A method and system for intelligent cutting control of working face

Abstract

This invention discloses an intelligent cutting control method and system for working faces, relating to the field of intelligent control technology. The method includes the following steps: acquiring historical data from multi-source sensors of support equipment, the cutting host, and the conveying equipment; generating a multi-source fusion data training set; constructing an intelligent cutting parameter prediction model for the working face based on a deep learning model, and training the intelligent cutting parameter prediction model based on the multi-source fusion data training set; generating multi-source fusion data; inputting the multi-source fusion data into the trained intelligent cutting parameter prediction model to obtain predicted cutting parameters, generating cutting control commands based on the predicted cutting parameters, and constructing a confidence evaluation mechanism to quantitatively evaluate the reliability of the predicted cutting parameters, and deciding whether to execute the cutting control commands based on the quantitative evaluation results. This invention achieves intelligent and adaptive control of the underground mining cutting process, significantly improving operational efficiency, equipment stability, and system safety.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to an intelligent cutting control method and system for working surfaces. Background Technology

[0002] In current underground mining operations, the cutting process of the working face still heavily relies on the experience and manual intervention of operators. While existing technologies have partially achieved functions such as automatic height adjustment of the cutting host and automatic relocation of support equipment, they generally lack dynamic collaborative control and global optimization capabilities for the entire cutting process. Traditional control methods often employ fixed rules or PID feedback control strategies, which are ill-suited to adapting to changes in rock strata and fluctuations in equipment status under complex geological conditions.

[0003] Furthermore, existing technologies have not effectively integrated heterogeneous data from multiple sources, such as the movements of support equipment, operating parameters of the cutting host, and load of conveying equipment, nor have they attempted to construct a behavioral logic model for manual cutting and perform self-learning optimization. This results in limited intelligence levels in existing technologies, hindering improvements in operational efficiency and safety. Therefore, there is an urgent need in this field for a novel intelligent cutting control method with deep perception, autonomous decision-making, and continuous evolution capabilities. Summary of the Invention

[0004] To address the aforementioned shortcomings in the existing technology, this invention provides an intelligent cutting control method and system for the working surface.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A method for intelligent cutting control of a working surface is provided, comprising the following steps:

[0007] Acquire historical data from multiple sources of sensors in support equipment, cutting host, and conveying equipment;

[0008] The historical data from multiple sensors are processed by timestamp alignment, data cleaning, feature extraction and data standardization to generate a multi-source fusion data training set.

[0009] A working face intelligent cutting parameter prediction model is constructed based on a deep learning model, and the working face intelligent cutting parameter prediction model is trained based on a multi-source fusion data training set;

[0010] Collect real-time data from multiple sources of sensors in support equipment, cutting host and conveying equipment, and perform timestamp alignment, data cleaning, feature extraction and data standardization on the real-time data of multiple sources of sensors to generate multi-source fusion data;

[0011] Multi-source fusion data is input into the trained intelligent cutting parameter prediction model of the working face to obtain predicted cutting parameters. Cutting control commands are generated based on the predicted cutting parameters. A confidence evaluation mechanism is constructed to quantitatively evaluate the reliability of the predicted cutting parameters. Based on the quantitative evaluation results, a decision is made on whether to execute the cutting control commands.

[0012] Furthermore, the historical data from multiple sensors is timestamped and aligned using the following expression:

[0013]

[0014] in: For timestamp alignment in time Historical data from multiple sensors at the location In time Historical data from multiple sensors at the location In time Historical data from multiple sensors at the location Used as a time number.

[0015] Furthermore, feature extraction includes calculating the operating frequency of the support equipment, the average operating time of the support equipment, the proportion of operating types of the support equipment, the speed change rate of the cutting host, the average load current of the cutting host, the change rate of the height difference between the left and right rollers of the cutting host, the speed change rate of the conveying equipment, the average current of the conveying equipment, and the frequency change rate of the conveying equipment.

