Mine fully mechanized mining equipment state recognition based on deep learning and wi-fi6 transmission method

Abstract

This invention discloses a deep learning-based method for identifying the status of fully mechanized mining equipment and transmitting data via Wi-Fi 6. The invention relates to the field of mine equipment status monitoring technology, and includes: constructing a spatiotemporal-vibration joint database by collecting equipment operating parameters and environmental vibration data; analyzing vibration multipath effects using a Rayleigh fading model and Monte Carlo simulation to calculate path loss values; constructing a space-state-multipath matrix using a deep learning model and generating a multidimensional feature map of the equipment; filtering abnormal equipment based on health scores and analyzing the abnormality scores; implementing a three-level dynamic response by comparing with preset thresholds: a normal mode and an enhanced identification mode; and employing a sliding window statistical and priority-based rate limiting mechanism for high-concurrency scenarios. By evaluating identification accuracy, communication stability, and resource consumption, the threshold is adaptively adjusted and the multipath compensation strategy is optimized. This method significantly improves the accuracy of underground equipment status identification and communication reliability while ensuring resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of mine equipment status monitoring technology, specifically to a method for identifying the status of fully mechanized mining equipment and transmitting data via Wi-Fi 6 based on deep learning. Background Technology

[0002] The stability and efficiency of underground communication systems in coal mines are crucial for ensuring safe production and improving operational efficiency. However, due to the unique characteristics of the underground coal mine environment, such as narrow roadways, complex geological structures, severe electromagnetic interference, and significant multipath effects, traditional wireless communication technologies are insufficient to meet the needs of underground communication. Existing underground communication systems largely rely on wired transmission or early wireless communication technologies, which have significant shortcomings in terms of coverage, data transmission rate, and anti-interference capabilities, making it impossible to achieve real-time monitoring and efficient communication of underground equipment status. Furthermore, with the development of intelligent coal mines, the demand for status identification and concurrent data transmission of fully mechanized mining equipment is increasing, and traditional communication methods also show significant limitations in integrating with programmable scheduling systems, public network telephones, and underground wireless communication.

[0003] Currently, although some studies have attempted to introduce Wi-Fi technology into underground coal mine communication, early Wi-Fi technologies still struggle to adapt to the complex underground environment in terms of bandwidth, coverage, and resistance to multipath effects. Especially in scenarios involving high-density device access and concurrent transmission of large amounts of data, the performance of existing technologies is further limited. Based on this, a deep learning-based method for identifying the status of fully mechanized mining equipment and transmitting Wi-Fi 6 data is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying the status of fully mechanized mining equipment and transmitting data via Wi-Fi 6 based on deep learning.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A deep learning-based method for identifying the status of fully mechanized mining equipment and transmitting via Wi-Fi 6 includes: real-time acquisition of key operating parameters and environmental vibration data of fully mechanized mining equipment through cameras and various types of sensors deployed underground, forming a spatiotemporal-vibration joint database; obtaining vibration multipath path loss values ​​through multipath effect analysis; inputting the spatiotemporal-vibration joint database data into a preset deep learning model, outputting equipment status baselines, constructing a space-state-multipath matrix that integrates vibration multipath characteristics, and extracting a set of baseline feature vectors.

[0007] Based on the extracted baseline feature vector set, key features are activated through weighted nonlinear transformation and ReLU function; a convolutional neural network is introduced for iterative training to generate a multidimensional feature map of the equipment; based on the equipment status index, vibration multipath path loss value and preset retransmission limit, the equipment health score is obtained; and the multidimensional feature map of the equipment is filtered to obtain the multidimensional feature map of abnormal equipment.

[0008] The multi-dimensional feature map of abnormal devices is mapped to the abnormal score and compared with the preset initial switching threshold to realize the abnormal classification. The corresponding identification mode is matched according to the abnormal classification. It also includes the identification and control of concurrent abnormal reporting under the enhanced identification mode.

