Mine fan ventilation control method and system based on digital twinning

Abstract

The invention relates to the technical field of mine ventilation control, in particular to a mine fan ventilation control method and system based on digital twinning. The method comprises the following steps: carrying out data preprocessing on acquired mine monitoring data; performing feature extraction on the preprocessed mine monitoring data; utilizing a random forest algorithm to predict the relationship between the data characteristics and the fan air volume; a digital twinborn model is used for resolving the relation between the data characteristics and the fan air volume, and a fan ventilation control strategy is obtained; and generating a PLC control instruction based on the fan ventilation control strategy. And through multi-source data fusion and deep analysis, the real-time ventilation requirement of the mine can be accurately grasped. Based on random forest algorithm prediction and digital twinborn model calculation, dynamic and accurate adjustment of the air volume of the draught fan is achieved, it is ensured that the concentration of harmful gas in all underground areas is maintained within a safety threshold value, the oxygen content is sufficient, an ideal environment is created for personnel operation and equipment operation, and the safety and reliability of a ventilation system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mine ventilation control, and in particular to a mine fan ventilation control method and system based on digital twin. Background Technique

[0002] In today's mine mining industry, with the continuous deepening and expansion of mining activities, the mine ventilation system faces many challenges that need to be solved urgently, and the demand for advanced ventilation control technology is becoming more and more urgent.

[0003] From the perspective of environmental complexity, the depth and scope of mine mining continue to expand, and the underground environment has become extremely complex. On the one hand, the risk of generation and accumulation of harmful gases has increased significantly. For example, toxic and harmful gases such as carbon monoxide and nitrogen dioxide, as well as exhaust gases emitted by diesel equipment during operation, not only seriously threaten the lives and health of miners, but also pose major hidden dangers to safe production. On the other hand, multi-level mining makes the gas distribution and ventilation requirements of each level vary greatly, and the traditional ventilation mode is difficult to meet such complex differential requirements.

[0004] In terms of ventilation mode, the traditional mine ventilation system mostly adopts a fixed ventilation mode and lacks the ability to adjust in real time according to actual working conditions. This mode cannot adapt to the dynamic changes in the number of people entering the well with the production progress, as well as the special requirements of different working scenarios and personnel densities for air quality and oxygen content. It not only causes a large amount of energy waste, but also is difficult to ensure good ventilation effects and cannot effectively cope with the complex and changeable underground environment.

[0005] From the perspective of safety and energy-saving requirements, the safety, reliability and energy-saving of the mine ventilation system are the keys to ensuring production. Ensuring the safety and comfort of the underground working environment, avoiding safety accidents caused by poor ventilation, and at the same time reducing ventilation energy consumption to reduce operating costs are important goals for the development of the industry. However, the existing ventilation technologies have obvious deficiencies in these aspects and are difficult to meet the actual production needs.

[0006] Therefore, to solve the above problems and improve the overall performance of the mine ventilation system, an innovative technical solution is urgently needed, which can comprehensively consider multiple factors such as the number of underground personnel, equipment operation status, and gas concentration in each level, realize real-time dynamic adjustment of the ventilation volume, and achieve precise ventilation with the help of advanced control means and monitoring technologies, providing strong support for the safe production and sustainable development of the mine. Summary of the Invention

[0007] To solve the problems mentioned above, the present invention provides a mine fan ventilation control method and system based on digital twin.

[0008] In the first aspect, a method for dynamically controlling the air volume of a mine fan based on monitoring data provided by the present invention adopts the following technical solutions:

[0009] A mine fan ventilation control method based on digital twin, comprising:

[0010] Obtain mine monitoring data;

[0011] Perform data preprocessing on the obtained mine monitoring data;

[0012] Extract features from the preprocessed mine monitoring data;

[0013] Use the random forest algorithm to predict the relationship between data features and fan air volume;

[0014] Use the digital twin model to solve the relationship between data features and fan air volume, and obtain the fan ventilation control strategy;

[0015] Generate PLC control instructions based on the fan ventilation control strategy.

[0016] Furthermore, the obtaining of mine monitoring data includes static monitoring data, dynamic monitoring data, and equipment monitoring data. Among them, the static monitoring data includes data of underground personnel, simultaneous operating power data of diesel engines, and air supply volume data; the dynamic data includes concentrations of harmful gases, oxygen concentration, and carbon dioxide concentration in the mine; the equipment monitoring data includes motor bearing and motor temperature data, vibration data, and electrical parameter data.

[0017] Furthermore, the data preprocessing of the obtained mine monitoring data includes missing data processing and abnormal data processing. Among them, the K-nearest neighbor algorithm is used to fill in the missing data, and the Z-score method is used to identify abnormal data.

