A method for selecting test points for flowmeter devices based on an optimized test matrix

Abstract

This invention discloses a method for selecting test points for flowmeter devices based on an optimized test matrix. It collects multiphase flow parameters and test points under different conditions as training and testing data, establishing a model combining artificial neural networks and genetic algorithms. The performance of the neural network is reflected in the mean square error of the training results. An intelligent learning algorithm is used to optimize the test evaluation matrix, reducing the selection of duplicate test points, maximizing test point coverage, and achieving automatic test point selection. This reduces the influence of human factors and allows for more targeted analysis of factors affecting flowmeter measurement performance, improving the reliability of flowmeter test evaluation. This invention makes the evaluation of standard devices more intelligent and digital, resulting in more scientific and reasonable test evaluation results.

Description

Technical Field

[0001] This invention relates to the field of gas-liquid two-phase and oil-gas-water multiphase flow, and is particularly applicable to the automatic selection of test points in multiphase flow meter standard devices. Background Technology

[0002] Multiphase flow meters differ from traditional single-phase flow meters. They typically incorporate multiple measurement sensors and dedicated software programs, making the testing and evaluation of their metering performance unique and complex. Multiphase flow conditions are complex, with flow points composed of two or three phases from an oil, gas, or water medium in varying proportions. Factors such as temperature, pressure, water content, and gas content must also be considered, along with flow patterns, flow regimes, and phase transitions. Therefore, a multidimensional test matrix is ​​determined comprehensively based on the flow meter's measurement range, influencing factors, and the testing capabilities of standard multiphase flow meter devices. This results in a large number of test points and complex arrangements.

[0003] Currently, there is no unified method for selecting the test evaluation matrix for multiphase flowmeters, either domestically or internationally. The selection of test points for multiphase flowmeters is mainly done manually, which cannot be automated and suffers from problems such as significant human influence, low automation levels, low testing and evaluation efficiency, and inability to rationally allocate test points. Automatic selection of test points for multiphase flowmeter standard devices is a crucial step in achieving intelligent calibration of these devices. With the development of big data and artificial intelligence technologies, we have conducted technological research on this aspect, applying artificial neural networks and fuzzy algorithms to the intelligent selection of calibration points for multiphase flowmeter standard devices for the first time. Using an artificial neural network intelligent backpropagation algorithm, combined with big data analysis of historical test data, and programmatically incorporating the Daqing multiphase flowmeter standard device test matrix selection rules, we have achieved automatic and precise optimization of the test evaluation matrix, realizing intelligent and automatic calibration of multiphase flowmeter standard devices and improving the testing and evaluation efficiency of the devices. Summary of the Invention

[0004] To address the challenges of low efficiency and low automation in manually selecting test points for multiphase flowmeter evaluation under complex and variable testing conditions with multiple parameters, this invention provides a method for selecting test points for flowmeter devices based on an optimized test matrix. Employing an artificial neural network intelligent backpropagation algorithm, the method achieves precise optimization of the multiphase flowmeter test evaluation matrix, enabling automatic selection of flowmeter test points and effectively ensuring the scientific and efficient nature of the flowmeter performance evaluation method.

[0005] To achieve the above objectives, the present invention provides the following technical solution: 1. A method for selecting test points for a flow meter device based on an optimized test matrix, characterized by comprising the following steps:

[0006] S1: A three-layer BP neural network training model is combined with a genetic algorithm;

[0007] S2: Establish the database required for solving the multiphase flow parameter model, extract the flow experimental data from the large database, filter the required data, and establish a multi-parameter database;

[0008] S3: Optimize the parameters of the initial weight matrix of the BP neural network using a genetic algorithm, calculate the fitness of each individual using multiphase flow features in the database, determine the optimal weight matrix through fitness evaluation, and establish a genetic neural network point selection model.

[0009] S4: Use the temperature, pressure, and phase fraction parameters in the multiphase flow testing device as the input layer of the neural network; use the test points as the output labels for model training.

[0010] S5: Use gradient search to adjust the connection strength between input nodes and hidden nodes, and between hidden nodes and output nodes. Use pre-trained weights to fine-tune the model. Use the mean squared error loss function to calculate the model loss and combine it with the Adam optimization algorithm to update the parameters. Use the training function train to optimize the test matrix algorithm and adjust parameters such as the learning coefficient.

[0011] S6: Use the database to create a test dataset to evaluate the network model, and use the standard root mean square error as the evaluation metric.

[0012] As a preferred embodiment of the present invention, in step S1, the values ​​of each parameter in the network are determined according to empirical formulas to build an initial neural network model. Then, the initial weights are used as input to the genetic algorithm. After selection, crossover, and mutation operations, the fitness value is calculated, and the best set of weight matrices is input into the neural network model. The error is calculated, the weights of each hidden layer are adjusted, and the prediction result is output.

[0013] As a preferred embodiment of the present invention, the database required for establishing the multiphase flow parameter solution model in step S2, and the extraction of flow experimental data from the large database to filter the required data to establish a multi-parameter database, are necessary to ensure the accuracy of the neural network iteration results, which require a large amount of experimental data for calculation and analysis.