[0009] The actual recognition effect and multipath compensation effect under different recognition modes are comprehensively evaluated, including the status recognition accuracy, wireless communication stability and resource consumption of each abnormal device. Based on the evaluation results, the initial switching threshold is automatically adjusted, and the device adaptability of the abnormal recognition and compensation strategy is intelligently optimized according to the multipath compensation performance under different modes.

[0010] As a further aspect of the present invention: a spatiotemporal-vibration joint database is formed, and vibration multipath path loss values ​​are obtained through multipath effect analysis, including: aligning key operating parameters and environmental vibration data by timestamp and equipment spatial location, constructing a spatiotemporal index for each data entry, with timestamp as the primary key and equipment ID and spatial coordinates as auxiliary indexes, and obtaining spatiotemporal-vibration joint features through alignment and integration;

[0011] A time-series database is used to store sensor data, and a relational database is used to manage device metadata. A spatiotemporal-vibration joint database is obtained by linking multi-source data through a unified timestamp.

[0012] Based on a spatiotemporal-vibration joint database, the vibration acceleration time series of each device at different time points and spatial locations is extracted. Then, according to the geometry of the downhole environment and the relative positional relationship between the sensor and the device, the number and distribution of typical multipath propagation paths are determined. Using a pre-set Rayleigh fading statistical model, the envelope amplitude distribution of each path is calculated. Combining the actual measured small displacement changes caused by vibration, the relative phase and amplitude attenuation factors of each path are adjusted, and Monte Carlo simulation is performed on the instantaneous synthesized signal of multiple paths. Through statistical analysis of the simulation results, the additional loss distribution characteristics of the multipath are obtained, and thus the vibration multipath loss value L is obtained. i , where i is the index of the fully mechanized mining equipment and i = (1,2,...,n), and n is the number of fully mechanized mining equipment in the mine, where n is a positive integer.

[0013] As a further aspect of the present invention: the original multi-parameter data from the spatiotemporal-vibration joint database is input into a pre-trained deep learning model, and the state S of each device is obtained through the device detection branch of the deep learning model. i ∈{0,1}, where 0 indicates that the device is in normal condition and 1 indicates that the device is in abnormal condition;

[0014] The baseline feature vector regression branch in the deep learning model reads the actual coordinates (x, y, y) of the device in the downhole three-dimensional space from the spatiotemporal-vibration joint database. i ,y i ,z i According to the vibration multipath loss value L corresponding to this equipment; i And the preset retransmission limit prediction M i For each device, retain a small-dimensional vector Δ i =[L i M i Based on pre-stored images of the entire mine operation area, a fixed-size sparse 3D mesh is established, with each mesh cell corresponding to a certain spatial voxel. For the i-th device, its coordinates (x, y, z) are set. i ,y i ,z i Mapping to the nearest voxel unit serves as the position index dimension of the matrix; based on the voxel unit, a state level dimension and a multipath influence dimension are assigned to each device, forming a three-dimensional index at the corresponding position (x, y, z, S) in the three-dimensional matrix. i In the blank, fill in the corresponding radial quantity Δ for the device. i This yields the space-state-multipath matrix.

[0015] As a further aspect of the present invention: Based on the space-state-multipath matrix, a set of baseline feature vectors is extracted, including: expanding the three-dimensional space-state-multipath matrix into a one-dimensional vector to obtain the set of baseline feature vectors at the current time t. Among them, S i L represents the device state corresponding to device i. i M represents the vibration multipath loss value corresponding to device i. i C is the preset maximum number of retransmissions for device i. i Let T be the coding rate adjustment factor for device i, and T be the transpose symbol.

