A method for predicting viscosity of oil-water emulsion based on neural network

Abstract

The present invention relates to the field of oil-water emulsification, and discloses a method for predicting the viscosity of an oil-water emulsion based on a neural network. The present invention adopts a method combining experiments with machine learning to provide a dynamic prediction method for the viscosity of oil-water stirred emulsions under different shear rates, temperatures, and water contents, adopts a conical plate variable volume turbulent drag reduction evaluation device designed independently to stir and prepare oil-water emulsions under different water contents, and obtains the viscosity of the oil-water emulsion under multi-factor conditions through real-time testing, and predicts and evaluates the viscosity of the oil-water emulsion by combining a Bayesian regularization algorithm with a BP neural network algorithm, thereby solving the problems of low prediction accuracy and poor universality of the viscosity of the oil-water emulsion, and improving the prediction accuracy of the viscosity of the oil-water emulsion. The method is suitable for multi-variable, large-sample oil-water emulsification, and has good industrial guidance significance and application prospects for oilfield exploitation and crude oil processing.

Description

Technical Field

[0001] The invention relates to the field of oil-water emulsification, and in particular to a method for predicting the viscosity of an oil-water emulsion based on a neural network during oilfield exploitation. Background Art

[0002] In the field of oilfield exploitation, oil-water emulsions (oil-in-water, water-in-oil, or oil-in-water-in-oil emulsions) are usually formed during oilfield water injection and chemical flooding. Accurate prediction of the apparent viscosity of oil-water emulsions will directly affect the power and heat consumption of subsequent exploitation processes. Therefore, accurate prediction of the viscosity of oil-water emulsions is of great significance for reducing energy consumption during the exploitation process.

[0003] The apparent viscosity of oil-water emulsion is related to factors such as shear rate, temperature, relative content of two phases, pH value, inorganic salts, and mineralization. At present, the research on oil-water emulsification mainly adopts a combination of experimental and theoretical methods. In the experimental part, oil-water emulsions with different water contents are prepared in a stirring device, and a certain amount of prepared oil-water emulsions are added to a high-temperature and high-pressure rheometer. The apparent viscosity of the oil-water emulsion is tested by controlling the high-temperature and high-pressure rheometer under different shear rates, temperatures, pressures, etc., and then the influence of water content, shear rate, temperature, and pressure on the viscosity of the oil-water emulsion is analyzed, and finally the apparent viscosity of the emulsion is predicted. In the theoretical part, the relationship between the apparent viscosity of the emulsion and each influencing factor is determined based on experimental data, and linear or nonlinear regression is performed to obtain a prediction model for the viscosity of the emulsion. However, due to the many factors affecting the apparent viscosity of the oil-water emulsion and the complex change rules, the existing emulsion viscosity prediction model has limited applicability, is not universal, and there is also an inaccurate viscosity prediction. Therefore, for the prediction problem of oil-water emulsion viscosity, developing a universal emulsion viscosity prediction method based on neural network has important theoretical value and guiding significance. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of existing oil-water emulsion prediction methods and provide an oil-water emulsion viscosity prediction method based on a BP neural network with a Bayesian regularization algorithm. The problems of low prediction accuracy and poor universality of oil-water emulsion viscosity prediction are solved by combining the Bayesian regularization algorithm with the BP neural network algorithm.

[0005] In order to solve the above technical problems, the present invention provides a method for predicting the viscosity of an oil-water emulsion based on a neural network, comprising the following steps:

[0006] (1) An oil-water emulsification experiment was carried out using a self-designed conical plate variable volume turbulent drag reduction evaluation device to stir and prepare oil-water emulsions under different water contents. The schematic diagram and actual picture of the device are shown in Figure 1-2 As shown;

[0007] (2) Control the rotation speed and water bath temperature of the conical plate in the conical plate variable volume turbulence drag reduction evaluation device, so as to measure the viscosity of the oil-water emulsion under different water content, shear rate and temperature conditions in real time;

[0008] (3) Take 50 ml of the stable oil-water emulsion formed by stirring and add it to a high-temperature and high-pressure rheometer. Set the temperature and shear rate to be the same as those of the stirring device (i.e., the conical plate variable volume turbulent drag reduction evaluation device) used to prepare the emulsion, so as to test the viscosity of the oil-water emulsion under different pressure conditions;

[0009] (4) Record and organize the viscosity of oil-water emulsions under different water contents, shear rates, and temperatures obtained from experimental tests, and establish a test set for machine learning;

[0010] (5) Analyze the data in the test set and preprocess the data. First, logarithmize the viscosity data of the oil-water emulsion, and then normalize the shear rate, temperature, water content, and viscosity.