[0018] Furthermore, the feature extraction of the preprocessed mine monitoring data includes first performing dimensionality reduction feature extraction through the PCA principal component analysis method, extracting time series features based on wavelet transform, and further extracting features using a multi-layer autoencoder. After feature fusion and selection of the features extracted by PCA principal component analysis, wavelet transform, and autoencoder respectively, the final data features are obtained.

[0019] Furthermore, the use of the random forest algorithm to predict the relationship between data features and fan air volume includes constructing decision trees using the CART algorithm, using the mean squared error as a measure of node splitting, forming a random forest by constructing N decision trees, evaluating each decision tree using out-of-bag data, using the out-of-bag error as an estimate of the performance of the random forest model on unseen data, and finally tuning the parameters of the random forest through random search, so as to obtain the relationship between data features and fan air volume.

[0020] Further, the method for calculating the relationship between data features and the air volume of the fan by using the digital twin model to obtain the ventilation control strategy of the fan includes separately constructing a physical ventilation model and a data-driven model based on the digital twin, and using a weighted fusion method to fuse the physical ventilation model and the data-driven model. With the air quality and ventilation energy consumption of the mine set as the goals, the relationship between data features and the air volume of the fan is calculated by defining an optimization objective function and setting constraint conditions, so as to simulate the operating state of the ventilation system.

[0021] Further, generating a PLC control instruction based on the ventilation control strategy of the fan includes converting the air volume control requirement in the ventilation control strategy of the fan into a speed control requirement of the fan motor, establishing the relationship between the air volume and the motor speed according to the performance curve of the fan, converting the motor speed control requirement into a control instruction recognizable by the PLC, and finally sending the generated PLC control instruction to the PLC controller of the mine ventilation system through the industrial Ethernet to achieve real-time control of the fan.

[0022] In a second aspect, a mine fan ventilation control system based on digital twins includes:

[0023] A data acquisition module configured to acquire mine monitoring data;

[0024] A preprocessing module configured to perform data preprocessing on the acquired mine monitoring data;

[0025] A feature extraction module configured to extract features from the preprocessed mine monitoring data;

[0026] A prediction module configured to predict the relationship between data features and the air volume of the fan by using the random forest algorithm;

[0027] A calculation module configured to calculate the relationship between data features and the air volume of the fan by using the digital twin model to obtain the ventilation control strategy of the fan;

[0028] A control module configured to generate a PLC control instruction based on the ventilation control strategy of the fan.

[0029] In a third aspect, the present invention provides a computer-readable storage medium storing multiple instructions, which are suitable for being loaded and executed by a processor of a terminal device to perform the above-mentioned method for controlling a mine fan ventilation based on digital twins.

[0030] In a fourth aspect, the present invention provides a terminal device including a processor and a computer-readable storage medium. The processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, which are suitable for being loaded and executed by the processor to perform the above-mentioned method for controlling a mine fan ventilation based on digital twins.

[0031] In summary, the present invention has the following beneficial technical effects:

[0032] 1. Through multi-source data fusion and in-depth analysis, the real-time ventilation requirements of the mine can be accurately grasped. Based on the prediction of the random forest algorithm and the calculation of the digital twin model, the dynamic and accurate adjustment of the fan air volume is realized, ensuring that the concentration of harmful gases in each area underground is maintained within the safety threshold, with sufficient oxygen content, creating an ideal environment for personnel operation and equipment operation, and significantly improving the safety and reliability of the ventilation system.

[0033] 2. Bid farewell to the energy waste of the traditional fixed ventilation mode, and intelligently regulate the fan speed according to the actual working conditions. While ensuring the ventilation effect, the fan energy consumption is greatly reduced. Taking the centrifugal fan as an example, the speed is accurately adjusted according to the linear relationship between the air volume and the speed. Through actual application verification, compared with the traditional ventilation method, the mine operation cost is effectively reduced, contributing to green mining.

[0034] 3. Through the real-time analysis of equipment monitoring data, a comprehensive perception of the operating status of key equipment such as fans is realized. By monitoring the motor bearing temperature, vibration data and electrical parameters, potential equipment failure hazards can be detected in a timely manner, early warnings can be issued and maintenance can be arranged to avoid production interruptions caused by sudden equipment failures. According to statistics, after the application of this technology, the continuity and stability of mine production are effectively guaranteed. Brief Description of the Drawings

[0035] Figure 1 is a schematic diagram of a mine fan ventilation control method based on digital twin in Embodiment 1 of the present invention;

[0036] Figure 2 is a schematic diagram of a mine fan ventilation control system based on digital twin in Embodiment 1 of the present invention. Detailed Embodiment

[0037] The present invention will be further described in detail below with reference to the accompanying drawings.

