Improved method and system for predicting wind pressure drop on turn platform of return ramp

Abstract

The application relates to the technical field of mine ventilation, in particular to a turning platform wind pressure drop prediction method and system for a turning platform of a turning ramp, which comprises the following steps: obtaining initial data sets of ventilation of a turning platform of a turning ramp to obtain standard data sets; determining ventilation local resistance parameters according to the standard data sets to establish a ventilation local resistance equation of the turning platform of the turning ramp; combining a neural network structure and the ventilation local resistance equation to establish a loss function, training and evaluating the neural network structure by using the loss function to obtain a candidate wind pressure drop prediction model; and determining a final wind pressure drop prediction model of the turning platform of the turning ramp according to an evaluation result of the candidate wind pressure drop prediction model to realize prediction and analysis of the wind pressure drop of the turning platform of the turning ramp. The application introduces physical constraints into a BP algorithm to construct a prediction model, and the model strategy can adapt to different mine turning ramps to realize wind pressure drop prediction of a turning platform of a complex turning ramp.

Description

Technical Field

[0001] This invention relates to the field of mine ventilation technology, specifically to an improved BP algorithm-based method and system for predicting wind pressure drop on turning platforms of ramps. Background Technology

[0002] In mine ventilation systems, the turnaround ramp is the core passageway. Its local resistance characteristics have a significant impact on the overall energy consumption of the mine ventilation system, making it an important link in ventilation system optimization and energy conservation.

[0003] Currently, the calculation of wind pressure drop in turning platforms of ramp-type turn-off tunnels mainly relies on empirical formulas such as the local resistance coefficient method. This involves consulting relevant manuals to obtain the local resistance coefficient, and then combining it with parameters such as wind speed and air density for calculation. However, this traditional method has many limitations in practical applications: there are complex nonlinear relationships between geometric features such as turning angle, radius of curvature to roadway width ratio, and cross-sectional dimensions, making it difficult to accurately reflect the actual wind pressure drop based solely on empirical formulas. Besides geometric parameters, environmental and structural factors such as support protrusion, relative height difference coefficient, air density, and airflow disturbance all affect the wind pressure drop, further increasing the difficulty and complexity of the calculation. Empirical formulas are derived from specific working conditions and have poor adaptability to complex and variable actual working conditions. It is difficult to find a perfectly matching empirical formula for accurate calculation in practical applications. Furthermore, due to limitations imposed by site conditions, the measurement results have low timeliness, failing to meet the needs of dynamic adjustment and intelligent control of mine ventilation systems in a timely manner, and thus not well adapting to the requirements of intelligent ventilation development.

[0004] Therefore, it is necessary to propose a multi-dimensional prediction method based on the BP algorithm to promote the intelligent development of mine ventilation and further realize the accurate prediction of local resistance of mine ventilation. Summary of the Invention

[0005] To address the shortcomings of existing methods and the limitations of practical applications, and in order to achieve efficient and accurate prediction of wind pressure for turnaround ramp platforms, we effectively integrate multi-dimensional features to continuously optimize the intelligent model and enable its practical application.

[0006] In a first aspect, this invention provides an improved BP algorithm for predicting wind pressure drop on a turnaround ramp platform. The method includes the following steps: obtaining an initial data set of ventilation data for the turnaround ramp platform; processing the initial data set using a data standard processing model to obtain a standardized data set of ventilation data for the turnaround ramp platform; determining local resistance parameters of ventilation based on the standardized data set; establishing a local resistance equation for ventilation of the turnaround ramp platform based on the local resistance parameters; constructing a neural network structure; establishing a loss function by combining the neural network structure and the local resistance equation; training and evaluating the neural network structure using the loss function based on the standardized dataset to obtain candidate wind pressure drop prediction models; determining the final wind pressure drop prediction model for the turnaround ramp platform based on the evaluation results of the candidate wind pressure drop prediction models; and performing predictive analysis of wind pressure drop on the turnaround ramp platform based on the final wind pressure drop prediction model.

[0007] This invention includes data preprocessing, model building, training, and other stages. It can fully consider the complex characteristics of the ventilation system of the turnaround ramp platform. Through the combination of physical constraints and data-driven approaches, as well as a multi-model selection strategy, the optimal wind pressure drop prediction model is finally obtained. This model can more accurately capture the influence of various factors in the ventilation system on wind pressure drop, thereby significantly improving the accuracy of the prediction results.

[0008] Optionally, obtaining the initial data set for ventilation of the turnaround ramp platform includes: acquiring a three-dimensional model of the tunnel, and obtaining three-dimensional spatial data of the turnaround ramp platform based on the three-dimensional model; collecting ventilation information of the turnaround ramp platform, and obtaining ventilation parameter data of the turnaround ramp platform over a period of time based on the ventilation information; and integrating the three-dimensional spatial data and the ventilation parameter data to obtain the initial data set for ventilation of the turnaround ramp platform. This invention's data set integrates multi-dimensional data, which can reveal the operating rules of the ventilation system from different perspectives, providing more comprehensive information for subsequent modeling and prediction.

[0009] Optionally, the step of setting a data standard processing model to process the initial dataset to obtain a standardized dataset for ventilation of the turnaround ramp platform includes: analyzing the mean and standard deviation of the initial dataset based on the initial dataset; establishing a data standardization processing model based on the mean and standard deviation; and processing the initial dataset using the data standardization processing model to obtain a standardized dataset for ventilation of the turnaround ramp platform.

[0010] The data standardization processing model satisfies the following relationship;

[0011] ,

[0012] in, For standardized data, X represents the original data in the initial dataset. The mean, The standard deviation is denoted as .

[0013] This invention standardizes the data, converting it into dimensionless standardized data to eliminate dimensional differences, thus more accurately reflecting the impact of different characteristics on the ventilation system.

[0014] Optionally, determining the ventilation local resistance parameter based on the standardized dataset includes: constructing an expression for the ventilation local resistance parameter; and determining the ventilation local resistance parameter using the expression for the ventilation local resistance parameter and the standardized dataset.