[0011] (6) Process the data in the test set. After normalization, randomly select 80% of the test set as the training set and 20% of the test set as the validation set according to the 8:2 principle.

[0012] (7) Determine the number of neurons in the input layer, output layer, and hidden layer;

[0013] (8) Setting the activation function of the Bayesian-BP neural network;

[0014] (9) Prediction value restoration of Bayesian-BP neural network;

[0015] (10) Establish the neural network structure, see Figure 3 As shown;

[0016] (11) Establish and train the Bayesian-BP neural network model. The training process is shown in Figure 4 As shown;

[0017] (12) Verification and evaluation of the Bayesian-BP neural network model. The mean square error iteration effect is shown in Figure 5 As shown;

[0018] (13) Prediction and evaluation of oil-water emulsion viscosity. The prediction results are shown in Figure 6-11 shown.

[0019] In step (5), logarithmic processing is performed on the viscosity data in the test set, and the calculation expression is:

[0020]

[0021] Where η iis the viscosity value of the i-th oil-water emulsion, is the viscosity value after logarithmic processing.

[0022] In step (5), the shear rate, temperature, water content and the viscosity of the oil-water emulsion after logarithmic processing in the test set are uniformly normalized, and the calculation expression is:

[0023]

[0024] In the formula, x i is the shear rate, temperature, moisture content or viscosity after logarithmic processing of the i-th test set; x max 、x min are the maximum and minimum values ​​of the experimental sample data respectively; a and b are constants, which are 1 and -1 respectively; y i is the normalized value of the i test sets. After normalization, the data is in the range of [-1,1].

[0025] In step (7), the preset range of the number of neurons in the hidden layer is preliminarily determined by the following formula:

[0026]

[0027] Where n is the number of neurons in the input layer of the neural network; m is the number of neurons in the hidden layer of the neural network; l is the number of neurons in the output layer of the neural network; α is a constant, ranging from [1,10].

[0028] In step (8), the S-type function is used as the activation function in the neural network, and its calculation expression is:

[0029]

[0030] The step (9) is to restore the value of the output neuron of the Bayesian-BP neural network, and its calculation expression is:

[0031]

[0032] In the formula, is the neuron value of the output layer of the i-th test set; are the maximum and minimum values ​​of the output layer neurons in the test set, respectively; a and b are constants, which are 1 and -1, respectively; Predicted emulsion viscosity for the ith test set.

[0033] The step (10) adopts a three-layer neural network, namely, an input layer, a hidden layer, and an output layer. Set the input vector to x = (x1, x2, x3) T , where x i =(xi1 ,x i2 ,...,x ij ), x i is the numerical vector of the i-th (i=1, 2, 3, respectively representing shear rate, temperature, and moisture content) test set after preprocessing, x ij is the value of the i-th test set preprocessed in the j-th test set. Set the output variable to y = (y1, y2, ..., y j ), where y j is the predicted value of the jth test set. The execution process between the input layer and the hidden layer is Where j = 1, 2, ..., 6, is the activation function determined in step (8), ω ij is the connection weight between the input layer and the hidden layer. The execution process between the hidden layer and the output layer is Among them, α and β are the first fitting parameter and the second fitting parameter respectively, ω k is the connection weight between the hidden layer and the output layer.

[0034] Compared with the existing prediction method of oil-water emulsion viscosity mathematical regression, the beneficial effect of the present invention is that it solves the problems of many influencing factors and complex changing rules of oil-water emulsion viscosity, and adopts a Bayesian regularization method combined with a BP neural network algorithm to solve the problems of low prediction accuracy and poor universality of oil-water emulsion viscosity prediction, thereby improving the prediction accuracy of oil-water emulsion viscosity. In addition, the method is suitable for multi-variable and large-sample oil-water emulsification situations, and has good industrial guiding significance and application prospects for oilfield exploitation and crude oil processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Purpose of the drawings: In order to more clearly illustrate the embodiments and technical solutions of the present invention, the drawings required for use in the embodiments will be simply marked and introduced below.

[0036] Figure 1 It is a schematic diagram of the structure of the conical plate variable volume test domain system and the torque test system.

[0037] Figure 2 This is a real-life picture of a conical plate variable volume turbulence drag reduction evaluation device.

[0038] Figure 3 It is a neural network structure diagram for predicting the viscosity of oil-water emulsion.

