A coal mine main fan frequency converter speed regulation method based on random wind speed field prediction

Abstract

The application provides a coal mine main fan frequency converter speed regulation method based on random wind speed field prediction. First, wind measuring points are arranged in underground stoping working faces, tunneling working faces and air shafts. According to various data monitored by the wind measuring points, information prediction is carried out, and a graph neural network is established to calculate air supply. According to the real-time air volume calculated by the graph neural network, the relationship between the control frequency of the main fan frequency converter and the real-time air volume is obtained, and the frequency converter control frequency is calculated. The beneficial effect is that the present application takes the random field as the theoretical basis, takes each key ventilation detection in the underground as the vertex of the random field, takes the ventilation detection result as the attribute of the vertex, establishes an underground ventilation control random field model, then establishes a field data prediction model to predict the future information of the wind field, thereby guiding the ventilation regulation of the main fan and effectively improving the response speed of the ventilation control.

Description

Technical Field

[0001] This invention relates to the field of technology, and more specifically, to a method for speed regulation of a frequency converter for a main ventilation fan in a coal mine based on random wind speed field prediction. Background Technology

[0002] The underground ventilation system in coal mines is one of the core systems for safe production. It is divided into two types: forced ventilation and exhaust ventilation. The exhaust ventilation process is as follows: fresh air enters various underground roadways through the main shaft, auxiliary shaft, or dedicated air supply shaft. Simultaneously, the main ventilation at the return air shaft opening operates, creating negative pressure to draw out stale air from the underground roadways, thus achieving air circulation. The core working locations underground are the longwall mining face and the tunneling face, therefore, ventilation safety in these core work areas is crucial.

[0003] The main ventilation fan in a coal mine is a critical piece of equipment for ensuring safe underground production. Currently, there are two control methods for main ventilation fans: fixed-frequency and variable-frequency. The fixed-frequency method sets the operating frequency of the main ventilation fan to 50Hz, with the motor drawing out polluted air from the mine at its rated speed. The airflow is mainly controlled by the damper angle; this method is not energy-efficient. The variable-frequency control method adjusts the fan speed, thereby regulating the airflow in the main ventilation shaft. Adjustment methods include frequency regulation and damper regulation. To further save energy, the damper is typically opened to its maximum capacity while the motor frequency is adjusted.

[0004] Currently, the main ventilation fan is adjusted using either rated operation or manual adjustment, lacking the capability for globally intelligent operation. The common practice is to measure wind speed and harmful gas concentrations at the underground apex, then notify the surface operator by phone to control the main ventilation fan's airflow. This method, relying solely on single-point underground monitoring, is insufficient for comprehensive control in complex underground roadways. Furthermore, because coal mine ventilation systems are characterized by significant time delays—meaning that after the main ventilation fan is adjusted, a considerable amount of time is required underground to achieve the desired airflow—the overall system adjustment delay is substantial.

[0005] Currently, control methods for main ventilation fans include PID control and derivative algorithms that combine some intelligent algorithms with PID, such as fuzzy PID control and neural network PID control. These methods can provide a certain degree of intelligent control, but their principle still relies on intelligent adjustment based on the pre-existing airflow demand. This airflow demand is typically a single-point demand, which may negatively impact other demand points. Furthermore, current control methods are based on post-event adjustment, meaning they adjust the ventilation fan according to the airflow at a specific point in time. Due to the large time delay in ventilation systems, it takes a considerable amount of time after the main ventilation fan is adjusted before the ventilation situation at the airflow request point changes. Summary of the Invention

[0006] Based on at least one of the above-mentioned technical problems, this invention proposes a method for speed regulation of the frequency converter of the main ventilation fan in coal mines based on random wind speed field prediction.

[0007] A method for speed regulation of a frequency converter for a main ventilation fan in a coal mine based on random wind speed field prediction, comprising the following steps:

[0008] S1. Determine the wind measurement points and collect their data. Wind measurement points 1 and 2 will monitor the ventilation of the two longwall faces in real time. Wind measurement point 3 will monitor the ventilation of the tunnel face in real time. Wind measurement point 4 will monitor the ventilation of the ventilation shaft in real time. Collect the wind speed V, air volume C, temperature T, gas concentration W, and carbon dioxide concentration G data of each wind measurement point, and set the detection point number as i. The detection results at time t are: Vi(t), Ci(t), Ti(t), Wi(t), Gi(t), i = 1, 2, 3, 4.