[0016] As a further aspect of the present invention: generating a multi-dimensional feature map of devices, including: a set of baseline feature vectors for each device i at the current time obtained in step one, and the spatial coordinates (x, y) of the devices. i ,y i ,z iThe eigenvectors in the baseline eigenvector set are multiplied by a preset first set of weight matrices through a weighted nonlinear affine transformation, and the corresponding preset bias vectors are added. The ReLU activation function is then applied element-wise to the affine transformation result. The output of the previous activation step is used as the input to the next affine transformation, and the ReLU activation operation is repeated. The output of the final mapping layer is denoted as the intermediate eigenvector. Where l represents the l-th layer currently being calculated, and l = 1, 2, ..., L⁻¹, and L represents the total number of layers included in the weighted nonlinear affine transformation design. This represents the output of the l-th layer;

[0017] Intermediate representation of all devices According to spatial coordinates (x i ,y i ,z i The initial feature map F is obtained by mapping the type identifier and the type identifier onto a sparse 3D mesh and stitching them together. (0) ; with F (0) As input, the system iteratively updates the data using a convolutional neural network. In each iteration, it extracts the corresponding vibration multipath loss value L for each device. i Generate channel-level gain ΔW i , represents the weight vector that the i-th device needs to dynamically adjust on each feature channel, in the initial feature map F. (0) Multiply the channel by 1+ΔW i The final device multidimensional feature map F is obtained. (L) .

[0018] As a further aspect of the present invention: the selection principle for the multidimensional feature map of abnormal equipment is as follows: in the multidimensional feature map of equipment, for each equipment i, the equipment status index S is extracted from its corresponding position. i Vibration multipath loss value L i and preset maximum number of retransmissions M i The device health score (SP) for each device is obtained by weighted summation. i And assign the corresponding device health score SP to each device. i Compared with the preset device health threshold, the device will meet the device health score SP. i Device index i that is below the preset device health threshold is collected into the abnormal device set S. abn For each i∈S abn In the device multidimensional feature map F (L) The spatial coordinates (x) of the device are located in the middle. i ,y i ,z i Using the spatial coordinates as the center, extract the small tensors formed by the neighborhood of the device as the multidimensional feature map of the device's anomalous features. Where j is the index of the abnormal device and j = (1,2,...m), and m is the number of abnormal devices obtained from the device multidimensional feature map.

[0019] As a further aspect of the present invention: matching the corresponding identification pattern according to the anomaly classification includes: for each anomaly device, a multi-dimensional feature map of the anomaly device. By performing global average pooling, it is compressed into a one-dimensional feature vector V. j ;

[0020] The one-dimensional eigenvector V j The input is fed into a separate fully connected layer, and a scalar anomaly score AS is generated through linear transformation and nonlinear activation function. j The linear transformation is based on the formula: z j =σ·V j +u; where z j The intermediate output of the linear transformation is represented by σ and u, which are independent preset weights and preset biases of the fully connected layer. The nonlinear activation is based on the formula: AS j = sigma(z j ); where sigma is the Sigmoid function, which limits the score to the interval [0,1];

[0021] After obtaining the score AS for each abnormal device j Then, it is compared with the preset initial switching threshold T. low and T high Compare;

[0022] When AS j ≤T low When this occurs, it is marked as a low-level anomaly, and the normal identification mode is maintained;

[0023] When T low <AS j ≤T high When this occurs, it is marked as a medium-level anomaly, and a minor compensation is performed while maintaining the normal identification pattern.

[0024] When AS j >T high When an anomaly is detected, it is marked as a high-level anomaly and automatically switches to enhanced recognition mode.

[0025] As a further aspect of the present invention: using a sliding time window ΔT of fixed length win The total number of devices identified as high-level anomalies within this window is recorded as the sliding window device statistics value m. abn (t), according to the formula: m abn (t)=|{j∣AS j (t')≥Thigh ,t'∈[t-ΔT win ,t]}|;where, AS j (t') represents the anomaly score of the j-th device at time t', where t'∈[t-ΔT]. win [,t] represents all times t' that are backtracked from the current time t to t-ΔT. win All states within this interval are included in the statistics;

[0026] The sliding window device statistics m abn (t) and the preset concurrent reporting threshold m th The preset concurrent reporting threshold m is compared. th Indicates any ΔT win The window displays the maximum number of high-level anomalous devices allowed to enter enhanced mode.

