A method for predicting the posture of a flight of flight of a flight of a flight of a flight of a flight of a flight

Abstract

The application discloses a kind of scraper conveyor posture prediction method, belong to equipment monitoring technical field, method includes: S1, acquisition multi-source data, including equipment operating state data, support posture data and equipment physical structure parameters;S2, to the equipment operating state data, support posture data of acquisition time synchronization, and then calculate the cutting bottom depth of support chute and the advance state vector of push support;S3, according to the cutting bottom depth and advance state vector of calculation, through the first sub-network extraction posture data and the time sequence feature of support posture data when coal winning machine cuts bottom;Through second sub-network extraction equipment physical structure parameters structure feature;S4, the time sequence feature and structure feature extracted are spliced, and the posture prediction value of scraper conveyor is output by posture prediction model.This method solves the problems of low prediction accuracy, poor real-time performance and inability to dynamically optimize in the prior art by fusing multi-source heterogeneous data and constructing a deep learning prediction model.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of equipment monitoring, and particularly relates to a scraper conveyor posture prediction method. BACKGROUND

[0002] In the process of comprehensive mechanized mining, the scraper conveyor, as the key equipment for coal transportation, directly affects the transportation efficiency and equipment safety. At present, the posture monitoring method of the scraper conveyor includes:

[0003] (1) Based on the spline interpolation method and the Kalman filtering algorithm, the real-time position and posture of the scraper conveyor are accurately simulated, and the reliability of unmanned mining is improved.

[0004] (2) By obtaining the shape of the scraper conveyor and the structural parameters of the coal mining machine, the pitch angle obtained by the strapdown inertial navigation system and the double-axis inclination sensor during the actual operation of the coal mining machine is reversely mapped to the distance walked by the coal mining machine on the scraper conveyor at a certain moment, so as to position the coal mining machine. The inclination sensor needs to be installed on the chute of the scraper conveyor, and this method is less used in actual application. At present, the inclination sensor is mainly installed on the support, and this method also has limitations.

[0005] (3) The position of the scraper conveyor is accurately identified by calculating the stroke value of the hydraulic support and the slope difference of the curved section, and the positioning accuracy is improved. This method is simple and easy to use, but the precision is greatly affected by the terrain fluctuation.

[0006] The above-mentioned existing methods have the disadvantages of poor adaptability, difficult implementation, lack of verification and feedback mechanism for prediction results, and inability to dynamically adjust the prediction model according to the actual working conditions, resulting in continuous accumulation of prediction errors after long-term operation. SUMMARY

[0007] In view of the above-mentioned deficiencies in the prior art, the scraper conveyor posture prediction method provided by the present application solves the problems of poor adaptability, difficult implementation, lack of verification and feedback mechanism for prediction results, and inability to dynamically adjust the prediction process according to the actual working conditions, resulting in continuous accumulation of prediction errors after long-term operation.

[0008] In order to achieve the above-mentioned application purposes, the technical scheme adopted by the present application is as follows: a scraper conveyor posture prediction method, comprising the following steps:

[0009] S1, collecting multi-source data, including equipment running state data, support posture data and equipment physical structure parameters;

[0010] S2, time synchronizing the collected equipment running state data and support posture data, and then calculating the cutting bottom depth of the support chute and the advance state vector of the push chute support;

[0011] S3. Based on the calculated cutting depth and the advance state vector, the first sub-network is used to extract the time series features of the shearer's cutting posture data and the support posture data, including the posture angle change characteristics when the support is pushed, the correlation characteristics between the push rod length and the cutting depth, and the influence characteristics of the adjacent supports on the current support.

[0012] extracting structural features of physical structural parameters of the device through the second sub-network;

[0013] S4. The extracted time series features and structural features are spliced ​​together, and the attitude prediction value of the scraper conveyor is output through the attitude prediction model.