[0016] Furthermore, the intelligent cutting parameter prediction model for the working face includes a multi-scale feature extraction layer, a device collaborative attention fusion layer, a temporal coding layer, a global attention decision layer, and a fully connected layer connected in sequence.

[0017] The multi-scale feature extraction layer includes three independent multi-scale convolutional branches, which process the data of the support equipment, the cutting host, and the conveying equipment in parallel. Each multi-scale convolutional branch contains multiple convolutional channels with different kernel sizes, which are used to capture the equipment operation modes under different time windows through convolution operations to obtain the multi-scale features of the support equipment, the cutting host, and the conveying equipment respectively. The multi-scale features of the support equipment, the cutting host, and the conveying equipment are fused through adaptive weighting operations to obtain the multi-scale fused features of the support equipment, the cutting host, and the conveying equipment respectively.

[0018] The equipment collaborative attention fusion layer is used to generate the final fusion collaborative information enhancement features by calculating the interactive attention weights between the support equipment, the cutting host, and the conveying equipment based on the multi-scale fusion characteristics of the support equipment, the cutting host, and the conveying equipment.

[0019] The temporal coding layer is used to perform temporal modeling of the final fused collaborative information enhancement features through a bidirectional long short-term memory network to obtain bidirectional hidden state features at all time steps;

[0020] The global attention decision layer is used to extract temporal information related to the current cutting decision from the bidirectional hidden state features at all time steps to obtain a globally weighted context vector;

[0021] Fully connected layers are used to map the global weighted context vector to predicted cut parameters.

[0022] Furthermore, the specific processing procedure of the device collaborative attention fusion layer is as follows:

[0023] The interaction attention score between the support equipment, the cutting host, and the conveying equipment is calculated using the following expression:

[0024]

[0025] in: For the first The device for the first Attention score for each device For the first Multi-scale fusion features of individual devices For the projection matrix of the query, For the first Multi-scale fusion features of individual devices Let be the projection matrix of the key. This is the matrix transpose operator. For the first The dimensions of multi-scale fusion features of each device;

[0026] The calculated interaction attention scores among the support equipment, cutting host, and conveying equipment are normalized to obtain the interaction attention weights among them, expressed as follows:

[0027]

[0028] in: For the first The device and the first Interaction attention weights between devices It is the symbol for an exponential function. For the first The device for the first Attention score for each device;

[0029] Based on the interaction attention weights among the support equipment, cutting host, and conveying equipment, a fusion collaborative information enhancement feature for the support equipment, cutting host, and conveying equipment is generated, and its expression is as follows:

[0030]

[0031] in: For the first Enhanced features of integrated and collaborative information from individual devices The projection matrix of the value;

[0032] The fusion and collaboration information enhancement features of support equipment, cutting host and conveying equipment are spliced ​​and fused to generate the final fusion and collaboration information enhancement features.

[0033] Furthermore, the intelligent cutting parameter prediction model for the working face is trained based on the multi-source fusion data training set. The specific process is as follows:

[0034] A loss function for predicting intelligent cutting parameters of the working face is constructed by combining cross-entropy loss and cutting process optimization loss;

[0035] Based on a multi-source fusion data training set, the intelligent cutting parameter prediction model for the working face is trained using the loss function of the model.

[0036] Furthermore, the loss function of the intelligent cutting parameter prediction model for the working face is expressed as follows:

[0037] ,

[0038]

[0039] in: The loss function value for the intelligent cutting parameter prediction model of the working face. The value of the cross-entropy loss function. For balance coefficient, To optimize the loss function value for the cutting process, This is a weighting coefficient for the cutting efficiency loss. To reduce cutting efficiency, This is the weighting coefficient for equipment stability loss. For equipment stability loss, This is the weighting coefficient for energy consumption loss. This refers to energy consumption losses.

[0040] A working face intelligent cutting control system applied to the above method includes a data acquisition module, a data preprocessing module, a large model training module, and an intelligent inference module;

[0041] The data acquisition module is used to acquire historical data from multi-source sensors of support equipment, cutting host and conveying equipment, collect real-time data from multi-source sensors of support equipment, cutting host and conveying equipment, and transmit the historical data and real-time data of multi-source sensors to the data preprocessing module.

