BP neural network-based slurry conveying pipeline siltation critical flow velocity prediction method

Abstract

The invention provides a BP neural network-based slurry conveying pipeline siltation critical flow velocity prediction method, and relates to the technical field of fluid conveying. Comprising the following steps: S1, determining four items of conveying pipe diameter, particle size, solid material density and volume fraction as main factors influencing the flow velocity change of the slurry conveying pipeline; s2, introducing a Mixup algorithm to perform enhancement processing on the original data; s3, in the process of predicting the sedimentation critical flow velocity of the slurry conveying pipeline, introducing a BP neural network, constructing an intelligent prediction model, and continuously learning, mining and expanding rule features contained in data; s4, deducing a more accurate flow velocity predicted value by using a grey correlation analysis algorithm when the sedimentation critical flow velocity of the slurry conveying pipeline is predicted by depending on the BP neural network model; according to the method, accurate prediction of the deposition critical flow velocity is realized, a powerful guarantee is provided for safe operation and efficiency improvement of the slurry conveying pipeline, and the method has important significance for guaranteeing safe operation of the slurry conveying pipeline.

Description

Technical Field

[0001] This invention relates to the field of fluid transport technology, and in particular to a method for predicting the critical flow velocity of siltation in slurry transport pipelines based on a BP neural network. Background Technology

[0002] In industrial production and transportation, pipeline transport has become a crucial method for transporting slurry materials. However, the complex and variable flow behavior of slurry in pipelines leads to frequent slurry accumulation problems, which not only reduce transport efficiency and increase energy consumption but may also cause safety accidents such as pipeline blockage and rupture. The critical slurry accumulation velocity, as a key parameter in pipeline transport systems, measures the flow stability and slurry accumulation tendency of slurry in pipelines, providing better guidance for pipeline slurry prevention and treatment.

[0003] Elkarii M et al. proposed a flow prediction model based on CFD (fluid dynamics). This model combines fluid dynamics and computational principles with particle-particle interactions and particle-wall interactions to construct a computational mathematical model for slurry transport in pipelines. By substituting historical data into the model, the critical flow velocity of the pipeline can be predicted. However, CFD simulations require a certain degree of simplification of the actual physical model, which may lead to the neglect of some minor factors in velocity prediction, resulting in significant deviations in the prediction results.

[0004] He Peng et al. proposed a flow velocity prediction model based on machine learning, establishing an extended two-fluid phase-separation model for pipelines to clarify the various factors affecting the critical flow velocity for pipeline sedimentation. They constructed an intelligent prediction model for critical flow velocity by combining the minimum pressure gradient method and a backpropagation neural network. By inputting historical data corresponding to each factor into the model for learning, critical flow velocity prediction can be achieved. However, the predictive performance of the machine learning model is highly dependent on the quality of the training data. The difficulty of data collection varies under different operating conditions, resulting in significant limitations in the application of this model.

[0005] Zhang Xinsheng et al. proposed a flow velocity prediction method based on SSA-CNN (Sparrow Search Algorithm-Convolutional Neural Network). After identifying the various factors affecting the critical flow velocity in a pipeline, they selected the main factors as input features. Based on an optimized convolutional neural network, an intelligent prediction model was established, which continuously learns the input features and outputs the final prediction result. However, the SSA-CNN model, combining the Sparrow Search Algorithm and the convolutional neural network, has high complexity, resulting in poor real-time performance of flow velocity prediction. He Peng et al. proposed a flow velocity prediction model based on grey support vector regression. They used the minimum pressure gradient method to construct a mathematical model for calculating the critical flow velocity and collected a large number of key parameters as input samples for flow velocity prediction. A grey relational analysis algorithm was used to assign corresponding weights to each class of samples. By inputting the data samples and weight information into the prediction model based on the support vector regression algorithm, the critical flow velocity prediction result can be obtained. However, the kernel function of the support vector regression algorithm has different applicability to different datasets and prediction problems, resulting in poor reliability of this prediction model.