[0038] Embodiment 1

[0039] Referring to Figure 1 , a mine fan ventilation control method based on digital twin in this embodiment includes:

[0040] Obtain mine monitoring data;

[0041] Perform data preprocessing on the obtained mine monitoring data;

[0042] Extract features from the preprocessed mine monitoring data;

[0043] Use the random forest algorithm to predict the relationship between data features and fan air volume;

[0044] Use the digital twin model to calculate the relationship between data features and fan air volume, and obtain the fan ventilation control strategy;

[0045] Generate PLC control instructions based on the fan ventilation control strategy.

[0046] Specifically:

[0047] S1. Obtain mine monitoring data, including static monitoring data, dynamic monitoring data, and equipment monitoring data:

[0048] Static monitoring data:

[0049] Data of personnel entering the mine: Record information such as the number of personnel currently entering the mine and their distribution areas. It can be obtained through the attendance system or personnel positioning system at the mine entrance.

[0050] Simultaneous operating power data of diesel engines: Statistically calculate the total power of diesel engines operating simultaneously in the mine. Diesel engines consume oxygen and produce harmful gases when working in the mine, and the power is related to the gas production volume. It can be obtained through the power monitoring device of the diesel engine.

[0051] Air supply volume data: Measure the air supply volume in each area of the mine to reflect the working state of the current ventilation system. It can be calculated using a wind speed sensor and the cross-sectional area of the ventilation duct.

[0052] Dynamic monitoring data:

[0053] Concentration of harmful gases in the mine: Such as the concentration of harmful gases such as carbon monoxide (CO) and hydrogen sulfide (H 2 S). These gases are extremely harmful to human health and need to be monitored in real time. Gas sensors can be used for detection.

[0054] Oxygen concentration: Oxygen is a necessary condition for maintaining human respiration. Too low oxygen concentration in the mine can cause personnel to suffocate due to lack of oxygen. Data is obtained through an oxygen sensor.

[0055] Carbon dioxide concentration: Too high carbon dioxide concentration can affect the respiratory function of personnel. Gas sensors are also used for monitoring.

[0056] Equipment monitoring data:

[0057] Motor bearing and motor temperature data: Motors generate heat during operation. Too high temperature may cause motor damage. The temperature of the motor bearing and the motor body is monitored through temperature sensors.

[0058] Vibration data: Abnormal motor vibration may indicate equipment failure or unstable installation. Vibration sensors are used to collect the vibration data of the motor.

[0059] Electrical parameter data: Include parameters such as the current, voltage, and power of the motor. These parameters reflect the operating state and energy consumption of the motor and can be measured through electrical instruments.

[0060] S2. Data cleaning

[0061] Missing data handling: The K-nearest neighbor algorithm is used for data filling. For a sample x with missing values, calculate its Euclidean distance from all other samples x i (i = 1, 2,..., n) (where m is the number of features). Select the K nearest samples, and fill the missing values according to the average of the corresponding feature values of these K samples .

[0062] Outlier data handling: The Z-score method is used to identify outliers. For each data point x in the dataset i , calculate its Z-score value (where μ is the mean of the dataset and σ is the standard deviation of the dataset). If |Z i | > 3, then consider this data point as an outlier, and it can be deleted or corrected using the average of adjacent normal data

[0063] Data normalization: The min-max normalization method is adopted to map the data to the interval [0, 1]. For the original data x, the normalized result where x min and x max are the minimum and maximum values of this feature data respectively

[0064] S3. Feature extraction:

[0065] Step 1: Before feature extraction, the preprocessed mine monitoring data needs to be divided into a training set and a test set according to a certain ratio, generally 80:20 or 70:30. Let the preprocessed data matrix be X, with dimensions n×m, where n is the number of samples and m is the number of features. Divide X into the training set X train and the test set X test , and at the same time, the corresponding labels (if in a supervised learning scenario) are also divided accordingly

[0066] Step 2: Preliminary dimensionality reduction based on principal component analysis (PCA)

[0067] Data centering: For the training set X train , calculate the mean vector of each column where n train is the number of training set samples, and x i is the i-th sample vector. Subtract the mean vector from each row of X train to obtain the centered data matrix X c_train .

[0068] Covariance matrix calculation: Calculate X c_train 's covariance matrix

[0069]

[0070] Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix C, i.e., C = VΛVT ， where V is an orthogonal matrix composed of eigenvectors, and Λ is a diagonal matrix composed of eigenvalues.

[0071] Principal component selection: Sort the eigenvalues in descending order and calculate the cumulative variance contribution rate When R(k) reaches 0.9, determine to select the eigenvectors corresponding to the first k largest eigenvalues to form the projection matrix P.

[0072] Data projection: Multiply the centralized training set X c_train by the projection matrix P to obtain the dimensionality-reduced training set Y train = X c train P. For the test set X test , first perform centering (using the mean vector μ calculated from the training set), and then multiply by the projection matrix P to obtain the dimensionality-reduced test set Y test .