[0042] The data preprocessing module is used to perform timestamp alignment, data cleaning, feature extraction and data standardization on historical data from multiple sources sensors to generate a multi-source fusion data training set, and then transmit the multi-source fusion data training set to the large model training module. The module performs timestamp alignment, data cleaning, feature extraction and data standardization on real-time data from multiple sources sensors to generate multi-source fusion data, and then transmits the multi-source fusion data to the intelligent inference module.

[0043] The large model training module is used to build a prediction model for intelligent cutting parameters of the working face based on the deep learning model, and to train the prediction model for intelligent cutting parameters of the working face based on the multi-source fusion data training set.

[0044] The intelligent inference module is used to input multi-source fused data into the trained intelligent cutting parameter prediction model of the working face to obtain predicted cutting parameters, generate cutting control commands based on the predicted cutting parameters, and build a confidence evaluation mechanism to quantitatively evaluate the reliability of the predicted cutting parameters, and decide whether to execute the cutting control commands based on the quantitative evaluation results.

[0045] The beneficial effects of this invention are as follows:

[0046] (1) This invention achieves global perception and collaborative control of the entire working face production system by deeply integrating the data of support equipment, cutting host and conveying equipment, and capturing their complex spatiotemporal correlation by the intelligent cutting parameter prediction model of the working face. This makes the method provided by this invention make overall decisions from a global perspective rather than isolated single-machine control, thereby significantly improving the overall operation efficiency and resource recovery rate.

[0047] (2) This invention provides an intelligent cutting control system for a working face, including a data acquisition module, a data preprocessing module, a large model training module, and an intelligent inference module. The data acquisition module can acquire historical data from multiple sources of sensors of the support equipment, the cutting host, and the conveying equipment, and acquire real-time data from multiple sources of sensors of the support equipment, the cutting host, and the conveying equipment. It also transmits the historical data and real-time data of the multiple sources of sensors to the data preprocessing module. The data preprocessing module can perform timestamp alignment, data cleaning, feature extraction, and data standardization on the historical data of the multiple sources of sensors, generate a multi-source fusion data training set, and transmit the multi-source fusion data training set to the large model training module for multi-source sensor data training. Real-time data undergoes timestamp alignment, data cleaning, feature extraction, and data standardization to generate multi-source fused data, which is then transmitted to the intelligent inference module. The large model training module is used to construct a working face intelligent cutting parameter prediction model based on a deep learning model and train the model on the multi-source fused data training set. The intelligent inference module inputs the multi-source fused data into the trained working face intelligent cutting parameter prediction model to obtain predicted cutting parameters, generates cutting control commands based on the predicted cutting parameters, and constructs a confidence evaluation mechanism to quantitatively evaluate the reliability of the predicted cutting parameters. Based on the quantitative evaluation results, it decides whether to execute the cutting control commands. Attached Figure Description

[0048] Figure 1 A schematic diagram of a working face intelligent cutting control method;

[0049] Figure 2 This is a schematic diagram of a working face intelligent cutting control system. Detailed Implementation

[0050] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0051] like Figure 1 As shown, a method for intelligent cutting control of a working face includes steps S1-S5, as detailed below:

[0052] S1. Acquire historical data from multiple sensors of the support equipment, cutting host, and conveying equipment.

[0053] In an optional embodiment of the present invention, the present invention acquires historical data from multiple sources of sensors on the support equipment, cutting host, and conveying equipment from the working face through a data acquisition module. In a typical application scenario of the present invention, the support equipment is specifically a hydraulic support, the cutting host is specifically a coal mining machine, and the conveying equipment is specifically a scraper conveyor. Specifically, the present invention needs to acquire historical data from multiple sources of sensors through a sensor network deployed on each piece of equipment: for the hydraulic support, its action timing data is acquired through displacement sensors; for the coal mining machine, its operating speed and load current are acquired through speed sensors and current sensors, and its height difference is acquired through height sensor data of the left and right drums; for the scraper conveyor, corresponding data is acquired through speed sensors, current sensors, and frequency sensors.