[0039] Figure 4 It is a training flow chart for oil-water emulsion viscosity prediction.

[0040] Figure 5 It is the iteration graph of the mean square error of the training set.

[0041] Figure 6 This is a comparison chart of the relationship between viscosity and water content of the training set.

[0042] Figure 7 It is the predicted regression plot of the training set.

[0043] Figure 8 is the prediction error histogram of the training set.

[0044] Fig. 9 This is a comparison chart of the relationship between the viscosity and shear rate of the verification set.

[0045] Fig.10 It is the predicted regression plot of the validation set.

[0046] Fig.11 is the prediction error histogram of the validation set.

[0047] Figure 1 In the figure: (1) variable rotating conical circular plate; (2) variable cylindrical test domain; (3) variable cylindrical infiltrant; (4) open transparent polycarbonate cylindrical bowl; (5), (6), (7) stainless steel sleeve; (8) coupling, (9) rotating shaft, (10) torque sensor, (11) rotating shaft, (12) temperature sensor, (13) cone-plate rotating shaft. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of the specification, not all of the embodiments. Based on one or more embodiments of the specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the embodiments of this specification.

[0049] The instruments, reagents, materials, software, etc. involved in the following embodiments, unless otherwise specified, are all conventional instruments, reagents, materials, etc. in the prior art and can be obtained through regular commercial channels. The experimental methods, detection methods, etc. involved in the following embodiments, unless otherwise specified, are all conventional experimental methods, detection methods, etc. in the prior art.

[0050] The present invention will be further described below:

[0051] The present invention adopts a method combining experiments and machine learning to provide a dynamic prediction method for the viscosity of oil-water emulsion under different shear rates, temperatures, and water contents. The specific implementation method is as follows:

[0052] 1. Oil-water emulsification device and experimental implementation method:

[0053] The conical plate variable volume turbulence drag reduction evaluation device, wherein the variable cylindrical test domain (2) has a diameter of 134 mm and a height of 19 mm, the variable rotating conical circular plate (1) has a diameter of 130 mm, an edge thickness of 1 mm, and a thickness of 2 mm at the center of the conical circular plate. The device geometry is shown in FIG. Figure 1 During the experiment, the stirring speed n (or shear rate) of the motor (11) is controlled by a computer. ), and after the oil-water emulsification is completed, the torque sensor (10) feeds back the real-time torque value M (or shear stress τ), thereby calculating the apparent viscosity η of the emulsion, and the calculation expression is:

[0054]

[0055] (1) An oil-water emulsification experiment was carried out using a self-designed conical plate variable volume turbulent drag reduction evaluation device to stir and prepare oil-water emulsions with different water contents.

[0056] Before the experiment, prepare the oil-water emulsions with different water contents (40% to 90%). To ensure the accuracy of the experimental test, the crude oil needs to be dehydrated before the experiment. =6000s -1 , set the water bath test temperature (T = 40 ~ 80 ° C), the sampling interval is 5s, and start stirring after standing for 5 minutes. After stirring for 10 minutes, the oil-water emulsification in the transparent variable cylindrical test area (2) is completed.

[0057] (2) The rotation speed and water bath temperature of the conical plate in the conical plate variable volume turbulence drag reduction evaluation device are controlled to obtain the viscosity of the oil-water emulsion under different water content, shear rate and temperature conditions in real time.

[0058] Change the shear rate between 0 and 1000 s -1 Within the range, the shear rate is gradually changed according to the linear law, from low to high or from high to low, and the test obtains the change data of fluid shear stress and shear viscosity with shear rate. The data results are analyzed to obtain the change relationship between the viscosity of oil-water emulsion and different shear rates, different water contents, and different temperatures. Among them, the shear stress τ is calculated by the anti-torque value M measured by the torque sensor, and the shear rate It can be calculated by the rotation speed n of the cone plate. In the cone plate variable volume turbulence drag reduction evaluation device, its calculation expression is:

[0059]

[0060] The viscosity of 192 groups of oil-water emulsions under different water content, shear rate and temperature conditions obtained from experimental tests were recorded and organized.

[0061] (3) 50 ml of the stable oil-water emulsion formed by stirring was added to a high-temperature and high-pressure rheometer to measure the viscosity of the oil-water emulsion under different pressure conditions.

[0062] If it is necessary to consider the effect of different pressures on the viscosity of the oil-water emulsion under formation conditions, take 50 ml of the stable oil-water emulsion formed by stirring and add it to the high-temperature and high-pressure rheometer. Set the temperature and shear rate to be the same as those of the stirring device (i.e., the conical plate variable volume turbulent drag reduction evaluation device) used to prepare the emulsion originally, so as to test and obtain the viscosity values ​​of the oil-water emulsion under different pressure conditions.