[0073] Step 3: Wavelet transform to extract time series features (for dynamic monitoring data)

[0074] Data screening: Screen out the dynamic monitoring data (such as time series data of the change of harmful gas concentration in the mine over time, the change of oxygen concentration over time, etc.) from the dimensionality-reduced data Y train and Y test , and denote them as Y train_time and Y test_time .

[0075] Wavelet basis selection: Select the Daubechies wavelet as the wavelet basis because it has good compact support and vanishing moment properties, and can effectively capture the local features of the signal.

[0076] Wavelet decomposition: Perform 3-layer wavelet decomposition on each time series data in Y train_time to obtain the approximation coefficient cA 3 and the detail coefficients cD 1 , cD 2 , cD 3 . The wavelet decomposition formula is , where φ 3，k (t) is the scaling function, and ψ i，k (t) is the wavelet function.

[0077] Feature extraction: Extract features from the approximation coefficients and detail coefficients, including mean, variance, and energy. For example, the energy of the detail coefficients at the i-th layer Combine the extracted features into a new feature vector to obtain the features W of the training set after wavelet transform train . For Y test_time Perform the same wavelet decomposition and feature extraction operations to obtain the features W of the test set after wavelet transform test .

[0078] Step 4: The autoencoder further extracts features

[0079] Model construction: Construct a multi-layer autoencoder. The dimension of the input layer is the number of features of Y train (the dimension after PCA dimensionality reduction), and the dimension of the hidden layer is set to half of the dimension of the input layer. The output h of the encoder is h = f(W 1 x + b 1 ), where W 1 is the weight matrix of the encoder, b 1 is the bias vector, and f uses the ReLU activation function. The output of the decoder where W 2 is the weight matrix of the decoder, b 2 is the bias vector, and g uses the linear activation function.

[0080] Model training: Use the mean squared error loss function

[0081] to train the autoencoder, and use the Adam optimization algorithm to update the parameters of the model. The number of training epochs is set to 100.

[0082] Feature extraction: After training, input Y train into the encoder to obtain the output of the hidden layer, which is the features A of the training set extracted by the autoencoder train . Input Y test into the encoder to obtain the features A of the test set extracted by the autoencoder test .

[0083] Step 5: Feature fusion

[0084] Fuse the features after PCA dimensionality reduction, the features extracted by wavelet transform, and the features extracted by the autoencoder. For the training set, concatenate Y train , W train and A train column-wise into a new feature matrix F train . For the test set, concatenate Y test , W test and A test column-wise into the new feature matrix F test .

[0085] Step 6: Feature Selection

[0086] Use the chi-square test to perform feature selection on the fused feature matrix F train (assuming a supervised learning scenario with corresponding class labels y train ).

[0087] Calculate the chi-square value: For each feature in F train , calculate its chi-square value with the class label y train where O is the actual observed value and E ij is the expected observed value. ij

[0088] Feature screening: Select the top 80% of the features with the highest chi-square values as the final training set features F train_final . According to the feature indices selected from the training set, select the corresponding features from F test to obtain the final test set features F test_final .

[0089] S4. Use the random forest algorithm to predict the relationship between data features and fan air volume

[0090] 1. Before using the random forest algorithm, the data after feature extraction needs to be prepared. Assume that the data matrix obtained after feature extraction is X, with dimensions n×p, where n is the number of samples and p is the number of features; the corresponding fan air volume label vector is y, with dimensions n×1.

[0091] Divide the dataset into a training set (X train , y train ) and a test set (X test , y test ) in a ratio of 7:3 or 8:2. To ensure the generalization ability of the model, stratified sampling can be used, especially when there is class imbalance in the data, to ensure that the proportions of each class in the training set and test set are similar to those in the original dataset.

[0092] 2. Random forest model construction and training

[0093] 2.1 Basic principle of random forest

[0094] Random forest is an ensemble learning model composed of multiple decision trees. In the construction process of each decision tree, the methods of bootstrap sampling and random feature selection are adopted to increase the diversity and generalization ability of the model.

[0095] 2.2 Bootstrap sampling

[0096] ​Randomly draw n samples from the training set X train with replacement to form a new training subset The corresponding label subset is Since it is sampling with replacement, approximately of the samples will appear in The samples not drawn form the out-of-bag data (OOB).

[0097] 2.3 Random Feature Selection

[0098] When constructing each node of each decision tree, instead of considering all p features, randomly select m features (m < p). Usually select the optimal feature from these m features for node splitting.

[0099] 2.4 Decision Tree Construction

[0100] The construction of the decision tree uses the CART (Classification and Regression Trees) algorithm. For regression problems, the mean squared error (MSE) is commonly used as the measure for node splitting.