[0054] S2. Perform timestamp alignment, data cleaning, feature extraction, and data standardization on historical data from multiple sensors to generate a multi-source fusion data training set.

[0055] In an optional embodiment of the present invention, the present invention uses a data preprocessing module to perform timestamp alignment, data cleaning, feature extraction, and data standardization on historical data from multiple sources, generating a multi-source fusion data training set. In the data preprocessing module, the historical data from multiple sources is first timestamp aligned to synchronize all historical data to a unified timeline, as expressed by:

[0056]

[0057] in: For timestamp alignment in time Historical data from multiple sensors at the location In time Historical data from multiple sensors at the location In time Historical data from multiple sensors at the location Used as a time number.

[0058] The specific data cleaning process is as follows: First, based on the physical limits of the equipment, a threshold range is set to directly remove distorted outliers; second, the Z-score statistical method is used to identify and process hidden outlier data points that deviate from the normal distribution, thereby ensuring the quality and consistency of historical data.

[0059] Feature extraction includes calculating the action frequency of the support equipment, the average action duration of the support equipment, the proportion of action types of the support equipment, the speed change rate of the cutting host, the average load current of the cutting host, the change rate of the height difference between the left and right rollers of the cutting host, the speed change rate of the conveying equipment, the average current of the conveying equipment, and the frequency change rate of the conveying equipment.

[0060] Specifically, this invention performs data standardization processing on historical data from multi-source sensors through Min-Max normalization to eliminate the influence of dimensions.

[0061] S3. Construct a working face intelligent cutting parameter prediction model based on a deep learning model, and train the working face intelligent cutting parameter prediction model based on a multi-source fusion data training set.

[0062] In an optional embodiment of the present invention, the present invention receives a multi-source fused data training set through a large model training module and inputs it into the intelligent cutting parameter prediction model for the working face for training. During training, the parameters of the intelligent cutting parameter prediction model for the working face are updated by minimizing the loss function of the intelligent cutting parameter prediction model for the working face to obtain the trained intelligent cutting parameter prediction model for the working face.

[0063] The intelligent cutting parameter prediction model for the working face includes a multi-scale feature extraction layer, a device collaborative attention fusion layer, a temporal coding layer, a global attention decision layer, and a fully connected layer connected in sequence.

[0064] The multi-scale feature extraction layer includes three independent multi-scale convolutional branches, which process the data of the support equipment, the cutting host, and the conveying equipment in parallel. Each multi-scale convolutional branch contains multiple convolutional channels with different kernel sizes, which are used to capture the equipment operation modes under different time windows through convolution operations to obtain the multi-scale features of the support equipment, the cutting host, and the conveying equipment respectively. The multi-scale features of the support equipment, the cutting host, and the conveying equipment are fused through adaptive weighting operations to obtain the multi-scale fused features of the support equipment, the cutting host, and the conveying equipment respectively.

[0065] The expression for the convolution operation is:

[0066]

[0067] in: For the first The device in the Feature maps under each convolutional channel For the equipment number, , This refers to the numbering of the convolution channels under the multi-scale convolution branch. This is a 1D convolution operation operator. For the first Input data from each device For the first The device in the Convolution weights under each convolution channel The kernel size is the convolution kernel size. For the first kernel size per convolution channel , For the first The device in the Bias term under each convolutional channel;

[0068] The expression for the adaptive weighting operation is:

[0069]

[0070] in: For the first Multi-scale fusion features of individual devices This refers to the number of convolution channels in the multi-scale convolution branch. For the first The device in the Adaptive weight coefficients under each convolutional channel;

[0071] The equipment collaborative attention fusion layer is used to generate the final fusion collaborative information enhancement features by calculating the interactive attention weights between the support equipment, the cutting host, and the conveying equipment based on the multi-scale fusion characteristics of the support equipment, the cutting host, and the conveying equipment.