[0014] Furthermore, in step S2, the time synchronization processing of the equipment operation status data and the bracket posture data is performed, including:

[0015] S201, performing clock synchronization on a data acquisition device for collecting device operation status data based on the NTP protocol;

[0016] S202: Based on clock synchronization, the trigger position of the support pushing and sliding action is used as a reference to align the timestamps of the posture data of each support, and align the spatial positions of the shearer cutting data corresponding to the support;

[0017] S203 : Based on the alignment of the timestamp and the spatial position, unified resampling is performed to obtain time-synchronized device operation status data and bracket posture data.

[0018] Furthermore, the step S202 is specifically as follows:

[0019] When the bracket completes the pulling action and triggers the completion flag, the current global timestamp is recorded and bracket position , and when the push-slide action flag is ready to be triggered, the time for preparing the push-slide work is recorded as ;

[0020] Data points of any bracket attitude sensor , take the time lie in and The data values ​​between them are used, and the posture data of the bracket is obtained through median filtering to complete the timestamp alignment of the posture data of each bracket and the spatial position alignment of the coal cutting machine bottom data; wherein, Indicates the corrected timestamp, Indicates the data value collected by the attitude sensor.

[0021] Furthermore, in step S2, the cutting depth of the support chute is for:

[0022]

[0023] wherein, represents the vertical height of the rocker arm end, , represents the cutting radius of the shearer drum, represents the height of the machine body connection point from the floor, represents the rocker arm length, represents the shearer machine body pitch angle, represents the inclination angle of the rocker arm relative to the machine body;

[0024] The method for determining the advance state vector of the pushover support is:

[0025] The advance state threshold is set ;

[0026] When , the advance state is left support advance;

[0027] When , the advance state is right support advance;

[0028] When , the advance state is no offset;

[0029] The determined advance state is one-hot encoded to obtain the advance state vector .

[0030] Further, in the step S3, the first sub-network is an LSTM model, the LSTM model extracts the time sequence features of the input data through a set number of neurons, and outputs the hidden state of the last time step;

[0031] The input data is , represents the number of samples, represents the time step, represents the features of each time step, including the attitude angle, the cutting bottom depth, the push rod length, the attitude angle of the left and right supports, the three dimensions of the advance state vector, and the position of the current support; the attitude angle includes the pitch angle, the roll angle and the heading angle.

[0032] Further, in the step S3, the second sub-network is a PSFC network, and the PSFC network includes a dimension expansion layer, a first full connection layer, a ReLU activation function, a second full connection layer and an output layer connected in turn;

[0033] The dimension expansion layer expands the dimension of the input data by constructing a derivative explicit coding of the implicit relationship of the physical parameters;

[0034] The input data of the PSFC network includes support width, chute length, maximum horizontal bending angle, maximum vertical bending angle and shearer shoe spacing;

[0035] After the dimension expansion of the dimension expansion layer, the data input to the first full connection layer further includes horizontal direction maximum offset, vertical direction maximum fluctuation and the number of chutes covered by the shoe.

[0036] Further, in the step S4, the attitude prediction model includes a third full connection layer and a fourth full connection layer connected in sequence.

[0037] Further, the loss function of the attitude prediction model is:

[0038]

[0039]

[0040] In the formula, represents a prediction error term, represents a physical constraint term, represents a total constraint weight, represents a position constraint term, represents an angle constraint term.

[0041] Further, the angle constraint term is:

[0042]

[0043]

[0044]

[0045] In the formula, represents a horizontal direction constraint term, represents a vertical direction constraint term, and respectively represent constraint weight coefficients of and , respectively, represents a maximum horizontal bending angle of the scraper conveyor, represents a maximum vertical bending angle of the scraper conveyor, represents a vertical angle difference of the scraper conveyor, represents a vertical angle difference of the scraper conveyor, subscript represents the chute sequence number of the scraper conveyor, represents the total number of chutes of the scraper conveyor.

[0046] Further, the position constraint term is:

[0047]

[0048]

[0049]

[0050] In the formula, represents a horizontal position constraint, represents a vertical position constraint, and respectively represent and constraint weight coefficients of and respectively represent a horizontal offset and a vertical offset of the scraper conveyor, and respectively represent a horizontal maximum offset and a vertical maximum offset of the scraper conveyor, and subscript represents a number of a push-off section of the scraper conveyor, represents a total number of push-off sections of the scraper conveyor.

