An oil-water separation filter screen feed parameter optimization method based on a neural network

Abstract

The application provides an oil-water separation filter screen feed parameter optimization method based on a neural network, which comprises the following steps: (1) collecting original data of an oil-water separation experiment and performing data preprocessing; (2) establishing an oil-water separation filter screen feed parameter optimization model based on a BP neural network; (3) initializing the weight and threshold of the prediction model obtained in step (2) and setting key parameters; (4) performing optimization training and error evaluation on the optimization model obtained in step (2) to determine the best optimization model; (5) optimizing the oil-water separation filter screen feed parameters by using the best optimization model obtained in step (4); wherein the optimized feed parameters include any one of a feed temperature, a feed pH or a membrane pressure. The optimization method optimizes and adjusts the best feed parameters for separating different oil-containing wastewater through model inversion, saves economic and time costs in actual test of the performance of a separation material, and improves the oil-water separation efficiency.

Description

Technical Field

[0001] This invention belongs to the field of oil-water separation research and relates to a method for optimizing feed parameters, and more particularly to a method for optimizing feed parameters of an oil-water separation filter based on neural networks. Background Technology

[0002] In recent years, intelligent oil-water separation materials capable of responding to pre-wetting and other mechanisms have emerged in the field of oil-water separation materials research. These materials promise to achieve simultaneous oil / water separation in specially designed separation devices based on only one type of material. Such materials exhibit excellent oil-water separation performance and self-cleaning properties; for example, stainless steel mesh with an asymmetric titanium dioxide coating possesses pre-wetting characteristics. Previous researchers have prepared such materials in the laboratory; their surface has an asymmetric titanium dioxide coating, thus exhibiting superamphilicity. After pre-wetting with oil or water, they become hydrophobic and oleophobic, enabling efficient separation of oily wastewater.

[0003] However, when using mesh or membrane-based oil-water separation materials, including stainless steel mesh with an asymmetric titanium dioxide coating, to separate actual oily wastewater, pollutants in the water often accumulate on the mesh surface and in the pores, causing surface fouling, a decrease in water flux, and blockage, thus severely affecting the material's separation performance. Besides pollutants in the water, water flux is also affected by various process variables, such as feed temperature, feed pH, oil concentration, membrane pressure, separation time, separation volume, and hydraulic diameter of the particles.

[0004] In addition, in recent years, artificial neural network technology has been widely used to predict chemical parameters and variables under different physicochemical conditions, as well as various membrane filtration processes. It has good data simulation and optimization capabilities for complex nonlinearities and can solve the problem of optimizing feed parameters for oil-water separation filters.

[0005] Therefore, how to use artificial neural network technology to optimize the feed parameters for separating different types of oily wastewater, so as to save the economic and time costs required when actually testing the performance of separation materials, and thus improve the oil-water separation efficiency, has become an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for optimizing the feed parameters of an oil-water separation filter based on a neural network. This optimization method uses model inversion to optimize and adjust the optimal feed parameters for separating different oily wastewaters, saving economic and time costs when actually testing the performance of the separation material, and significantly improving the oil-water separation efficiency.

[0007] To achieve this objective, the present invention adopts the following technical solution:

[0008] This invention provides a method for optimizing the feed parameters of an oil-water separation filter based on a neural network. The optimization method includes the following steps:

[0009] (1) Collect raw data from the oil-water separation experiment and perform data preprocessing;

[0010] (2) An optimization model for the feed parameters of the oil-water separation filter screen was established based on a BP neural network;

[0011] (3) Initialize the weights and thresholds of the prediction model obtained in step (2) and set the key parameters;

[0012] (4) Perform optimization training and error evaluation on the optimized model obtained in step (2) to determine the best optimized model;

[0013] (5) Optimize the feed parameters of the oil-water separation filter using the optimal optimization model obtained in step (4).

[0014] The optimized feed parameters in step (5) include any one of feed temperature, feed pH or membrane pressure.

[0015] This invention introduces a BP neural network with strong fitting ability into the optimization method of feed parameters for oil-water separation filters. By optimizing and adjusting the optimal feed parameters for separating different oily wastewaters through model inversion, the invention saves economic and time costs when actually testing the performance of separation materials, significantly improves oil-water separation efficiency, and provides a new approach for the optimization research of feed parameters for oil-water separation materials.

[0016] Preferably, the raw data in step (1) includes input variables and output variables.