[0027] When the sliding window device statistics value m abn (t) satisfies a condition greater than or equal to the preset concurrent reporting threshold m. th This indicates a high-concurrency anomaly reporting scenario. From then until the end of the next sliding window, the system enters a high-concurrency rate limiting state.

[0028] As a further aspect of the present invention: Strengthening the control of concurrent anomaly reporting under the identification mode includes: based on the anomaly score AS of the abnormal device. j and device health score SP j The priority YP of each abnormal device is obtained by weighted summation. j Based on the priority YP of each abnormal device. j Sort all faulty devices in descending order and generate a priority queue;

[0029] Based on the generated priority queue, select the top m... abn If a device malfunctions, it is designated as an enhanced identification device in this sliding window and enters the enhanced identification mode. The remaining devices are postponed to the next sliding window for re-evaluation.

[0030] As a further aspect of the present invention: the initial switching threshold is automatically adjusted, including: calculating the state recognition accuracy, wireless communication stability and resource overhead scores for each abnormal device j within a preset period; for each abnormal device, a comprehensive performance score is obtained by weighting the recognition accuracy and subtracting the resource consumption and stability penalty in the corresponding recognition mode; the scores of the regular recognition mode and the enhanced recognition mode of all devices are averaged to obtain two global averages, and the difference between the two is compared.

[0031] If the global average score in the enhanced recognition mode is greater than that in the regular recognition mode, the enhanced recognition mode is deemed to be of excellent quality; the switching threshold T originally used for graded switching is automatically reduced by a preset downward adjustment value.low and T high ;

[0032] If the global average score in the enhanced recognition mode is less than or equal to that in the regular recognition mode, the switching threshold T will be automatically increased by a preset adjustment value. low and T high This reduces the trigger range of the enhanced recognition mode.

[0033] The beneficial effects of this invention are:

[0034] (1) This invention, through deep integration of vibration multipath effect analysis and equipment status identification, innovatively quantifies the multipath loss value caused by downhole equipment vibration, and accurately characterizes the channel dynamic characteristics by combining the Rayleigh fading model and Monte Carlo simulation; constructs a space-state-multipath three-dimensional matrix, associating equipment coordinates, status labels and multipath quantities, to provide structured input for deep learning. Based on weighted nonlinear affine transformation and ReLU activation, a multidimensional feature map of the equipment is generated, and the feature weights are dynamically optimized through channel-level gain, significantly improving the model's expressive power; at the same time, a dynamic hierarchical mechanism is designed: abnormal equipment is screened based on equipment health scores, and a three-level identification mode is triggered by the abnormal scores, and sliding window concurrent control and priority scheduling are introduced to realize on-demand resource allocation; by comprehensively evaluating the identification accuracy, communication stability and resource overhead, the switching threshold is automatically optimized to form an adaptive closed loop; the high bandwidth and low latency characteristics of Wi-Fi 6 support real-time backhaul of multi-source data, ensuring efficient system collaboration. Attached Figure Description

[0035] The invention will now be further described with reference to the accompanying drawings.

[0036] Figure 1 This is a schematic diagram of the overall method flow of the deep learning-based method for identifying the status of fully mechanized mining equipment and transmitting data via Wi-Fi 6 in this invention.