[0015] The expression for the local resistance parameter of ventilation satisfies the following relationship;

[0016] ,

[0017] in, To support the protrusion, This is the distance from the outermost edge of the support structure to the center of the tunnel cross-section. This is the distance from the original excavated wall of the tunnel to the center of the circle;

[0018] ,

[0019] in, This is the relative elevation difference coefficient. For the longitudinal elevation difference of the turning platform, The width of the alleyway.

[0020] The ventilation local resistance parameter expression of this invention comprehensively considers the influence of factors such as the protrusion of the support structure and the relative height difference of the turning platform on the ventilation local resistance, and can more comprehensively consider the combined influence of multiple factors, thereby improving the accuracy and reliability of the ventilation local resistance calculation formula. Optionally, the step of establishing the ventilation local resistance equation for the turning platform of the ramp based on the ventilation local resistance parameters includes: obtaining the support protrusion and relative height difference coefficients using the ventilation local resistance parameter expression; and using the original historical data in the initial data set, fitting and determining the foundation local resistance coefficient through multiple linear regression analysis. and each influence coefficient The specific values; based on the support protrusion, the relative height difference coefficient, and the fitted coefficients, the ventilation local resistance equation of the turning platform of the ramp is established;

[0021] The local resistance equation for ventilation satisfies the following relationship;

[0022] ,

[0023] in, This is a reference value for local wind pressure drop calculated based on physical equations. Based on the local drag coefficient, for Influence coefficient, To support the protrusion, for Influence coefficient, This is the relative elevation difference coefficient. for and The synergistic influence coefficient For airflow density, This refers to the airflow velocity.

[0024] The ventilation local resistance equation of this invention provides a quantitative basis for the optimized design of the ventilation system of the turning platform of the ramp. By analyzing and adjusting the parameters in the equation, the changes in ventilation local resistance under different design schemes can be predicted.

[0025] Optionally, the step of building a neural network structure and establishing a loss function based on the neural network structure and the ventilation local resistance equation includes: determining the number of layers and nodes of the neural network, introducing a nonlinear activation function of the BP algorithm, building the neural network topology, and initializing the weights and bias parameters of the neural network; introducing physical constraint equations, analyzing the residuals of the airflow continuity equation and the wind pressure drop residuals based on the physical constraint equations; and establishing a loss function based on the neural network structure, the residuals of the airflow continuity equation, the wind pressure drop residuals, and the ventilation local resistance equation.

[0026] The loss function satisfies the following relationship:

[0027] ,

[0028] in, For loss function, For neural network predictions and actual observed values The mean square error between them For predicted values, For the true value, To balance the weights, To reduce the residual weight for wind pressure, The physical consistency residual of wind pressure drop (defined as the neural network prediction value) Calculated values ​​from physical equations (the deviation between them) For the continuous residual weight of airflow, It represents the continuous residual of the airflow.

[0029] This invention introduces a nonlinear activation function of the BP algorithm, enabling the neural network to learn complex nonlinear relationships in ventilation local resistance data, better capture complex nonlinear features, and thus improve the prediction accuracy of ventilation local resistance.

[0030] Optionally, the step of training and evaluating the neural network structure using the loss function based on the standardized data set to obtain a candidate wind pressure drop prediction model includes: establishing a training model with the objective of minimizing the loss function; employing an iterative optimization algorithm to iteratively solve the training model based on the standardized data set, updating the weights and bias parameters of the neural network until convergence conditions are met; setting a validation evaluation function to evaluate the model's performance during the iteration process; analyzing the prediction performance of the wind pressure drop prediction model based on the validation evaluation function; and determining the corresponding candidate wind pressure drop prediction model by combining the convergence of the loss function optimization and the prediction performance. The loss function of this invention measures the difference between the model's predicted value and the actual value. By minimizing the loss function, the model's prediction can be made closer to the actual situation, thereby improving the accuracy and reliability of model training.

[0031] Optionally, determining the final wind pressure drop prediction model for the turnaround ramp platform based on the evaluation results of the candidate wind pressure drop prediction models includes: extracting different combinations of input features from the standardized dataset; selecting a wind pressure drop prediction model based on the combinations of input features; and the wind pressure drop prediction model satisfying the following relationship.

[0032] ,

[0033] in, For wind pressure drop prediction models, For input data, These are geometric feature vectors extracted based on the 3D model of the tunnel. The width of the alleyway, For the longitudinal elevation difference of the turning platform, To support the protrusion, For airflow velocity, For airflow density, specifically, the three-dimensional geometric features of the turning platform. This refers to vector data obtained after feature extraction from a 3D model, which includes at least one or more combinations of turning radius, turning angle, tunnel centerline length, and cross-sectional contraction rate. These geometric features can quantitatively characterize the complex influence of the spatial morphology of the turning platform on the airflow field. By extracting different combinations of input features and developing a model, this invention allows for optimization and adjustment of model parameters based on the impact of different features on model performance, making the model's prediction results closer to actual observations.

[0034] Optionally, the step of selecting a wind pressure drop prediction model based on input feature combinations and determining the final wind pressure drop prediction model for the turnaround ramp platform includes: training multiple candidate wind pressure drop prediction models based on different input feature combinations to construct a candidate model set; constructing a hierarchical decision tree model, using the prediction performance index output by the verification evaluation function and the physical constraint residual value as the discrimination node of the hierarchical decision tree model; and using the hierarchical decision tree model to screen and analyze the candidate model set, determining the output optimal performance model as the final wind pressure drop prediction model for the turnaround ramp platform. The hierarchical decision tree model of this invention provides a systematic and structured framework for wind pressure drop prediction models, enabling a clearer and more organized analysis of the impact of different factors on model selection, and improving the scientific rigor and rationality of the wind pressure drop prediction model for the turnaround ramp platform.