[0017] Optionally, the input variables include water flux, feed pH, oil concentration, membrane pressure, separation time, separated volume, and hydraulic diameter of particles; the output variables include feed temperature.

[0018] Optionally, the input variables include water flux, feed temperature, oil concentration, membrane pressure, separation time, separated volume, and hydraulic diameter of particles; the output variable includes feed pH.

[0019] Optionally, the input variables include water flux, feed temperature, feed pH, oil concentration, separation time, separated volume, and hydraulic diameter of particles; the output variables include membrane pressure.

[0020] Preferably, the data preprocessing in step (1) includes data partitioning and normalization processes performed sequentially.

[0021] Preferably, the data partitioning includes randomly dividing the original data into training set data, test set data, and validation set data.

[0022] Preferably, the data partitioning function includes the dividerand function.

[0023] Preferably, the ratio of the training set data, test set data, and validation set data is (70-90):(5-15):(5-15), for example, it can be 70:15:15, 75:15:10, 75:10:15, 80:10:10, 85:10:5, 85:5:10, or 90:5:5, but it is not limited to the listed values. Other unlisted values ​​within this range are also applicable.

[0024] Preferably, the normalization process includes normalizing the input variables to the range of [-1, 1] to avoid their values ​​being too large and affecting model training.

[0025] Preferably, the normalization process uses the mapminmax function.

[0026] Specifically, the expression for the mapminmax function is:

[0027] X ni =(x i -x min ) / (x max -x min (1)

[0028] Where, x i It is the input variable, X ni It is the normalized input variable, x max It is the maximum value in the input variable dataset, x min It is the minimum value in the input variable dataset.

[0029] Preferably, the topology of the optimization model in step (2) is divided into at least 3 layers, for example, 3, 4, 5, 6, 7, 8, 9 or 10 layers, more preferably 4 layers, but not limited to the listed values, other unlisted values ​​within this range are also applicable.

[0030] Preferably, the topology of the optimization model in step (2) includes an input layer, a first hidden layer, a second hidden layer, and an output layer.

[0031] Preferably, the activation function from the input layer to the first hidden layer includes the tan-sigmoid function.

[0032] Preferably, the activation function from the first hidden layer to the second hidden layer includes the tan-sigmoid function.

[0033] Specifically, the formula for the tan-sigmoid function is:

[0034] F(x)=(e +x -e -x ) / (e +x +e -x (2)

[0035] Preferably, the activation function from the second hidden layer to the output layer includes a purelin or ReLU function to adjust the effect of the tan-sigmoid function on the network convergence.

[0036] Specifically, the expression for the purelin function is:

[0037] Y = x (3)

[0038] Specifically, the expression for the ReLU function is:

[0039] Relu(x) = max(0,x) (4)

[0040] Preferably, the function used for initialization in step (3) includes the train function.

[0041] This invention uses the train function in MATLAB, which by default makes the initial weights Gaussian distributed.

[0042] Preferably, the key parameters in step (3) include the maximum number of training iterations, the learning rate, and the sum of squared network errors.

[0043] Preferably, the optimization training in step (4) is performed using the Levenberg-Marquardt algorithm.

[0044] In this invention, the Levenberg-Marquardt algorithm has a large memory requirement and a fast convergence speed. Using it as a training function can avoid directly calculating the Hessian matrix when correcting network connection weights at an approximate second-order training rate, thereby reducing the amount of computation in training and significantly saving training time.

[0045] Preferably, the error assessment in step (4) is performed using the mean square error, average error, and sum of squared errors function.

[0046] Specifically, this invention uses mean squared error (MSE) to measure the predictive performance of a neural network on validation set data, and the expression for MSE is:

[0047] MSE=∑(J model,i -J exp,i ) 2 / h (5)

[0048] Among them, J model,i J is the predicted value of the i-th normalized flux. exp,i These are the corresponding observed values.

[0049] Specifically, this invention uses the average error (AVE) to evaluate the accuracy of predictions made by a single artificial neural network, and the expression for AVE is:

[0050] Error(%)=[(J model,i -J exp,i ) / J exp,i ]×100 (6)

[0051] AVE(%)=[∑Error / n]×100 (7)

[0052] Among them, J model,i J is the predicted value of the i-th normalized flux. exp,i is the corresponding observation value, and n is the number of observation values.