[0037] Figure 2 This is a schematic diagram of the screening process of the multidimensional feature map of abnormal equipment in step two of the present invention;

[0038] Figure 3 This is a schematic diagram of the matching method for identifying patterns in step three of the present invention;

[0039] Figure 4 This is a schematic diagram of the method for automatically adjusting the initial switching threshold in step four of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Example 1

[0042] Please see Figure 1 As shown, this invention is a method for identifying the status of fully mechanized mining equipment and transmitting data via Wi-Fi 6 based on deep learning, comprising the following steps:

[0043] Step 1: Using cameras and various sensors deployed underground, key operating parameters and environmental vibration data of the fully mechanized mining equipment are collected in real time, forming a spatiotemporal-vibration joint database. Multipath effect analysis is used to obtain vibration multipath path loss values. The raw multi-parameter data is input into a preset deep learning model, which outputs equipment state baselines and constructs a space-state-multipath matrix that integrates vibration multipath characteristics. A set of baseline feature vectors is extracted. Specifically, the fully mechanized mining equipment includes coal mining machines, hydraulic supports, etc., and each piece of equipment is assigned a unique equipment number (equipment ID) and its type, location, and other basic information are registered in the system. Key operating parameters include temperature, current, and pressure; environmental vibration data includes vibration acceleration and equipment displacement amplitude.

[0044] All data are aligned by timestamp and device spatial location. A spatiotemporal index is constructed for each data point. Each data point uses timestamp as the primary key and device ID and spatial coordinates as secondary indexes. Through alignment, the above multi-source data are integrated into spatiotemporal-vibration joint features.

[0045] A time-series database is used to store sensor data, and a relational database is used to manage device metadata. A spatiotemporal-vibration joint database is obtained by linking multi-source data through a unified timestamp.

[0046] Based on the obtained spatiotemporal-vibration joint database, the vibration acceleration time series of each device at different time points and spatial locations is extracted. Then, according to the geometric dimensions of the downhole environment and the relative positional relationship between the sensor and the device, the number and distribution of typical multipath propagation paths are determined. Next, using a pre-set Rayleigh fading statistical model, which assumes that the signal amplitude on each reflection or scattering path follows a zero-mean, Gaussian distribution complex random variable, the envelope amplitude distribution of each path is calculated. Based on this, combined with the actual measured small displacement changes caused by vibration, the relative phase and amplitude attenuation factors of each path are adjusted, and Monte Carlo simulation is performed on the instantaneous synthesized signal of multiple paths. Finally, through statistical analysis of the simulation results, the distribution characteristics of the additional loss of the multipath caused by vibration are obtained, and thus the vibration multipath loss value L is obtained. i , where i is the index of the fully mechanized mining equipment and i = (1,2,...,n), and n is the number of fully mechanized mining equipment in the mine, where n is a positive integer;

[0047] It should be noted that the Rayleigh fading statistical model is a mature existing technology, and will not be described in detail in this embodiment;

[0048] The raw multi-parameter data from the spatiotemporal-vibration joint database are input into a pre-trained deep learning model, as shown in the convolutional network. The state S of each device is obtained through the device detection branch of the deep learning model. i ∈{0,1}, where 0 indicates that the device is in normal condition and 1 indicates that the device is in abnormal condition;

[0049] Furthermore, the baseline feature vector regression branch in the deep learning model reads the actual coordinates (x, y, y) of the device in the downhole three-dimensional space from the spatiotemporal-vibration joint database. i ,y i ,z i According to the vibration multipath loss value L corresponding to this equipment; i And the preset retransmission limit prediction M i For each device, retain a small-dimensional vector Δ i =[L i M i This is used to characterize the vibration-driven impact of the device on the wireless channel at the current moment; and a fixed-size sparse three-dimensional mesh is established based on pre-stored images of the entire mine operation area, with each mesh cell corresponding to a certain spatial voxel. For the i-th device, its coordinates (x... i ,y i ,z iThe data is mapped to the nearest voxel unit, serving as the position index dimension of the matrix. Based on the voxel unit, a state level dimension and a multipath influence dimension are assigned to each device, forming a total three-dimensional index. The corresponding position (x, y, z, S) in the three-dimensional matrix is ​​then filled with the multipath quantity Δ corresponding to that device. i This yields the three-dimensional space-state-multipath matrix;

[0050] Furthermore, the aforementioned three-dimensional space-state-multipath matrix is ​​expanded into a one-dimensional vector to obtain the set of baseline feature vectors at the current time t. Among them, S i L represents the device state corresponding to device i. i M represents the vibration multipath loss value corresponding to device i. i C is the preset maximum number of retransmissions for device i. i Let T be the coding rate adjustment factor for device i, and T be the transpose symbol.