[0009] S2. Predict the monitoring information of each wind measurement point. Let the current time be t, and the input information of the i-th detection point be:

[0010] X i (t)=[V i (t), C i (t), T i (t), W i (t), G i (t)] (1)

[0011] The specific steps are as follows:

[0012] 21) The prediction model uses the LSTM architecture of PyTorch, with an input sequence length of n=20 and a prediction length of m;

[0013]

[0014] 22) The LSTM settings are as follows: input length = 20, input dimension = 5, number of LSTM layers = 1, output length m, output dimension = 5.

[0015] The output length is the predicted length, reflecting the time-delay characteristics of ventilation regulation. The calculation method for m is as follows:

[0016]

[0017] In the formula, T f The sampling time can be set; T l The time it takes for the polluted air to travel from the furthest wind measurement point underground to the main ventilation fan is measured on-site.

[0018] 3) Establish the above LSTM neural network for all wind measurement points to predict the monitoring data of each wind measurement point;

[0019] 4) Training the LSTM neural network involves collecting historical data, creating data samples, and training the network. The loss function is the mean squared error function.

[0020] S3. Calculation of main ventilator air volume based on graph neural network: A random field model P(Y|X) for the air measurement points is established using graph neural network, where X represents the input information of all air measurement points; Y represents the air volume of each air measurement point, and the air volume Y takes a value between 0 and 1; when it is 0, it means no air volume is needed, and when it is 1, it means the maximum air volume is needed. The random field model P(Y|X) essentially represents a probability value, which can be approximated by graph neural network.

[0021] Therefore, the following steps are established:

[0022] 31) Establish the input information X = [X1, X2, X3, X4] for the graph neural network, which represents the real-time detection values ​​of each wind measurement point, where,

[0023] X i =[V i (t), C i (t), T i (t), W i (t), G i (t)] (3)

[0024] 32) Establish the output information of the graph neural network, Y = [Y1, Y2, Y3, Y4], which represents the air volume requirement of each wind measurement point;

[0025] 33) Establish a random field model P(Y|X) for the air volume at each monitoring point, that is, the required air volume at each location under the monitoring data conditions at each wind measurement point, 0≤P(Y|X). i |X)≤1, where 0 represents no air volume required and 1 represents the maximum air volume. This model is approximated using a graph neural network.

[0026] 34) Based on the random field model, a graph neural network is established for each wind measurement point. The graph neural network consists of a 4-layer graph network structure.

[0027] 35) Based on the probability values ​​of air volume demand at each wind measurement point obtained by the graph neural network, and by introducing a fully connected neural network, the air supply volume D of the main ventilation fan is calculated.

[0028] 36) To train the graph neural network, collect real-time monitoring data from downhole wind measurement points, create a sample set, and use the mean squared error loss function during training until the neural network converges. Then the neural network will have the ability to calculate real-time air volume.

[0029] S4. Determine the decision-making method for the main ventilation fan inverter control: The inverter control frequency f is linearly related to the air volume.

[0030] Sexual relationship: f(t)=kD(t)(7)

[0031] In the formula, k is a constant. It can be tested directly on-site. When the inverter control frequency is 50Hz, the air volume is at its maximum (D). max ;

[0032] S5. Determine the inverter control frequency decision method, the steps of which are as follows:

[0033] 51) Set the current time as t, and determine the prediction length m according to equation (2);

[0034] 52) Based on the LSTM prediction, the values ​​at time m of each wind measurement point are:

[0035] X i (t+m)=[V i (t+m),C i (t+m), T i (t+m), W i (t+m), G i (t+m)](8)

[0036] 53) Determine the air volume D(m) at time m using a graph neural network.

[0037] 54) Determine the inverter control frequency f(m) according to equation (7);

[0038] 55) From the current time t until time t+m, the frequency converter executes the control frequency f(m).

[0039] 56) Repeat step one at time t+m to re-predict and re-execute the control frequency.

[0040] More preferably, in step S3, the graph neural network is divided into four layers. The input values ​​of each layer are the feature value Fi and the link matrix E. After the input values ​​are processed with the coal seam weights W, the output feature value F is obtained. i+1 The link matrix E remains constant, i = 1, 2, 3, 4; the input feature values ​​F of each graph neural network layer. i Weight settings, feature output value F i+1 for:

[0041]

[0042] Where p×q represents a matrix with p rows and q columns of eigenvalues;

[0043] Each layer of the graph convolutional neural network F i The specific calculation method is as follows:

[0044] a. The first layer input F1 is the feature matrix of each wind measurement point:

[0045]

[0046] b. Determine the link matrix E, where E is the link matrix for each wind measurement point.