[0051] The beneficial effects of the present application are:

[0052] (1) The scraper conveyor morphology is affected by nonlinear factors such as push-off timing, support cooperative action, and coal seam resistance. Traditional physical models are difficult to accurately depict (such as the friction at the hinge of the middle trough, and the coupling effect of multiple push-off supports). The present application can automatically learn the implicit mapping between multiple variables through data driving through the LSTM model, which is particularly suitable for processing the dynamic relationship between the push-off sequence of the support and the shape of the scraper.

[0053] (2) The scraper conveyor will be affected by the left and right supports when pushing off, and will be offset to the direction of the pushed-off support, thereby affecting the attitude change of the scraper conveyor. In the method of the present application, the relative positions of the push-off support and the adjacent support are quantified through left support advance, right support advance, and no offset, and the accuracy of obtaining the attitude change data of the scraper conveyor is improved by quantifying the physical characteristics.

[0054] (3) Based on the LSTM model for extracting the time sequence characteristics of the attitude data, the present application also designs a PSFC network to extract the structural characteristics of the physical structure parameters of the equipment, and splices the time sequence characteristics as the basis for subsequent attitude prediction. Compared with the traditional method of only extracting time sequence characteristics to predict attitude data, the physical structure characteristics of the equipment are extracted through a separate PSFC network, which improves the accuracy of the attitude prediction results while reducing the network complexity.

[0055] (4) The attitude prediction model of the scraper conveyor is trained by combining the loss function of physical constraint, so that the model can automatically adjust parameters according to actual errors, the method provided by the application limits the accuracy of long-term operation, so that the attitude prediction result conforms to the mechanical motion law, and the reliability of the method in actual production is enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The attitude prediction method flowchart of the scraper conveyor provided by the application is shown. DETAILED DESCRIPTION

[0057] The specific embodiments of the application are described below to facilitate those skilled in the art to understand the application, but it should be clear that the application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the application defined and determined by the appended claims, and all the application and creation utilizing the concept of the application are within the scope of protection.

[0058] The application embodiment provides a scraper conveyor attitude prediction method, as shown in Figure 1 The method comprises the following steps:

[0059] S1, collecting multi-source data, including device running state data, support attitude data and device physical structure parameters;

[0060] S2, time synchronizing the collected device running state data and support attitude data, and then calculating the cutting bottom depth of the support chute and the advance state vector of the pushing support;

[0061] S3, according to the calculated cutting bottom depth and advance state vector, extracting the time sequence characteristics of the attitude data and support attitude data when the coal mining machine cuts the bottom through the first sub-network, including the attitude angle change characteristics when the support is pushed, the correlation characteristics of the push rod length and the cutting bottom depth, and the influence characteristics of the adjacent support on the current support;

[0062] The structure characteristics of the device physical structure parameters are extracted through the second sub-network;

[0063] S4, splicing the extracted time sequence characteristics and structure characteristics, and outputting the attitude prediction value of the scraper conveyor through the attitude prediction model.

[0064] In step S1 of the embodiment of the present application, the collected device running state data includes shearer body inertial navigation data, inertial measurement unit data, shearer attitude data, support attitude data, support push rod stroke data, and the position of the support beside the push-and-pull support; wherein the shearer body inertial navigation data includes acceleration data and angular velocity data obtained through the inertial measurement unit, the shearer attitude data includes body attitude and left and right drum bottoming state data obtained in real time through the high-precision angle sensor, the support attitude data includes floor pitch angle, roll angle and heading angle data measured by the inclinometer, the support push rod stroke data is collected by the magnetostrictive displacement sensor, and the lead state of the left and right supports of the push-and-pull support is determined through the record of the push-and-pull action.

[0065] In step S1 of the embodiment of the present application, the collected support attitude data is the attitude data when the support is pushed and pulled for the previous N times, which is used to analyze the attitude change trend, and N = support length / single push-and-pull length.