[0035] Secondly, this invention also provides an improved BP algorithm-based wind pressure drop prediction system for turnaround ramp platforms. This system can efficiently execute the improved BP algorithm-based wind pressure drop prediction method for turnaround ramp platforms provided by this invention. The system includes an input device, a processor, an output device, and a memory, wherein the input device, processor, output device, and memory are interconnected. The memory includes a computer-readable storage medium as described in the first aspect of this invention. The memory stores a computer program, which includes program instructions. The processor is configured to call the program instructions. The improved BP algorithm-based wind pressure drop prediction system for turnaround ramp platforms provided by this invention has a compact structure, strong applicability, and greatly improves operating efficiency. Attached Figure Description

[0036] Figure 1 The flowchart of the improved BP algorithm for predicting wind pressure drop on a turning platform of a ramp is shown below.

[0037] Figure 2 This is a schematic diagram illustrating the analysis and comparison of the wind pressure drop prediction model for the turnaround ramp platform of the present invention.

[0038] Figure 3 This is a schematic diagram of the wind pressure drop prediction system for a turning ramp platform using the improved BP algorithm of the present invention. Detailed Implementation

[0039] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0040] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0041] Please see Figure 1 To achieve efficient and accurate prediction of wind pressure on turning platforms of ramp-type turns, and to promote continuous model optimization and practical application, this invention provides an improved BP algorithm for predicting wind pressure drop on turning platforms of ramp-type turns. The method includes the following steps:

[0042] S1. Obtain the initial data set for ventilation of the turnaround ramp platform, set up a data standard processing model to process the initial data set, and obtain a standardized data set for ventilation of the turnaround ramp platform. The implementation steps and specific contents are as follows:

[0043] First, obtain the initial data set for ventilation of the turning platform of the ramp.

[0044] A three-dimensional model of the tunnel is acquired, and based on this model, three-dimensional spatial data of the turning platform of the ramp is obtained. In this embodiment, radar is used to perform a comprehensive scan of the tunnel, and the three-dimensional model of the tunnel is reconstructed based on the scan data. This method allows for accurate acquisition of the tunnel's spatial parameters. Subsequently, the spatial data is processed and stored in the system storage, ensuring both the authenticity and accuracy of the data source and providing data support for subsequent analysis.

[0045] The ventilation information of the turnaround ramp platform is collected, and the ventilation parameter data of the turnaround ramp platform over a period of time is obtained based on the ventilation information.

[0046] After acquiring the three-dimensional spatial data, ventilation information of the turning platform of the ramp is collected to obtain ventilation parameter data over a period of time. In this embodiment, a barometer is used to measure the wind pressure drop of the platform, and key parameters such as air volume and wind speed are measured simultaneously. After the measurement is completed, the ventilation parameter data is stored in the system memory for subsequent retrieval and analysis.

[0047] By integrating the aforementioned three-dimensional spatial data and ventilation parameter data, an initial data set for the ventilation of the turnaround ramp platform is obtained. After acquiring the three-dimensional spatial data and ventilation parameter data separately, the two types of data are integrated to finally obtain the initial data set for the ventilation of the turnaround ramp platform, laying the foundation for subsequent analysis and prediction of the platform's ventilation conditions.

[0048] Then, a data standard processing model is set up to process the initial dataset to obtain a standardized dataset of ventilation for the turning platform of the ramp.

[0049] The first step is to analyze the mean and standard deviation of the initial dataset. Statistical analysis is performed on the initial dataset to calculate the mean and standard deviation. In this embodiment, the mean and standard deviation are calculated separately from data obtained from multiple measurements using statistical methods, providing a basis for establishing the standardization processing model.

[0050] The second step is to establish a data standardization model based on the mean and standard deviation calculated in the first step. The data standardization model satisfies the following relationship:

[0051] ,

[0052] in, For standardized data, X represents the original data in the initial dataset. The mean, The standard deviation is denoted as .

[0053] The above-mentioned standardized processing model can transform the initial data into a data format with unified standards.

[0054] The third step involves processing the initial dataset using a data standardization model to obtain a standardized dataset for ventilation of the turnaround ramp platform. In this embodiment, the data standardization model established in the second step is used to process all data in the initial dataset. In actual implementation, a processor can be used to standardize the three-dimensional spatial data of the tunnel and the ventilation parameter data. After processing, the standardized data is stored in the system storage, thus obtaining a standardized dataset for ventilation of the turnaround ramp platform, providing a standardized and unified data foundation for subsequent analysis and research.

[0055] S2. Determine the ventilation local resistance parameters based on the standardized dataset, and establish the ventilation local resistance equation for the turning platform of the ramp. The specific steps and implementation details are as follows:

[0056] First, the local resistance parameters of ventilation are determined based on a standardized dataset.

[0057] In this embodiment, expressions for local ventilation resistance parameters are constructed, which mainly include expressions for roadway support protrusion and relative elevation difference coefficients.

[0058] The expression for the protrusion of roadway support satisfies the following relationship;

[0059] ,

[0060] in, To support the protrusion, This is the distance from the outermost edge of the support structure to the center of the tunnel cross-section. This is the distance from the original excavated wall of the tunnel to the center of the circle.

[0061] The expression for roadway support protrusion is mainly used to quantify the influence of roadway support structure on local ventilation resistance. Among them, support protrusion is an important indicator for measuring the degree of protrusion of support structure. The distance from the outermost edge of the support structure to the center of the roadway cross-section reflects the spatial position of the support structure. The distance from the original excavated wall of the roadway to the center reflects the initial spatial morphology of the roadway.

[0062] The expression for the relative height difference coefficient satisfies the following relationship;

[0063] ,

[0064] in, This is the relative elevation difference coefficient. For the longitudinal elevation difference of the turning platform, The width of the alleyway.

[0065] The relative elevation difference coefficient expression can describe the relative relationship between the longitudinal elevation difference of the turning platform and the roadway width. The relative elevation difference coefficient reflects the relative magnitude of the longitudinal elevation difference of the turning platform on the roadway width scale; the longitudinal elevation difference of the turning platform reflects the height change of the turning platform in the vertical direction; the roadway width is an important parameter of the roadway space size.