[0006] Due to the significant advantages of BP neural network technology in nonlinear modeling and prediction, it was introduced into the analysis of critical flow velocity for siltation in slurry transport pipelines, exploring a novel flow velocity prediction method based on BP neural networks. Through theoretical analysis, the main factors affecting the critical flow velocity for siltation were identified. An intelligent prediction model based on BP neural networks was established, and the prediction results of the critical flow velocity for pipeline siltation were obtained by learning from sample data. Summary of the Invention

[0007] To address the aforementioned problems, this invention provides a method for predicting the critical flow velocity for siltation in slurry transport pipelines based on a BP neural network. This invention achieves accurate prediction of the critical flow velocity for siltation, providing strong support for the safe operation and efficiency improvement of slurry transport pipelines, and is of great significance for ensuring the safe operation of slurry transport pipelines.

[0008] To solve the above problems, the technical solution adopted by the present invention is as follows:

[0009] The method for predicting the critical flow velocity of slurry transport pipeline siltation based on BP neural network includes the following steps: S1. Determine the four main factors affecting the flow velocity change of slurry transport pipeline: pipeline diameter, particle size, solid material density, and volume fraction, and collect corresponding data samples; S2. Introduce the Mixup algorithm to enhance the original data and generate an expanded dataset that satisfies the principle of minimizing neighborhood risk; S3. In the process of predicting the critical flow velocity of slurry transport pipeline siltation, introduce BP neural network to construct an intelligent prediction model, continuously learn and mine the regular features contained in the expanded data, and provide a basis for the prediction of critical flow velocity; S4. When predicting the critical flow velocity of slurry transport pipeline siltation using the BP neural network model, use the grey relational analysis algorithm to calculate the weights of each critical flow velocity influencing factor, and input them into the prediction model simultaneously to derive a more accurate flow velocity prediction value.

[0010] Preferably, in the actual data augmentation process of step S2, the nearest neighbor risk rule is used to define the expected risk estimation formula as follows: .

[0011] In the formula, Indicates expected risk. Represents the target data point. Indicates neighboring data points. This represents the total number of data points in the current dataset. Indicates the data point number. This indicates a neighborhood distribution.

[0012] Preferably, in step S2, with the goal of minimizing the expected risk, a dataset based on sampling from a neighboring distribution can be generated; .

[0013] In the formula, This represents an augmented dataset. Indicates the data dimension.

[0014] For the newly generated augmented dataset, its neighborhood distribution Mixup can be defined as follows:

[0015] .

[0016] In the formula, Indicates the neighborhood distribution Mixup. This indicates the number of the newly generated sample point selected randomly. This indicates mixed weights.

[0017] Preferably, the mixing weight is a parameter used to control the degree of mixing between the original data and the new data, and its value range is:

[0018] .

[0019] In the formula, Indicates the beta distribution, represents two hyperparameters; a portion of newly generated data is randomly selected from the Mixup distribution to form an augmented dataset, which is used for subsequent prediction of critical flow velocities for pipeline siltation.

[0020] Preferably, in step S3, the data corresponding to each variable is imported into the corresponding input layer neuron node of the BP neural network, and then, combined with the connection weights and thresholds, the gradient descent method is used to control the hidden layer nodes to complete forward and backward propagation learning.

[0021] Preferably, construct Figure 3 The BP neural network prediction model shown imports the data corresponding to each variable into the corresponding input layer neuron node, and then combines the connection weights and thresholds, using the gradient descent method to control the hidden layer nodes to complete forward and backward propagation learning.

[0022] Preferably, Figure 3 In the BP neural network prediction model shown, the number of nodes in the input layer is determined by the number of variables affecting the critical flow velocity, while the number of nodes in the hidden layer is adjustable; the optimal number of nodes in the hidden layer is determined using the following formula.

[0023] .