[0072] The specific processing steps of the device collaborative attention fusion layer are as follows:

[0073] The interaction attention score between the support equipment, the cutting host, and the conveying equipment is calculated using the following expression:

[0074]

[0075] in: For the first The device for the first Attention score for each device The projection matrix for the query is obtained through training. For the first Multi-scale fusion features of individual devices The projection matrix of the key is obtained through training. This is the matrix transpose operator. For the first The dimensions of multi-scale fusion features of each device;

[0076] The calculated interaction attention scores among the support equipment, cutting host, and conveying equipment are normalized to obtain the interaction attention weights among them, expressed as follows:

[0077]

[0078] in: For the first The device and the first Interaction attention weights between devices It is the symbol for an exponential function. For the first The device for the first Attention score for each device;

[0079] Based on the interaction attention weights among the support equipment, cutting host, and conveying equipment, a fusion collaborative information enhancement feature for the support equipment, cutting host, and conveying equipment is generated, and its expression is as follows:

[0080]

[0081] in: For the first Enhanced features of integrated and collaborative information from individual devices The projection matrix with values ​​is obtained through training;

[0082] The fusion and collaboration information enhancement features of support equipment, cutting host and conveying equipment are spliced ​​and fused to generate the final fusion and collaboration information enhancement features.

[0083] The temporal coding layer is used to perform temporal modeling of the final fused collaborative information enhancement features through a bidirectional long short-term memory network to obtain bidirectional hidden state features at all time steps;

[0084] The global attention decision layer is used to extract temporal information related to the current cutting decision from the bidirectional hidden state features at all time steps to obtain a globally weighted context vector;

[0085] The data processing procedure for the global attention decision layer is as follows:

[0086] Based on the bidirectional hidden state features at all time steps, the attention score at all time steps is calculated, and its expression is:

[0087]

[0088] in: In time Attention score at the location, The first learnable parameter matrix, This is the matrix transpose operator. The symbol for the hyperbolic tangent function is... For the second learnable parameter matrix, In time The bidirectional hidden state features at the location, For the third learnable parameter matrix, The context vector is specifically a weighted sum of the bidirectional hidden state features across all time steps.

[0089] Softmax normalization is applied to the attention scores at all time steps to obtain the attention weights at all time steps;

[0090] The global weighted context vector is calculated based on the attention weights at all time steps, and its expression is as follows:

[0091]

[0092] in: For the globally weighted context vector, The total number of time steps. In time Attention weights at each location;

[0093] Fully connected layers are used to map the global weighted context vector to predicted cut parameters, and their expression is:

[0094]

[0095] in: To predict cutting parameters, The Softmax activation function is used. This is the weight matrix of the output layer. This is the bias term for the output layer.

[0096] This invention trains a predictive model for intelligent cutting parameters of the working face based on a multi-source fusion data training set. The specific process is as follows:

[0097] The loss function of the intelligent cutting parameter prediction model for the working face is expressed as follows:

[0098] ,

[0099]

[0100] in: The loss function value for the intelligent cutting parameter prediction model of the working face. The value of the cross-entropy loss function. For balance coefficient, To optimize the loss function value for the cutting process, This is a weighting coefficient for the cutting efficiency loss. To reduce cutting efficiency, This is the weighting coefficient for equipment stability loss. For equipment stability loss, This is the weighting coefficient for energy consumption loss. This refers to energy consumption losses.

[0101] The expression for the cutting efficiency loss is:

[0102]

[0103] in: The number of samples in the multi-source fusion data training set. The sample number is the training set of the multi-source fusion data. For the first The actual cutting time of each sample under the predicted cutting parameters. For the first The baseline value for the cutting time of each sample.

[0104] The expression for equipment stability loss is:

[0105]

[0106] in: For the first The standard deviation of the load current during the cutting process for each sample. For the first The standard deviation of the equipment vibration amplitude during the cutting process for each sample.