[0101] Let the sample set of the current node be D. For a certain feature s and splitting point, divide D into D left ={(x, y) ∈ D | x j ≤ s} and D right ={(x, y) ∈ D | x j > s} two parts. Then the mean squared error after division is:

[0102] where n left and n right are the number of samples in D left and D right respectively, n = n left + n right , MSE left and MSE right are the mean squared errors of the sample labels in D left and D right respectively:

[0103]

[0104] where and are the means of the sample labels in D left and D right respectively.

[0105] Select the feature j and splitting point s that minimize MSE(j, s) for node splitting, and repeat this process until the stopping condition is met (such as the number of samples in the node is less than a certain threshold, the depth of the tree reaches the maximum, etc.).

[0106] 2.5 Random Forest Training

[0107] Repeat the above processes of bootstrap sampling, random feature selection, and decision tree construction to build N decision trees, forming a random forest RF = {T 1 , T 2 , …, T N}.

[0108] 3. Model Evaluation and Optimization

[0109] 3.1 Out-of-Bag Error (OOB Error) Evaluation

[0110] During the training process, evaluate each decision tree using the out-of-bag data. For each sample x i , find those decision trees that did not use this sample for training, and use these decision trees to predict x i to obtain the predicted value The out-of-bag error is then:

[0111]

[0112] The out-of-bag error can be used as an estimate of the model's performance on unseen data and is used to preliminarily judge the model's generalization ability.

[0113] 3.2 Test Set Evaluation

[0114] Use the test set (X test , y test ) to evaluate the trained random forest model. Commonly used evaluation metrics include mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R 2 ).

[0115]

[0116]

[0117] where n test is the number of samples in the test set, is the predicted value of the model for the i-th sample,

[0118] is the mean of the test set sample labels.

[0119] 3.3 Model Optimization

[0120] Parameter Tuning: Methods such as Grid Search or Random Search can be used to tune the parameters of the random forest, such as the number of decision trees N, the number of features m randomly selected for each node, the maximum depth of the tree, etc.

[0121] Feature Importance Analysis: The random forest can calculate the importance of each feature. By sorting the feature importances, unimportant features can be removed, reducing the complexity of the model and improving the generalization ability of the model. The importance I of feature j j can be calculated in the following way:

[0122] For each decision tree T k , calculate the out-of-bag data prediction error OOBError under the original features k ; then randomly shuffle the values of feature j in the out-of-bag data and calculate the prediction error OOBError' again k . Then the importance of feature j in decision tree T k is

[0123] ΔOOBError k = OOBError' k - OOBError k . Average the importance of feature j in all decision trees to obtain the final importance of feature j:

[0124]

[0125] 4. Creative Expansion: Combine Bayesian Optimization and Feature Interaction Analysis

[0126] 4.1 Bayesian Optimization of Random Forest Parameters

[0127] Traditional grid search and random search are less efficient in parameter tuning, especially in high-dimensional parameter spaces. Bayesian optimization is a global optimization method based on probability models. It can adaptively select the next set of parameters to be evaluated based on historical parameter evaluation results, thus finding the optimal parameter combination more quickly.

[0128] Bayesian optimization approximates the objective function (such as out-of-bag error) by constructing a surrogate model (such as a Gaussian process) and selects the next set of parameters according to the posterior distribution of the surrogate model. The specific steps are as follows:

[0129] Initialize a set of parameter combinations, train and evaluate the random forest model to obtain the objective function value.

[0130] Construct a Gaussian process model based on this historical data to predict the mean and variance of the objective function under different parameter combinations.

[0131] Use a sampling function (such as the expected improvement function) to select the next set of most promising parameter combinations.

[0132] Train and evaluate the new parameter combinations, and update the Gaussian process model.

[0133] Repeat the above steps until the stopping condition is met (such as reaching the maximum number of iterations).

[0134] 4.2 Feature Interaction Analysis

[0135] In addition to considering the importance of individual features, the interaction effects between features on the prediction of the fan air volume can also be analyzed. A feature interaction detection method based on random forest can be used, such as calculating the conditional feature importance to discover the interactions between features.

[0136] Given features j and l, calculate the importance I of feature j when the value of feature l is fixed. j|l . If I j|l differs significantly from the ordinary importance I of feature j j , it indicates that there is an interaction between features j and l. In this way, more information hidden in the data can be mined, further improving the prediction performance of the model.

[0137] 5. Predicting the Fan Air Volume

[0138] Input the new data after preprocessing and feature extraction into the trained random forest model. For each decision tree T i , obtain the predicted value The final predicted value of the random forest is the average of the predicted values of all decision trees:

[0139] Through the above steps, the relationship between data features and the fan air volume can be predicted more accurately using the random forest algorithm. At the same time, the performance and interpretability of the model are further improved through creative expansion methods.

