degree with the characteristic curve, the comprehensive opening degree of the control valve participating in the dynamic simulation calculation can be obtained. Therefore, the accuracy of the configured characteristic curve determines the accuracy and realism of the model simulation results.

[0045] This invention provides a method for optimizing the dynamic simulation characteristic curve of a control valve, which specifically includes the following steps.

[0046] S1: Obtain historical data of the field control valve, relevant parameters of the simulation control valve, and relevant physical property parameters, and calculate the overall opening degree of the simulation control valve under the corresponding historical data;

[0047] Acquire historical data for field control valves, including the actual valve opening (SetOP) and valve inlet pressure (P). in Valve outlet pressure (P) out The relevant historical data of valve flow rate (F); the relevant parameters of the simulated control valve are the valve Cv value of the simulated control valve; the relevant physical property parameters are the material flow density value (ρ).

[0048] Based on the acquired historical data of the field control valve, relevant parameters of the simulation control valve, and relevant physical property parameters, the overall opening degree (RealOP) of the simulation control valve under the corresponding historical data is calculated. The calculation formula is as follows:

[0049]

[0050] S2: Based on the actual opening degree (SetOP) of the field control valve and the comprehensive opening degree (RealOP) of the simulated control valve, the data is preprocessed using an improved dynamic core-based KMeans clustering method;

[0051] By using an improved KMeans clustering method with dynamically adjusted cluster centers, the actual opening (SetOP) data of field control valves and the comprehensive opening (RealOP) data of simulated control valves are preprocessed. This not only removes abnormal noise data, but also clusters a large number of similar data into the same cluster, and uses the cluster center of the cluster to represent these similar data, thereby reducing the data duplication rate.

[0052] The K-means algorithm is a commonly used clustering algorithm. Given a value of K and K initial cluster centroids, each data sample is assigned to the cluster represented by the nearest cluster centroid. After all data samples are assigned, the centroid of each cluster is recalculated by averaging the values ​​of all data samples within that cluster. This process is repeated iteratively until no or the minimum number of objects are reassigned to different clusters, the cluster centroids change very little, or the specified number of iterations is reached. The number of cluster centroids, K, is a hyperparameter in the K-means algorithm that is difficult to determine and needs to be chosen empirically. Furthermore, the clustering results are significantly affected by outliers, and the results may not be globally optimal but rather locally optimal, resulting in low reliability.

[0053] The calculation steps are as follows:

[0054] 1. Set the number of cores K, and randomly select K data points as the initial cores, denoted as K0.

[0055] 2. Define the loss function:

[0056] 3. Let t = 0, 1, 2, ... be the number of iterations. For each sample x... i Assign it to the nearest class center. The i-th class cluster at step t is:

[0057]

[0058] 4. For each cluster, recalculate the cluster center. The i-th cluster center at iteration t+1 is:

[0059]

[0060] 5. Repeat steps 3 and 4 until J converges;

[0061] This invention proposes an improved KMeans clustering method with dynamic core adjustment, based on KMeans clustering. This method can dynamically adjust the number of cores (K) and effectively eliminate the influence of outliers. See details below. Figure 3 The calculation steps are as follows:

[0062] S301: Set the data deviation tolerance Δ and the data filtering tolerance (%) σ;

[0063] S302: Set the number of cluster centers K, and randomly select K data points as initial cluster centers, denoted as Kcenter.

[0064] S303: Define the loss function:

[0065] S304: Let t = 0, 1, 2, ... be the number of iterations. For each sample x... i Assign it to the nearest class center. The i-th class cluster at step t is:

[0066]

[0067] S305: For each cluster, recalculate the cluster center. The i-th cluster center at iteration t+1 is:

[0068]

[0069] S306: Repeat steps 4 and 5 until J converges;

[0070] S307: Determine the value of each sample x iCheck if the distance to its class core is less than the data deviation tolerance Δ, record the number N of class cores that do not meet the condition, and reset the number of class cores K = K + N;

[0071] S308: Repeat steps 2 to 7 until all samples x are obtained. i The distance to the center of its class is less than the data deviation tolerance Δ;

[0072] S309: Calculate the percentage of samples within each class out of the total number of samples. If the percentage of samples within a class is less than the data filtering tolerance (%) σ, then remove that class.