[0107] The expression for energy consumption loss is:

[0108]

[0109] in: For the first Total energy consumption of each sample For the first Effective cutting energy consumption per sample.

[0110] S4. Collect real-time data from multiple sensors of the support equipment, cutting host and conveying equipment, and perform timestamp alignment, data cleaning, feature extraction and data standardization on the real-time data of the multiple sensors to generate multi-source fusion data.

[0111] In an optional embodiment of the present invention, the present invention collects real-time data from multiple sources of sensors of support equipment, cutting host and conveying equipment through a data acquisition module, and then performs timestamp alignment, data cleaning, feature extraction and data standardization on the real-time data of multiple sources of sensors through a data preprocessing module to generate multi-source fusion data.

[0112] S5. Input the multi-source fusion data into the trained intelligent cutting parameter prediction model of the working face to obtain the predicted cutting parameters, generate cutting control commands based on the predicted cutting parameters, and build a confidence evaluation mechanism to quantitatively evaluate the reliability of the predicted cutting parameters, and decide whether to execute the cutting control commands based on the quantitative evaluation results.

[0113] In an optional embodiment of the present invention, the present invention receives multi-source fused data through an intelligent inference module and inputs it into a trained intelligent cutting parameter prediction model for the working surface to obtain predicted cutting parameters, and generates cutting control commands based on the predicted cutting parameters. To ensure safety, the present invention incorporates a confidence evaluator within the intelligent inference module to analyze the probability distribution predicted by the trained intelligent cutting parameter prediction model for the working surface, the completeness of the input data, and compares it with historical patterns and expert rules. Only when the confidence level is higher than a preset threshold will the generated control command be sent to the equipment controller, thereby executing the cutting control command.

[0114] like Figure 2 As shown, an intelligent cutting control system for the working face applied to the above method includes a data acquisition module, a data preprocessing module, a large model training module, and an intelligent inference module.

[0115] In an optional embodiment of the present invention, the data acquisition module is used to acquire historical data from the multi-source sensors of the support equipment, the cutting host and the conveying equipment, acquire real-time data from the multi-source sensors of the support equipment, the cutting host and the conveying equipment, and transmit the historical data and real-time data of the multi-source sensors to the data preprocessing module.

[0116] In an optional embodiment of the present invention, the data preprocessing module is used to perform timestamp alignment, data cleaning, feature extraction and data standardization on the historical data of multi-source sensors to generate a multi-source fusion data training set, and transmit the multi-source fusion data training set to the large model training module. The module then performs timestamp alignment, data cleaning, feature extraction and data standardization on the real-time data of multi-source sensors to generate multi-source fusion data, and transmits the multi-source fusion data to the intelligent inference module.

[0117] In an optional embodiment of the present invention, the large model training module is used to construct a working face intelligent cutting parameter prediction model based on a deep learning model, and to train the working face intelligent cutting parameter prediction model based on a multi-source fusion data training set.

[0118] In an optional embodiment of the present invention, the intelligent inference module is used to input multi-source fusion data into the trained intelligent cutting parameter prediction model of the working face to obtain predicted cutting parameters, generate cutting control commands based on the predicted cutting parameters, construct a confidence evaluation mechanism to quantitatively evaluate the reliability of the predicted cutting parameters, and decide whether to execute the cutting control commands based on the quantitative evaluation results.