[0066] Then, based on the ventilation local resistance parameters, the ventilation local resistance equation of the turning platform of the ramp was established.

[0067] The support protrusion and relative height difference coefficient can be obtained using the expression for the ventilation local resistance parameter. After constructing the expression for the ventilation local resistance parameter, relevant data from the standardized dataset are substituted into the above two expressions for calculation, which can accurately determine the support protrusion and relative height difference coefficient. Based on this, unstandardized original historical data from the initial dataset (500 sets of historical working condition data are selected in this embodiment), including measured physical quantities such as wind pressure drop, flow velocity, and density, are used to perform multiple linear regression fitting to determine the key coefficients in the physical equation. Since original physical quantities are used for regression, the obtained coefficients have clear physical meaning: the foundation local resistance coefficient is determined through fitting calculation. The influence coefficient of support protrusion is 0.12. The relative elevation difference influence coefficient is 1.05. The synergistic influence coefficient is 0.88. The value is 0.42. Substituting these specific coefficient values ​​into the equation yields a reference value for wind pressure drop with physical constraints. The aforementioned local resistance parameters for ventilation provide crucial data support for the subsequent analysis and optimization of the ventilation system of the turnaround ramp platform.

[0068] The ventilation local resistance equation for the turning platform of the ramp-type turnout, established based on the support protrusion and relative elevation difference coefficient, satisfies the following relationship;

[0069] ,

[0070] in, This is a reference value for local wind pressure drop calculated based on physical equations. Based on the local drag coefficient, for Influence coefficient, To support the protrusion, for Influence coefficient, This is the relative elevation difference coefficient. for and The synergistic influence coefficient For airflow density, This refers to the airflow velocity.

[0071] The above equations not only consider the influence of support protrusion and relative height difference coefficient on local ventilation resistance, but also consider the interaction between the two through a synergistic influence coefficient. This multi-factor synergistic consideration method is more in line with the complex situation of actual roadway ventilation, helps to obtain a more realistic wind pressure drop change law, and improves the accuracy of prediction results.

[0072] The ventilation local resistance parameters, from the perspectives of the prominence of the roadway support structure and the relative relationship between the longitudinal elevation difference of the turning platform and the roadway width, can comprehensively reflect the physical characteristics of the turning platform of the ramp. When improving the BP algorithm to predict wind pressure drop, using relevant parameters as input features allows the ventilation local resistance equation to more accurately capture the influence of the roadway structure on ventilation local resistance and wind pressure drop. S3. A neural network structure is built, and a loss function is established by combining the neural network structure and the ventilation local resistance equation. Based on a standardized dataset, the loss function is used to train and evaluate the neural network structure to obtain a candidate wind pressure drop prediction model. The specific steps and implementation content are as follows:

[0073] First, a neural network structure is built, and a loss function is established by combining the neural network structure with the local resistance equation of ventilation.

[0074] The number of layers and nodes in the neural network is determined, and the backpropagation (BP) algorithm's nonlinear activation function is introduced to construct the neural network topology. The weights and bias parameters of the neural network are then initialized (in this embodiment, the weights are initialized using the Xavier initialization method, and the bias parameters are initialized to 0). In this embodiment, the processor runs the BP algorithm, processing the relevant data according to its nonlinear activation function to construct the key elements of the neural network structure. After acquiring the parameters, they are stored in the system memory. This parameter data provides the data foundation for subsequent model training, further ensuring the consistency and accuracy of the entire construction process of the wind pressure drop prediction model for the turning ramp platform.

[0075] Based on the neural network parameters obtained above, a neural network with a specific structure is constructed. This neural network consists of an input layer, four hidden layers, and an output layer. The computational relationships between the layers are as follows:

[0076] Input layer, input data is represented as It is the initial input of the entire neural network, providing the basic data for the calculation of subsequent layers.

[0077] Output of hidden layer 1 It is calculated using the following formula:

[0078] ,

[0079] in, This is the output of hidden layer 1, where ReLU is a non-linear activation function. for The weight, For input data, for The bias parameters.

[0080] Nonlinear activation functions can introduce nonlinear characteristics into neural networks and enhance the network's expressive power; the weight matrix of hidden layer 1 determines the degree of influence of input data on the output of hidden layer 1; the bias parameters of hidden layer 1 are mainly used to adjust the baseline value of the output.

[0081] Output of hidden layer 2 The calculation formula is: ,

[0082] in, This is the output of hidden layer 2, where ReLU is a non-linear activation function. for The weight, Output for hidden layer 1 for The bias parameters.

[0083] The weight matrix of hidden layer 2 reflects the contribution of the output of hidden layer 1 to the output of hidden layer 2.

[0084] Output of hidden layer 3 Calculated by the following formula:

[0085] ,

[0086] in, This is the output of hidden layer 3, where ReLU is a non-linear activation function. for The weight, Output for hidden layer 2. for The bias parameters.

[0087] Output of hidden layer 4 The calculation formula is:

[0088] ,

[0089] in, This is the output of hidden layer 4, where ReLU is a non-linear activation function. for The weight, Output for hidden layer 3. for The bias parameters.

[0090] Predicted values ​​of the output layer Calculated using the following formula:

[0091] ,

[0092] in, The output layer prediction value, For the weights of the output layer, Output for hidden layer 4. These are the bias parameters for the output layer.

[0093] The weight matrix of the output layer determines the output of hidden layer 4. The impact on the final predicted value; the bias parameters of the output layer are used to adjust the baseline value of the output result.

[0094] The above steps completed the construction of a neural network topology based on the BP algorithm nonlinear activation function, laying the foundation for subsequent tasks such as wind pressure drop prediction.

[0095] Physical constraint equations are introduced, and the residuals of the airflow continuity equation and the wind pressure drop residuals are analyzed based on these equations.