[0024] In the formula, This represents the optimal number of hidden layer nodes. These represent the number of nodes in the network input layer and output layer, respectively. Represents a constant.

[0025] Preferably, after the basic structure of the BP neural network prediction model in step S3 is determined, the network model is trained using partial sample data; after the training convergence requirement is met, a critical flow velocity prediction model with optimal prediction performance is obtained, and then the expanded data samples are used to learn, and the final prediction result can be output.

[0026] Preferably, in step S4, the data sequence is expanded for each critical flow velocity influencing factor, and a homogenization transformation is performed to generate a data subsequence; the correlation between the parent sequence and the subsequence is analyzed using the following formula.

[0027] .

[0028] in: In the formula, These represent the mother sequence and the newborn child sequence, respectively. Represents the correlation coefficient. Indicates the data sampling time point. Indicates the absolute difference. This represents the two extreme values ​​within the range of the absolute difference. This represents the resolution parameter.

[0029] Preferably, the weighting coefficients can be obtained by averaging the correlation coefficient data series.

[0030] .

[0031] In the formula, Represents the weighting coefficients. Indicates the time sampling period. Indicates the sequence length;

[0032] By replacing the initial connection weights in the network model with weight coefficients, and then relearning the expanded critical flow velocity influencing factor data samples, the predicted critical flow velocity value for pipeline siltation can be automatically obtained.

[0033] The beneficial effects of this invention are as follows:

[0034] Compared with existing technologies, a method for predicting the critical flow velocity for siltation in slurry transport pipelines, based on a BP neural network, is designed. This method, through in-depth analysis of the main factors affecting the critical flow velocity, expands the data sample using the Mixup algorithm, and combines grey relational analysis to determine the weights of each factor, achieving accurate prediction of the critical flow velocity for siltation. This provides strong support for the safe operation and efficiency improvement of slurry transport pipelines and is of great significance for ensuring their safe operation. Attached Figure Description

[0035] Figure 1 This is a diagram illustrating the specific Beta distribution of the present invention.

[0036] Figure 2 This is a schematic diagram illustrating the neuron computation principle of this invention.

[0037] Figure 3 This is a diagram of the BP neural network prediction model of the present invention.

[0038] Figure 4 This is a diagram of the test device for the slurry conveying pipeline of the present invention.

[0039] Figure 5 This diagram illustrates the accuracy of model prediction results under different hidden layer node conditions according to the present invention.

[0040] Figure 6 This is a graph showing the change in network training error in this invention.

[0041] Figure 7 The graph shows the critical flow velocity prediction results of different methods in this invention.

[0042] Figure 8This is a comparison chart of the relative errors of the prediction results of the various methods in this invention. Detailed Implementation

[0043] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0044] In the analysis of critical flow velocity for siltation in slurry transport pipelines, the autoregressive moving average model is used for prediction. However, this model can only capture the linear relationship of historical data, leading to a relatively large error in the final prediction result. Therefore, a method for predicting critical flow velocity for siltation in slurry transport pipelines based on a backpropagation (BP) neural network is proposed. Starting from the calculation principle of critical flow velocity for siltation in slurry transport pipelines, the main factors affecting the critical flow velocity are identified and used as input variables for subsequent prediction analysis. Data is collected for each influencing factor, and the Mixup algorithm is applied to expand the data sample. A backpropagation neural network is used to establish an intelligent prediction model based on propagation learning. The influence weights of each factor are derived through grey relational analysis. The expanded sample and weight information are input into the network model to obtain the predicted value of the critical flow velocity for pipeline siltation. Experimental results show that the relative error of the prediction result of this method is always less than 0.5%, achieving accurate prediction and analysis of pipeline siltation.

[0045] See attached document Figure 1 -Appendix Figure 8 A method for predicting the critical flow velocity for siltation in slurry transport pipelines based on a BP neural network includes the following steps:

[0046] Step 1.1 Determine the factors affecting the critical flow velocity of pipeline siltation.

