Method for determining slurry conveying pressure parameters of dredging hydraulic filling air-entrapping boosting pipeline

Abstract

The invention discloses a method for determining slurry conveying pressure parameters of a dredging and hydraulic reclamation air-entrapping boosting pipeline, and belongs to the technical field of dredging and hydraulic reclamation engineering. The problems that in the prior art, pressure parameters are difficult to determine, and gas filling boosting cannot be suitable for different projects are solved, the data rejection amount of each parameter is judged, key information is prevented from being lost, and the training accuracy of the convolutional neural network model is ensured; the attention mechanism is introduced to automatically learn the weights of the slurry parameter characteristics and the pipeline parameter characteristics, the channel attention mechanism is utilized to distribute different weights for different types of characteristic channels, and the space attention mechanism is combined to monitor and focus on the characteristic parameters corresponding to the important weights in real time; and when the verification result exceeds a preset error threshold value, the convolutional neural network model focuses on the training of the key features by adjusting the data proportion of the feature parameters corresponding to the important weights, so that the gas filling parameters can be accurately predicted under different working conditions.

Description

Technical Field

[0001] This invention relates to the field of dredging and reclamation engineering technology, specifically to a method for determining the pressure parameters of slurry transport in a dredging and reclamation aerated propulsion pipeline. Background Technology

[0002] In dredging and reclamation projects, pipeline slurry transportation is a common construction method. Traditional pipeline slurry transportation relies on pump pressure to propel the slurry flow, but this method has many problems when transporting long distances and large volumes, such as high energy consumption, low transportation efficiency, and severe pipeline wear. To solve these problems, aeration technology has been gradually applied to pipeline slurry transportation; aeration involves injecting a certain amount of gas into the pipeline to change the flow characteristics of the slurry, thereby improving transportation efficiency and reducing energy consumption.

[0003] However, it is currently difficult to accurately predict the impact of changes in slurry concentration and gas ratio on the conveying pressure, resulting in low conveying efficiency; and the inability to dynamically adjust the gas injection parameters affects the stability and safety of the conveying process.

[0004] Therefore, it does not meet the existing requirements, so we propose a method for determining the slurry transport pressure parameters of dredging, filling and aeration-assisted pipelines. Summary of the Invention

[0005] The purpose of this invention is to provide a method for determining the slurry delivery pressure parameters of dredging, reclamation, and aerated concrete propulsion pipelines. By judging the amount of data discarded for each parameter, key information loss is avoided, ensuring the accuracy of convolutional neural network model training. An attention mechanism is introduced to automatically learn the weights of slurry and pipeline parameter features. A channel attention mechanism is used to assign different weights to different types of feature channels. A spatial attention mechanism is combined to monitor and focus on the feature parameters corresponding to important weights in real time. When the verification result exceeds a preset error threshold, the data proportion of the feature parameters corresponding to important weights is adjusted, making the convolutional neural network model more focused on training key features. This ensures accurate prediction of aerated concrete parameters under different operating conditions, solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for determining slurry transport pressure parameters in a dredging, reclamation, and aerated propulsion pipeline, the method being based on the influence of slurry concentration, gas ratio, and pipeline parameter characteristics on the transport propulsion effect, the method comprising the following steps:

[0008] Collect the characteristics of different types of pipeline parameters, as well as the parameters of the dredged slurry and aeration parameters, as data samples; the slurry parameters include the concentration of slurry and the proportion of gas, and the aeration parameters include the pumping pressure and the gas injection pressure;

[0009] The collected data samples are cleaned and preprocessed, and it is determined whether the number of data samples needs to be supplemented based on the cleaning results. Then, the slurry parameter characteristics and pipeline parameter characteristics that have the greatest impact on the gasification parameter characteristics are extracted from the data samples using the correlation analysis method.

[0010] A convolutional neural network model is constructed based on data samples. The slurry parameter features and pipeline parameter features are used as input features to train the convolutional neural network model and predict the gas filling parameters.

[0011] The accuracy of the prediction results is verified based on the characteristics of the gas filling parameters, and the weights of the convolutional neural network model are adjusted according to the verification results.

[0012] The new slurry concentration, gas concentration, and corresponding pipeline parameters are obtained and input into a trained convolutional neural network model for prediction to obtain the corresponding pumping pressure and gas injection pressure.

[0013] Furthermore, based on the cleaning results, it is determined whether the number of data samples needs to be increased, specifically as follows:

[0014] Based on the training requirements of the convolutional neural network model, the removal thresholds for each parameter in the data samples are preset.

[0015] Extract the cleaned data sample, compare the amount of data before and after cleaning for each parameter in the data sample, and obtain the amount of data to be removed for each parameter.

[0016] The data removal evaluation coefficient corresponding to each parameter is calculated using the removal amount corresponding to each parameter. The data removal evaluation coefficient corresponding to each parameter is compared with the preset removal threshold to obtain the comparison results.