[0051] Please see Figure 2 As shown, step two: Based on the baseline feature vector set generated in step one, key features are activated through weighted nonlinear transformation and ReLU function; a convolutional neural network is introduced for iterative training to generate a multidimensional feature map of the equipment; based on the equipment status index, vibration multipath path loss value and preset retransmission limit, the equipment health score is obtained; and the multidimensional feature map of the equipment is filtered to obtain the multidimensional feature map of abnormal equipment.

[0052] Based on the set of baseline feature vectors for each device i at the current time obtained in step one, and the spatial coordinates (x, y) of the device... i ,y i ,z i The system employs a weighted nonlinear affine transformation to multiply the feature vectors in the baseline feature vector set by a preset first set of weight matrices and add the corresponding preset bias vectors. Then, it applies the ReLU activation function element-wise to the affine transformation result. The output of the previous activation step is used as the input for the next affine transformation, repeating the weight multiplication, bias addition, and ReLU activation operations. Each ReLU activation layer clears some unnecessary feature components to zero, making the hidden layer representation naturally sparse. The output of the final mapping layer is denoted as the intermediate feature vector. Where l represents the l-th layer currently being calculated, and l = 1, 2, ..., L⁻¹, and L represents the total number of layers included in the weighted nonlinear affine transformation design. This represents the output of the l-th layer;

[0053] Intermediate representation of all devices According to spatial coordinates (x i ,y i ,z iThe data, along with the type identifier, is mapped onto a sparse 3D mesh, forming an initial feature map F of shape A×H×D×E. (0) Where A is the number of voxel grid cells in the X direction, H is the number of voxel grid cells in the Y direction, D is the number of voxel grid cells in the Z direction, and E is the number of channels carried by each voxel; and F... (0) As input, the system iteratively updates the data using a convolutional neural network. In each iteration, it extracts the corresponding vibration multipath loss value L for each device. i Generate channel-level gain ΔW i , represents the weight vector that the i-th device needs to dynamically adjust on each feature channel, and then in the initial feature map F (0) Multiply the channel by 1+ΔW i The final device multidimensional feature map F is obtained. (L) ;

[0054] Furthermore, in the multidimensional feature map of the equipment, for each equipment i, the equipment status index S is first extracted from its corresponding location. i Vibration multipath loss value L i and preset maximum number of retransmissions M i The device health score (SP) for each device is obtained by weighted summation. i And assign the corresponding device health score SP to each device. i Compared with the preset device health threshold, the device will meet the device health score SP. i Device index i that is below the preset device health threshold is collected into the abnormal device set S. abn For each i∈S abn In the device multidimensional feature map F (L) The spatial coordinates (x) of the device are located in the middle. i ,y i ,z i Using the spatial coordinates as the center, extract the small tensors formed by the neighborhood of the device as the multidimensional feature map of the device's anomalous features. Where j is the index of the abnormal device and j = (1,2,...m), and m is the number of abnormal devices obtained from the device multidimensional feature map.

[0055] In this embodiment, the multidimensional feature map of abnormal equipment is constructed based on the spatiotemporal-vibration joint database generated in step one and the equipment health score extracted in step two, providing multidimensional feature support for anomaly classification.

[0056] Please see Figure 3As shown, step three: map the multi-dimensional feature map of the abnormal device to the abnormal score and compare it with the preset initial switching threshold to realize the abnormal classification. Match the corresponding identification mode according to the abnormal classification. It also includes the identification and control of concurrent abnormal reporting under the enhanced identification mode.