[0073] See Figure 4 The clustering results obtained by the KMeans clustering algorithm were compared with those obtained by the improved dynamic core adjustment KMeans clustering method. In this embodiment, the data bias tolerance Δ of the improved dynamic core adjustment KMeans clustering method was set to 0.5, and the data filtering tolerance (%) σ was set to 1%. The final number of cores K before removing outliers was 25. Since the initial number of cores K in the KMeans clustering algorithm could not be determined, the number of cores K in the KMeans clustering algorithm was also set to 25. The KMeans clustering method had 5 outlier cores, and the percentage of data in these cores was less than 1% of the total data. These were considered outlier cores and should be removed to reduce the impact of abnormal noise data.

[0074] In this embodiment, the KMeans clustering algorithm is improved by dynamically adjusting the number of clusters by setting the cluster core filtering tolerance and the maximum deviation distance of the cluster cores. This makes the obtained cluster cores more representative of the original distribution characteristics of the data, and at the same time, it can effectively remove outliers. Outliers will affect the optimization results of the simulated valve characteristic curve, causing the optimization results to deviate from the normal value range and resulting in unreliable optimization results.

[0075] S3: Based on the given opening degree of the field control valve, the cluster data obtained by clustering is divided into n intervals, and the intervals with no data in adjacent intervals are merged into a large empty interval. The initial characteristic curve is obtained based on the data of each merged interval.

[0076] Generally, the setOP (setoff) range of the field control valve is between 0% and 100%. This range is divided into n equal intervals, and the cluster cores are assigned to these intervals based on their setOP data. Due to varying degrees of interval division, some intervals may contain no data. Intervals with no data in adjacent intervals are merged into an empty interval, denoted as n'. A data partitioning diagram is shown below. Figure 5As shown, in this embodiment, the opening degree (SetOP) of the field control valve is divided into 10 intervals. The 60-70 and 70-80 intervals are merged into one empty interval because there is no data.

[0077] After merging the intervals, the origin (0,0) is used as the starting point of the initial characteristic curve for the first interval, and the curve is passed through the pre-set initial slope k of the initial characteristic curve for each interval. i Obtain the initial slope k i The initial characteristic curves within each interval are used to obtain the linear initial characteristic curve y = k within the range of 0–100% of the field control valve opening (SetOP). i *x+b i (i = 1 to n').

[0078] Wherein, the initial slope k is set i The methods include: presetting an initial k based on empirical data. i The value is a fixed value, usually chosen as 1; or the center of each cluster obtained by clustering is calculated by the least squares method.

[0079] S4: Use the sum of the distances from each data point to the initial characteristic curve as the optimization function, and calculate the slope k of the initial characteristic curve within each merged interval. i (i = 1 to n') and the number of intervals n before merging are used as optimization parameters. An optimization algorithm is used to perform optimization operations to obtain the slope k of the characteristic curve of each interval when the total distance is minimized. i And the number of intervals, n;

[0080] The calculation of the distance from each data point to the initial characteristic curve is as follows: The distances from each data point (x, y, y) obtained through clustering are... i y i x in ) i Substituting into the initial characteristic curve, we can obtain the ordinate y corresponding to the initial characteristic curve. i '=k i *x i +b i The ordinate y under the initial characteristic curve i 'and the actual ordinate y i Subtract and take the absolute value |y i -y i '|' will give you the distance from the data point to the initial characteristic curve.

[0081] By substituting all data points into the initial characteristic curve, we can obtain the sum of the distances from each data point to the initial characteristic curve, sum|y. i -y i '|, and sum|y of the distances. i -yi '| As the optimization function of the optimization algorithm, the slope k of the characteristic curve in each merged interval is...' i (i = 1 to n') and the number of intervals n are used as optimization parameters for the optimization algorithm. The optimization algorithm is used to optimize the parameters to obtain the slope k of the characteristic curve of each interval when the sum of the absolute distances is minimized. i And the number of intervals n. It should be noted that the number of intervals n' after merging is related to the number of intervals n initially. Therefore, when optimizing and adjusting the number of intervals n initially, the number of intervals n' after merging is adjusted accordingly.

[0082] There are many optimization algorithms, including Nelder-Mead Simplex, Principal AXIS (PRAXIS), Constrained Optimization BY Linear Approximations (COBYLA), BOBYQA, NEWUOA, and based on Subplex (SBPLX). By comparing various optimization algorithms, the Principal AXIS algorithm requires relatively fewer iterations to converge, has higher accuracy, and the optimized characteristic curve is more representative. In this embodiment, the Principal AXIS algorithm is selected for optimization. The optimized characteristic curve is shown in Figure 1(b).

[0083] See Figure 6 The figure shows a comparison of three data points: the actual flow rate of the control valve in the field, the flow rate obtained from the dynamic simulation of the control valve when its original characteristic curve is linear, and the flow rate obtained from the dynamic simulation of the control valve after the characteristic curve has been optimized. It is evident that after optimization, the flow rate obtained from the simulation is very close to the actual flow rate of the control valve in the field. The dynamic simulation characteristic curve optimization method for control valves provided by this invention significantly improves the accuracy of dynamic simulation of control valves.