[0053] Specifically, this invention uses the sum of squared errors (SSE) function to characterize the error performance of the BP neural network, and the expression for SSE is:

[0054]

[0055] Where N is the number of sample pairs, and L is the number of network output nodes. It is the expected output value of the k-th node when inputting the p-th time. It is the output value of the k-th node when inputting the p-th time. It is the error between the output value of the k-th node and the expected value when inputting for the p-th time.

[0056] Preferably, the method for determining the optimal optimization model in step (4) includes: repeatedly adjusting the number of neurons in the hidden layer during the optimization training process of the optimization model, and determining the optimal number of neurons in the hidden layer in combination with the error evaluation results, thereby obtaining the optimal optimization model.

[0057] Preferably, the number of neurons in the hidden layer is 1-20, for example, 1, 2, 4, 6, 8, 10, 12, 14, 16, 18 or 20, but is not limited to the listed values. Other unlisted values ​​within this range are also applicable.

[0058] Preferably, the optimization process in step (5) includes: collecting the characteristic parameters and water flux of the oily wastewater to be separated, inputting them into the optimal optimization model obtained in step (4), and obtaining the optimized value of the feed parameters to guide the setting of the feed parameters, thereby realizing the optimization of the feed parameters of the oil-water separation filter during the oil-water separation process.

[0059] In this invention, the characteristic parameters include feed temperature, feed pH, oil concentration, membrane pressure, and hydraulic diameter of particles.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] This invention introduces a BP neural network with strong fitting ability into the optimization method of feed parameters for oil-water separation filters. By optimizing and adjusting the optimal feed parameters for separating different oily wastewaters through model inversion, the invention saves economic and time costs when actually testing the performance of separation materials, significantly improves oil-water separation efficiency, and provides a new approach for the optimization research of feed parameters for oil-water separation materials. Attached Figure Description

[0062] Figure 1 This is the optimization model architecture diagram in the optimization method provided in Example 1;

[0063] Figure 2 This is a diagram of the BP neural network structure in the optimization method provided in Example 1. Detailed Implementation

[0064] The technical solution of the present invention will be further illustrated below through specific embodiments. Those skilled in the art should understand that the embodiments described are merely illustrative of the present invention and should not be construed as limiting the invention in any way.

[0065] Example 1

[0066] This embodiment provides a method for optimizing the feed parameters of an oil-water separation filter based on a neural network. The optimization method includes the following steps:

[0067] S1. Collect raw data from the oil-water separation experiment to construct a data sample, including feed temperature, feed pH, oil concentration, membrane pressure, and hydraulic diameter of particles; use water flux, feed pH, oil concentration, membrane pressure, separation time, separation volume, and hydraulic diameter of particles as input variables, and feed temperature as output variable.

[0068] The specific process of the oil-water separation experiment is as follows: Four sets of experiments were designed, each set with a single characteristic parameter as the independent variable, taking five different values, and the experiments were carried out respectively. The water flow data and corresponding feed temperature of the oil-water separation filter were recorded every 5 minutes. A total of 17 time nodes of oil-water separation filter feed temperature data were recorded, and a total of 340 data points were collected.

[0069] S2. Using the divideand function in MATLAB, the original data is randomly divided into training set data, test set data and validation set data in a ratio of 80:10:10, that is, 272 sets of data are used as training set data, 34 sets of data are used as test set data and 34 sets of data are used as validation set data.

[0070] S3. The input variables are normalized to the range [-1, 1] using the mapminmax function, and the expression of the mapminmax function is:

[0071] X ni =(x i -x min ) / (x max -x min (1)

[0072] Where, x i It is the input variable, X ni It is the normalized input variable, x max It is the maximum value in the input variable dataset, x min It is the minimum value in the input variable dataset;

[0073] S4. Establishing a system based on a BP neural network, as shown below. Figure 1 The optimization model for the feed parameters of the oil-water separation filter screen is shown. The topology of the optimization model is divided into four layers, including an input layer, a first hidden layer, a second hidden layer, and an output layer (see...). Figure 2 The activation function from the input layer to the first hidden layer is the tan-sigmoid function, the activation function from the first hidden layer to the second hidden layer is the tan-sigmoid function, and the activation function from the second hidden layer to the output layer is the ReLU function. The formula for the tan-sigmoid function is:

[0074] F(x)=(e +x -e -x ) / (e +x +e -x (2)

[0075] The expression for the ReLU function is:

[0076] Relu(x) = max(0,x) (3)

[0077] S5. Initialize the weights and thresholds of the optimized model obtained in step S4 using the train function in MATLAB, and set key parameters, including the maximum number of training iterations, learning rate, and sum of squared network errors.