[0140] S5. Use the digital twin model to solve the relationship between data features and the fan air volume, and obtain the fan ventilation control strategy, where

[0141] 1. Digital Twin Model Construction

[0142] Physical Model Construction

[0143] Ventilation Network Topology Modeling: Abstract the mine ventilation system as a network topology structure composed of nodes and branches. Nodes represent the intersection points of ventilation roadways, fan positions, chambers, etc., and branches represent ventilation roadways. Represented by graph theory, let G=(V, E) be the ventilation network topology graph, where V={v 1 , v 2 , …, vn} is a set of nodes, and \(E = \{e\) 1 , e 2 , …, e m} is a set of branches. For each branch \(e\) i , its length \(L\) i , cross-sectional area \(S\) i , friction resistance coefficient \(\alpha\) i and other physical parameters are defined.

[0144] Fan characteristic modeling: The performance curve of a fan can usually be fitted with a polynomial function. Assume that the relationship between the air pressure \(H\) and the air volume \(Q\) of the fan can be expressed as

[0145] \(H = a\) 0 + a 1 \(Q + a\) 2 \(Q\) 2 + … + a k \(Q\) k , where \(a\) 0 , \(a\) 1 , …, \(a\) k are coefficients obtained through fan performance tests. At the same time, considering the relationship between the efficiency \(\eta\) of the fan and the air volume \(Q\), it can be fitted as

[0146] \(\eta = b\) 0 + b 1 \(Q + b\) 2 \(Q\) 2 + … + b l \(Q\) l .

[0147] Airflow modeling: According to the principles of fluid mechanics, the airflow in a ventilation roadway follows the law of conservation of energy and the law of conservation of mass. For branch \(e\) i , its resistance \(h\) i can be calculated using the Darcy - Weisbach formula: where \(\lambda\) i is the resistance coefficient of the branch, and \(Q\) i is the air volume in the branch. The law of conservation of mass is expressed as where \(I\) in and \(I\) out are the sets of branches flowing into and out of a certain node respectively.

[0148] Data - driven model construction

[0149] Machine learning model selection: The long short - term memory network (LSTM) is used to capture the dynamic characteristics of the mine ventilation system. The calculation formulas for the input gate \(i\) t , forget gate \(f\) t , output gate \(o\) t and cell state \(C\) t of the LSTM unit are as follows:

[0150] i t = σ(W ii x t + W hi h t-1 + b i )

[0151] f t = σ(W if x t + W hf h t-1 + b f )

[0152] o t = σ(W io x t + W ho h t-1 + b o )

[0153]

[0154] h t = o t ⊙ tanh(C t )

[0155] Among them, x t is the input vector at time t, h t-1 is the hidden state at the previous time, W is the weight matrix, b is the bias vector, σ is the sigmoid function, tanh is the hyperbolic tangent function, and ⊙ is element-wise multiplication.

[0156] Model training: Use historical monitoring data as the training set, with data features (such as harmful gas concentration, number of personnel, etc.) as the input and the fan air volume and other ventilation parameters as the output to train the LSTM model. The mean squared error loss function is adopted where N is the number of samples, y i is the true value, is the predicted value. Use stochastic gradient descent (SGD) or its improved algorithms (such as the Adam algorithm) to update the model parameters.

[0157] Model fusion

[0158] Adopt the method of weighted fusion to fuse the physical model and the data-driven model. Let the output of the physical model be y p , and the output of the data-driven model be y d , and the fused output y = ω p y p + ω d y d , where ω p and ωd is the weight coefficient, and ω p + ω d = 1. The weight coefficient can be optimized by cross - validation to minimize the error of the fusion model on the validation set.

[0159] 2. Solve the relationship between data features and fan air volume

[0160] Definition of the optimization objective function

[0161] Taking the goal of ensuring the air quality in the mine and reducing ventilation energy consumption, define the optimization objective function J. Let C g,k be the concentration of the k - th harmful gas, be the safety concentration threshold of this harmful gas; Q i be the air volume of the i - th fan, P i be the power of the i - th fan, and be the penalty coefficient. Then the objective function can be expressed as:

[0162]

[0163] where K is the number of types of harmful gases, and M is the number of fans.

[0164] Setting of constraint conditions

[0165] Air volume balance constraint: For each node in the ventilation network, the air volume flowing into the node is equal to the air volume flowing out of the node, that is

[0166] Fan performance constraint: The air volume and air pressure of the fan need to satisfy its performance curve, that is and 0 ≤ Q i ≤ Q i,max where Q i，max is the maximum air volume of the i - th fan.

[0167] Gas concentration constraint: The concentration of harmful gases in each area of the mine cannot exceed the safety threshold, that is.