[0084] The above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method of optimizing a dynamic simulation characteristic curve of a control valve, characterized in that Comprising the following steps: S1: obtaining the historical data of the field regulating valve, the related parameters of the simulation regulating valve and the related physical parameters, and calculating the comprehensive opening degree of the simulation regulating valve corresponding to the historical data; S2: based on the real opening degree of the field regulating valve and the comprehensive opening degree data of the simulation regulating valve, the data is pretreated through the improved dynamic class center adjustment KMeans clustering method, and the steps of the improved dynamic class center adjustment KMeans clustering method are: S201: setting data deviation tolerance Δ and data filtering tolerance σ; S202: Set the number of class centers K, randomly select K data as initial class centers, denoted as ; S203: Define the loss function: ; S204: Let t = 0, 1, 2,... be the iteration step number, for each sample x i , assign it to the nearest class center, the ith class cluster at the tth iteration is: ； S205: for each cluster, the cluster center is recalculated, and the i-th cluster center in the t+1 step iteration is: ； S206: repeat steps S204 and S205 until J converges; S207: judging whether the distance from each sample x i to the class center to which it belongs is less than the data deviation tolerance Δ, recording the number of class centers N that do not meet the condition, and resetting the number of class centers K = K + N; S208: repeat steps S202-S207 for all samples x i the distance to its belonging class center is smaller than the data bias tolerance Δ; S209: count the percentage of the number of samples in each class in the total number of samples, and if the percentage of the number of samples in each class is less than the data filtering tolerance σ, the cluster center is removed; S3: According to the given opening degree of the field regulating valve, the cluster obtained class center data is equally divided into n intervals, and the interval with no data in the adjacent interval is merged into a large empty interval, the number of intervals after merging is n', and the linear initial characteristic curve y=k i *x+b i is set as the initial slope, i=1~n'; i ​ S4: take the distance sum of each data point to the initial characteristic curve as an optimization function, take the initial characteristic curve slope k i of each interval after merging, and the interval number n before merging as optimization parameters, and use an optimization algorithm to perform optimization operation to obtain the interval characteristic curve slope k i and the interval number n under the condition of minimum distance sum.

2. The optimization method of claim 1, wherein, The historical data of the field regulating valve includes the related historical data of the real opening degree of the regulating valve, the inlet pressure of the valve, the outlet pressure of the valve and the flow rate of the valve; the related parameters of the simulation regulating valve are the Cv value of the simulation regulating valve; and the related physical parameters are the mass flow density value.

3. The optimization method of claim 2, wherein, The calculation formula of the comprehensive opening degree of the simulation regulating valve is: where F is the valve flow rate; P in is the valve inlet pressure; P out is the valve outlet pressure; Cv is the Cv value of the simulated control valve; and p is the mass flow density value.

4. The optimization method of claim 1, wherein, In step S3, obtaining the initial characteristic curve is specifically: Taking the coordinate origin (0, 0) as the starting point of the first interval characteristic curve, the initial slope k i of each interval characteristic curve is preset i , the initial characteristic curve in each interval is obtained, and thus the linear initial characteristic curve y=k i *x+b i in the field regulating valve opening interval of 0~100% is obtained, i=1~n'; wherein, the manner of setting the initial slope k i includes: presetting the value of the initial k i as a fixed value according to empirical data, or calculating and obtaining the cluster center of each cluster obtained by clustering through the least square method.

5. The optimization method of claim 1, wherein, In step S4, the distance from each data point to the characteristic curve is calculated as follows: the distances from each data point (x, y, y) obtained through clustering are... i y i x in ) i Substituting into the initial characteristic curve, we can obtain the ordinate y corresponding to the initial characteristic curve. i '=k i * x i +b i The ordinate y under the initial characteristic curve i 'and the actual ordinate y i Subtract and take the absolute value |y i -y i By substituting '|' into the initial characteristic curve, we can obtain the distance from the data point to the initial characteristic curve. Substituting all data points into the initial characteristic curve, we can obtain the sum of the distances from all data points to the initial characteristic curve, sum|y. i -y i '|.

6. The optimization method of claim 1, wherein, In step S4, the optimization algorithm includes one or more of Nelder-Mead Simplex, PRincipal AXIS, ConstrainedOptimization BY Linear Approximations, BOBYQA, NEWUOA and based on Subplex.

Images