Method for predicting high-pressure gas standard k coefficient of Coriolis mass flowmeter based on hybrid model
By establishing a gas scale k coefficient prediction model based on physical model and combining with neural network models for training, the problems of insufficient complexity and efficiency of high-pressure gas testing in the existing technology are solved, and efficient and accurate prediction of the gas scale k coefficient of the Coriolis mass flowmeter is achieved, and the quality control and efficiency of the production process are improved.
Patent Information
- Application Number
- CN202411604755.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-12
- Publication Date
- 2025-03-25
AI Technical Summary
In the prior art, obtaining the high-pressure gas scale k coefficient of the Coriolis mass flowmeter requires complex and difficult high-pressure gas tests, resulting in insufficient calculation efficiency and accuracy of the gas scale k coefficient.
By obtaining the water standard and gas standard data of multiple test samples, a prediction model based on physical models is established, combined with the neural network model, the gas standard k coefficient prediction model is trained to predict the gas standard k coefficient of the Coriolis mass flowmeter.
The prediction accuracy and calculation efficiency of the gas scale k coefficient are improved, the calibration process is simplified, the test time and cost are reduced, and the quality control accuracy and production efficiency in the production process are improved.
Smart Images
Figure CN119692162A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of flow meter calibration, and in particular to a method for predicting a high pressure gas calibration k coefficient of a Coriolis mass flow meter based on a hybrid model. Background Art
[0002] Different fluid properties will affect the measurement accuracy and performance of the Coriolis Mass Flow Meter. Therefore, the Coriolis Mass Flow Meter needs to be calibrated with water and gas. Water calibration is suitable for application scenarios where liquid mass flow needs to be measured, such as chemical, petroleum, pharmaceutical and other industries. The density of liquid is usually much higher than that of gas. As a standard liquid medium, water has a relatively stable density and is suitable for calibrating flow meters. Through water calibration, reliable measurement accuracy can be obtained under the working conditions of measuring liquid mass flow. Gas calibration is suitable for application scenarios where gas mass flow needs to be measured, such as the measurement of gas media such as natural gas and air. The density of gas is much lower than that of liquid, which means that the mass flow rate of gas will be lower at the same volume flow rate; gas is compressible and its density changes with temperature and pressure; the flow characteristics of gas are different from those of liquid, especially under high pressure and high temperature conditions, the flow characteristics of gas are more complicated. Therefore, gas calibration needs to be carried out under specific temperature and pressure conditions to ensure that the flow meter can accurately reflect the measurement results under actual working conditions.
[0003] In the prior art, it is necessary to select a water standard test or a gas standard test according to the application scenario of the flow meter, and calculate a constant coefficient k by calibrating the standard values at different flow points. The water standard and gas standard methods are the same, but the k coefficients of the water standard and gas standard are different. If you want to obtain the gas standard k coefficient, you must conduct a high-pressure gas standard experiment. The gas standard station has very high configuration requirements and is not easy to test, especially the high-pressure gas test.
[0004] Therefore, it is necessary to provide a method for predicting the high-pressure gas standard k-factor of a Coriolis mass flowmeter based on a hybrid model, so as to improve the calculation efficiency and accuracy of the gas standard k-factor. Summary of the invention
[0005] The present invention provides a method for predicting a high-pressure gas-mark k coefficient of a Coriolis mass flowmeter based on a hybrid model, comprising: obtaining a plurality of test samples, wherein the test samples include water-mark test data and gas-mark test data of a plurality of sample Coriolis mass flowmeters under the same test environment; establishing a physical model based on the plurality of test samples, wherein the independent variables of the physical model include a water-mark k coefficient and a correction coefficient, and the dependent variables of the physical model include a calculated value of the gas-mark k coefficient, and the correction coefficient is determined based on a plurality of water-mark parameters and a plurality of gas-mark parameters; establishing a gas-mark k coefficient prediction model; training the gas-mark k coefficient prediction model based on the physical model, wherein the gas-mark k coefficient prediction model is a neural network model, and the input of the gas-mark k coefficient prediction model includes operating parameters and a water-mark k coefficient; obtaining the current operating parameters and the water-mark k coefficient of the Coriolis mass flowmeter; and determining the gas-mark k coefficient of the Coriolis mass flowmeter according to the current operating parameters and the water-mark k coefficient of the Coriolis mass flowmeter through the water-mark k coefficient prediction model.