[0078] S6. The Levenberg-Marquardt algorithm is used to optimize and train the model obtained in step S4, and the mean squared error (MSE), average error (AVE), and sum of squared errors (SSE) are used for error evaluation. The expression for MSE is as follows:

[0079] MSE=∑(Jmodel,i -J exp,i ) 2 / h (4)

[0080] The expression for AVE is:

[0081] Error(%)=[(J model,i -J exp,i ) / J exp,i ]×100 (5)

[0082] AVE(%)=[∑Error / n]×100 (6)

[0083] Among them, J model,i J is the predicted value of the i-th normalized flux. exp,i These are the corresponding observations, and n is the number of observations.

[0084] The expression for SSE is:

[0085]

[0086] Where N is the number of sample pairs, and L is the number of network output nodes. It is the input value of the k-th node when inputting for the p-th time. It is the output value of the k-th node when inputting the p-th time. It is the error between the output value of the k-th node and the expected value at the p-th input.

[0087] S7. During the optimization training of the optimized model, the number of neurons in the hidden layer is repeatedly adjusted within the range of 1-20. The optimal number of neurons in the hidden layer is determined based on the error evaluation results, thus obtaining the optimal optimized model. The optimal optimized model has 4 neurons in the first hidden layer and 15 neurons in the second hidden layer (see...). Figure 2 At this point, the errors in the training set, validation set, and test set data reach a small value, which can be used to optimize the feed temperature of the oil-water separator filter.

[0088] S8. Optimize the feed temperature of the oil-water separation filter using the optimal optimization model obtained in step S7. The optimization process includes: collecting 17 sets of data on water flux, feed pH, oil concentration, membrane pressure, separation time, separation volume, and hydraulic diameter of particles in the oily wastewater to be separated; inputting the characteristic parameters and water flux into the optimal optimization model obtained in step S7 to obtain the optimized feed temperature value, which is used to guide the setting of the actual feed temperature, thereby optimizing the feed parameters of the oil-water separation filter during the oil-water separation process.

[0089] Example 2

[0090] This embodiment provides a method for optimizing the feed parameters of an oil-water separation filter based on a neural network. The optimization method changes the input variables in step S1 to water flux, feed temperature, oil concentration, membrane pressure, separation time, separation volume, and hydraulic diameter of particles, and changes the output variable to feed pH. The remaining steps and conditions are the same as in embodiment 1, so they will not be described in detail here.

[0091] In this embodiment, the optimized value of the feed pH is used to guide the setting of the actual feed pH, thereby optimizing the feed parameters of the oil-water separation filter during the oil-water separation process.

[0092] Example 3

[0093] This embodiment provides a method for optimizing the feed parameters of an oil-water separation filter based on a neural network. The optimization method changes the input variables in step S1 to water flux, feed temperature, feed pH, oil concentration, separation time, separation volume, and hydraulic diameter of particles, and changes the output variable to membrane pressure. The remaining steps and conditions are the same as in embodiment 1, so they will not be described in detail here.

[0094] In this embodiment, the optimized value of the obtained membrane pressure is used to guide the setting of the actual membrane pressure, thereby optimizing the feed parameters of the oil-water separation filter during the oil-water separation process.

[0095] Example 4

[0096] This embodiment provides a method for optimizing the feed parameters of an oil-water separation filter based on a neural network. The optimization method is the same as that in Embodiment 1, except that the ratio of the randomly divided data in step S2 is changed to 90:5:5, that is, 306 sets of data are used as training data, 17 sets of data are used as test data, and 17 sets of data are used as validation data. Therefore, it will not be described in detail here.

[0097] In this embodiment, the optimized value of the feed temperature is used to guide the setting of the actual feed temperature, thereby optimizing the feed parameters of the oil-water separation filter during the oil-water separation process.

[0098] Example 5

[0099] This embodiment provides a method for optimizing the feed parameters of an oil-water separation filter based on a neural network. The optimization method, in addition to changing the activation function from the second hidden layer to the output layer in step S4 to a purelin function, has the following expression for the purelin function:

[0100] Y = x (8)

[0101] The remaining steps and conditions are the same as in Example 1, so they will not be repeated here.