[0017] From the comparison results, obtain the number of samples whose evaluation coefficient exceeds the removal threshold after data removal;

[0018] If the number of samples exceeding the removal threshold is one, then it is determined that no data addition is needed.

[0019] When the number of samples exceeding the removal threshold is greater than one, it is determined that data addition is required.

[0020] Furthermore, the removal thresholds for various parameters in the data samples pre-defined for training based on the convolutional neural network model are compensated and adjusted to obtain the pre-defined removal thresholds, including:

[0021] Extract data samples from each item;

[0022] Retrieve historical training data of the convolutional neural network model;

[0023] The evaluation coefficients of each data sample are removed from the historical training data of the convolutional neural network model.

[0024] The standard deviation of the data elimination evaluation coefficient corresponding to each data sample is obtained based on the data elimination evaluation coefficient of each data sample.

[0025] Retrieve the number of times data needs to be added to the data samples from the historical training data of the convolutional neural network model;

[0026] For each data sample, retrieve the number of data samples that need to be added for each data sample increment;

[0027] The maximum difference in the number of samples to be added is obtained based on the number of samples that need to be added each time the data sample is increased.

[0028] Retrieve the two training moments corresponding to the maximum difference in the number of samples increased from the historical training data, and obtain the training time interval based on the two training moments;

[0029] The data is removed by removing the standard deviation of the evaluation coefficient, the maximum difference in the number of samples added to the historical training data, and the removal thresholds for various parameters in the data samples corresponding to the current training of the convolutional neural network model based on the training time interval.

[0030] Furthermore, the data is removed by removing the standard deviation of the evaluation coefficient, the maximum difference in the number of samples added to the historical training data, and the removal thresholds for various parameters in the data samples corresponding to the current training of the convolutional neural network model, including:

[0031] Retrieve the number of samples corresponding to each two consecutive data additions and the time interval corresponding to each two consecutive training iterations from the historical training data where data samples need to be added.

[0032] The ratio of the increase in the number of samples between each two adjacent data points to the time interval between each two adjacent training points is used to obtain the first sample time ratio.

[0033] The average value of the first sample time ratio is obtained by adding the corresponding first sample time ratio to each two adjacent data points;

[0034] Retrieve data from historical training data, including the maximum difference in the number of samples and their corresponding training time intervals;

[0035] The second sample time ratio is obtained by increasing the number of samples in the historical training data and comparing it with the corresponding training time interval.

[0036] The time decay coefficient is obtained by using the ratio of the first sample time to the ratio of the second sample time.

[0037] The preset removal thresholds for each parameter in the data sample corresponding to the current training of the convolutional neural network model are obtained by combining the standard deviation of the data removal evaluation coefficient with the time decay coefficient and the preset removal thresholds for each parameter in the data sample based on the training requirements of the convolutional neural network model.

[0038] Furthermore, the accuracy of the prediction results is verified based on the characteristics of the gas filling parameters, and the weights of the convolutional neural network model are adjusted according to the verification results, including:

[0039] Based on historical training data of the convolutional neural network model, an error threshold is preset;

[0040] An attention mechanism is introduced into the convolutional neural network model to automatically learn feature weights;

[0041] Different weights are assigned to the slurry parameter feature channels and the pipeline parameter feature channels based on channel attention, and the feature parameters corresponding to the important weights are monitored in real time based on spatial attention.

[0042] When the validation results of the convolutional neural network model exceed the error threshold, the proportion of data corresponding to the feature parameters of important weights in the data samples is adjusted so that the convolutional neural network model focuses on training the feature parameters corresponding to important weights.

[0043] Furthermore, it was determined that data augmentation was needed for the data sample, specifically as follows:

[0044] From the comparison results, retrieve the data sample types and data removal evaluation coefficients corresponding to the data sample increase required;

[0045] Based on the evaluation coefficients removed from the data, determine the amount of data that needs to be supplemented to the data sample.

[0046] According to the corresponding data volume, increase the number of data samples for the corresponding data sample type.

[0047] Furthermore, the accuracy of the prediction results is verified based on the characteristics of the gas filling parameters, and the weights of the convolutional neural network model are adjusted according to the verification results. This also includes:

[0048] The characteristics of each group of slurry parameters and pipeline parameters in the data sample, as well as the corresponding aeration parameters, are sorted and divided.

[0049] The gradient analysis algorithm is used to calculate the gradient of each group of slurry parameter characteristics and pipeline parameter characteristics on the output aeration parameters. The importance of the influence of slurry parameter characteristics and pipeline parameter characteristics on aeration parameters is evaluated based on the gradient results.

[0050] The evaluation results show that the larger the absolute value of the gradient, the greater the influence of this parameter feature on the gas filling parameters output by the convolutional neural network model.