[0006] Furthermore, the water mark test data includes the water mark driving frequency, the water mark medium temperature and the medium pressure in the water mark tube; the gas mark test data includes the gas mark driving frequency, the gas mark medium temperature and the medium pressure in the gas mark tube.
[0007] Furthermore, determining the correction coefficient based on the multiple water mark parameters and the multiple gas mark parameters includes: establishing a multivariate regression sub-model, wherein the physical model includes the multivariate regression sub-model, the dependent variables of the multivariate regression sub-model include multiple water mark parameters and multiple gas mark parameters, the multiple water mark parameters include water mark driving frequency, water mark medium temperature and medium pressure in water mark tube, the multiple gas mark parameters include gas mark driving frequency, gas mark medium temperature and medium pressure in gas mark tube, and the independent variable of the multivariate regression sub-model is the correction coefficient; based on the multiple test samples, determining multiple fitting parameters of the multivariate regression sub-model; determining the correction coefficient based on the multiple water mark parameters and the multiple gas mark parameters through the multivariate regression sub-model.
[0008] Furthermore, the physical model is: Among them, k 气 k is the calculated value of the gas standard k coefficient, 水 is the water mark k coefficient, α1 is the correction coefficient, a, b and c are fitting parameters, f 气 is the gas standard driving frequency, f 水 is the water mark driving frequency, T 气 is the gas standard medium temperature, T 水 is the water standard medium temperature, P 气 is the medium pressure in the gas calibration tube, P 水 It is the medium pressure in the water mark pipe.
[0009] Furthermore, the operating parameters include operating water temperature, operating air temperature, operating water pressure, operating air pressure, water mark driving frequency and air mark driving frequency.
[0010] Furthermore, the gas standard k coefficient prediction model includes an input layer, a first layer, a second layer and an output layer, wherein the first layer includes 10 neurons, the second layer includes 20 neurons, and the third layer includes 1 neuron.
[0011] Furthermore, based on the physical model, a gas mark k coefficient prediction model is trained, including: S11, initializing the weights and bias values of the neural network; S12, obtaining training samples, forward propagating the input feature vectors of the training samples, and obtaining the outputs of the first layer, the second layer and the output layer of the gas mark k coefficient prediction model; S13, calculating the loss value based on the outputs of the physical model and the first layer, the second layer and the output layer of the gas mark k coefficient prediction model; S14, judging whether the preset end condition is met according to the loss value, if so, completing the training, if not, executing S15; S15, updating the weights and bias values of the first layer, the second layer and the output layer of the gas mark k coefficient prediction model according to the loss value through the stochastic gradient descent algorithm, and executing S12.
[0012] Furthermore, based on the outputs of the first layer, the second layer and the output layer of the physical model and the gas standard k coefficient prediction model, a loss value is calculated, including: for each training sample, calculating the gas standard k coefficient calculation value corresponding to the training sample through the physical model; calculating the error penalty value corresponding to the training sample according to the gas standard k coefficient prediction value output by the output layer of the gas standard k coefficient prediction model and the gas standard k coefficient calculation value corresponding to the training sample; calculating the loss value according to the error penalty value corresponding to the training sample.
[0013] Furthermore, the loss function of the gas standard k coefficient prediction model is: Among them, L(W,b) is the loss value, M is the total number of training samples included in a training batch, is the mean square error between the predicted value and the actual value, is the predicted value of the gas standard k coefficient of the i-th training sample output by the output layer of the gas standard k coefficient prediction model, y i is the true value of the gas standard k coefficient of the i-th training sample, ||W‖2+‖b‖2 is the regularization term, W is the bias matrix, b is the weight vector, is the error penalty term, k 气 is the calculated value of the gas standard k coefficient corresponding to the training sample calculated by the physical model, λ1 is the regularization coefficient, and λ2 is the physical model similarity penalty coefficient.