[0066] In step S1 of the embodiment of the present application, the collected physical structure parameters include shearer size data (swing arm length and body size), support size data (width and floor size), and scraper conveyor size data (body height, push-and-pull section width and push-and-pull section gap).

[0067] In step S2 of the embodiment of the present application, a double-dimension alignment method based on support position and time stamp is adopted, and the trigger position of the support push-and-pull is used as the reference to realize multi-source data synchronization in combination with the NTP protocol.

[0068] Specifically, in step S2 of the embodiment, the time synchronization processing of the device running state data and the support attitude data is performed, including:

[0069] S201, synchronizing the clock of the data collection device for collecting the device running state data and the support attitude data based on the NTP protocol;

[0070] S202, on the basis of the clock synchronization, taking the trigger position of the support push-and-pull action as the reference, aligning the time stamp of each support attitude data, and aligning the spatial position of the support corresponding shearer cutting bottom data;

[0071] S203, on the basis of the time stamp and spatial position alignment, performing unified resampling to obtain the time-synchronized device running state data and support attitude data.

[0072] In step S201 of the embodiment, each data collection device is synchronized with the ground clock server through the NTP protocol to correct the local clock deviation; wherein the time stamp correction formula is:

[0073]

[0074] In the formula, represents the corrected global uniform timestamp, represents the device local original timestamp, represents the clock offset calculated by the NTP protocol , wherein is the request time sent by the client, is the server receiving time, is the server response time, is the client receiving response time.

[0075] The step S202 in the embodiment is specifically:

[0076] When the support completes the pull support action and triggers the completion action flag, the current global timestamp and the support position are recorded, and the time for preparing to trigger the push and pull action flag is recorded as ; wherein the support position is determined by the support number or absolute coordinates;

[0077] For the data points of any support attitude sensor , the time is located between the data values of and , and the attitude data of the support is obtained through median filtering to complete the timestamp alignment of each support attitude data and the spatial position alignment of the coal mining machine cutting bottom data; wherein represents the corrected timestamp, represents the data value collected by the attitude sensor.

[0078] In the embodiment, the cutting bottom data is calculated from the attitude data of the coal mining machine. When the front roller of the coal mining machine is cutting the bottom of the N support, the N-20 support may be preparing to push and pull.

[0079] In the embodiment, the spatial alignment of the attitude data of each support is mainly to align the attitude data of the support with the cutting bottom data of the roller of the coal mining machine. The position of the coal mining machine refers to the position of the center point of the coal mining machine, which corresponds to a support number. The support numbers corresponding to the rollers on the left and right sides of the coal mining machine are not the support numbers of the coal mining machine. The support numbers corresponding to the left and right rollers can be calculated from the support numbers of the coal mining machine, thereby realizing the spatial alignment of the attitude data of the support. For time alignment, generally, the cutting bottom time of the support is earlier than the pull support time of the support, and the push and pull time of the support. Since the push and pull calculation is required, the data needs to be aligned in time.

[0080] In step S2 in the embodiment of the present application, the cutting depth of the support chute is calculated according to the inclination angle of the shearer boom and the state of the machine body, in combination with the size data of the shearer, through a geometric algorithm; specifically, the cutting depth is defined as the vertical distance of the cutting point at the end of the shearer boom relative to the bottom surface of the scraper conveyor chute, and the cutting depth of the support chute is :

[0081]

[0082] wherein, the vertical height of the end of the boom, , the cutting radius of the shearer drum, the height of the machine body connection point from the floor, the length of the boom, the pitch angle of the shearer machine body, the inclination angle of the boom relative to the machine body.

[0083] In the embodiment, the relative position of the push chute and the adjacent chute is quantified by three states of “left chute leading / right chute leading / no offset”; specifically, the method for determining the leading state vector of the push chute is as follows:

[0084] The leading state threshold is set ;

[0085] When , the leading state is left chute leading;

[0086] When , the leading state is right chute leading;

[0087] When , the leading state is no offset;

[0088] The determined leading state is one-hot encoded to obtain the leading state vector .