[0096] Physical constraint equations can effectively analyze the residuals of the airflow continuity equation and the wind pressure drop equation. The residuals reflect the degree of deviation between the model's prediction results and the actual physical laws. By calculating and analyzing the residuals, we can evaluate the model's simulation effect on the physical process and provide a basis for the subsequent construction of the loss function.

[0097] The formula for calculating the residual of the airflow continuity equation is as follows:

[0098] ,

[0099] in, The residuals of the airflow continuity equation are... For ventilation air volume, The cross-sectional area of ​​the tunnel, This refers to the airflow velocity.

[0100] The airflow continuity equation residual measures the difference between the ventilation volume and the product of the tunnel cross-sectional area and the airflow velocity; the ventilation volume reflects the volume of gas passing through the tunnel per unit time; the tunnel cross-sectional area determines the size of the space through which the gas passes; and the airflow velocity reflects the speed at which the gas flows in the tunnel. In this embodiment, the processor calculates the airflow continuity equation residual based on the above calculation formulas and data, and stores it in the system memory.

[0101] By combining the above neural network structure, the residuals of the airflow continuity equation, the residuals of wind pressure drop, and the local resistance equation of ventilation, a loss function is constructed that comprehensively considers the model prediction error and the degree of satisfaction of physical constraints.

[0102] The loss function described above closely integrates the prediction results of the neural network with actual physical laws, which can guide the model to better fit the real data during the training process while following physical laws.

[0103] Therefore, the above loss function satisfies the following relationship:

[0104] ,

[0105] in, For loss function, For neural network predictions and actual observed values The mean square error between them For predicted values, For the true value, To balance the weights, To reduce the residual weight for wind pressure, The physical consistency residual of wind pressure drop (defined as the neural network prediction value) Calculated values ​​from physical equations (the deviation between them) For the continuous residual weight of airflow, This represents the continuous residual of the airflow. In this embodiment, to balance the model's prediction accuracy with the effectiveness of physical constraints, a balancing weight was set based on experimental verification. Wind pressure drop residual weight Continuous airflow residual weights .

[0106] The loss function is a metric for measuring the overall performance of the model; mean squared error reflects the accuracy of the model's predictions; balancing weights are mainly used to coordinate the relative importance between the model's prediction errors and the physical constraint residuals; weights This reflects the contribution of wind pressure drop residuals to the loss function; weights This demonstrates the contribution of the residuals in the airflow continuity equation to the loss function.

[0107] In this embodiment, the processor, combined with the aforementioned loss function and various parameter values, can efficiently complete numerical calculations, further ensuring the effective operation of the loss function and providing technical support for subsequent model training and optimization.

[0108] Then, based on a standardized dataset, the neural network structure is trained and evaluated using a loss function to obtain candidate wind pressure drop prediction models. A training model is established with the goal of minimizing the loss function, and a processor is introduced to obtain the result of minimizing the loss function through the processor and the training model. To find an optimal set of neural network parameters that minimizes the loss function, a training model is established in this embodiment. This model reduces the error between the model's predicted values ​​and the true values ​​by adjusting the weights and bias parameters of each layer of the neural network, while better satisfying the physical constraints. Its mathematical expression satisfies the following relationship:

[0109] ,

[0110] in, This represents a model that minimizes the loss function. Indicates minimization. For the weights of each layer, For the bias parameters of each layer, For loss function, For input data, This represents the actual wind pressure drop.

[0111] Minimizing the loss function represents the goal of the entire optimization process; the minimization operation is to find the condition that minimizes the loss function; the weight parameters of each layer of the neural network determine how the input data is passed and transformed between layers; the bias parameters of each layer are used to adjust the baseline value of the output of each layer; the loss function integrates the error between the predicted value and the true value, as well as the residuals of physical equations such as ventilation local resistance and airflow continuity, which helps to comprehensively measure the model performance; the input data mainly includes standardized feature vectors such as roadway geometric parameters and ventilation parameters, and the actual wind pressure drop is an important reference standard for measuring the accuracy of model prediction.

[0112] After completing the model establishment and training process for minimizing the loss function, this embodiment further sets up a verification evaluation function to more comprehensively and accurately evaluate the predictive performance of the trained model and ensure the reliability and effectiveness of the model in practical applications.

[0113] To quantify the prediction performance of the loss model on the validation set, the mean squared error is introduced as a validation evaluation function in this embodiment, and its mathematical expression is as follows:

[0114] ,

[0115] in, Mean square error, The total number of samples in the validation set, Let i be the predicted wind pressure drop value for the i-th sample. Let be the actual wind pressure drop value of the i-th sample.

[0116] Mean squared error (MSE) is a commonly used metric to measure the difference between predicted and true values; a smaller value indicates better model prediction performance. The total number of samples in the validation set is the number of data samples used to evaluate model performance. By calculating the average of the squared differences between the predicted and true values ​​of all samples, the model's prediction accuracy on the entire validation set can be comprehensively reflected.

[0117] The predictive performance of the model is analyzed based on the processor and validation evaluation function. The convergence of the loss function optimization and the predictive performance are then combined to obtain corresponding candidate wind pressure drop prediction models. In this embodiment, the processor, combined with the aforementioned validation evaluation function, is used to analyze the predictive performance of the current trained model. Specifically, the prediction error of each sample is calculated sequentially according to the mean squared error (MSE) formula, and the MSE value of the entire validation set is finally obtained. The MSE value provides a direct understanding of the model's prediction bias on the validation set, allowing for the determination of whether the model is overfitting or underfitting, and providing an important basis for further model optimization.

[0118] When a trained model satisfies the convergence condition of minimizing the loss function during training and achieves the preset predictive performance through validation evaluation function analysis, it is identified as a corresponding candidate wind pressure drop prediction model. The minimization result reflects the model's optimization level on the training data, while the predictive performance reflects the model's generalization ability on the validation dataset. A model exhibiting a small loss value on the training data and a low mean squared error on the validation set indicates good performance. Therefore, by comprehensively considering these two factors, a high-performance candidate wind pressure drop prediction model can be obtained.