[0047]

[0048] If the element in the i-th row and j-th column of matrix E is 1, it means that the i-th wind measurement point and the j-th wind measurement point are related; if it is 0, it means that they are not related. In this patent, all wind measurement points are related.

[0049] c. The feature values ​​output of each layer of the graph neural network are:

[0050] F i+1 =EF i W i (6)

[0051] d. According to the above graph neural network, the output F5 is actually the probability value of the wind volume demand of each wind measurement point, and the random field model P(Y|X) of the wind volume of each detection point.

[0052] e. Introduce a fully connected neural network to calculate the air supply volume of the main ventilator. F5 uses a fully connected neural network to fit the relationship between the required air volume at each point and the air supply volume D of the main ventilator. Thus, given the real-time data of each air measurement point, the air supply volume D of the main ventilator can be calculated.

[0053] Beneficial effects: This invention is based on random fields, with each key ventilation detection point in the mine serving as the vertex of the random field. The results of the ventilation detection are used as the attributes of the vertex to establish a random field model for underground ventilation control. Then, a prediction model for the field data is established to predict the future information of the wind field, thereby guiding the ventilation adjustment of the main ventilation fan and effectively improving the response speed of ventilation control. Attached Figure Description

[0054] Fig. 1 A schematic diagram of the distribution of downhole ventilation and wind measurement points according to the present invention is shown;

[0055] Fig. 2 A schematic diagram of the random field model of the present invention is shown;

[0056] Fig. 3 A schematic diagram of the wind measurement point graph neural network structure of the present invention is shown. Detailed Implementation

[0057] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0058] like Figs. 1-3 The method for speed control of a frequency converter for a main ventilation fan in a coal mine based on random wind speed field prediction includes the following steps:

[0059] S1. Determine the wind measurement points and collect their data. Wind measurement points 1 and 2 will monitor the ventilation of the two longwall faces in real time. Wind measurement point 3 will monitor the ventilation of the tunnel face in real time. Wind measurement point 4 will monitor the ventilation of the ventilation shaft in real time. Collect the wind speed V, air volume C, temperature T, gas concentration W, and carbon dioxide concentration G data of each wind measurement point, and set the detection point number as i. The detection results at time t are: Vi(t), Ci(t), Ti(t), Wi(t), Gi(t), i = 1, 2, 3, 4.

[0060] S2. Predict the monitoring information of each wind measurement point. Let the current time be t, and the input information of the i-th detection point be:

[0061] X i (t)=[V i (t), C i (t), T i (t), W i (t), G i (t)] (1)

[0062] The specific steps are as follows:

[0063] 21) The prediction model uses the LSTM architecture of PyTorch, with an input sequence length of n=20 and a prediction length of m;

[0064]

[0065] 22) The LSTM settings are as follows: input length = 20, input dimension = 5, number of LSTM layers = 1, output length m, output dimension = 5.

[0066] The output length is the predicted length, reflecting the time-delay characteristics of ventilation regulation. The calculation method for m is as follows:

[0067]

[0068] In the formula, T f The sampling time can be set; T l The time it takes for the polluted air to travel from the furthest wind measurement point underground to the main ventilation fan is measured on-site.

[0069] 3) Establish the above LSTM neural network for all wind measurement points to predict the monitoring data of each wind measurement point;

[0070] 4) Training the LSTM neural network involves collecting historical data, creating data samples, and training the network. The loss function is the mean squared error function.

[0071] S3. Calculation of main ventilator air volume based on graph neural network: A random field model P(Y|X) for the air measurement points is established using graph neural network, where X represents the input information of all air measurement points; Y represents the air volume of each air measurement point, and the air volume Y takes a value between 0 and 1; when it is 0, it means no air volume is needed, and when it is 1, it means the maximum air volume is needed. The random field model P(Y|X) essentially represents a probability value, which can be approximated by graph neural network.