[0089] In step S3 in the embodiment of the present application, the first sub-network is an LSTM model, which extracts the time sequence features of the input data through a set number of neurons and outputs the hidden state of the last time step;

[0090] Specifically, the LSTM model in the embodiment is set with 1 layer of hidden layer, 64 neurons per layer, and the input data is , wherein, the number of samples, represents the features of each time step, including the attitude angle of the current support, the cutting bottom depth, the push rod length, the attitude angle of the left and right supports, the three dimensions of the state vector and the position of the advance; the attitude angle includes the pitch angle, the roll angle and the heading angle.

[0091] In this embodiment, taking 3 as an example, the LSTM model outputs the hidden state of the last time step (shape (N, 64)), and the time sequence features of the 3 times of pushing and pulling data are extracted, including the attitude angle change feature of the support pushing and pulling, the correlation feature of the push rod length and the cutting bottom depth, and the influence feature of the side support on the current support.

[0092] In this embodiment, the LSTM model solves the gradient disappearance and gradient explosion problems of the traditional recurrent neural network (RNN) through the gating mechanism, the forgetting gate determines which information to discard, the input gate controls the addition of new information, the output gate determines the output content, and the cell state is responsible for long-term information transmission, which can effectively process the time sequence features of the coal mining machine cutting bottom data and the historical attitude data of the support, and learn the change rule of the data over time.

[0093] In step S3 of the embodiment of the application, the second sub-network is a PSFC network, and the PSFC network comprises a dimension expansion layer, a first full connection layer, a ReLU activation function, a second full connection layer and an output layer connected in sequence.

[0094] The dimension expansion layer expands the dimension of the input data by constructing an implicit relationship derived from an explicit coding physical parameter, so that the PSFC network learns more rich structure information.

[0095] The input data of the PSFC network includes the support width, the length of the push section, the maximum horizontal bending angle, the maximum vertical bending angle and the distance between the sheaves of the coal mining machine.

[0096] After the dimension expansion of the dimension expansion layer, the data input to the first full connection layer further includes the maximum horizontal offset, the maximum vertical fluctuation and the number of push sections covered by the sheave, wherein the maximum horizontal offset is the maximum allowable misplacement of adjacent push sections in the horizontal direction, and is represented as The maximum vertical fluctuation is the maximum allowable fluctuation of adjacent push sections in the vertical direction, and is represented as The number of push sections covered by the sheave is the number of middle sections affected by the coal mining machine .

[0097] In this embodiment, in the PSFC network, the first full connection layer maps the 8-dimensional input to 16-dimensional, and preliminarily extracts the linear / nonlinear relationship between the parameters; the ReLU activation function introduces nonlinearity to avoid the pure linear transformation that cannot model complex physical constraints; the second full connection layer maps the 16-dimensional features to 32-dimensional, and outputs the final structure features.

[0098] In this embodiment, the PSFC network performs linear transformation and activation operations on the physical structure parameters, converting them into a representation consistent with the dimensions of other feature vectors, and then splicing them with the time series features output by the LSTM network, so that the subsequent posture prediction model can comprehensively consider the impact of the physical characteristics of the equipment on the scraper conveyor posture when predicting the posture data.

[0099] In step S4 of an embodiment of the present invention, the posture prediction model includes a third fully connected layer and a fourth fully connected layer connected in sequence, mapping the feature vector of the spliced ​​temporal features and structural features into a posture prediction value of the scraper conveyor, including pitch angle, roll angle, heading angle and offset.

[0100] In this embodiment of the present invention, a transfer learning algorithm is used to update the parameters of the posture prediction model based on the prediction error. The Adam optimizer (learning rate 0.001, batch size 64) is used for training for 100 epochs, and an early stopping strategy is applied to prevent overfitting.

[0101] In this embodiment, the physical constraint term is introduced to optimize the loss function, and the loss function of the posture prediction model is obtained. for:

[0102]

[0103]

[0104] Where, represents the forecast error term, represents the physical constraint term, represents the total constraint weight, represents the position constraint, Represents an angular constraint.