[0051] The weights of the convolutional neural network model are adjusted based on the evaluation results of feature importance.

[0052] Furthermore, based on correlation analysis, the slurry parameter characteristics and pipeline parameter characteristics that have the greatest impact on the gasification parameter characteristics are extracted from the data sample, specifically:

[0053] The Pearson correlation coefficient between slurry parameter characteristics, pipeline parameter characteristics and aeration parameter characteristics was calculated based on the Pearson correlation analysis algorithm.

[0054] By setting a preset coefficient threshold, the closer the absolute value of the correlation coefficient of the parameter feature in the calculation results is to the coefficient threshold, the stronger the correlation of the parameter feature. The slurry parameter feature and pipeline parameter feature with the strongest correlation with the gas filling parameter feature are obtained.

[0055] Furthermore, the collected data samples are preprocessed, specifically as follows:

[0056] The Z-score method is used to identify and remove outliers in the data samples, interpolation is used to fill in missing data, and duplicate data in the data samples are deleted.

[0057] The slurry parameter characteristics, pipeline parameter characteristics, and aeration parameter characteristics are normalized so that all parameters are scaled to a specified range.

[0058] Furthermore, the method also includes:

[0059] By installing pressure sensors and flow meters on the pipeline, the changes in various parameters during the transportation process can be monitored in real time;

[0060] Compare the changes in various parameters with the prediction results to determine whether the gas refueling parameters given in the prediction results are accurate.

[0061] Based on the comparison results, determine whether dynamic adjustment of the pressurization parameters is necessary.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] 1. In this invention, by setting the removal thresholds for each parameter and calculating the data removal evaluation coefficient, the amount of data removal for each parameter that exceeds the reasonable range can be accurately identified, thereby determining whether data samples need to be supplemented to avoid the loss of key information; and data samples are added in a targeted manner according to the data types that exceed the removal threshold, so as to restore the integrity and representativeness of the data and ensure the accuracy and reliability of the training of the convolutional neural network model.

[0064] 2. In this invention, by introducing an attention mechanism, the convolutional neural network model can automatically learn and dynamically adjust the weights of slurry parameter features and pipeline parameter features, thereby more accurately capturing features that have a key impact on the prediction of gas filling parameters. A channel attention mechanism is used to assign different weights to different types of feature channels, and a spatial attention mechanism is combined to monitor and focus on the feature parameters corresponding to important weights in real time. When the verification result exceeds a preset error threshold, the proportion of data corresponding to the feature parameters of important weights in the data samples is adjusted, making the convolutional neural network model focus more on training key features, thereby improving its prediction accuracy and generalization ability. This ensures that gas filling parameters can be accurately predicted under different working conditions, thereby optimizing the delivery process, reducing energy consumption, and improving engineering efficiency. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating the method for determining the slurry transport pressure parameters in the dredging, filling, and aeration-assisted pipeline of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] To address the technical problem in existing technologies where it is difficult to accurately determine the slurry delivery pressure parameters in aerated propulsion pipelines, making existing aerated propulsion systems unsuitable for different engineering scenarios, please refer to [link / reference needed]. Figure 1 This embodiment provides the following technical solution:

[0068] A method for determining the slurry transport pressure parameters in a dredging, reclamation, and aerated propulsion pipeline is based on the influence of slurry concentration, gas ratio, and pipeline parameter characteristics on the transport propulsion effect. The method includes the following steps:

[0069] Collect different types of pipeline parameter characteristics, as well as the parameters of the dredged slurry and aeration parameters, as data samples; slurry parameters include slurry concentration and gas content, and aeration parameters include pumping pressure and gas injection pressure; slurry parameters also include slurry density, viscosity, solid particle content, and particle size distribution; pipeline parameter characteristics include pipeline inner diameter, length, material, roughness, and laying method such as horizontal or inclined; aeration parameters include gas injection flow rate, gas injection pressure, and gas content.

[0070] The collected data samples were cleaned and preprocessed, and based on the cleaning results, it was determined whether the number of data samples needed to be increased. Then, correlation analysis was used to extract the slurry parameter characteristics and pipeline parameter characteristics that had the greatest impact on the aeration parameter characteristics from the data samples, specifically:

[0071] The Z-score method is used to identify and remove outliers from the data samples. Interpolation is used to fill in missing data and duplicate data is removed. Normalization is performed on the slurry parameters, pipeline parameters, and aeration parameters to scale all parameters to a specified range. Data cleaning removes outliers and missing values ​​from the data samples. Normalization scales the data to the [0,1] range to improve the training efficiency and convergence speed of the neural network.