[0014] As a preferred embodiment of the present invention, in step S3, an initial trial population is randomly formed. Through expected evolution, the fitness function of the MSE is adopted, and the fitness value of the population is obtained through pre-input from the database to provide support for the selection operator. Then, the initial weights of the BP neural network are obtained through exchange, mutation and multiple operations.

[0015] As a preferred embodiment of the present invention, in step S4, the number of processing units in the intermediate layer of the connection model adjustment network is established by establishing the correspondence between the input parameters and the test flow point and the correspondence between the flow field coefficient in the pipeline. The Monte Carlo gradient algorithm is used to search for local minima, and the principle machine that can iteratively approximate any function is used for feedback calculation through the backpropagation algorithm of the artificial neural network.

[0016] As a preferred embodiment of the present invention, the standard root mean square error is used as an evaluation function for the impact of each input variable on the performance of the neural network. Considering the stability and convergence of the iterative calculation process, a linear rectified function is used as the basic function to set the excitation threshold to ensure the convergence of the calculation during the iterative process.

[0017] As a preferred embodiment of the present invention, in conjunction with the model described in claims 1-6, the neural network is combined with the pre-genetic algorithm. First, the genetic algorithm is used to perform matrix optimization to obtain the optimal weights, which are then input into the BP neural network model. The error is calculated, the weight matrix parameters of each hidden layer are adjusted, and the prediction results are output.

[0018] Beneficial Effects: This invention adopts the above-mentioned scheme, using an intelligent learning algorithm to optimize the test evaluation matrix, reducing the selection of duplicate test points, maximizing the coverage of test points, achieving automatic selection of test points, reducing the influence of human factors, and more specifically analyzing the factors affecting the flowmeter's metering performance, thereby improving the reliability of flowmeter test evaluation. Simultaneously, the automatic selection of test points can be linked with the device's automatic control system, ultimately achieving comprehensive automation, improving the efficiency of device test evaluation, and shortening the calibration time for a single flowmeter from the original (3-4) days to (2-3) days. This invention makes the standard device more intelligent and digital in evaluation, resulting in more scientific and reasonable test evaluation results. Attached Figure Description

[0019] Figure 1 This is a system diagram of the automatic selection method for test points of the flowmeter standard device of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0022] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] Please see Figure 1 The present invention provides a technical solution: The technical content of the present invention patent is as follows:

[0025] A three-layer backpropagation (BP) neural network training model is combined with a genetic algorithm. The initial neural network model is built by determining the values ​​of each parameter in the network based on empirical formulas. Then, the initial weights are used as input to the genetic algorithm. After selection, crossover, and mutation operations, the fitness value is calculated, and the best set of weight matrices is input into the neural network model. The error is calculated using the test set in the database, the weights of each hidden layer are adjusted, and the trained model is used to output the prediction results.

[0026] To ensure the accuracy of neural network iteration results, a large amount of experimental data is required for calculation and analysis. This invention establishes a database required for solving multiphase flow parameters, extracts flow experimental data from the large database, and filters the required data to establish a multi-parameter database.

[0027] The parameters of the initial weight matrix of the BP neural network are optimized by a genetic algorithm, an initial population is randomly formed, and the selection operator, which obtains the fitness value of the population through expected evolution, obtains the weights of the BP neural network through exchange, mutation and multiple operations.

[0028] Using temperature, pressure, and phase fraction parameters in a multiphase flow testing device as the input layer of a neural network, the number of processing units in the intermediate layer of the network is adjusted by establishing the correspondence between the input parameters and the test flow rate points and the flow field coefficients in the pipe. The Monte Carlo gradient algorithm is used to search for local minima, and the model is trained by feedback calculation through the backpropagation algorithm of an artificial neural network that can iteratively approximate any function, thus obtaining the fitted curve.

[0029] Gradient search techniques are used to adjust the connection strength between input nodes and hidden layer nodes, and between hidden layer nodes and output nodes. The training function `train` is used to optimize the test matrix algorithm and adjust parameters such as the learning coefficients. The root mean square error (RMSE) is used as an evaluation function to assess the impact of each input variable on the neural network performance. Considering the stability and convergence of the iterative computation process, the Rectified Linear Unit (ReLU) function is used as the basic function to set the activation threshold, ensuring computational convergence during the iteration process.

[0030] Combining the above models can overcome the defect of backpropagation algorithm being prone to getting trapped in local extrema, while improving the performance of the test matrix and increasing the training speed. The neural network is combined with the pre-genetic algorithm. First, the genetic algorithm is used to optimize the matrix to obtain the optimal weights, which are then input into the BP neural network model. Then, the gradient descent algorithm is used to fine-tune the weight matrix parameters of each hidden layer, and the adjusted model is used to output the prediction results.