[0014] Furthermore, the weights and bias values of the first layer, the second layer and the output layer of the gas standard k coefficient prediction model are updated based on the following formula: Among them, W i is the weight matrix of the i-th layer, b i is the bias value of the i-th layer, is the derivative of the loss value with respect to the weight of the i-th layer, is the derivative of the loss value with respect to the bias value of the i-th layer, and α2 is the learning rate.
[0015] Compared with the prior art, the method for predicting the high-pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model provided by the present invention has at least the following beneficial effects:
[0016] 1. This method combines the advantages of physical models and neural network models. It introduces physical laws and mechanisms through physical models, and then learns and fits complex nonlinear relationships through neural network models, thereby improving the prediction accuracy of gas standard k coefficient. The neural network model has good generalization ability and can accurately predict new and unseen operating parameters and water standard k coefficients by learning the data characteristics of a large number of test samples, so that the water standard k coefficient prediction model can maintain a high prediction accuracy under different environments and conditions. The calibration process of the existing Coriolis mass flowmeter may require complex tests and measurements, and this method can greatly simplify the calibration process and reduce test time and cost through the prediction of the water standard k coefficient prediction model. An accurate water standard k coefficient prediction model can make the quality control in the production process more accurate and efficient, avoid production delays or product quality problems caused by inaccurate calibration, and thus improve overall production efficiency. Through the prediction of the water standard k coefficient prediction model, enterprises can more accurately understand the needs of the production process, thereby optimizing resource allocation, reducing unnecessary waste, and improving resource utilization efficiency.
[0017] 2. By including the water mark driving frequency, water mark medium temperature, medium pressure in the water mark tube, and the corresponding gas mark parameters, this method can more comprehensively consider the factors affecting the performance of the Coriolis mass flowmeter, thereby improving the accuracy of the prediction. By establishing a multivariate regression sub-model to determine the correction coefficient, this method can more accurately reflect the relationship between the water mark parameters and the gas mark parameters, thereby further improving the accuracy of the prediction. The physical model clarifies the relationship between the water mark k-factor, the gas mark k-factor, and the correction coefficient, making the prediction process more transparent and explainable.
[0018] 3. This method can make full use of the advantages of neural networks in dealing with complex nonlinear relationships and improve the accuracy and generalization of predictions. The loss function not only considers the mean square error between the predicted value and the actual value, but also introduces regularization terms and error penalty terms to balance the complexity of the model and the prediction accuracy, prevent overfitting, and ensure the consistency of the model output with the calculated value of the physical model. Through clear training steps and conditional judgments (such as whether the preset end conditions are met), this method can ensure that the neural network model is fully trained while avoiding unnecessary waste of computing resources. This method takes into account a variety of operating parameters, so that the water standard k coefficient prediction model can be applied to the calibration and prediction of Coriolis mass flowmeters under different environments and conditions, which improves the practicality and flexibility of the water standard k coefficient prediction model. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:
[0020] Figure 1 is a flow chart of a method for predicting a high pressure gas standard k-factor of a Coriolis mass flowmeter based on a hybrid model according to some embodiments of this specification;
[0021] Figure 2 is a schematic diagram of the structure of the gas standard k coefficient prediction model shown in some embodiments of this specification;
[0022] Figure 3 is a schematic diagram of the change of loss value in the process of training the gas standard k coefficient prediction model according to some embodiments of this specification;
[0023] Figure 4 Schematic diagram of the prediction error of the gas standard k coefficient according to some embodiments of the present specification. DETAILED DESCRIPTION
[0024] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.
[0025] Figure 1 is a flow chart of a method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to some embodiments of this specification, such as Figure 1As shown, the method for predicting the high-pressure gas standard k-factor of a Coriolis mass flowmeter based on a hybrid model may include the following steps.
[0026] Step 110, obtaining a plurality of test samples.
[0027] The test samples include water calibration test data and gas calibration test data of multiple sample Coriolis mass flowmeters under the same test environment, that is, the test environment for the same sample Coriolis mass flowmeter during water calibration and gas calibration is the same. The test environment of any two sample Coriolis mass flowmeters can be different. The water calibration test data includes the water calibration drive frequency, the water calibration medium temperature, the medium pressure in the water calibration tube and the water calibration k coefficient. The gas calibration test data includes the gas calibration drive frequency, the gas calibration medium temperature, the medium pressure in the gas calibration tube and the gas calibration k coefficient.