[0047] To address the issue of siltation in slurry transport pipelines, we reviewed current research findings on critical flow velocities in pipelines and selected three of the most commonly used methods for calculating critical flow velocities, as shown in formulas (1), (2), and (3).

[0048] (1)

[0049] (2)

[0050] (3)

[0051] In the formula, This indicates the critical flow velocity for pipe siltation. This represents Durand's velocity factor. This represents the diameter of the slurry delivery pipe. This represents the density parameter values ​​of liquid, material, and slurry. Represents volume fraction. Represents the drag coefficient. It represents the acceleration due to gravity.

[0052] As can be seen from the above calculation principle of the critical flow velocity for pipeline siltation, the main factors affecting the flow velocity change in slurry conveying pipelines include the pipe diameter, particle size, solid material density, and volume fraction. Therefore, the simplified formula for calculating the critical flow velocity of a pipeline is defined as follows:

[0053] (4)

[0054] As can be seen from formula (4), changes in the values ​​of each factor will cause a significant change in the critical flow velocity. These four factors are considered as the main variables required for predicting the critical flow velocity of slurry transport pipeline siltation, and corresponding data samples are collected.

[0055] Step 1.2 Establish a Mixup data augmentation processing scheme.

[0056] To better uncover the inherent features and patterns in the original data samples, the Mixup algorithm is introduced to augment the original data, generating an expanded dataset that satisfies the principle of minimizing neighborhood risk. In the actual data augmentation process, the principle of minimizing neighborhood risk is applied, and the expected risk estimation formula is defined as follows:

[0057] (5)

[0058] In the formula, Indicates expected risk. Represents the target data point. Indicates neighboring data points. This represents the total number of data points in the current dataset. Indicates the data point number. This indicates a neighborhood distribution.

[0059] The advantages of introducing the Mixup algorithm to augment the original data are as follows: 1. Traditional data augmentation easily destroys the physical meaning of the data, while Mixup generates mixed weights for the original sample pairs according to the beta distribution, constructing new samples. 2. The actual operating conditions of slurry transport pipelines are continuous, while the original experimental data are mostly discrete typical operating conditions. The augmented data generated by Mixup covers "atypical but reasonable" intermediate operating conditions, enabling the trained BP neural network to more accurately cope with continuously changing operating conditions in actual engineering, reducing prediction errors caused by operating conditions exceeding the training sample range. 3. It makes the augmented dataset more evenly distributed, avoiding excessive bias of the network towards a certain type of sample during training.

[0060] To minimize the expected risk, a dataset based on sampling from a neighboring distribution can be generated.

[0061] (6)

[0062] In the formula, This represents an augmented dataset. Indicates the data dimension.

[0063] For the newly generated augmented dataset, its neighborhood distribution Mixup can be defined as follows:

[0064] (7)

[0065] In the formula, Indicates the neighborhood distribution Mixup. This indicates the number of the newly generated sample point selected randomly. This indicates mixed weights.

[0066] Mixing weights are parameters used to control the degree of mixing between the original and new data, and their value range is as follows:

[0067] (8)

[0068] In the formula, Indicates the beta distribution, This represents two hyperparameters. The specific details of the Beta distribution are as follows: Figure 1 As shown.

[0069] By randomly selecting a portion of newly generated data from the Mixup distribution, an augmented dataset can be formed for subsequent prediction of critical flow velocities for pipeline siltation.

[0070] The advantages of using newly generated data in the Mixup distribution are: 1. Randomly selecting a portion of the newly generated data ensures that the size of the augmented dataset meets the network training requirements (reaching a total sample size of 200, conforming to the parameter setting of "original 100 → expanded to 200" in Table 2), while avoiding redundant computation, ensuring that the model has real-time prediction capabilities in practical engineering applications. 2. Random selection maintains the statistical randomness of the augmented data, ensuring that its distribution characteristics are consistent with the original data, making the patterns learned by the BP neural network more closely resemble the real critical flow velocity influence mechanism, rather than the features of artificially selected data.