[0057] S1: Map the multi-dimensional feature map of the abnormal device to an anomaly score and compare it with a preset initial switching threshold to achieve anomaly classification, including: for each abnormal device's multi-dimensional feature map... The feature vector V is compressed into a one-dimensional feature vector V through the Global Average Pooling (GAP) operation. j ;

[0058] The one-dimensional eigenvector V j The input is passed to a fully connected layer, which processes the one-dimensional feature vector V. j The input is fed into a separate fully connected layer, and a scalar anomaly score AS is generated through linear transformation and nonlinear activation function. j The linear transformation is based on the formula: z j =σ·V j +u; where z j The intermediate output of the linear transformation is σ and u, which are independent preset weights and preset biases of the fully connected layer, and have no shared relationship with the weight matrix in step two; the nonlinear activation is based on the formula: AS j = sigma(z j ); where sigma is the Sigmoid function, which limits the score to the interval [0,1];

[0059] After obtaining the score AS for each abnormal device j Then, it is compared with the preset initial switching threshold T. low and T high Compare;

[0060] When AS j ≤T low When this occurs, it is marked as a low-level anomaly, and the normal identification mode is maintained;

[0061] When T low <AS j ≤T high When an anomaly is detected, it is marked as a medium-level anomaly. While maintaining the normal identification mode, a lightweight dynamic compensation strategy is added based on the preset adjustment range, such as slightly adjusting the upper limit of the number of retransmissions.

[0062] When AS j >T highWhen an anomaly is detected, it is marked as a high-level anomaly and automatically switches to an enhanced recognition mode. This enhanced recognition mode includes increasing the number of training rounds, dynamically adjusting the learning rate, batch size, and other hyperparameters based on the equipment's vibration multipath loss value L. i Real-time communication quality feedback allows for dynamic adjustment of compensation parameters;

[0063] S2: When the system automatically switches to enhanced detection mode, it also includes the identification and control of high-concurrency anomaly reporting, including:

[0064] The identification criteria for high-concurrency anomaly reporting are specifically: a sliding time window ΔT of fixed length. win The total number of devices identified as high-level anomalies within this window is recorded as the sliding window device statistics value m. abn (t), according to the formula: m abn (t)=|{j∣AS j (t')≥T high ,t'∈[t-ΔT win ,t]}|;where, AS j (t') represents the anomaly score of the j-th device at time t', where t'∈[t-ΔT]. win [,t] represents all times t' that are backtracked from the current time t to t-ΔT. win All states within this interval are included in the statistics;

[0065] The sliding window device statistics m abn (t) and the preset concurrent reporting threshold m th The preset concurrent reporting threshold m is compared. th Indicates any ΔT win The window displays the maximum number of high-level anomalous devices allowed to enter enhanced mode.

[0066] When the sliding window device statistics value m abn (t) satisfies a condition greater than or equal to the preset concurrent reporting threshold m. th This indicates a high-concurrency anomaly reporting scenario. From then until the end of the next sliding window, the system enters a high-concurrency rate limiting state.

[0067] Anomaly score AS based on abnormal device j and device health score SP j The priority YP of each abnormal device is obtained by weighted summation. j Based on the priority YP of each abnormal device. j Sort all faulty devices in descending order and generate a priority queue;

[0068] Based on the generated priority queue, select the top m... abnOne abnormal device is recorded as an enhanced identification device set in this slide window, and the remaining devices are postponed to the next slide window for re-evaluation;

[0069] The abnormal reporting control process is as follows: within the remaining time of this sliding window, only devices in the enhanced recognition device set can enter the enhanced recognition mode, increase the training rounds and compensation intensity, and initiate high-priority reporting to the monitoring platform. After the window ends, the enhanced recognition device set is re-counted and updated according to the same process.