[0168] Solving the optimization algorithm

[0169] Use the particle swarm optimization algorithm (PSO) to solve the optimization problem. Let the position vector of the particle X=(Q 1 , Q 2 ,..., Q M ) represent the air volumes of each fan, and the velocity vector V=(v 1 , v 2 ,..., v M ). The update formula of the particle is as follows: v i，d (t + 1)=w×v i，d (t)+c 1 ×r 1×(p best，i，d -x i，d (t))+c 2 ×r 2 ×(g best，d -x i，d (t))。

[0170] x i，d (t + 1) = x i，d (t) + v i，d (t + 1)

[0171] where w is the inertia weight, c 1 and c 2 are acceleration constants, r 1 and r 2 are random numbers between 0 and 1, p best，i is the historical optimal position of the i-th particle, g best is the global optimal position, t is the number of iterations, and d represents the dimension.

[0172] 3. Develop the ventilation control strategy for the fan

[0173] Dynamic partition control: According to the regional functions and the distribution of harmful gases in the mine, the mine is divided into different ventilation partitions. For each partition, an independent ventilation control strategy is developed based on the calculated optimal air volume. For example, the personnel operation area, the coal storage bunker area, etc. are divided into different partitions respectively. For the personnel operation area, the oxygen supply and the dilution of harmful gases are mainly ensured, and for the coal storage bunker area, the dust concentration is mainly controlled.

[0174] Real-time feedback regulation: Use the digital twin model to simulate the operation status of the ventilation system in real time, and compare the simulation results with the actual monitoring data. If there is a deviation between the two, the air volume and the operation status of the fan are adjusted in a timely manner. For example, when the harmful gas concentration in a certain area is actually monitored to be higher than the simulated value, the air supply volume in this area is increased, and the optimization calculation is carried out again.

[0175] Emergency response strategy: Aiming at possible emergencies in the mine (such as fires, gas explosions, etc.), an emergency ventilation control strategy is developed. When an emergency occurs, according to the type and location of the event, the operation mode of the fan is quickly adjusted to ensure that harmful gases can be discharged in time and provide good ventilation conditions for personnel evacuation and rescue work. For example, when a fire occurs, the air reversal technology is adopted to introduce fresh air into the fire area and discharge the smoke and harmful gases out of the mine at the same time.

[0176] S6. Convert the strategy into motor parameters

[0177] The fan ventilation control strategy usually gives ventilation parameters such as the target air volume and air pressure, while the PLC needs to control the operating parameters of the fan motor, such as speed, voltage, current, etc. First, it is necessary to establish the mapping relationship between the ventilation parameters and the motor parameters.

[0178] Air volume - speed relationship: For a centrifugal fan, according to the similarity law, when the geometric shape and air density of the fan remain unchanged, the air volume Q is proportional to the speed n, that is, Q 1 / Q 2 =n 1 / n 2 . Assuming that the current fan speed n 1 and air volume Q 1 are known, as well as the target air volume Q 2 required by the control strategy, then the target speed can be calculated

[0179] Air pressure - speed relationship: The air pressure H is proportional to the square of the speed, that is, H 1 / H 2 =(n 1 / n 2 ) 2 . This relationship can be used to further verify whether the target speed can meet the air pressure requirement.

[0180] 2. Conversion of motor parameters to PLC signals

[0181] The PLC usually controls the speed regulation device of the motor (such as a frequency converter) by outputting voltage or current signals through an analog output module. It is necessary to determine the signal range that the motor speed regulation device can accept, such as a 0 - 10V voltage signal or a 4 - 20mA current signal, and the corresponding motor speed range for this signal range.

[0182] Assume that the motor speed regulation device accepts a 0 - 10V voltage signal, and the corresponding motor speed range is 0 - 1500r / min. If the calculated target speed is n target , then the corresponding PLC output voltage

[0183] 3. Instruction encoding and distribution

[0184] Encode the calculated PLC output signal value and convert it into a digital quantity that the PLC can recognize and process. If the resolution of the analog output module of the PLC is 12 bits (i.e., 0 - 4095), then convert the above calculated voltage value U into the corresponding digital quantity Finally, send the encoded instruction to the PLC controller through an industrial communication network (such as Ethernet, Profibus, etc.).

[0185] Example 2

[0186] This embodiment provides a mine fan ventilation control system based on digital twin, as Figure 2 shown, Figure 2 which shows a mine ventilation control scheme, covering three parts: ventilation stage division and related parameter setting, ventilation equipment control architecture, and gas monitoring in each middle section.

[0187] In terms of ventilation stage division and related parameter setting, it is divided into four stages. Each stage corresponds to different main fan air supply volumes (increasing from 40% to 100%), different ranges of the number of people entering the well (increasing from 0 - 39% to 80 - 100%), and the parameter of the diesel equipment is given as 4m 3 / kw.min, as well as the corresponding PIC frequency (increasing from 20HZ to 50HZ).