[0071] Innovative steps: After constructing the BP neural network prediction model, instead of directly using the initial connection weights of the network, the weight coefficients of each influencing factor are calculated through grey relational analysis (Formula 10-12), and these weights are used to replace the initial connection weights to retrain the network to output the critical flow velocity prediction value.

[0072] The advantages are as follows: 1. Traditional BP neural networks rely on data-driven gradient descent for weight updates, which cannot intuitively reflect the contribution of each influencing factor to the critical flow velocity, resulting in a lack of physical meaning in the model's prediction results. Grey relational analysis, by calculating the correlation coefficient between the parent and child sequences, clarifies the importance ranking of "pipe diameter weight 0.3 > slurry density weight 0.25 > flow velocity weight 0.2" (Table 2), making the weight distribution of the BP neural network consistent with engineering theory. 2. It enables the BP neural network to be trained from a starting point "closer to the optimal solution". 3. It can weaken the interference of single outlier samples, making the weight coefficients more robust.

[0073] Step 1.3 Construct a BP neural network flow velocity prediction model.

[0074] In the process of predicting the critical flow velocity for siltation in slurry transport pipelines, a BP neural network is introduced to construct an intelligent prediction model. Figure 2 The neural computation principle shown continuously learns and mines the regular features contained in the expanded data, providing a basis for the prediction of critical flow velocity.

[0075] Figure 2 In the formula, the parameters are These represent the network input information, output information, and weight parameters, respectively. This represents the activation function. Indicates an additional signal.

[0076] Considering that critical velocity prediction involves four key variables, a construction is made. Figure 3 The BP neural network prediction model shown is a BP neural network topology diagram. The data corresponding to each variable is imported into the corresponding input layer neuron node. Then, combined with the connection weights and thresholds, the gradient descent method is used to control the hidden layer nodes to complete the forward and backward propagation learning.

[0077] The advantages of choosing the above approach include: 1. During forward propagation, the input layer transmits data on various influencing factors to the hidden layer. Combining the weight coefficients and thresholds obtained from grey relational analysis, a nonlinear transformation is achieved through the ReLU activation function (Table 2), and finally, the output layer outputs the predicted critical flow velocity. 2. Backpropagation calculates the error between the predicted and actual values ​​using gradient descent and updates the connection weights of each neuron along the network layers. For example, when the predicted critical flow velocity is higher than the actual value, the model will reduce the weights of positively correlated factors such as "pipe diameter" and increase the weights of negatively correlated factors such as "density" through backpropagation, gradually reducing the error. Figure 6 The error curves show that after 1000 iterations, the errors of both the training and validation sets converge to below 0.0001, proving that the method can effectively optimize network parameters and enable the model prediction accuracy to meet engineering requirements.

[0078] choose Figure 3The network prediction model and related methods shown have the following advantages: the four neurons in the input layer correspond to the four core influencing factors (Formula 4) of "pipe diameter, particle size, solid material density, and volume fraction", with no redundant input variables, avoiding the "curse of dimensionality" caused by too many input dimensions; the one neuron in the output layer directly outputs the critical flow velocity prediction value, the network structure is simple and the computational efficiency is high.

[0079] Figure 3 In the BP neural network prediction model shown, the number of nodes in the input layer is determined by the number of variables affecting the critical flow velocity, while the number of nodes in the hidden layer is adjustable. In order to reveal the inherent laws of high-quality data samples, the optimal number of nodes in the hidden layer is determined by formula (9).

[0080] (9)

[0081] In the formula, This represents the optimal number of hidden layer nodes. These represent the number of nodes in the network input layer and output layer, respectively. Represents a constant.