[0072] The Pearson correlation coefficients between slurry and pipeline parameters and aeration parameters were calculated using the Pearson correlation analysis algorithm. A threshold value was preset for the coefficients; the closer the absolute value of the correlation coefficient for a parameter feature was to the threshold, the stronger the correlation. The slurry and pipeline parameters with the strongest correlation to the aeration parameters were identified. Based on the correlation analysis, feature variables that significantly influence the prediction of delivery pressure parameters were selected to reduce input dimensionality and improve the generalization ability of the convolutional neural network model. The Pearson correlation coefficient between each slurry and pipeline parameter and the aeration parameters was calculated. The formula is:

[0073]

[0074] Where X and Y represent the feature and target outputs, respectively; and Let it be expressed as their mean.

[0075] Based on the training requirements of the convolutional neural network model, the removal thresholds for each parameter in the data samples are preset. Data samples after cleaning are extracted, and the data volume before and after cleaning for each parameter is compared to obtain the removal amount for each parameter. The data removal evaluation coefficient for each parameter is calculated using the removal amount, and this coefficient is compared with the preset removal threshold to obtain the comparison results. From the comparison results, the number of samples whose data removal evaluation coefficient exceeds the removal threshold is obtained. If only one sample exceeds the threshold, it is determined that no data addition is needed. If more than one sample exceeds the threshold, data addition is determined. From the comparison results, the data sample type and data removal evaluation coefficient corresponding to the data sample requiring addition are retrieved. Based on the data removal evaluation coefficient, the amount of data to be added to the data sample is obtained. Data addition is performed on the corresponding data sample type according to the corresponding data volume.

[0076] The beneficial effects achieved by the above are as follows: by setting the removal thresholds for each parameter in the preset parameters and calculating the data removal evaluation coefficient, it is possible to accurately identify when the amount of data removal for each parameter exceeds the reasonable range, and then determine whether it is necessary to supplement the data samples to avoid the loss of key information; and to add data samples in a targeted manner according to the data types that exceed the removal threshold, thereby restoring the integrity and representativeness of the data and ensuring the accuracy and reliability of the training of the convolutional neural network model.

[0077] Specifically, the removal thresholds for various parameters in the data samples pre-defined for training requirements based on the convolutional neural network model are compensated and adjusted to obtain the pre-defined removal thresholds, including:

[0078] Extract data samples from each item;

[0079] Retrieve historical training data of the convolutional neural network model;

[0080] The evaluation coefficients of each data sample are removed from the historical training data of the convolutional neural network model.

[0081] The standard deviation of the data elimination evaluation coefficient corresponding to each data sample is obtained based on the data elimination evaluation coefficient of each data sample.

[0082] Retrieve the number of times data needs to be added to the data samples from the historical training data of the convolutional neural network model;

[0083] For each data sample, retrieve the number of data samples that need to be added for each data sample increment;

[0084] The maximum difference in the number of samples to be added is obtained based on the number of samples that need to be added each time the data sample is increased.

[0085] Retrieve the two training moments corresponding to the maximum difference in the number of samples increased from the historical training data, and obtain the training time interval based on the two training moments;

[0086] The data is removed by removing the standard deviation of the evaluation coefficient, the maximum difference in the number of samples added to the historical training data, and the removal thresholds for various parameters in the data samples corresponding to the current training of the convolutional neural network model based on the training time interval.

[0087] The technical effects of the above solution are as follows: First, extract the data samples from the current training and retrieve the historical training data of the Convolutional Neural Network (CNN) model as a reference for threshold compensation adjustment. Obtain the "data removal evaluation coefficient" for each data sample from the historical training data, calculate its standard deviation (reflecting the degree of coefficient fluctuation), and use it to measure the stability of the data samples. Statistically count the "number of data augmentations" in historical training and the "number of sample data" corresponding to each augmentation, and calculate the maximum difference in the number of samples added, reflecting the degree of abrupt change in the scale of data augmentation. Retrieve the two training moments corresponding to the "maximum difference in the number of samples" and calculate their time interval to evaluate the impact of data augmentation frequency on model training. Combining the three indicators of "standard deviation of data removal evaluation coefficient," "maximum difference in the number of data augmented samples," and "training time interval," dynamically compensate and adjust the removal threshold of each parameter in the current training data sample to make the threshold adapt to the current data distribution and training state. By using the evaluation coefficient fluctuation (standard deviation) and data augmentation behavior (sample number difference, time interval) in the historical data, the removal threshold is adjusted in real time to avoid model training deviation caused by changes in data distribution. To avoid the problems of "over-removing valid data" or "retaining noisy data" caused by fixed thresholds, this method improves the purity and representativeness of the data samples. By dynamically adjusting the threshold based on the frequency and scale of data augmentation, it reduces training oscillations caused by fluctuations in data augmentation strategies, accelerating model convergence. The adaptive threshold adjustment mechanism enables the model to better learn data features, reduces the risk of overfitting, and improves generalization performance on data with different distributions. It eliminates the need for manually preset fixed thresholds; automatic compensation and adjustment based on historical data reduces manual intervention costs and improves training efficiency.