[0105] In this embodiment, when calculating the angle constraint term, it is assumed that the scraper conveyor consists of n sections, and the predicted posture parameters of each section include the horizontal angle (heading angle) , the vertical angle (pitch angle) is , the angle difference between adjacent sliding sections includes the horizontal angle difference and vertical angle difference ; Based on this, at the maximum horizontal bending angle and maximum vertical bending angle Based on the angle constraint for:

[0106]

[0107]

[0108]

[0109] In the formula, represents a horizontal direction constraint term, represents a vertical direction constraint term, and respectively represent and constraint weight coefficients of represents a maximum horizontal bending angle of the scraper conveyor, represents a maximum vertical bending angle of the scraper conveyor, represents a vertical angle difference of the scraper conveyor, represents a vertical angle difference of the scraper conveyor, subscript represents a number of the flight of the scraper conveyor, represents a total number of flights of the scraper conveyor.

[0110] In the embodiment, when calculating the position constraint term, the three-dimensional coordinates of each flight of the scraper conveyor are set as , the horizontal position offset is defined as the projection distance difference of adjacent flights in the horizontal plane, the vertical displacement offset is defined as the projection distance difference of adjacent flights in the vertical plane, and then the displacement offset of adjacent flights includes the horizontal offset and the vertical offset ; according to the flight length L and the maximum bending angle , , the maximum offset allowed by the mechanical structure includes the horizontal maximum offset and the vertical maximum offset ; based on this, the position constraint term is:

[0111]

[0112]

[0113]

[0114] In the formula, represents a horizontal position constraint, represents a vertical position constraint, and respectively represent and constraint weight coefficients of and respectively represent the horizontal offset and the vertical offset of the scraper conveyor, and respectively represent the horizontal maximum offset and the vertical maximum offset of the scraper conveyor, subscript represents the serial number of the pan of the scraper conveyor, represents the total number of pans of the scraper conveyor.

[0115] In the loss function provided by the embodiment of the present application, the horizontal and vertical angle difference and the position difference between each pair of adjacent pans are calculated by designing a physical constraint term, and the max function is used to calculate the penalty value for the angle difference and the position exceeding the physical limit; on this basis, the penalty values of all adjacent pan pairs are summed to obtain the total physical constraint term, and the weight coefficient is used to control the strength of the physical constraint.

[0116] The loss function provided by the present application ensures that when the predicted angle difference and position difference of adjacent pans are within the physical allowable range, the constraint term is 0, which does not affect the training; when the difference exceeds the physical limit, the constraint term produces a penalty proportional to the amount of exceeding. At the same time, the importance of the prediction accuracy and the physical constraint can be balanced by adjusting the weight coefficient.

[0117] The principles and implementation manners of the present application are described in the specific embodiments in the present application, and the above embodiment descriptions are only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will be changed, and the above description should not be understood as the limitation of the present application.

[0118] Those skilled in the art will realize that the embodiments described herein are for the purpose of understanding the principles of the present application and should be understood as not limiting the scope of the present application to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspiration disclosed in the present application without departing from the essence of the present application, and these modifications and combinations are still within the scope of protection of the present application.

Claims

1. A method for predicting the posture of a scraper conveyor, characterized in that: The method comprises the following steps: S1, collecting multi-source data, including device running state data, support posture data and device physical structure parameters; S2, time synchronizing the collected device running state data and support posture data, and then calculating the cutting bottom depth of the support chute and the advance state vector of the pushing support; S3, according to the calculated cutting bottom depth and advance state vector, extracting the time sequence features of the posture data and support posture data when the coal mining machine cuts the bottom through the first sub-network, including the posture angle change features when the support is pushed, the correlation features of the push rod length and the cutting bottom depth, and the influence features of the adjacent support on the current support; extracting the structure features of the device physical structure parameters through the second sub-network; S4, concatenating the extracted time sequence features and structure features, and outputting the posture prediction value of the scraper conveyor through the posture prediction model.

2. The scraper conveyor pose prediction method of claim 1, wherein, In the step S2, the device running state data and the support posture data are time synchronized, including: S201, synchronizing the clock of the data collection device for collecting the device running state data based on the NTP protocol; S202, based on the clock synchronization, aligning the time stamps of the support posture data and the spatial positions of the corresponding coal mining machine cutting bottom data with the trigger position of the support pushing action as the reference; S203, based on the time stamp and spatial position alignment, uniformly resampling to obtain the time synchronized device running state data and support posture data.