[0082] After the basic structure of the BP neural network prediction model is determined, the network model is trained using partial sample data. Once the training convergence requirement is met, a critical flow velocity prediction model with optimal prediction performance is obtained. Then, by learning from the expanded data samples, the final prediction result can be output.

[0083] Step 1.4 Generate the predicted critical flow velocity value for siltation in the slurry transport pipeline.

[0084] When predicting the critical flow velocity for siltation in slurry transport pipelines using a BP neural network model, the weights of various critical flow velocity influencing factors are calculated using a grey relational analysis algorithm and simultaneously input into the prediction model to derive more accurate flow velocity prediction values.

[0085] The grey relational analysis algorithm has the following advantages: 1. By calculating the correlation coefficient between the parent and child sequences, grey relational analysis clarifies the importance ranking of "pipe diameter weight 0.3 > slurry density weight 0.25 > flow velocity weight 0.2" (Table 2), making the weight distribution of the BP neural network consistent with engineering theory. 2. The weight coefficients derived from the analysis are based on the calculation principle of critical flow velocity (Formulas 1-4), reflecting the true influence of each factor. Using them as the initial weights of the network allows the BP neural network to start training from a point "closer to the optimal solution". 3. By calculating the average correlation coefficient of multiple sets of data (Formula 12), grey relational analysis can weaken the interference of single outlier samples, making the weight coefficients more robust.

[0086] For each critical flow velocity influencing factor, the data sequence is expanded and homogenized to generate a subsequence. The correlation between the parent sequence and the subsequence is analyzed using formula (10), yielding the following:

[0087] (10)

[0088] in:

[0089] (11)

[0090] In the formula, These represent the mother sequence and the newborn child sequence, respectively. Represents the correlation coefficient. Indicates the data sampling time point. Indicates the absolute difference. This represents the two extreme values ​​within the range of the absolute difference. This represents the resolution parameter.

[0091] The weighting coefficients can be obtained by averaging the correlation coefficient data series.

[0092] (12)

[0093] In the formula, Represents the weighting coefficients. Indicates the time sampling period. Indicates the sequence length.

[0094] By replacing the initial connection weights in the network model with weight coefficients, and then relearning the expanded critical flow velocity influencing factor data samples, the predicted critical flow velocity value for pipeline siltation can be automatically obtained.

[0095] Replacing the initial connection weights in the network model with weight coefficients has the following advantages: The high weight coefficient (0.3) of the pipe diameter will make the network focus on the influence of the pipe diameter on the critical flow velocity in the early stage of training, avoiding directional deviation caused by random initial weights. According to the training error curve, the model can converge within 1000 iterations, which is much faster than the traditional network with random initial weights.

[0096] The experimental process is analyzed below.

[0097] 2.1 Experimental Environment

[0098] After the critical velocity prediction method based on BP neural network was designed, an indoor test device for slurry conveying pipelines was built to verify the practical application effect of the method. Figure 4 As shown.

[0099] from Figure 4Seven pipe sections were selected from the data, and the corresponding test data are shown in Table 1. Based on this, a critical sedimentation velocity prediction experiment was conducted.

[0100] Table 1 Test Data

[0101]

[0102] The BP neural network prediction method, the SSA-CNN-based method, and the machine learning-based method were applied to the experiment to predict the critical flow velocity for siltation in the target pipeline. The results were compared with the critical flow velocity data in actual operation to demonstrate the accuracy of the prediction results of each method.

[0103] 2.2 Network Model Parameter Settings

[0104] The BP neural network model is the core of the BP neural network prediction method for predicting the critical flow velocity for pipeline siltation. To determine optimal network model parameters, the number of hidden layer network nodes was initially increased from 0 to 10, and the change in the model's prediction accuracy was observed. Figure 5 The statistical results are shown.

[0105] from Figure 5 It can be seen that when the number of hidden layer nodes reaches 8, the prediction accuracy reaches 0.9, and then tends to stabilize. Therefore, the optimal number of hidden layer network nodes is set to 8.