[0028] Specifically, the water standard k coefficient or the gas standard k coefficient is calculated by calibrating the standard values at different flow points. m It is proportional to the phase time difference Δt between the upstream and downstream signals, as follows:
[0029] Q m =k(Δt+Δt0)=kΔt+Q m0
[0030] Among them, Δt0 is the zero point offset time. When the flow rate is zero, the phase time difference caused by factors such as system error reflects the error of the flow meter near the zero point, Q m0 It is the current zero value of the Coriolis mass flowmeter.
[0031] The specific calibration method is as follows:
[0032] The mass flow rate Q of the standard device at different mass flow points is measured mi , and record the phase time difference Δt of the upstream and downstream signals of the Coriolis mass flowmeter at the corresponding flow point i ;
[0033] Calculate the water scale k factor or gas scale k factor and zero point Q using the least squares method m0 :
[0034]
[0035] Where n is the number of measurement points.
[0036] The main calibration parameters of the Coriolis mass flowmeter include the calibrated mass flow point Q m, operating frequency f and phase time difference Δt. The calibrated flow points usually include maximum flow, minimum flow, 25% of maximum flow, 50% of maximum flow and 75% of maximum flow. The operating frequency refers to the vibration frequency of the symmetrical elbow in the Coriolis mass flowmeter. The phase time difference refers to the time difference between the zero crossing points of the phases of the two upstream and downstream detection coils. By accurately controlling and measuring these parameters, the measurement accuracy and reliability of the flowmeter under different flow conditions can be ensured.
[0037] Step 120: Establish a physical model based on multiple test samples.
[0038] The independent variables of the physical model include a water-mark k coefficient and a correction coefficient, the dependent variables of the physical model include a calculated value of a gas-mark k coefficient, and the correction coefficient is determined based on a plurality of water-mark parameters and a plurality of gas-mark parameters.
[0039] Preferably, the correction coefficient is determined based on a plurality of water mark parameters and a plurality of gas mark parameters, including:
[0040] Establish a multivariate regression sub-model, wherein the physical model includes the multivariate regression sub-model, the dependent variables of the multivariate regression sub-model include multiple water mark parameters and multiple gas mark parameters, the multiple water mark parameters include the water mark driving frequency, the water mark medium temperature and the medium pressure in the water mark tube, the multiple gas mark parameters include the gas mark driving frequency, the gas mark medium temperature and the medium pressure in the gas mark tube, and the independent variable of the multivariate regression sub-model is the correction coefficient;
[0041] Based on multiple experimental samples, multiple fitting parameters of the multivariate regression sub-model are determined;
[0042] The correction coefficients are determined based on multiple water-marked parameters and multiple gas-marked parameters through a multivariate regression submodel.
[0043] As a preferred embodiment, the physical model is:
[0044]
[0045] Among them, k 气 k is the calculated value of the gas standard k coefficient, 水 is the water mark k coefficient, α1 is the correction coefficient, a, b and c are fitting parameters, f 气 is the gas standard driving frequency, f 水 is the water mark driving frequency, T 气 is the gas standard medium temperature, T 水 is the water standard medium temperature, P 气 is the medium pressure in the gas calibration tube, P 水 It is the medium pressure in the water mark pipe.
[0046] The maximum accuracy of the physical model under 500 test samples is 0.43%, and the average error is 0.3%.
[0047] Step 130, establishing a gas standard k-factor prediction model.
[0048] In order to further improve the prediction accuracy and generalization ability, the gas standard k coefficient prediction model is introduced. Figure 2 is a schematic diagram of the structure of the gas standard k coefficient prediction model shown in some embodiments of this specification, such as Figure 2 As shown, preferably, the gas standard k coefficient prediction model includes an input layer, a first layer, a second layer and an output layer, wherein the first layer includes 10 neurons, the second layer includes 20 neurons, and the third layer includes 1 neuron.