[0119] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method of intelligent cutting control of a working face, characterized in that, The method comprises the following steps: obtaining multi-source sensor historical data of the supporting equipment, the cutting host and the conveying device; timestamp alignment, data cleaning, feature extraction and data standardization processing are performed on the multi-source sensor historical data to generate a multi-source fusion data training set; feature extraction includes calculating the action frequency of the supporting equipment, the average action duration of the supporting equipment, the action type proportion of the supporting equipment, the speed change rate of the cutting host, the average load current of the cutting host, the change rate of the height difference between the left and right rollers of the cutting host, the speed change rate of the conveying device, the average current of the conveying device and the frequency change rate of the conveying device; a working face intelligent cutting parameter prediction model is constructed based on a deep learning model, and the working face intelligent cutting parameter prediction model is trained based on the multi-source fusion data training set; the working face intelligent cutting parameter prediction model comprises a multi-scale feature extraction layer, a device collaborative attention fusion layer, a time sequence encoding layer, a global attention decision layer and a full connection layer connected in turn; the multi-scale feature extraction layer comprises three independent multi-scale convolution branches, which respectively and in parallel process the data of the supporting equipment, the cutting host and the conveying device; each multi-scale convolution branch comprises a plurality of convolution channels with different convolution kernel sizes, which are used to simultaneously capture the device running modes in different time windows through convolution operation to respectively obtain multi-scale features of the supporting equipment, the cutting host and the conveying device, and respectively fuse the multi-scale features of the supporting equipment, the cutting host and the conveying device through adaptive weighting operation to respectively obtain multi-scale fusion features of the supporting equipment, the cutting host and the conveying device; the device collaborative attention fusion layer is used to calculate the interaction attention weight between the supporting equipment, the cutting host and the conveying device based on the multi-scale fusion features of the supporting equipment, the cutting host and the conveying device to generate final fusion collaborative information enhancement features; the time sequence encoding layer is used to perform time sequence modeling on the final fusion collaborative information enhancement features through a bidirectional long short-term memory network to obtain bidirectional hidden state features of all time steps; the global attention decision layer is used to extract time sequence information related to the current cutting decision from the bidirectional hidden state features of all time steps to obtain a global weighted context vector; the full connection layer is used to map the global weighted context vector to a predicted cutting parameter; multi-source sensor real-time data of the supporting equipment, the cutting host and the conveying device are collected, and timestamp alignment, data cleaning, feature extraction and data standardization processing are performed on the multi-source sensor real-time data to generate multi-source fusion data; the multi-source fusion data is input into the trained working face intelligent cutting parameter prediction model to obtain a predicted cutting parameter, a cutting control instruction is generated based on the predicted cutting parameter, a confidence evaluation mechanism is constructed to quantitatively evaluate the reliability of the predicted cutting parameter, and whether to execute the cutting control instruction is decided based on the quantitative evaluation result.

2. The method of face intelligent cutting control of claim 1, wherein, The timestamp alignment of the multi-source sensor historical data is expressed as: wherein: is the multi-source sensor historical data at time after timestamp alignment, is the multi-source sensor historical data at time after timestamp alignment, is the multi-source sensor historical data at time after timestamp alignment, is the time number.

3. The method of face intelligent cutting control of claim 1, wherein, The specific processing process of the device collaborative attention fusion layer is: The interaction attention score between the supporting equipment, the cutting host and the conveying device is calculated, and its expression is: wherein: is a number of devices, is an attention score of the device for the device, is a multi-scale fused feature of the is a projection matrix for the query, is a multi-scale fused feature of the device, is a projection matrix for the key, is a transpose operator for the matrix, is a dimension of the multi-scale fused feature of the device. The interaction attention scores between the support equipment, the cutting host and the conveying device are normalized to obtain interaction attention weights between the support equipment, the cutting host and the conveying device, and the expression is wherein: is the interaction attention weight between the th device and the th device, is the exponential function symbol, is the attention score of the th device to the th device; Based on the interaction attention weights between the support equipment, the cutting host and the conveying device, a fusion collaborative information enhanced feature of the support equipment, the cutting host and the conveying device is generated, and the expression is wherein: is a fusion coordination information enhancement feature of the device, is a projection matrix of values; The fusion collaborative information enhanced features of the support equipment, the cutting host and the conveying device are spliced and fused to generate a final fusion collaborative information enhanced feature.

4. The method of face intelligent cutting control of claim 1, wherein, The working face intelligent cutting parameter prediction model is trained based on the multi-source fusion data training set, and the specific process is: A loss function of the working face intelligent cutting parameter prediction model is constructed by combining the cross-entropy loss and the cutting process optimization loss; Based on the multi-source fusion data training set, the working face intelligent cutting parameter prediction model is trained by using the loss function of the working face intelligent cutting parameter prediction model.