[0088] Specifically, the data is removed by removing the standard deviation of the evaluation coefficient, the maximum difference in the number of samples added to the historical training data, and the removal thresholds for various parameters in the data samples corresponding to the current training of the convolutional neural network model, including:

[0089] Retrieve the number of samples corresponding to each two consecutive data additions and the time interval corresponding to each two consecutive training iterations from the historical training data where data samples need to be added.

[0090] The ratio of the increase in the number of samples between each two adjacent data points to the time interval between each two adjacent training points is used to obtain the first sample time ratio.

[0091] The average value of the first sample time ratio is obtained by adding the corresponding first sample time ratio to each two adjacent data points;

[0092] Retrieve data from historical training data, including the maximum difference in the number of samples and their corresponding training time intervals;

[0093] The second sample time ratio is obtained by increasing the number of samples in the historical training data and comparing it with the corresponding training time interval.

[0094] The time decay coefficient is obtained by using the ratio of the first sample time to the ratio of the second sample time.

[0095] The time decay coefficient is obtained by the following formula:

[0096]

[0097] Where α represents the time decay coefficient; R 01p and R 02 These represent the average of the first sample time ratio and the second sample time ratio, respectively; λ represents the adjustment coefficient, and the adjustment coefficient ranges from 0.5 to 1.3; specifically, λ represents the relative disturbance degree between extreme fluctuations and steady-state rhythms, and is used to quantify the disturbance sensitivity of the training system. The time decay law of the impact of the reaction shock.

[0098] The preset removal thresholds for each parameter in the data sample corresponding to the current training of the convolutional neural network model are obtained by combining the standard deviation of the data removal evaluation coefficient with the time decay coefficient and the preset removal thresholds for each parameter in the data sample based on the training requirements of the convolutional neural network model.

[0099] The preset removal thresholds for each parameter in the data sample are obtained using the following formula:

[0100] Y = Y p *[1+(1-α)*(μ c +k*σ c )]

[0101] Where Y represents the threshold for removing various parameters from the data samples corresponding to the current training of the convolutional neural network model; p This represents the threshold values ​​for removing various parameters from the data samples predefined for training the convolutional neural network model; α represents the time decay coefficient; μ c σ represents the average value of the evaluation coefficients after data removal. c This represents the standard deviation of the evaluation coefficient after removing data. Specifically, (μ) c +k*σ c This is used to amplify the impact of data fluctuations, making the threshold more sensitive to data changes. (1-α)*(μ) c +k*σ c The threshold α represents the dynamic correction amount of the current data. The smaller α is, the higher the weight of the current data characteristics, forcing the threshold to adapt to the noise level of the current data first, rather than relying on the initial baseline. Ultimately, the threshold Y dynamically balances the "initial setting" and the "actual data / training state", simulating the law of "parameter adaptive adjustment to maintain stable operation" in a physical system.

[0102] The technical effects of the above solution are as follows: the ratio of the number of samples to the time interval between two adjacent data augmentations (the first sample time ratio) reflects the scale efficiency of data augmentation per unit time. The average value of the first sample time ratio represents the long-term stable level of the scale efficiency of data augmentation (similar to the steady-state characteristics of a system). The ratio of the maximum difference in the number of samples in data augmentation to its time interval (the second sample time ratio) represents the extreme fluctuation of the scale efficiency of data augmentation (equivalent to the transient impact on a system). Through the above technical solution, the dynamic impact of data augmentation and the characteristic fluctuations of data samples are mapped to the "transient interference" and "noise feedback" of the physical system. By simulating "parameter adaptive correction" through mathematical formulas, the elimination threshold is adapted to the dynamic changes in the training process, thus conforming to the objective law that "system parameters need to be dynamically adjusted with the environment to optimize performance".