3. The scraper conveyor pose prediction method of claim 2, wherein, The step S202 is specifically: When the bracket completes the pull bracket action and triggers the completion action flag, record the current global timestamp and the bracket position , and when preparing to trigger the push and pull action flag, record the time when the push and pull work is prepared ; Data points of any support attitude sensor , take time Data values between and , and obtain the attitude data of the support through median filtering, complete the timestamp alignment of each support attitude data and the spatial position alignment of the coal winning machine cutting bottom data; wherein, represents the corrected timestamp, represents the data value collected by the attitude sensor.

4. The scraper posture prediction method according to claim 1, characterized by, The cutting depth of the support chute in the step S2 is: : wherein represents the vertical height of the rocker arm tip, , represents the cutting radius of the shearer drum, represents the height of the machine body connection point from the floor, represents the length of the rocker arm, represents the pitch angle of the shearer machine body, represents the inclination angle of the rocker arm relative to the machine body; The method for determining the advance state vector of the pushing support is: Setting an advance state threshold ; When the lead state is a left frame lead; When the lead state is right frame lead; When the lead state is unbiased; The determined lookahead state is one-hot encoded to obtain an lookahead state vector .

5. The scraper conveyor pose prediction method of claim 4, wherein, In the step S3, the first sub-network is an LSTM model, which extracts the time sequence features of the input data through a certain number of neurons and outputs the hidden state of the last time step; The input data is , represents the number of samples, represents the time step, represents the features of each time step, including the attitude angle, the cutting depth, the push rod length, the attitude angle of the left and right supports, the three dimensions of the state vector and the position of the current support; the attitude angle includes the pitch angle, the roll angle and the heading angle.

6. The scraper conveyor pose prediction method of claim 1, wherein, In the step S3, the second sub-network is a PSFC network, which comprises a dimension expansion layer, a first full connection layer, a ReLU activation function, a second full connection layer and an output layer connected in sequence; The dimension expansion layer expands the dimension of the input data by constructing the implicit relationship of the derived explicit coding physical parameters; The input data of the PSFC network includes the support width, the length of the chute section, the maximum horizontal bending angle, the maximum vertical bending angle and the distance between the coal mining machine shoes; After the dimension expansion of the dimension expansion layer, the data input to the first full connection layer also includes the maximum horizontal offset, the maximum vertical fluctuation and the number of chute sections covered by the shoes.

7. The scraper conveyor pose prediction method of claim 1, wherein, In the step S4, the posture prediction model comprises a third full connection layer and a fourth full connection layer connected in sequence.

8. The scraper conveyor pose prediction method of claim 1, wherein, Loss function of the pose prediction model is: wherein denotes a prediction error term, denotes a physical constraint term, denotes a total constraint weight, denotes a position constraint term, denotes an angle constraint term.

9. The scraper conveyor pose prediction method of claim 8, wherein, The angle constraint term is: wherein, represents a horizontal direction constraint term, represents a vertical direction constraint term, and respectively represent and a constraint weight coefficient of represents a maximum horizontal bending angle of the scraper conveyor, represents a maximum vertical bending angle of the scraper conveyor, represents a vertical angle difference of the scraper conveyor, represents a vertical angle difference of the scraper conveyor, subscript represents a number of the flight of the scraper conveyor, represents a total number of the flights of the scraper conveyor.

10. The scraper conveyor pose prediction method of claim 8, wherein, The position constraint term is: In the formula, represents a horizontal position constraint, represents a vertical position constraint, and respectively represent and a constraint weight coefficient of and respectively represent a horizontal offset and a vertical offset of the scraper conveyor, and respectively represent a horizontal maximum offset and a vertical maximum offset of the scraper conveyor, and subscript represents a number of the chain flight of the scraper conveyor, represents a total number of the chain flights of the scraper conveyor.

Images