5. The method of face intelligent cutting control of claim 4, wherein, The loss function of the working face intelligent cutting parameter prediction model, and the expression is ， wherein: is a loss function value of the working face intelligent cutting parameter prediction model, is a cross-entropy loss function value, is a balance coefficient, is a cutting process optimization loss function value, is a weight coefficient of the cutting efficiency loss, is a cutting efficiency loss, is a weight coefficient of the equipment stability loss, is an equipment stability loss, is a weight coefficient of the energy consumption loss, is an energy consumption loss.

6. A face intelligent cutting control system applied to the method of any one of claims 1-5, characterized in that, It includes a data acquisition module, a data preprocessing module, a large model training module and an intelligent reasoning module; The data acquisition module is used to acquire multi-source sensor historical data of the support equipment, the cutting host and the conveying device, collect multi-source sensor real-time data of the support equipment, the cutting host and the conveying device, and transmit the multi-source sensor historical data and the multi-source sensor real-time data to the data preprocessing module; The data preprocessing module is used for timestamp alignment, data cleaning, feature extraction and data standardization processing of the multi-source sensor historical data, to generate a multi-source fusion data training set, and transmit the multi-source fusion data training set to the large model training module; the multi-source sensor real-time data is timestamped, cleaned, feature extracted and standardized, to generate multi-source fusion data, and transmit the multi-source fusion data to the intelligent reasoning module; feature extraction includes calculating the action frequency of the support equipment, the average action time of the support equipment, the action type proportion of the support equipment, the speed change rate of the cutting host, the average load current of the cutting host, the change rate of the height difference between the left and right rollers of the cutting host, the speed change rate of the conveying device, the average current of the conveying device and the frequency change rate of the conveying device; The large model training module is configured to construct a working face intelligent cutting parameter prediction model according to a deep learning model, and train the working face intelligent cutting parameter prediction model based on a multi-source fusion data training set; the working face intelligent cutting parameter prediction model comprises, in sequence, a multi-scale feature extraction layer, a device collaborative attention fusion layer, a time sequence coding layer, a global attention decision layer and a full connection layer; the multi-scale feature extraction layer comprises three independent multi-scale convolution branches, which respectively and in parallel process data of the support equipment, the cutting host and the conveying equipment; each multi-scale convolution branch comprises a plurality of convolution channels with different convolution kernel sizes, which are configured to simultaneously capture the running modes of the equipment in different time windows through convolution operation to respectively obtain multi-scale features of the support equipment, the cutting host and the conveying equipment, and respectively fuse the multi-scale features of the support equipment, the cutting host and the conveying equipment through adaptive weighting operation to respectively obtain multi-scale fusion features of the support equipment, the cutting host and the conveying equipment; the device collaborative attention fusion layer is configured to calculate the interaction attention weights between the support equipment, the cutting host and the conveying equipment according to the multi-scale fusion features of the support equipment, the cutting host and the conveying equipment to generate final fusion collaborative information enhancement features; The time sequence coding layer is configured to perform time sequence modeling on the final fusion collaborative information enhancement features through a bidirectional long short-term memory network to obtain bidirectional hidden state features of all time steps; The global attention decision layer is configured to extract time sequence information related to the current cutting decision from the bidirectional hidden state features of all time steps to obtain a global weighted context vector; and the full connection layer is configured to map the global weighted context vector to a predicted cutting parameter; The intelligent reasoning module is configured to input the multi-source fusion data into the trained working face intelligent cutting parameter prediction model to obtain a predicted cutting parameter, generate a cutting control instruction based on the predicted cutting parameter, construct a confidence evaluation mechanism, quantitatively evaluate the reliability of the predicted cutting parameter, and decide whether to execute the cutting control instruction based on the quantitative evaluation result.

Images