[0103] Existing solutions often employ fixed thresholds or simple linear adjustments, which cannot handle dynamic changes in data characteristics during training (such as sample distribution shifts and noise intensity fluctuations). The aforementioned technical solutions utilize the standard deviation (σ) of historical data. c The data augmentation method uses the sample-to-time ratio (Rsample) and time decay coefficient (α) to dynamically adjust the removal threshold, allowing it to automatically adjust with data distribution. This technique analyzes the data augmentation's sample-to-time ratio (Rsample). 01p R 02The quantification of the attenuation effect of data augmentation on the threshold makes threshold adjustments more consistent with actual training conditions. When the scale of data augmentation fluctuates significantly, the time decay coefficient (α) automatically reduces threshold sensitivity, minimizing training oscillations caused by abrupt changes in data distribution and accelerating convergence. Existing solutions lack this "shock-resistant" mechanism, making them prone to gradient explosion or repeated fluctuations in model performance. This is further addressed by combining the mean (μ) of the data removal evaluation coefficient. c ) and standard deviation (σ) c The above technical solution can accurately identify the boundary between "data that needs to be removed" and "valid data," avoiding misjudgments based on fixed thresholds and improving data quality. The initial threshold (Y) is adjusted by α. p ) and real-time data feedback (μ c σ c By adjusting the weights of the data, the above-mentioned technical solutions achieve a dynamic balance between "following prior knowledge" and "adapting to new data," reducing the risk of overfitting. Traditional solutions rely on manually preset thresholds, which are difficult to adapt to complex and ever-changing real-world scenarios. The average of the ratio of the first sample time to the long-term and short-term characteristics of the data (R0) is used. 01p The second sample time ratio (R) reflects the long-term stable characteristics of data augmentation. 02 This approach captures extreme fluctuations, and the combined α value more comprehensively characterizes training dynamics. Existing solutions typically consider only a single dimension (such as current data statistics), easily ignoring the impact of historical training information on the threshold. The aforementioned technical solution automatically calculates the removal threshold through a mathematical model, reducing the cost of manual trial and error, and is particularly suitable for large-scale datasets or real-time training scenarios. Traditional methods require repeated manual adjustment of the threshold, which is inefficient and relies on experience. The robustness enhancement time decay mechanism makes the threshold robust to anomalous data augmentation behavior. Even if accidental large-scale data augmentation occurs, the threshold change can be smoothed through the decay effect of α, avoiding system collapse.

[0104] A convolutional neural network (CNN) model is constructed based on data samples. The required number of features and the dimension of the input layer are selected. The dimension of the input layer is determined by the number of features; for example, if there are 20 features, the input layer size is 20. The slurry parameter features and pipeline parameter features are used as input features to train the CNN model. Multiple convolutional kernels are used to perform convolution operations on the input features. Each convolutional layer contains multiple kernels, and the size of the kernels can be adjusted according to the feature dimension; for example, a kernel of size 3 can be used to extract local features of the slurry and pipeline parameters. Pooling layers are added after the convolutional layers to reduce the feature dimension and computational cost. After further extracting higher-level features using convolutional layers, fully connected layers are added to integrate the extracted features and output the final gasification parameters. The number of neurons in the output layer is determined by the dimension of the target output. For example, if two gasification parameters are predicted, such as pumping pressure and gas injection pressure, the output layer has 2 neurons.

[0105] The accuracy of the prediction results is verified based on the characteristics of the gas filling parameters. The weights of the convolutional neural network model are then adjusted based on the verification results, including:

[0106] The data samples were organized and categorized for each group of slurry and pipeline parameter features, along with their corresponding aeration parameters. Gradient analysis was used to calculate the gradient of each group of slurry and pipeline parameter features with respect to the output aeration parameters. The importance of these features on the aeration parameters was assessed based on the gradient results. The larger the absolute value of the gradient, the greater the influence of that parameter feature on the aeration parameters output by the convolutional neural network model. Based on the assessment of feature importance, the weights of the convolutional neural network model were adjusted to enhance the influence of important features and weaken the influence of unimportant features.

[0107] The system also includes: pre-setting an error threshold based on historical training data of the convolutional neural network (CNN) model; introducing an attention mechanism into the CNN model to automatically learn feature weights; assigning different weights to slurry parameter feature channels and pipeline parameter feature channels based on channel attention, and monitoring the feature parameters corresponding to important weights in real time based on spatial attention, such as the gas demand corresponding to high-concentration areas; calculating the gradient of each set of slurry parameter features and pipeline parameter features with respect to the output gasification parameters using gradient analysis; the larger the absolute value of the gradient, the greater the impact of the feature on the output; adjusting the model weights based on the evaluation results of feature importance. For important features, their corresponding weights are increased; for unimportant features, their corresponding weights are decreased. When the validation results of the CNN model exceed the error threshold, the data proportion of feature parameters corresponding to important weights in the data samples is adjusted so that the CNN model focuses on training the feature parameters corresponding to important weights.

[0108] The beneficial effects achieved by the above are as follows: By introducing an attention mechanism, the convolutional neural network model can automatically learn and dynamically adjust the weights of slurry parameter features and pipeline parameter features, thereby more accurately capturing features that have a key impact on the prediction of gas filling parameters; by using a channel attention mechanism to assign different weights to different types of feature channels, and combining it with a spatial attention mechanism to monitor and focus on the feature parameters corresponding to important weights in real time; when the verification result exceeds the preset error threshold, by adjusting the data proportion of feature parameters corresponding to important weights in the data samples, the convolutional neural network model focuses more on training key features, thereby improving its prediction accuracy and generalization ability, ensuring that gas filling parameters can be accurately predicted under different working conditions, thereby optimizing the delivery process, reducing energy consumption, and improving engineering efficiency.