[0106] Prepare a large amount of experimental data to form a training set and a validation set, input them into the BP neural network model for learning, and observe the changes in the model's output error as follows. Figure 6 As shown.

[0107] from Figure 6 It can be seen that after applying the training set and validation set to the network model and completing 1000 consecutive iterations, both sets can meet the convergence requirements, proving that the network prediction model has good application performance.

[0108] Based on the above process, the parameters of the method in this paper are set as shown in Table 2.

[0109] Table 2. Method Parameters Table

[0110]

[0111] The previous text explained that the four main factors affecting the flow velocity of the slurry conveying pipeline are the pipe diameter, particle size, solid material density, and volume fraction. However, there are five variables here, and the reason why the variables are different is that the flow velocity is calculated and predicted by weighting the flow velocity together with other factors. The variables are related to the influencing factors, but they are not one-to-one correspondences.

[0112] 2.3 Analysis of Critical Velocity Prediction Results

[0113] Using a trained network prediction model, the critical flow velocity for siltation was predicted for seven experimental pipe sections, ultimately yielding... Figure 7 The statistical results are shown below. For easier comparison and analysis, the prediction results and actual critical velocity values ​​of the other two methods are also presented. Figure 7 middle.

[0114] from Figure 7 It can be seen that the BP neural network prediction method gives a critical velocity prediction value that is very close to the actual value, while the prediction results of the other two methods differ significantly from the actual values.

[0115] Relative error was selected as the evaluation index for... Figure 7 The critical velocity prediction results shown are further analyzed to reflect the accuracy of the prediction results of each method, and based on this, a system is established. Figure 8 The comparison images shown.

[0116] (13)

[0117] In the formula, This indicates the relative error of the critical flow velocity prediction result. This represents the predicted critical flow velocity value. This represents the actual critical flow velocity value.

[0118] from Figure 8 It can be seen that after applying the BP neural network prediction method, the relative error of the prediction results is always below 0.5%, with an average of only 0.21%. In contrast, the average relative errors of the other two prediction methods reached 1.4% and 2.05%, respectively. In conclusion, extending the BP neural network-based critical velocity prediction method to the treatment of slurry transport pipeline siltation will help to better prevent pipeline siltation.

[0119] Based on the above research, a method for predicting the critical flow velocity for siltation in slurry transport pipelines was designed, using a BP neural network as its core. This method, through in-depth analysis of the main factors affecting the critical flow velocity, expands the data sample using the Mixup algorithm, and combines grey relational analysis to determine the weights of each factor, achieving accurate prediction of the critical flow velocity for siltation. This provides strong support for the safe operation and efficiency improvement of slurry transport pipelines and is of great significance for ensuring the safe operation of slurry transport pipelines.

[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting a critical flow velocity of a slurry pipeline based on a BP neural network, characterized in that, It comprises the following steps: S1, determine the pipe diameter, particle size, solid material density, volume fraction of four items as the main factors affecting the flow velocity variation of slurry pipeline, and collect corresponding data samples; S2, introduce the Mixup algorithm to enhance the original data, generate an expanded data set that meets the principle of neighborhood risk minimization; S3, in the process of predicting the critical flow velocity of slurry pipeline deposition, introduce BP neural network to build an intelligent prediction model, and continuously learn and mine the characteristics of the expanded data to provide basis for the prediction of critical flow velocity; S4, rely on the BP neural network model to complete the prediction of the critical flow velocity of slurry pipeline deposition, use the grey correlation analysis algorithm to calculate the weight of each critical flow velocity influencing factor, input it into the prediction model at the same time, and deduce a more accurate flow velocity prediction value.