[0119] S4. Based on the evaluation results of the candidate wind pressure drop prediction models, determine the final wind pressure drop prediction model for the turnaround ramp platform. Based on the final wind pressure drop prediction model, perform the prediction and analysis of the wind pressure drop for the turnaround ramp platform. The specific steps and related content are as follows:

[0120] First, different combinations of input features are extracted from the standardized dataset.

[0121] To ensure that the wind pressure drop prediction model can accurately and effectively predict wind pressure drop, this embodiment trains and selects candidate models based on combinations of input features. Different combinations of input features reflect different operating conditions and influencing factors of the ramp ventilation system. By comprehensively considering relevant factors, a wind pressure drop prediction model that is more in line with the actual situation can be constructed.

[0122] To address the different combinations of input features in a ramp ventilation system, this embodiment uses a processor to construct multiple candidate wind pressure drop prediction models. The processor can quickly process large amounts of data and complex computational tasks. By analyzing and simulating different combinations of input features, multiple candidate wind pressure drop prediction models are trained and a candidate model set is constructed, which can improve the accuracy and efficiency of prediction.

[0123] The above wind pressure drop prediction model satisfies the following relationship;

[0124] ,

[0125] in, For wind pressure drop prediction models, For input data, These are geometric feature vectors extracted based on the 3D model of the tunnel. The width of the alleyway, For the longitudinal elevation difference of the turning platform, To support the protrusion, For airflow velocity, The airflow density. Specifically, the three-dimensional geometric features of the turning platform. This is vector data obtained after feature extraction from a 3D model, which includes at least one or more combinations of turning radius, turning angle, tunnel centerline length, and cross-sectional contraction rate. These geometric features can quantitatively characterize the complex influence of the spatial morphology of the turning platform on the airflow field. The output of the wind pressure drop prediction model, i.e., the predicted wind pressure drop value calculated by the model, can be used to evaluate the performance of the ramp ventilation system and optimize ventilation design.

[0126] The wind pressure drop prediction model is a mathematical model that includes specific algorithms and calculation rules. It can perform calculations based on input data and parameters to obtain corresponding prediction results.

[0127] The input data contains various information related to the ramp ventilation system.

[0128] The geometric feature vectors extracted from the 3D model of the tunnel include, but are not limited to, turning radius, turning angle, tunnel centerline length, and cross-sectional contraction rate. These geometric features collectively affect the local ventilation resistance by altering the degree of airflow field disturbance, the size of the vortex region, and energy loss. The tunnel width directly affects the airflow space and local ventilation resistance, and is one of the important factors affecting wind pressure drop.

[0129] The longitudinal elevation difference of the turning platform reflects the terrain change at the turning point of the ramp, which has a certain impact on airflow and wind pressure drop.

[0130] The faster the airflow velocity, the greater the local resistance to ventilation, and the greater the wind pressure drop will be.

[0131] Changes in airflow density affect the inertia and drag of the airflow, which in turn affects the wind pressure drop.

[0132] The above steps enable the scientific training of multiple candidate wind pressure drop prediction models suitable for inclined ramp ventilation systems, and clarify the input parameters and output results of the models, laying the foundation for subsequent model work.

[0133] Then, a hierarchical decision tree model was constructed to screen and analyze the candidate model set, and the final wind pressure drop prediction model for the turnaround ramp platform was determined.

[0134] To more accurately select a wind pressure drop prediction model suitable for a turnaround ramp platform, the embodiment uses a processor to construct a hierarchical decision tree model as an auxiliary selection tool.

[0135] The details of the hierarchical decision tree model are as follows:

[0136] Determine input features:

[0137] Input features are represented as ,in For the input feature set, These are different physical quantities in inclined roadway ventilation. These physical quantities include, but are not limited to, roadway width, longitudinal elevation difference, support protrusion, airflow velocity, and other characteristic quantities of inclined roadway ventilation obtained through actual measurement, which are important factors affecting air pressure drop.

[0138] Constructing the first-level decision tree:

[0139] A first-level decision tree is constructed to preliminarily classify the input features of the ramp ventilation system, thereby determining the general direction for selecting the wind pressure drop prediction model. Its mathematical expression is as follows:

[0140] ,

[0141] in, For the first level of output categories, This is the first-level decision tree. For input data.

[0142] Constructing subsequent hierarchical decision trees:

[0143] Based on the first-level decision tree, a second-level decision tree is further constructed, expressed as:

[0144] ,

[0145] in, For the second layer of output categories, This is the second-level decision tree. For input data, This is the first-level output category.

[0146] Following this pattern, the decision tree for each subsequent level is constructed, with the general formula as follows:

[0147] ,

[0148] in, For the first Layer output category, For the first Hierarchical decision tree For input data, For the first Layer output category.

[0149] The prediction performance index output by the validation evaluation function and the physical constraint residual value are used as the discriminant nodes of the hierarchical decision tree model. Specifically, when constructing the hierarchical decision tree model, the first layer of the decision tree uses the mean squared error (MSE) as the discriminant node to select models with an MSE less than a preset threshold (e.g., 2.0 Pa) to ensure prediction accuracy; the second layer of the decision tree uses the physical constraint residual ( The model with the best physical consistency among the remaining models is selected as the discrimination node to ensure compliance with the ventilation mechanism. Through the step-by-step construction of a hierarchical decision tree, the performance of different candidate models can be analyzed and classified in greater detail, thus providing a more accurate reference for the final wind pressure drop prediction model of the turnaround ramp platform. Subsequently, the hierarchical decision tree model is used to screen and analyze the candidate model set, and the model with the best output performance is determined as the final wind pressure drop prediction model for the turnaround ramp platform.

[0150] In this embodiment, a hierarchical decision tree model is combined with a set of candidate models to perform a comprehensive analysis of each candidate wind pressure drop prediction model. A processor is used to connect multiple candidate wind pressure drop prediction models with multi-layer decision tree discrimination logic to achieve rapid prediction of the optimal wind pressure drop prediction model for the turnaround ramp platform.