[0049] Step 140: training a gas standard k-factor prediction model based on the physical model.
[0050] The gas-mark k-coefficient prediction model is a neural network model, and the input of the gas-mark k-coefficient prediction model includes operating parameters and water-mark k-coefficient. The operating parameters include operating water temperature, operating air temperature, operating water pressure, operating air pressure, water-mark driving frequency, and gas-mark driving frequency.
[0051] In order to avoid the larger eigenvalues "eating up" the smaller eigenvalues, it is very necessary to keep the features at the same scale. This paper uses the normalization method to keep each input feature quantity at the same level, and the range of variation is kept between 0 and 1. For ease of description, the input feature vector is expressed as:
[0052] X (i) ={d1,d2,d3,d4,d5,d6,d7} i
[0053] Among them, X (i) is the input feature vector of the i-th training sample, d1 is the operating water temperature, d2 is the operating air temperature, d3 is the operating water pressure, d4 is the operating air pressure, d5 is the water mark driving frequency, d6 is the air mark driving frequency, and d7 is the water mark k coefficient.
[0054] Preferably, based on the physical model, the gas standard k coefficient prediction model is trained, including:
[0055] S11, initialize the neural network weights and bias values;
[0056] S12, obtaining training samples, forward propagating the input feature vectors of the training samples, obtaining the outputs of the first layer, the second layer and the output layer of the gas standard k coefficient prediction model, the input layer inputs the input feature vectors of the training samples into the gas standard k coefficient prediction model; the hidden layer trains and fits the neuron parameters, that is, the parameters of the gas standard k coefficient prediction model are trained in the direction of gradient descent, these parameters are mainly weights and biases, and can also include the selection of activation functions, etc., and the ReLU function is selected as the activation function;
[0057] S13, calculating the loss value based on the output of the first layer, the second layer and the output layer of the physical model and the gas standard k coefficient prediction model;
[0058] S14, judging whether the preset end condition is met according to the loss value, if so, completing the training, if not, executing S15;
[0059] S15. Update the weights and bias values of the first layer, the second layer and the output layer of the gas standard k coefficient prediction model according to the loss value through the stochastic gradient descent algorithm, and execute S12.
[0060] Specifically, the input feature vector of the training sample is forward propagated, and the outputs of the first layer, second layer and output layer of the gas standard k coefficient prediction model are obtained as follows:
[0061]
[0062] Among them, a 1 is the output of the first layer of the gas standard k coefficient prediction model, W 1 is the weight of the gas standard k coefficient prediction model, b 1 is the bias of the first layer of the gas standard k coefficient prediction model, ReLU() is the ReLU function, a 2 is the output of the second layer of the gas standard k coefficient prediction model, W 2 is the weight of the second layer of the gas standard k coefficient prediction model, b 2 is the bias of the second layer of the gas standard k coefficient prediction model, is the output of the output layer of the gas standard k coefficient prediction model, W 3 is the weight of the output layer of the gas standard k coefficient prediction model, b 3 The bias of the output layer of the gas standard k coefficient prediction model.
[0063] Preferably, the loss value is calculated based on the output of the first layer, the second layer and the output layer of the physical model and the gas standard k coefficient prediction model, including:
[0064] For each training sample, the gas standard k coefficient calculation value corresponding to the training sample is calculated through the physical model;
[0065] Calculate the error penalty value corresponding to the training sample according to the gas standard k coefficient prediction value output by the output layer of the gas standard k coefficient prediction model and the gas standard k coefficient calculation value corresponding to the training sample;
[0066] Calculate the loss value based on the error penalty value corresponding to the training sample.
[0067] Preferably, the loss function of the gas standard k coefficient prediction model is:
[0068]
[0069] Among them, L(W,b) is the loss value, M is the total number of training samples included in a training batch, is the mean square error between the predicted value and the actual value, is the predicted value of the gas standard k coefficient of the i-th training sample output by the output layer of the gas standard k coefficient prediction model, y i is the true value of the gas standard k coefficient of the i-th training sample, ||W||2+||b||2 is the regularization term, W is the bias matrix, b is the weight vector, is the error penalty term, which is used to constrain the output of the gas standard k coefficient prediction model to conform to the trend of the physical model. 气 is the calculated value of the gas-standard k coefficient corresponding to the training sample calculated by the physical model, λ1 is the regularization coefficient, and λ2 is the physical model similarity penalty coefficient. As an example only, the values of λ1 and λ2 can be 0.01.