[0102] In this embodiment, the optimized value of the feed temperature is used to guide the setting of the actual feed temperature, thereby optimizing the feed parameters of the oil-water separation filter during the oil-water separation process.

[0103] Therefore, this invention introduces a BP neural network with strong fitting ability into the optimization method of feed parameters for oil-water separation filters. By optimizing and adjusting the optimal feed parameters for separating different oily wastewaters through model inversion, the invention saves economic and time costs when actually testing the performance of separation materials, significantly improves oil-water separation efficiency, and provides a new approach for the optimization research of feed parameters for oil-water separation materials.

[0104] The applicant declares that the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention fall within the protection and disclosure scope of the present invention.

Claims

1. A method for optimizing feed parameters of an oil-water separation filter based on a neural network, characterized in that, The optimization method includes the following steps: (1) Collect the raw data of the oil-water separation experiment and perform data preprocessing; the data preprocessing includes data partitioning and normalization processing performed sequentially; the normalization processing includes normalizing the input variables to the range of [-1,1], and the function used in the normalization processing includes the mapminmax function; (2) An optimization model for the feed parameters of the oil-water separation filter is established based on a BP neural network; the topology of the optimization model is 4 layers, including an input layer, a first hidden layer, a second hidden layer and an output layer; wherein, the activation function from the input layer to the first hidden layer includes the tan-sigmoid function; the activation function from the first hidden layer to the second hidden layer includes the tan-sigmoid function; the activation function from the second hidden layer to the output layer includes the purelin or ReLU function; (3) Initialize the weights and thresholds of the prediction model obtained in step (2) and set the key parameters; (4) Perform optimization training and error evaluation on the optimized model obtained in step (2) to determine the best optimized model; wherein the optimization training is performed using the Levenberg-Marquardt algorithm; (5) Optimize the feed parameters of the oil-water separation filter using the optimal optimization model obtained in step (4); The optimized feed parameters in step (5) include any one of feed temperature, feed pH or membrane pressure.

2. The optimization method according to claim 1, characterized in that, The raw data in step (1) includes input variables and output variables.

3. The optimization method according to claim 2, characterized in that, The input variables include water flux, feed pH, oil concentration, membrane pressure, separation time, separated volume, and hydraulic diameter of particles. The output variables include the feed temperature.

4. The optimization method according to claim 2, characterized in that, The input variables include water flux, feed temperature, oil concentration, membrane pressure, separation time, separated volume, and hydraulic diameter of particles. The output variables include feed pH.

5. The optimization method according to claim 2, characterized in that, The input variables include water flux, feed temperature, feed pH, oil concentration, separation time, separated volume, and hydraulic diameter of particles. The output variables include membrane pressure.

6. The optimization method according to claim 2, characterized in that, The data partitioning includes randomly dividing the original data into training set data, test set data, and validation set data.

7. The optimization method according to claim 6, characterized in that, The data partitioning uses functions including the dividerand function.

8. The optimization method according to claim 6, characterized in that, The ratio of the amount of training set data, test set data, and validation set data is (70-90):(5-15):(5-15).

9. The optimization method according to claim 1, characterized in that, The initialization function used in step (3) includes the train function.

10. The optimization method according to claim 1, characterized in that, The key parameters mentioned in step (3) include the maximum number of training iterations, the learning rate, and the sum of squared network errors.

11. The optimization method according to claim 1, characterized in that, The error assessment in step (4) is performed using the mean square error, average error and sum of squared errors function.

12. The optimization method according to claim 1, characterized in that, The method for determining the optimal optimization model in step (4) includes: repeatedly adjusting the number of neurons in the hidden layer during the optimization training process of the optimization model, and determining the optimal number of neurons in the hidden layer in combination with the error evaluation results, thereby obtaining the optimal optimization model.

13. The optimization method according to claim 12, characterized in that, The number of neurons in the hidden layer is 1-20.

14. The optimization method according to claim 1, characterized in that, The optimization process described in step (5) includes: collecting the characteristic parameters and water flux of the oily wastewater to be separated, inputting them into the optimal optimization model obtained in step (4), and obtaining the optimized value of the feed parameters to guide the setting of the feed parameters, thereby realizing the optimization of the feed parameters of the oil-water separation filter during the oil-water separation process.

Images