[0063] (4) Record and organize the viscosity of oil-water emulsions under different water contents, shear rates, and temperatures obtained from experimental tests, and establish a test set for machine learning.

[0064] 2. Implementation method of oil-water emulsion viscosity prediction based on neural network:

[0065] In order to overcome the shortcomings of existing oil-water emulsion prediction methods, the present invention provides an oil-water emulsion viscosity prediction method using a BP neural network with a Bayesian regularization algorithm, which solves the problems of low oil-water emulsion viscosity prediction accuracy and poor universality by combining the Bayesian regularization algorithm with the BP neural network algorithm.

[0066] In order to solve the above technical problems, the specific implementation process of the Bayesian-BP neural network model, such as establishment, training, and prediction, is as follows:

[0067] (5) Analyze the data in the test set and pre-process the data. First, logarithmically process the viscosity data of the oil-water emulsion, and then normalize the shear rate, temperature, water content, and viscosity.

[0068] Based on the viscosity data of oil-water emulsion under different water content, shear rate and temperature conditions obtained from the oil-water emulsification experiment, the test set is preprocessed. Considering that the viscosity of the oil-water emulsion is a non-negative number, in order to avoid the predicted viscosity being a negative number during subsequent model training, the viscosity of the test set is logarithmically processed using the following formula:

[0069]

[0070] Where η i is the viscosity value of the i-th oil-water emulsion, is the viscosity value after logarithmic processing.

[0071] In order to eliminate the influence of dimension and speed up the convergence, the shear rate, temperature, water content and the viscosity of the oil-water emulsion after logarithmic processing of the test set are normalized using the following formula:

[0072]

[0073] In the formula, x i is the shear rate, temperature, moisture content or viscosity after logarithmic processing of the i-th test set; x max 、x min are the maximum and minimum values ​​of the experimental sample data respectively; a and b are constants, which are 1 and -1 respectively; y i is the normalized value of the i test sets. After normalization, the data is in the range of [-1,1].

[0074] (6) Process the data in the test set, and randomly select 80% of the test set as the training set and 20% of the test set as the validation set according to the 8:2 principle after normalization. Finally, a total of 154 sets of data are allocated to the training set and a total of 38 sets of data are allocated to the test set.

[0075] (7) Determine the number of neurons in the input layer, output layer, and hidden layer.

[0076] The preset range of the number of neurons in the hidden layer is preliminarily determined by the following formula:

[0077]

[0078] Where n is the number of neurons in the input layer; m is the number of neurons in the hidden layer of the neural network; l is the number of neurons in the output layer of the neural network; α is a constant, ranging from [1,10].

[0079] Since the number of input layer neurons in this embodiment is 3, namely shear rate, temperature, and water content, and the number of output layer neurons is 1, namely emulsion viscosity, this embodiment will determine the appropriate number of hidden layer neurons in the interval [3,12].

[0080] (8) Set the activation function of the Bayesian-BP neural network.

[0081] Using the S-type function as the activation function in the neural network, its calculation expression is:

[0082]

[0083] (9) Prediction value restoration of Bayesian-BP neural network.

[0084] Since the data preprocessing in step (5) performs a series of processing on the viscosity of the test set, the present invention performs a restoration process on the value of the output neuron of the Bayesian-BP neural network, and its calculation expression is:

[0085]

[0086] In the formula, is the neuron value of the output layer of the i-th test set; are the maximum and minimum values ​​of the output layer neurons in the test set, respectively; a and b are constants, which are 1 and -1, respectively; Predicted emulsion viscosity for the ith test set.

[0087] (10) Establishing a neural network structure

[0088] This embodiment adopts a three-layer neural network, namely, an input layer, a hidden layer, and an output layer. Figure 3 shown.

[0089] Set the input vector to x = (x1, x2, x3) T , where x i =(x i1 ,x i2 ,...,x ij ), x i is the numerical vector of the i-th (i=1, 2, 3, respectively representing shear rate, temperature, and moisture content) test set after preprocessing, x ij is the value of the i-th test set of the j-th test set after preprocessing.

[0090] Set the output variable to y = (y1, y2, ..., y j ), where y j is the predicted value of the jth test set.

[0091] The execution process between the input layer and the hidden layer is Where j = 1, 2, ..., 6, is the activation function determined in step (8), ω ij is the connection weight between the input layer and the hidden layer.