2. The method of claim 1, wherein the BP neural network-based prediction of the critical velocity of the slurry pipeline is based on the following equation: wherein, Vcrit is the critical velocity of the slurry pipeline, D is the diameter of the slurry pipeline, g is the gravitational acceleration, and p is the density of the slurry. In the actual data expansion process in step S2, the minimum neighborhood risk principle is used, and the expected risk estimation formula is defined as: ； wherein, represents the desired risk, represents the target data point, represents the neighboring data point, represents the total amount of data points in the current data set, represents the data point number, represents the neighboring distribution. 3.The method of claim 1, wherein, In step S2, in order to meet the minimum expected risk, a data set based on neighborhood distribution sampling can be generated; ； wherein denotes an enhanced data set, denotes a data dimension; The neighborhood distribution Mixup of the newly generated enhanced data set can be defined as: ； wherein, denotes the neighboring distribution Mixup, denotes the randomly selected newly generated sample point number, denotes the mixing weight.

4. The method of claim 3, wherein the BP neural network-based prediction of the critical velocity of the slurry pipeline is based on the following equation: wherein, Vcrit is the critical velocity of the slurry pipeline, D is the diameter of the slurry pipeline, g is the gravitational acceleration, and p is the density of the slurry. The mixing weight is a parameter used to control the mixing degree of the original data and the new data, and its value range is: ； wherein denotes a Beta distribution, denotes two hyperparameters; Randomly select part of the new data from the Mixup distribution to form an enhanced data set, which is used for subsequent prediction of the critical flow velocity of pipeline deposition.

5. The BP neural network-based method for predicting the critical velocity of a slurry pipeline according to claim 1, wherein, In step S3, the data corresponding to each variable is imported into the corresponding input layer neuron node of the BP neural network, and then combined with the connection weight and threshold value, the gradient descent method is used to control the hidden layer nodes to complete the forward propagation and backward propagation learning.

6. The BP neural network-based method for predicting the critical velocity of a slurry pipeline according to claim 5, wherein, The BP neural network topology structure diagram of the BP neural network prediction model is constructed, the data corresponding to each variable is imported into the corresponding input layer neuron node, and then combined with the connection weight and threshold value, the gradient descent method is used to control the hidden layer nodes to complete the forward propagation and backward propagation learning.

7. The BP neural network-based method for predicting the critical velocity of a slurry pipeline according to claim 6, wherein, In the BP neural network topology structure diagram of the BP neural network prediction model, the number of input layer nodes is determined by the number of critical flow velocity influencing variables, and the number of nodes in the hidden layer is adjustable; the optimal number of hidden layer nodes is determined by the following formula: ； In the formula, represents the optimal number of hidden layer nodes, respectively represent the number of network input layer nodes and output layer nodes, represents a constant.

8. The BP neural network-based method for predicting the critical velocity of a slurry pipeline according to claim 1, wherein, After the basic structure of the BP neural network prediction model in step S3 is determined, some sample data is applied to complete the network model training; after meeting the training convergence requirements, the critical flow velocity prediction model with optimal prediction performance is obtained, and then the expanded data samples are learned, and the final prediction result can be output. 9.The BP neural network-based method for predicting the critical velocity of a slurry pipeline according to claim 1, wherein, In step S4, for each critical flow velocity influencing factor expanded data sequence, a homogeneous transformation process is performed to generate a data sub-sequence; the correlation between the parent sequence and the sub-sequence is analyzed by the following formula: ； wherein: ; in the formula, respectively represent a mother sequence, a newborn daughter sequence, denotes a correlation coefficient, denotes a data sampling time point, denotes an absolute difference value, denotes two extreme values in the absolute difference value range, denotes a resolution parameter.

10. The BP neural network-based method for predicting the critical velocity of a slurry pipeline according to claim 9, wherein, The weight coefficient can be obtained by averaging the correlation coefficient data sequence; ； wherein denotes a weight coefficient, denotes a time sampling period, denotes a sequence length; The weight coefficient is used to replace the originally set connection weight in the network model, and then the expanded critical flow velocity influencing factor data samples are relearned, and the pipeline deposition critical flow velocity prediction value can be automatically obtained.

Images