[0070] Preferably, the weights and bias values of the first layer, the second layer and the output layer of the gas standard k coefficient prediction model are updated based on the following formula:
[0071]
[0072] Among them, W i is the weight matrix of the i-th layer, b i is the bias value of the i-th layer, is the derivative of the loss value with respect to the weight of the i-th layer, is the derivative of the loss value with respect to the bias value of the i-th layer, and α2 is the learning rate.
[0073] The loss value is finally converged in the direction of decreasing, and S12-S15 are repeatedly iterated to finally converge the loss value to a smaller value, wherein the preset end condition can be that the loss value converges to the preset loss value. At the same time, the change trend of the sample prediction loss value in the validation set is constantly observed during the training process. Figure 3 is a schematic diagram of the change of loss value during the training of the gas standard k coefficient prediction model shown in some embodiments of this specification, such as Figure 3 As shown in the figure, 500 experimental samples are divided into training set and validation set at a ratio of 9:1. After repeating the experimental samples for 300 times, the training is stopped. At this time, the loss values of the training set and validation set are already small, indicating that a strong generalization ability has been achieved. The blue line represents the change in the loss value of the training set, and the red line represents the change in the loss value of the validation set.
[0074] Step 150, obtaining the current operating parameters and water mark k coefficient of the Coriolis mass flowmeter.
[0075] Step 160 , determining the gas scale k coefficient of the Coriolis mass flowmeter according to the current operating parameters of the Coriolis mass flowmeter and the water scale k coefficient through a water scale k coefficient prediction model.
[0076] The beneficial effects of the method for predicting the high-pressure gas standard k-factor of the Coriolis mass flowmeter based on the hybrid model are explained below in combination with experimental data.
[0077] Figure 4 Schematic diagram of the prediction error of the gas-mark k coefficient according to some embodiments of the present specification. After the water-mark k coefficient prediction model is trained, the water-mark k coefficient prediction model is evaluated. The prediction error of the gas-mark k coefficient of the water-mark k coefficient prediction model for 50 samples that have not participated in the training is as follows: Figure 4 The maximum error of the gas-standard k-factor predicted by the water-standard k-factor prediction model is within 0.2%, which is more accurate than the physical model. In addition, without any gas-standard experiments, the prediction accuracy of the gas-standard k-factor meets the 0.2 level.
[0078] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.
Claims
1. A method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model, characterized in that: include: Acquire multiple test samples, wherein the test samples include water-standard test data and gas-standard test data of multiple sample Coriolis mass flowmeters under the same test environment; Based on the multiple test samples, a physical model is established, wherein the independent variables of the physical model include a water scale k coefficient and a correction coefficient, the dependent variable of the physical model includes a calculated value of a gas scale k coefficient, and the correction coefficient is determined based on multiple water scale parameters and multiple gas scale parameters; Establish a gas standard k-factor prediction model; Based on the physical model, a gas standard k coefficient prediction model is trained, wherein the gas standard k coefficient prediction model is a neural network model, and the input of the gas standard k coefficient prediction model includes operating condition parameters and water standard k coefficient; Obtain the current operating parameters and water mark k coefficient of the Coriolis mass flowmeter; The gas scale k coefficient of the Coriolis mass flowmeter is determined by the water scale k coefficient prediction model according to the current operating parameters of the Coriolis mass flowmeter and the water scale k coefficient.
2. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to claim 1, characterized in that: The water mark test data includes the water mark driving frequency, the water mark medium temperature and the medium pressure in the water mark pipe; The gas mark test data includes the gas mark driving frequency, the gas mark medium temperature and the medium pressure in the gas mark tube.
3. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to claim 2, characterized in that: Determining the correction coefficient based on the multiple water mark parameters and the multiple gas mark parameters includes: Establishing a multivariate regression sub-model, wherein the physical model includes the multivariate regression sub-model, the dependent variables of the multivariate regression sub-model include multiple water mark parameters and multiple gas mark parameters, the multiple water mark parameters include water mark driving frequency, water mark medium temperature and medium pressure in water mark tube, the multiple gas mark parameters include gas mark driving frequency, gas mark medium temperature and medium pressure in gas mark tube, and the independent variable of the multivariate regression sub-model is a correction coefficient; Based on the multiple test samples, determining multiple fitting parameters of the multivariate regression sub-model; The correction coefficient is determined based on the multiple water mark parameters and the multiple air mark parameters by the multiple regression sub-model.
4. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to claim 3, characterized in that: The physical model is: Among them, k 气 k is the calculated value of the gas standard k coefficient, 水 is the water mark k coefficient, α1 is the correction coefficient, a, b and c are fitting parameters, f 气 is the gas standard driving frequency, f 水 is the water mark driving frequency, T 气 is the gas standard medium temperature, T 水 is the water standard medium temperature, P 气 is the medium pressure in the gas calibration tube, P 水 It is the medium pressure in the water mark pipe.
5. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to any one of claims 1 to 4, characterized in that: The operating parameters include operating water temperature, operating air temperature, operating water pressure, operating air pressure, water mark driving frequency and air mark driving frequency.
6. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to any one of claims 1 to 4, characterized in that: The gas standard k coefficient prediction model includes an input layer, a first layer, a second layer and an output layer, wherein the first layer includes 10 neurons, the second layer includes 20 neurons, and the third layer includes 1 neuron.
7. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to claim 6, characterized in that: Based on the physical model, a gas standard k coefficient prediction model is trained, including: S11, initialize the neural network weights and bias values; S12, obtaining training samples, forward propagating the input feature vectors of the training samples, and obtaining the outputs of the first layer, the second layer and the output layer of the gas standard k coefficient prediction model; S13, calculating the loss value based on the output of the first layer, the second layer and the output layer of the physical model and the gas standard k coefficient prediction model; S14, judging whether the preset end condition is met according to the loss value, if so, completing the training, if not, executing S15; S15. Update the weights and bias values of the first layer, the second layer and the output layer of the gas standard k coefficient prediction model according to the loss value through the stochastic gradient descent algorithm, and execute S12.
8. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to claim 7, characterized in that: Based on the outputs of the first layer, the second layer and the output layer of the physical model and the gas standard k coefficient prediction model, the loss value is calculated, including: For each training sample, a calculated value of the gas standard k coefficient corresponding to the training sample is calculated by the physical model; Calculate the error penalty value corresponding to the training sample according to the gas standard k coefficient prediction value output by the output layer of the gas standard k coefficient prediction model and the gas standard k coefficient calculation value corresponding to the training sample; The loss value is calculated according to the error penalty value corresponding to the training sample.
9. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to any one of claims 1 to 4, characterized in that: The loss function of the gas standard k coefficient prediction model is: Among them, L(W,b) is the loss value, M is the total number of training samples included in a training batch, is the mean square error between the predicted value and the actual value, is the predicted value of the gas standard k coefficient of the i-th training sample output by the output layer of the gas standard k coefficient prediction model, y i is the true value of the gas standard k coefficient of the i-th training sample, ||W||2+||b||2 is the regularization term, W is the bias matrix, b is the weight vector, is the error penalty term, k 气 is the calculated value of the gas standard k coefficient corresponding to the training sample calculated by the physical model, λ1 is the regularization coefficient, and λ2 is the physical model similarity penalty coefficient.
10. The method for predicting the high pressure gas standard k coefficient of a Coriolis mass flowmeter based on a hybrid model according to any one of claims 1 to 4, characterized in that: The weights and bias values of the first layer, the second layer and the output layer of the gas standard k coefficient prediction model are updated based on the following formula: Among them, W i is the weight matrix of the i-th layer, b i is the bias value of the i-th layer, is the derivative of the loss value with respect to the weight of the i-th layer, is the derivative of the loss value with respect to the bias value of the i-th layer, and α2 is the learning rate.