[0092] The execution process between the hidden layer and the output layer is Among them, α and β are the first fitting parameter and the second fitting parameter respectively, ω k is the connection weight between the hidden layer and the output layer.

[0093] (11) Establish and train the Bayesian-BP neural network model.

[0094] According to the activation function set in step (8) and the neural network structure established in step (10), a Bayesian-BP neural network model is constructed with the numerical mean square error of the neurons in the output layer of the neural network as the optimization object. The training and prediction process is shown in Figure 4 shown.

[0095] The training set obtained in step (6) is imported into the workspace, initial values ​​are set for the fitting parameters and connection weights in the neural network, and training limiting conditions such as the maximum training rounds and the ideal mean square error are set. The training set is used to obtain the values ​​of the hidden layer neurons and the output layer neurons of the neural network layer by layer, and then the mean square error and the complex correlation coefficient between the values ​​of the output layer neurons and the sample values ​​are calculated to determine whether the training limiting conditions are met in this round of training. If not, the fitting parameters and connection weights are adjusted according to the training situation. If the training limiting conditions are met in this round of training, the training of the Bayesian-BP neural network model is terminated.

[0096] (12) Verification and evaluation of Bayesian-BP neural network model

[0097] The mean square error iteration effect is shown in Figure 5 As shown, Figure 5 It shows that the training stopped when the training number reached 5906 times due to the limit condition. In the 1698th round of training, the mean square error reached the minimum value of 0.004523. In the subsequent training, the mean square error fluctuated around 0.004523.

[0098] The comparison diagram of the relationship between the viscosity and water content of the training set is shown in Figure 6 As shown, a portion of the samples in the training set obtained in step (6) with consistent shear rate and temperature are taken, and a scatter plot of viscosity and water content is made based on the viscosity values ​​predicted from the portion of the samples and the experimental viscosity values, so as to better view the training effect of the Bayesian-BP neural network model.

[0099] The predicted regression plot of the training set is shown in Figure 7 As shown in the figure, for the Bayesian-BP neural network model obtained from the 1698th round of training, the viscosity predicted by the model is basically similar to the sample viscosity, and the predicted viscosity and the sample viscosity satisfy Among them, η i is the viscosity value of the i-th training set, is the i-th predicted viscosity value, and the regression coefficient is 0.99667.

[0100] The prediction error histogram of the training set is shown in Figure 8 As shown, the error between the data obtained by the Bayesian-BP neural network model and the sample data presents a normal distribution. The average error is calculated to be -0.00267 Pa·s, and the average relative error is 6.20%.

[0101] (13) Prediction and evaluation of oil-water emulsion viscosity. The prediction results are shown in Figure 6-11 shown.

[0102] The validation set obtained in step (6) is imported into the workspace, and the Bayesian-BP neural network model trained in step (11) can be used to directly predict the viscosity of the oil-water emulsion in the test set.

[0103] The comparison diagram of the relationship between viscosity and shear rate of the validation set is shown in Fig. 9 As shown, some samples with consistent temperature and moisture content in the verification set obtained in step (6) are selected, and a scatter plot of viscosity and temperature is made based on the predicted viscosity values ​​and experimental viscosity values ​​of these samples, so as to better view the prediction effect of the Bayesian-BP neural network model.

[0104] The prediction regression plot of the validation set is shown in Fig.10 As shown in FIG. 1 , the viscosity predicted by the Bayesian-BP neural network model trained in step (11) is basically similar to the sample viscosity, and the satisfaction between the predicted viscosity and the sample viscosity is Among them, η i is the viscosity value of the i-th training set, is the i-th predicted viscosity value, and the regression coefficient is 0.99811.

[0105] The prediction error histogram of the validation set is shown in Fig.11 As shown, the error between the data obtained by the Bayesian-BP neural network model and the sample data presents a normal distribution. The average error is calculated to be 0.00362 Pa·s, and the average relative error is 5.36%.

[0106] Compared with the existing prediction method of oil-water emulsion viscosity mathematical regression, the beneficial effect of the present invention is that it solves the problems of many influencing factors and complex changing rules of oil-water emulsion viscosity, and adopts a Bayesian regularization method combined with a BP neural network algorithm to solve the problems of low prediction accuracy and poor universality of oil-water emulsion viscosity prediction, thereby improving the prediction accuracy of oil-water emulsion viscosity. In addition, the method is suitable for multi-variable and large-sample oil-water emulsification situations, and has good industrial guiding significance and application prospects for oilfield exploitation and crude oil processing.