[0072] Therefore, the following steps are established:

[0073] 31) Establish the input information X = [X1, X2, X3, X4] for the graph neural network, which are the real-time detection values ​​of each wind measurement point, where X... i =[V i (t), C i (t), T i (t), W i (t), G i (t)] (3)

[0074] 32) Establish the output information of the graph neural network, Y = [Y1, Y2, Y3, Y4], which represents the air volume requirement of each wind measurement point;

[0075] 33) Establish a random field model P(Y|X) for the air volume at each monitoring point, that is, the required air volume at each location under the monitoring data conditions at each wind measurement point, 0≤P(Y|X). i |X)≤1, where 0 represents no air volume required and 1 represents the maximum air volume. This model is approximated using a graph neural network.

[0076] 34) Based on the random field model, a graph neural network is established for each wind measurement point. The graph neural network consists of a 4-layer graph network structure; the graph neural network is divided into four layers, and the input value of each layer is the feature value F. i The output eigenvalues ​​F are obtained by performing operations on the link matrix E, the input values, and the coal seam weights W. i+1 The link matrix E remains constant, i = 1, 2, 3, 4; the input feature values ​​F of each graph neural network layer. i Weight settings, feature output value F i+1 for:

[0077]

[0078] Where p×q represents a matrix with p rows and q columns of eigenvalues;

[0079] Each layer of the graph convolutional neural network F iThe specific calculation method is as follows:

[0080] a. The first layer input F1 is the feature matrix of each wind measurement point:

[0081]

[0082] b. Determine the link matrix E, where E is the link matrix for each wind measurement point:

[0083]

[0084] If the element in the i-th row and j-th column of matrix E is 1, it means that the i-th wind measurement point and the j-th wind measurement point are related; if it is 0, it means that they are not related. In this patent, all wind measurement points are related.

[0085] c. The feature values ​​output of each layer of the graph neural network are:

[0086] F i+1 =EF i W i (6)

[0087] d. According to the above graph neural network, the output F5 is actually the probability value of the wind volume demand of each wind measurement point, and the random field model P(Y|X) of the wind volume of each detection point.

[0088] e. Introduce a fully connected neural network to calculate the air supply volume of the main ventilator. F5 uses a fully connected neural network to fit the relationship between the required air volume at each point and the air supply volume D of the main ventilator. Thus, given the real-time data of each air measurement point, the air supply volume D of the main ventilator can be calculated.

[0089] 35) Based on the probability values ​​of air volume demand at each wind measurement point obtained by the graph neural network, and by introducing a fully connected neural network, the air supply volume D of the main ventilation fan is calculated.

[0090] 36) To train the graph neural network, collect real-time monitoring data from downhole wind measurement points, create a sample set, and use the mean squared error loss function during training until the neural network converges. Then the neural network will have the ability to calculate real-time air volume.

[0091] S4. Determine the decision-making method for the main ventilation fan inverter control: The inverter control frequency f is linearly related to the air volume.

[0092] Sexual relationship: f(t)=kD(t) (7)

[0093] In the formula, k is a constant. It can be tested directly on-site. When the inverter control frequency is 50Hz, the air volume is at its maximum (D). max ;

[0094] S5. Determine the inverter control frequency decision method, the steps of which are as follows:

[0095] 51) Set the current time as t, and determine the prediction length m according to equation (2);

[0096] 52) Based on the LSTM prediction, the values ​​at time m of each wind measurement point are:

[0097] X i (t+m)=[V i (t+m), C i (t+m), T i (t+m), W i (t+m), G i (t+m)] (8)

[0098] 53) Determine the air volume D(m) at time m using a graph neural network.

[0099] 54) Determine the inverter control frequency f(m) according to equation (7);

[0100] 55) From the current time t until time t+m, the frequency converter executes the control frequency f(m).

[0101] 56) Repeat step one at time t+m to re-predict and re-execute the control frequency.