[0151] The final mathematical expression for the wind pressure drop prediction model of the turnaround ramp platform is as follows:

[0152] ,

[0153] in, The output is the predicted wind pressure drop. For wind pressure drop prediction models, For input data, For decision trees.

[0154] Finally, the wind pressure drop prediction model for the turnaround ramp platform is used to predict and analyze the wind pressure drop of the turnaround ramp platform.

[0155] This application belongs to the field of mine ventilation technology, and proposes an improved BP algorithm model for predicting the air pressure drop of a turning platform on a ramp. Based on the above implementation, the model structure of this application is as follows:

[0156] The neural network structure consists of an input layer, four hidden layers, and an output layer. The input layer receives the roadway geometric parameters, ventilation parameters, and geometric feature vectors; the hidden layers contain neurons connected by nonlinear activation functions; and the output layer outputs the predicted air pressure drop value. The weights and bias parameters are initialized using appropriate methods and updated through an iterative optimization algorithm during training.

[0157] Physical constraints are introduced: a ventilation local resistance equation is established based on ventilation local resistance parameters, taking into account the influence of factors such as the protrusion of the support structure and the relative height difference of the turning platform; physical constraint residuals are introduced into the loss function, mainly including the physical consistency residual of wind pressure drop and the airflow continuity residual.

[0158] Loss function construction: By combining the mean square error of the local wind pressure drop reference value calculated by the process with the actual observed value and the physical constraint residual, the relative importance of the model prediction error and the physical constraint residual is coordinated by balancing the weights and residual weights.

[0159] Model training and evaluation: Train the neural network using a standardized dataset with the goal of minimizing the loss function; set a validation evaluation function to evaluate the model performance; combine the convergence and prediction performance of the loss function to determine candidate models.

[0160] Model optimization strategy: Construct a hierarchical decision tree model to verify the prediction performance index output by the evaluation function and the physical constraint residual value as the discrimination node, and screen and analyze the candidate model set to determine the model with the best performance.

[0161] Model Output and Application: The model outputs predicted wind pressure drop values ​​to assess ventilation system performance and optimize ventilation design; it is applicable to diverse working conditions of different mine turnaround ramps, providing support for ventilation system optimization and energy conservation.

[0162] To visually demonstrate the prediction effect of the final wind pressure drop prediction model for the turnaround ramp platform, this embodiment compares the measured wind pressure drop value with the AI ​​model's predicted value and generates a visualization image. Please refer to [link / reference] for details. Figure 2 At the same time, based on Figure 2The measured wind pressure drop values ​​and AI model prediction values ​​for different sample numbers can be obtained. Please refer to Table 1 for specific data information. It should be noted that Table 1 only shows 10 typical samples randomly selected from the verification dataset containing 500 historical operating conditions constructed in this embodiment, in order to verify the accuracy of the model prediction.

[0163] Table 1. Comparison of Measured and Predicted Values ​​of Wind Pressure Drop at Turning Platform of Turnaround Ramp

[0164]

[0165] The above Figure 2 It includes a wind pressure drop distribution map and a distribution map predicted by the model obtained in this invention, combined with... Figure 2 As can be clearly observed in Table 1, the white bars for samples 1 to 10 represent the measured wind pressure drop values, while the diagonal bars represent the AI ​​model predictions. The measured wind pressure drop values ​​and the AI ​​model predictions show a good matching relationship, and the degree of agreement between the bar heights directly reflects the closeness between the model predictions and the measured results. According to... Figure 2 As shown in Table 1, the distribution of measured wind pressure drop values ​​is in good agreement with the distribution of predicted values ​​obtained by the method of this invention. Furthermore, considering the model performance indicators, the mean absolute error... Average relative error Root mean square error Coefficient of determination The above indicators fully reflect the accuracy and feasibility of the improved BP algorithm of the present invention for predicting wind pressure drop on turning platforms of ramps, and enhance the practical application value of the method of the present invention in the field of wind pressure drop prediction for ramps.

[0166] In summary, the improved BP algorithm-based method for predicting wind pressure drop on inclined turning platforms introduces physical constraints into a neural network to construct the prediction model, improving the model's physical consistency and making it more applicable to mine inclined turning platforms, while significantly increasing computational efficiency. Furthermore, the improved BP algorithm-based method for predicting wind pressure drop on inclined turning platforms introduces physical constraints and data-driven approaches, providing a data foundation for calculating maximum ventilation resistance in mines and further providing strong technical support for the theoretical basis of mine ventilation.