[0188] In the part of the ventilation equipment control architecture, the main fan is connected to the core module through PLC, and the core module is respectively connected to two local fans (Local Fan 1# and Local Fan 2#). Each local fan also communicates with the core module through PLC, reflecting the centralized control of the core module over the main fan and local fans.

[0189] In terms of gas monitoring in each middle section, O 2 , CO, CO 2 , NO 2 and other gases are monitored in the 1st - 4th middle sections. These monitoring data may be used to assist in adjusting the ventilation control strategy.

[0190] A computer - readable storage medium stores multiple instructions, and the instructions are adapted to be loaded and executed by a processor of a terminal device for the described mine fan ventilation control method based on digital twin.

[0191] A terminal device includes a processor and a computer - readable storage medium. The processor is used to implement each instruction; the computer - readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor for the described mine fan ventilation control method based on digital twin.

[0192] The above are all preferred embodiments of the present invention. Without limiting the protection scope of the present invention accordingly, therefore: All equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A mine fan ventilation control method based on digital twin, characterized in that: include: Obtaining mine monitoring data; Perform data preprocessing on the acquired mine monitoring data; Extract features from the preprocessed mine monitoring data; Use the random forest algorithm to predict the relationship between data features and fan air volume; The digital twin model is used to solve the relationship between data characteristics and fan air volume to obtain the fan ventilation control strategy; Generate PLC control instructions based on fan ventilation control strategy.

2. A mine fan ventilation control method based on digital twin according to claim 1, characterized in that: The acquired mine monitoring data includes static monitoring data, dynamic monitoring data and equipment monitoring data, wherein the static monitoring data includes data of personnel entering the mine, data of diesel engine power running simultaneously and air supply volume data; the dynamic data includes concentrations of harmful gases, oxygen and carbon dioxide in the mine; the equipment monitoring data includes motor bearing and motor temperature data, vibration data and electrical parameter data.

3. A mine fan ventilation control method based on digital twin according to claim 2, characterized in that: The data preprocessing for the acquired mine monitoring data includes missing data processing and abnormal data processing, wherein the missing data is filled by using the K-nearest neighbor algorithm and the abnormal data is identified by using the Z-score method.

4. A mine fan ventilation control method based on digital twin according to claim 3, characterized in that: The feature extraction of the pre-processed mine monitoring data includes firstly extracting dimensionality reduction features through the PCA principal component analysis method, extracting time series features based on wavelet transform, and further extracting features by using a multi-layer autoencoder. The final data features are obtained by fusing and selecting the features extracted by the PCA principal component analysis, wavelet transform and autoencoder respectively.

5. A mine fan ventilation control method based on digital twin according to claim 4, characterized in that: The method of using a random forest algorithm to predict the relationship between data features and fan air volume includes using a CART algorithm to construct a decision tree, using mean square error as a measure of node splitting, forming a random forest by constructing N decision trees, and evaluating each decision tree using out-of-bag data, using the out-of-bag error as a performance estimate of the random forest model on unseen data, and finally tuning the parameters of the random forest through random search to obtain the relationship between data features and fan air volume.

6. A mine fan ventilation control method based on digital twin according to claim 5, characterized in that: The digital twin model is used to solve the relationship between data characteristics and fan air volume to obtain the fan ventilation control strategy, including respectively building a physical ventilation model and a data-driven model based on the digital twin, and integrating the physical ventilation model and the data-driven model using a weighted fusion method, setting goals based on mine air quality and ventilation energy consumption, and solving the relationship between data characteristics and fan air volume by defining an optimization objective function and setting constraints, thereby simulating the operating status of the ventilation system.

7. A mine fan ventilation control method based on digital twin according to claim 6, characterized in that: The method generates PLC control instructions based on the fan ventilation control strategy, including converting the air volume control requirement in the fan ventilation control strategy into the speed control requirement of the fan motor, establishing the relationship between the air volume and the motor speed according to the performance curve of the fan, converting the motor speed control requirement into a control instruction recognizable by the PLC, and finally sending the generated PLC control instruction to the PLC controller of the mine ventilation system through the industrial Ethernet to realize real-time control of the fan.

8. A mine fan ventilation control system based on digital twin, characterized in that: include: A data acquisition module configured to acquire mine monitoring data; A preprocessing module, configured to perform data preprocessing on the acquired mine monitoring data; A feature extraction module is configured to extract features from the preprocessed mine monitoring data; A prediction module, configured to predict the relationship between data features and fan air volume using a random forest algorithm; A solution module is configured to use the digital twin model to solve the relationship between the data characteristics and the fan air volume to obtain a fan ventilation control strategy; The control module is configured to generate PLC control instructions based on the fan ventilation control strategy.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method according to claim 1 .

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method as claimed in claim 1 .

Images