[0102] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for speed regulation of a frequency converter for a main ventilation fan in a coal mine based on random wind speed field prediction, characterized in that, The steps include: S1. Determine the wind measurement points and collect their data. Wind measurement points 1 and 2 will monitor the ventilation of the two longwall faces in real time. Wind measurement point 3 will monitor the ventilation of the tunnel face in real time. Wind measurement point 4 will monitor the ventilation of the ventilation shaft in real time. Collect the wind speed V, air volume C, temperature T, gas concentration W, and carbon dioxide concentration G data of each wind measurement point, and set the detection point number as i. The detection results at time t are: Vi(t), Ci(t), Ti(t), Wi(t), Gi(t), i=1,2,3,4. S2. Predict the monitoring information of each wind measurement point. Let the current time be t, and the input information of the i-th detection point be: （1） The specific steps are as follows: 21) The prediction model uses the LSTM architecture of PyTorch, with an input sequence length of n=20 and a prediction length of m; 22) The LSTM settings are as follows: input length = 20, input dimension = 5, number of LSTM layers = 1, output length m, output dimension = 5. The output length is the predicted length, reflecting the time-delay characteristics of ventilation regulation. The calculation method for m is as follows: （2） In the formula, The sampling time can be set. The time it takes for the polluted air to travel from the furthest wind measurement point underground to the main ventilation fan is measured on-site. 3) Establish the above LSTM neural network for all wind measurement points to predict the monitoring data of each wind measurement point; 4) Training the LSTM neural network involves collecting historical data, creating data samples, and training the network. The loss function is the mean squared error function. S3. Calculation of main fan air volume based on graph neural network, and establishment of random field model of air measurement point using graph neural network. Where X represents the input information of all wind measurement points; Y represents the air volume at each wind measurement point, with the air volume Y ranging from 0 to 1; when it is 0, it means no air volume is needed, and when it is 1, it means the maximum air volume is needed. (Random field model) Essentially, it represents a probability value, which can be approximated using a graph neural network; Therefore, the following steps are established: 31) Establishing the input information for the graph neural network , which are the real-time measured values ​​of each wind measurement point, among which, （3） 32) Establish the output information of the graph neural network. This is to determine the air volume requirements of each wind measurement point; 33) Establish a random field model for the air volume at each detection point. That is, the required air volume at each location under the monitoring data conditions of each wind measurement point. Where 0 represents no airflow required and 1 represents the maximum airflow, the model is approximated using a graph neural network; 34) Based on the random field model, a graph neural network is established for each wind measurement point. The graph neural network consists of a 4-layer graph network structure. 35) Based on the probability values ​​of air volume demand at each wind measurement point obtained by the graph neural network, and by introducing a fully connected neural network, the air supply volume D of the main ventilation fan is calculated. 36) To train the graph neural network, collect real-time monitoring data from downhole wind measurement points, create a sample set, and use the mean squared error loss function during training until the neural network converges. Then the neural network will have the ability to calculate real-time air volume. S4. Determine the decision-making method for the main ventilation fan inverter control: Inverter control frequency It has a linear relationship with air volume: （7） In the formula It is a constant. It can be tested directly on-site. When the inverter control frequency is 50Hz, the air volume is at its maximum. ; S5. Determine the inverter control frequency decision method, the steps of which are as follows: 51) Set the current time as t, and determine the prediction length m according to equation (2); 52) Based on the LSTM prediction, the values ​​at time m of each wind measurement point are: （8） 53) Determine the air volume D(m) at time m using a graph neural network. 54) Determine the inverter control frequency according to equation (7). (m); 55) From the current time t until time t+m, the frequency converter executes the control frequency. (m), 56) Repeat step one at time t + m to re-predict and re-execute the control frequency.

2. The method for speed regulation of coal mine main ventilation fan frequency converter based on random wind speed field prediction according to claim 1, characterized in that, In step S3, the graph neural network is divided into four layers, and the input value of each layer is the feature value F. i The output eigenvalues ​​F are obtained by performing operations on the link matrix E, the input values, and the coal seam weights W. i+1 The link matrix E remains constant, i=1,2,3,4; the input feature values ​​F of each graph neural network layer. i Weight settings, feature output value F i+1 for: in This represents a matrix with p rows and q columns of eigenvalues; Each layer of graph convolutional neural network The specific calculation method is as follows: a. The first layer input F1 is the feature matrix of each wind measurement point: （4） b. Determine the link matrix E, where E is the link matrix for each wind measurement point. （5） If the element in the i-th row and j-th column of matrix E is 1, it means that the i-th wind measurement point and the j-th wind measurement point are related; if it is 0, it means that they are not related. In the above description, all wind measurement points are related. c. The feature values ​​output of each layer of the graph neural network are: （6） d. Based on the above graph neural network, the output F5 is essentially the probability value of the air volume demand at each wind measurement point, and the random field model of the air volume at each measurement point. ; e. Introduce a fully connected neural network to calculate the air supply volume of the main ventilator. F5 uses a fully connected neural network to fit the relationship between the required air volume at each point and the air supply volume D of the main ventilator. Thus, given the real-time data of each air measurement point, the air supply volume D of the main ventilator can be calculated.

Images