[0167] Please see Figure 3In an optional embodiment, the present invention also provides an improved BP algorithm wind pressure drop prediction system for a turnaround ramp platform. This improved BP algorithm wind pressure drop prediction system for a turnaround ramp platform includes a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to call the program instructions and execute the specific steps of the improved BP algorithm wind pressure drop prediction method for a turnaround ramp platform and related embodiments provided by the present invention. The improved BP algorithm wind pressure drop prediction system for a turnaround ramp platform of the present invention has a complete structure and is objective and stable.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. An improved BP algorithm method for predicting wind pressure drop on a turning platform of a ramp, characterized in that, Includes the following steps: An initial data set of ventilation data for a turnaround ramp platform is obtained, and a data standardization processing model is set to process the initial data set to obtain a standardized data set of ventilation data for the turnaround ramp platform. The ventilation local resistance parameters are determined based on the standardized dataset, and the ventilation local resistance equation of the turnaround ramp platform is established based on the ventilation local resistance parameters. A neural network structure is constructed, and a loss function is established by combining the neural network structure and the ventilation local resistance equation. The neural network structure is trained and evaluated based on the standardized dataset using the loss function to obtain a candidate wind pressure drop prediction model. Based on the evaluation results of the candidate wind pressure drop prediction models, the final wind pressure drop prediction model for the turnaround ramp platform is determined. Based on this final model, the wind pressure drop prediction analysis for the turnaround ramp platform is implemented. The determination of local ventilation resistance parameters based on the standardized dataset includes: Construct expressions for local ventilation resistance parameters; The ventilation local resistance parameters are determined using the ventilation local resistance parameter expression and the standardized dataset. The expression for the local resistance parameter of ventilation satisfies the following relationship: ， in, To support the protrusion, This is the distance from the outermost edge of the support structure to the center of the tunnel cross-section. This is the distance from the original excavated wall of the tunnel to the center of the circle; ， in, This is the relative elevation difference coefficient. For the longitudinal elevation difference of the turning platform, The width of the alleyway; The ventilation local resistance equation for the turnaround ramp platform based on the ventilation local resistance parameters includes: The support protrusion and relative height difference coefficient are obtained using the expression for the local resistance parameter of ventilation. Based on the original historical data in the initial dataset, the basic local resistance coefficient and various influence coefficients were determined by using a multiple linear regression method. Based on the support protrusion, the relative height difference coefficient, and the influence coefficient, a ventilation local resistance equation is established for the turning platform of the turnaround ramp. The local resistance equation for ventilation satisfies the following relationship; ， in, This is a reference value for local wind pressure drop calculated based on physical equations. Based on the local drag coefficient, for Influence coefficient, To support the protrusion, for Influence coefficient, This is the relative elevation difference coefficient. for and The synergistic influence coefficient For airflow density, The airflow velocity; The construction of the neural network structure, and the establishment of the loss function based on the neural network structure and the ventilation local resistance equation, include: Determine the number of layers and nodes in the neural network, introduce the non-linear activation function of the BP algorithm, build the topology of the neural network, and initialize the weights and bias parameters of the neural network; Physical constraint equations are introduced, and the residuals of the airflow continuity equation and the wind pressure drop residuals are analyzed based on these equations. A loss function is established by combining the neural network structure, the residual of the airflow continuity equation, the residual of the wind pressure drop, and the ventilation local resistance equation; The loss function satisfies the following relationship: ， in, For loss function, For neural network predictions and actual observed values The mean square error between them To balance the weights, To reduce the residual weight for wind pressure, For the physical consistency residual of wind pressure drop, For the continuous residual weight of airflow, It represents the continuous residual of the airflow.

2. The improved BP algorithm method for predicting wind pressure drop on a turning platform of a ramp according to claim 1, characterized in that, The initial data set for obtaining ventilation information for the turnaround ramp platform includes: Obtain a three-dimensional model of the tunnel, and based on the three-dimensional model of the tunnel, obtain the three-dimensional spatial data of the turnaround ramp platform; Collect ventilation information of the turnaround ramp platform, and obtain ventilation parameter data of the turnaround ramp platform over a period of time based on the ventilation information; The initial data set for ventilation of the turnaround ramp platform is obtained by integrating the three-dimensional spatial data and the ventilation parameter data.

3. The improved BP algorithm method for predicting wind pressure drop on a turning platform of a ramp according to claim 2, characterized in that, The data standardization processing model is used to process the initial dataset to obtain a standardized dataset for ventilation of the turnaround ramp platform, including: The mean and standard deviation of the initial dataset are analyzed based on the initial dataset. A data standardization model is established based on the mean and the standard deviation. The initial dataset is processed using the data standardization processing model to obtain a standardized dataset for ventilation of the turnaround ramp platform. The data standardization processing model satisfies the following relationship: ， in, To standardize data, The original data in the initial dataset. The mean, The standard deviation is denoted as .

4. The improved BP algorithm method for predicting wind pressure drop on a turning platform of a ramp according to claim 1, characterized in that, The step of training and evaluating the neural network structure based on the standardized dataset and the loss function to obtain the candidate wind pressure drop prediction model includes: Establish a training model with the objective of minimizing the loss function; An iterative optimization algorithm is used to iteratively solve the training model based on the standardized dataset, updating the weights and bias parameters of the neural network until the convergence condition is met. Set up a verification and evaluation function to evaluate the performance of the model during the iteration process; The predictive performance of the wind pressure drop prediction model is analyzed based on the verification and evaluation function. Candidate wind pressure drop prediction models are determined based on the minimization results and the predicted performance.

5. The improved BP algorithm method for predicting wind pressure drop on a turning platform of a ramp according to claim 4, characterized in that, The process of determining the final wind pressure drop prediction model for the turnaround ramp platform based on the evaluation results of the candidate wind pressure drop prediction models includes: Extract different combinations of input features from the standardized dataset; Select a wind pressure drop prediction model based on the aforementioned input feature combination; The wind pressure drop prediction model satisfies the following relationship: ， in, For wind pressure drop prediction models, For input data, These are geometric feature vectors extracted based on the 3D model of the tunnel. The width of the alleyway, For the longitudinal elevation difference of the turning platform, To support the protrusion, For airflow velocity, For airflow density, specifically the It is vector data obtained after feature extraction from a three-dimensional model, which includes at least one or more combinations of turning radius, turning angle, tunnel centerline length, and cross-sectional contraction rate. Geometric features can quantitatively characterize the complex influence of the spatial morphology of the turning platform on the airflow field.

6. The improved BP algorithm method for predicting wind pressure drop on a turning platform of a ramp according to claim 5, characterized in that, The wind pressure drop prediction model selected based on the combination of input features includes: Based on different combinations of input features, multiple candidate wind pressure drop prediction models are trained to construct a candidate model set. A hierarchical decision tree model is constructed, and the prediction performance index output by the verification and evaluation function and the physical constraint residual value are used as the discrimination nodes of the hierarchical decision tree model. The candidate model set is screened and analyzed using the hierarchical decision tree model, and the output model with the best performance is determined as the final wind pressure drop prediction model for the turnaround ramp platform.

7. An improved BP algorithm-based wind pressure drop prediction system for turning platforms on ramps, characterized in that, The system includes a processor, input devices, output devices, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to call the program instructions to execute the improved BP algorithm for predicting wind pressure drop on a turning platform of a ramp as described in any one of claims 1-6.

Images