[0031] By adopting the above scheme and using intelligent learning algorithms to optimize the test evaluation matrix, the selection of duplicate test points is reduced, the test point coverage is maximized, and automatic selection of test points is achieved. This reduces the influence of human factors and allows for more targeted analysis of factors affecting flowmeter measurement performance, thereby improving the reliability of flowmeter test evaluation. Simultaneously, the automatic selection of test points can be linked with the device's automatic control system, ultimately achieving comprehensive automation and improving the efficiency of device test evaluation. The calibration time for a single flowmeter is shortened from the original (3-4) days to (2-3) days. This invention makes the standard device more intelligent and digital in evaluation, resulting in more scientific and reasonable test evaluation results.

[0032] In this embodiment, the technology of the present invention is applied to the actual liquid standard device of multiphase flowmeters of oil, gas and water in Daqing Oilfield. It provides an optimization scheme for the test evaluation matrix of the tested multiphase flowmeter and intelligent algorithm software, expands the applicable working conditions for multiphase flowmeter testing, improves the efficiency of test evaluation, and makes the test evaluation of multiphase flowmeters more accurate and reliable.

[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for selecting test points for a flowmeter device based on an optimized test matrix, characterized in that, Includes the following steps: S1: A three-layer BP neural network training model is combined with a genetic algorithm; S2: Establish the database required for solving the multiphase flow parameter model, extract the flow experimental data from the large database, filter the required data, and establish a multi-parameter database; S3: Optimize the parameters of the initial weight matrix of the BP neural network using a genetic algorithm, calculate the fitness of each individual using multiphase flow features in the database, determine the optimal weight matrix through fitness evaluation, and establish a genetic neural network point selection model. S4: Use the temperature, pressure, and phase fraction parameters in the multiphase flow testing device as the input layer of the neural network; Test points serve as output labels for model training; S5: Use gradient search to adjust the connection strength between input nodes and hidden nodes and between hidden nodes and output nodes. Use pre-trained weights to fine-tune the model. Use the mean squared error loss function to calculate the model loss and combine it with the Adam optimization algorithm to update the parameters. Use the training function train to optimize the test matrix algorithm and adjust parameters such as the learning coefficient. S6: Use the database to create a test dataset to evaluate the network model, and use the standard root mean square error as the evaluation metric.

2. The method for selecting test points for a flowmeter device based on an optimized test matrix according to claim 1, characterized in that, In step S1, the values ​​of each parameter in the network are determined according to empirical formulas to build an initial neural network model. Then, the initial weights are used as input to the genetic algorithm. After selection, crossover, and mutation operations, the fitness value is calculated to obtain the best set of weight matrices, which are then input into the neural network model. Gradient search technology is then used to fine-tune the model to improve its performance, adjust the weights of each hidden layer, and output the prediction results.

3. The method for selecting test points for a flowmeter device based on an optimized test matrix according to claim 1, characterized in that, In step S2, the database required for establishing the multiphase flow parameter solution model is established by extracting flow experimental data from the large database, filtering the required data, and establishing a multi-parameter database. This is to ensure that a large amount of experimental data is needed for calculation and analysis to ensure the accuracy of the neural network iteration results.

4. The method for selecting test points for a flowmeter device based on an optimized test matrix according to claim 1, characterized in that, In step S3, initial weights are randomly generated. The weights and topology of the neural network are encoded using binary encoding as the initial population input. The database in step S2 is used as the training set to calculate the fitness of each individual, which serves as the basis for the selection operator. The MSE function is used as the fitness function. The population is updated through roulette wheel selection, single-point crossover, and non-uniform mutation operations. The new population is used to evaluate fitness. After multiple calculations, the initial weights of the neural network are obtained.

5. The method for selecting test points for a flowmeter device based on an optimized test matrix according to claim 1, characterized in that, In step S4, the number of processing units in the intermediate layer of the connection model is established by the correspondence between the input parameters and the test flow points and the correspondence between the flow field coefficients in the pipeline. The Monte Carlo gradient algorithm is used to search for local minima, and the Adam optimization algorithm is used to update the parameters. The principle of iteratively approximating any function is performed by the backpropagation algorithm of the artificial neural network for feedback calculation.

6. The method for selecting test points for a flowmeter device based on an optimized test matrix according to claim 1, characterized in that, The standard root mean square error is used as an evaluation function for the impact of each input variable on the performance of the neural network. The loss function of the model is calculated using the training set, and the parameters of the model are adjusted by the Adam algorithm. Considering the stability and convergence of the iterative calculation process, the linear rectified function is used as the basic function to set the excitation threshold to ensure the convergence of the calculation during the iteration process.

7. The method for selecting test points for a flowmeter device based on an optimized test matrix as described in claim 1, characterized in that, In conjunction with the model described in claims 1-6, the neural network is combined with the pre-genetic algorithm. First, the genetic algorithm is used to optimize the matrix to obtain the optimal weights, which are then input into the BP neural network model. After the genetic algorithm initializes the weights, the model is optimized by GD fine-tuning. The weights are used as initial values, and the network parameters are updated by combining the Adam optimization algorithm to output the prediction results.

Images