Claims

1. A method for predicting viscosity of oil-water emulsion based on neural network, characterized in that: A dynamic prediction method for the viscosity of oil-water emulsion under different shear rates, temperatures, and water contents is provided by combining experiments with machine learning. The Bayesian regularization algorithm is combined with the BP neural network algorithm to solve the problems of low prediction accuracy and poor universality of the viscosity of oil-water emulsions. The method includes the following steps: (1) An oil-water emulsification experiment was carried out using a self-designed conical plate variable volume turbulent drag reduction evaluation device to stir and prepare oil-water emulsions with different water contents; (2) Control the rotation speed and water bath temperature of the conical plate in the conical plate variable volume turbulence drag reduction evaluation device, so as to measure the viscosity of the oil-water emulsion under different water content, shear rate and temperature conditions in real time; (3) If the effect of different pressures on the viscosity of the oil-water emulsion under formation conditions needs to be considered, 50 ml of the stable oil-water emulsion formed by stirring is taken and added to a high-temperature and high-pressure rheometer. The temperature and shear rate are set to be the same as those of the stirring device for the original emulsion preparation, so as to test the viscosity of the oil-water emulsion under different pressure conditions; (4) Record and organize the viscosity of oil-water emulsions under different water contents, shear rates, and temperatures obtained from experimental tests, and establish a test set for machine learning; (5) Analyze the data in the test set and preprocess the data. First, logarithmically process the viscosity data of the oil-water emulsion, and then normalize the shear rate, temperature, water content, and viscosity after logarithmic processing; (6) Process the data in the test set. After normalization, randomly select 80% of the test set as the training set and 20% of the test set as the validation set according to the 8:2 principle. (7) Determine the number of neurons in the input layer, output layer, and hidden layer; (8) Setting the activation function of the Bayesian-BP neural network; (9) Prediction value restoration of Bayesian-BP neural network; (10) Establishing a neural network structure; (11) Establish and train the Bayesian-BP neural network model; (12) Verification and evaluation of Bayesian-BP neural network model; (13) Prediction and evaluation of viscosity of oil-water emulsions.

2. The method for predicting viscosity of oil-water emulsion based on neural network according to claim 1, characterized in that: In the steps (1) and (2), the cone-shaped circular plate stirring shear rate in the oil-water emulsification experiment is constant. The water content of the tested oil-water emulsion ranges from 40% to 90%, the water bath test temperature ranges from T = 40 to 80°C, and the shear rate ranges from 0 to 1000s -1 .

3. The method for predicting viscosity of oil-water emulsion based on neural network according to claim 1, characterized in that: In step (5), logarithmic processing is performed on the viscosity data in the test set, and the calculation expression is:

4. The method for predicting viscosity of oil-water emulsion based on neural network according to claim 1, characterized in that: In step (5), the shear rate, temperature, water content and the viscosity of the oil-water emulsion after logarithmic processing in the test set are uniformly normalized, and the calculation expression is:

5. The method for predicting viscosity of oil-water emulsion based on neural network according to claim 1, characterized in that: The step (7) preliminarily determines the preset range of the number of neurons in the hidden layer, and its calculation expression is:

6. The method for predicting viscosity of oil-water emulsion based on neural network according to claim 1, characterized in that: In step (8), the S-type function is used as the activation function in the neural network, and its calculation expression is:

7. The method for predicting viscosity of oil-water emulsion based on neural network according to claim 1, characterized in that: The step (9) restores the value of the output neuron of the Bayesian-BP neural network, and its calculation expression is:

8. The method for predicting viscosity of oil-water emulsion based on neural network according to claim 1, characterized in that: The step (10) adopts a three-layer neural network, namely, an input layer, a hidden layer, and an output layer, and sets the input vector to be x=(x1, x2, x3) T , where x i =(x i1 ,x i2 ,...,x ij ), x i is the numerical vector of the i-th (i=1, 2, 3, respectively representing shear rate, temperature, and moisture content) test set after preprocessing, x ij The output variable is y=(y1,y2,...,y j ), where y j is the predicted value of the jth test set, and the execution process between the input layer and the hidden layer is Where j = 1, 2, ..., 6, is the activation function determined in step (8), ω ij is the connection weight between the input layer and the hidden layer, and the execution process between the hidden layer and the output layer is Among them, α and β are the first fitting parameter and the second fitting parameter respectively, ω k is the connection weight between the hidden layer and the output layer.

Images