[0109] The system acquires new slurry concentration, gas concentration, and corresponding pipeline parameters, and inputs them into a trained convolutional neural network model for prediction to obtain the corresponding pumping pressure and gas injection pressure. By installing pressure sensors and flow meters on the pipeline, the system monitors the changes in various parameters during the delivery process in real time. The changes in various parameters are compared with the prediction results to determine whether the gas filling parameters given by the prediction results are accurate. Based on the comparison results, the system determines whether the pressurization parameters need to be dynamically adjusted.

[0110] The beneficial effects achieved by the above methods are as follows: By using a convolutional neural network model to predict new slurry concentrations, gas ratios, and pipeline parameters, the corresponding pumping pressure and gas injection pressure can be accurately predicted. This allows engineers to optimize transportation parameters in advance during actual transportation, thereby improving transportation efficiency and reducing energy consumption. Simultaneously, by installing pressure sensors and flow meters on the pipeline for real-time monitoring and comparing the monitoring data with the prediction results, deviations between the predictions and actual conditions can be detected promptly. This allows engineers to dynamically adjust the gas injection parameters, ensuring the stability and safety of the transportation process and further improving the overall efficiency and economic benefits of the project. This method not only improves the accuracy and reliability of predictions but also enhances the adaptability and flexibility of the engineering system, making it of significant application value for dredging and reclamation projects.

[0111] Working principle: By collecting and preprocessing pipeline parameters, slurry parameters, and aeration parameters in dredging and reclamation projects as data samples, a convolutional neural network model is constructed for training and the corresponding aeration parameters are predicted. By comparing the real-time monitoring data with the prediction results, the aeration parameters are dynamically adjusted, which realizes the accurate prediction and dynamic optimization of the slurry transportation pressure parameters of the dredging and reclamation aeration booster pipeline, ensuring the efficient and stable operation of the transportation process.

[0112] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "possessing," or any other variations thereof are intended to cover non-exclusive possession, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0113] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that variations, modifications, substitutions and alterations may be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for determining the slurry transport pressure parameters of a dredging, reclamation, and aerated propulsion pipeline, characterized in that, The method is based on the influence of slurry concentration, gas ratio, and pipeline parameter characteristics on the delivery boosting effect. The method includes the following steps: Collect the characteristics of different types of pipeline parameters, as well as the parameters of the dredged slurry and aeration parameters, as data samples; the slurry parameters include the concentration of slurry and the proportion of gas, and the aeration parameters include the pumping pressure and the gas injection pressure; The collected data samples are cleaned and preprocessed, and it is determined whether the number of data samples needs to be supplemented based on the cleaning results. Then, the slurry parameter characteristics and pipeline parameter characteristics that have the greatest impact on the gasification parameter characteristics are extracted from the data samples using the correlation analysis method. A convolutional neural network model is constructed based on data samples. The slurry parameter features and pipeline parameter features are used as input features to train the convolutional neural network model and predict the gas filling parameters. The accuracy of the prediction results is verified based on the characteristics of the gas filling parameters, and the weights of the convolutional neural network model are adjusted according to the verification results. The new slurry concentration, gas concentration, and corresponding pipeline parameters are obtained and input into a trained convolutional neural network model for prediction to obtain the corresponding pumping pressure and gas injection pressure.

2. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 1, characterized in that, Based on the cleaning results, determine whether it is necessary to supplement the number of data samples, specifically: Based on the training requirements of the convolutional neural network model, the removal thresholds for each parameter in the data samples are preset. Extract the cleaned data sample, compare the amount of data before and after cleaning for each parameter in the data sample, and obtain the amount of data to be removed for each parameter. The data removal evaluation coefficient corresponding to each parameter is calculated using the removal amount corresponding to each parameter. The data removal evaluation coefficient corresponding to each parameter is compared with the preset removal threshold to obtain the comparison results. From the comparison results, obtain the number of samples whose evaluation coefficient exceeds the removal threshold after data removal; If the number of samples exceeding the removal threshold is one, then it is determined that no data addition is needed. When the number of samples exceeding the removal threshold is greater than one, it is determined that data addition is required.

3. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 2, characterized in that, The removal thresholds for various parameters in the pre-defined data samples based on the training requirements of the convolutional neural network model are compensated and adjusted to obtain the pre-defined removal thresholds, including: Extract data samples from each item; Retrieve historical training data of the convolutional neural network model; The evaluation coefficients of each data sample are removed from the historical training data of the convolutional neural network model. The standard deviation of the data elimination evaluation coefficient corresponding to each data sample is obtained based on the data elimination evaluation coefficient of each data sample. Retrieve the number of times data needs to be added to the data samples from the historical training data of the convolutional neural network model; For each data sample, retrieve the number of data samples that need to be added for each data sample increment; The maximum difference in the number of samples to be added is obtained based on the number of samples that need to be added each time the data sample is increased. Retrieve the two training moments corresponding to the maximum difference in the number of samples increased from the historical training data, and obtain the training time interval based on the two training moments; The data is removed by removing the standard deviation of the evaluation coefficient, the maximum difference in the number of samples added to the historical training data, and the removal thresholds for various parameters in the data samples corresponding to the current training of the convolutional neural network model based on the training time interval.

4. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 3, characterized in that, The data removal criteria include the standard deviation of the evaluation coefficient, the maximum difference in the number of samples added to the historical training data, and the removal thresholds for various parameters in the data samples corresponding to the current training of the convolutional neural network model based on the training time interval. Retrieve the number of samples corresponding to each two consecutive data additions and the time interval corresponding to each two consecutive training iterations from the historical training data where data samples need to be added. The ratio of the increase in the number of samples between each two adjacent data points to the time interval between each two adjacent training points is used to obtain the first sample time ratio. The average value of the first sample time ratio is obtained by adding the corresponding first sample time ratio to each two adjacent data points; Retrieve data from historical training data, including the maximum difference in the number of samples and their corresponding training time intervals; The second sample time ratio is obtained by increasing the number of samples in the historical training data and comparing it with the corresponding training time interval. The time decay coefficient is obtained by using the ratio of the first sample time to the ratio of the second sample time. The preset removal thresholds for each parameter in the data sample corresponding to the current training of the convolutional neural network model are obtained by combining the standard deviation of the data removal evaluation coefficient with the time decay coefficient and the preset removal thresholds for each parameter in the data sample based on the training requirements of the convolutional neural network model.

5. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 1, characterized in that, The accuracy of the prediction results is verified based on the characteristics of the gas filling parameters. The weights of the convolutional neural network model are then adjusted based on the verification results, including: Based on historical training data of the convolutional neural network model, an error threshold is preset; An attention mechanism is introduced into the convolutional neural network model to automatically learn feature weights; Different weights are assigned to the slurry parameter feature channels and the pipeline parameter feature channels based on channel attention, and the feature parameters corresponding to the important weights are monitored in real time based on spatial attention. When the validation results of the convolutional neural network model exceed the error threshold, the proportion of data corresponding to the feature parameters of important weights in the data samples is adjusted so that the convolutional neural network model focuses on training the feature parameters corresponding to important weights.

6. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 2, characterized in that, The determination requires data augmentation of the data sample, specifically: From the comparison results, retrieve the data sample types and data removal evaluation coefficients corresponding to the data sample increase required; Based on the evaluation coefficients removed from the data, determine the amount of data that needs to be supplemented to the data sample. According to the corresponding data volume, increase the number of data samples for the corresponding data sample type.

7. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 5, characterized in that, The accuracy of the prediction results is verified based on the characteristics of the gas filling parameters. The weights of the convolutional neural network model are adjusted according to the verification results. This also includes: The characteristics of each group of slurry parameters and pipeline parameters in the data sample, as well as the corresponding aeration parameters, are sorted and divided. The gradient analysis algorithm is used to calculate the gradient of each group of slurry parameter characteristics and pipeline parameter characteristics on the output aeration parameters. The importance of the influence of slurry parameter characteristics and pipeline parameter characteristics on aeration parameters is evaluated based on the gradient results. The evaluation results show that the larger the absolute value of the gradient, the greater the influence of this parameter feature on the gas filling parameters output by the convolutional neural network model. The weights of the convolutional neural network model are adjusted based on the evaluation results of feature importance.

8. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 1, characterized in that, Based on correlation analysis, the slurry parameter characteristics and pipeline parameter characteristics that have the greatest impact on the gasification parameter characteristics in the data sample are extracted, specifically: The Pearson correlation coefficient between slurry parameter characteristics, pipeline parameter characteristics and aeration parameter characteristics was calculated based on the Pearson correlation analysis algorithm. By setting a preset coefficient threshold, the closer the absolute value of the correlation coefficient of the parameter feature in the calculation results is to the coefficient threshold, the stronger the correlation of the parameter feature. The slurry parameter feature and pipeline parameter feature with the strongest correlation with the gas filling parameter feature are obtained.

9. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 8, characterized in that, The collected data samples were preprocessed, specifically as follows: The Z-score method is used to identify and remove outliers in the data samples, interpolation is used to fill in missing data, and duplicate data in the data samples are deleted. The slurry parameter characteristics, pipeline parameter characteristics, and aeration parameter characteristics are normalized so that all parameters are scaled to a specified range.

10. The method for determining the slurry transport pressure parameters of the dredging, reclamation, and aerated propulsion pipeline according to claim 1, characterized in that, The method further includes: By installing pressure sensors and flow meters on the pipeline, the changes in various parameters during the transportation process can be monitored in real time; Compare the changes in various parameters with the prediction results to determine whether the gas refueling parameters given in the prediction results are accurate. Based on the comparison results, determine whether dynamic adjustment of the pressurization parameters is necessary.

Images