[0070] Please see Figure 4 As shown, step four involves comprehensively evaluating the actual recognition and multipath compensation effects under different recognition modes, including the accuracy of status recognition for each abnormal device, wireless communication stability, and resource consumption. Based on the evaluation results, the initial handover threshold is automatically adjusted, and the device adaptability of the abnormal recognition and compensation strategy is intelligently optimized according to the multipath compensation performance under different modes; including:

[0071] Throughout the identification and transmission process, the key operating parameters, identification status, multipath compensation parameters, and corresponding wireless communication quality indicators such as signal strength RSSI, packet loss rate, latency, and throughput of each fully mechanized mining equipment are transmitted back to the ground monitoring platform in real time via the Wi-Fi 6 network. Each piece of data includes a unique device ID, timestamp, spatial location, current identification mode, status judgment result and its confidence level, current multipath compensation parameters including path loss, retransmission count, compensation gain, and resource consumption including CPU, memory, and bandwidth consumption.

[0072] For each abnormal device j, the state identification accuracy, wireless communication stability, and resource overhead score are calculated within a preset period. The state identification accuracy is the number of correct judgments divided by the total number of judgments. The wireless communication stability is calculated by combining the retransmission rate, packet loss rate, and RSSI fluctuation into a single stability score. The resource overhead score is calculated by weighting the resource overhead score based on the proportion of resource units used by the device and the average transmit power within the current period.

[0073] Each abnormal device is labeled with its three-dimensional indicators—state identification accuracy, wireless communication stability, and resource overhead score—in different modes, and stored in the evaluation database.

[0074] For each abnormal device, a comprehensive performance score is obtained by weighting the recognition accuracy score and subtracting the resource consumption and stability penalty in the corresponding recognition mode. The scores of the regular recognition mode and the enhanced recognition mode of all devices are averaged to obtain two global averages, and the difference between the two is compared.

[0075] It should be noted that the conventional identification mode includes low-level anomalies and medium-level anomalies;

[0076] If the global average score in the enhanced recognition mode is greater than that in the regular recognition mode, the enhanced recognition mode is deemed to be of excellent quality; the switching threshold T originally used for graded switching is automatically reduced by a preset downward adjustment value. low and T high This is so that more devices can enter enhanced mode;

[0077] If the global average score in the enhanced recognition mode is less than or equal to that in the regular recognition mode, the switching threshold T will be automatically increased by a preset adjustment value. low and T high This reduces the trigger range of enhanced recognition modes;

[0078] During adjustment, always ensure that the threshold is between the minimum and maximum boundaries set by the system;

[0079] Furthermore, the comprehensive evaluation of multipath compensation effectiveness includes: extracting the compensation parameters used in both modes and the corresponding communication stability and resource overhead scores of the small group of devices that performed the worst in this period; finding a new set of compensation parameters through an optimization algorithm so that the stability score of this group of abnormal devices is improved while the resource consumption does not exceed the original level; clustering and averaging these new parameters according to device type or geographical location to derive the optimal compensation parameters for each group of abnormal devices; and returning the optimized parameters to the matching of the identification pattern in step three to replace the old parameters.

[0080] The Wi-Fi 6 wireless communication group transmits the device status identification and multipath compensation results with device numbers back to the ground monitoring platform in real time, and records the vibration-related multipath compensation process in the database according to the device ID. The monitoring platform regularly performs statistical analysis on the identification results and communication quality of each device. Based on the adaptive learning algorithm, it automatically adjusts the parameters of the deep learning model and the multipath compensation mechanism, and continuously optimizes the overall performance of the status identification and wireless communication transmission of each device according to the vibration and data changes.

[0081] In this embodiment, a spatiotemporal-vibration joint database is constructed by collecting equipment operating parameters and environmental vibration data; vibration multipath effects are analyzed using the Rayleigh fading model and Monte Carlo simulation to calculate path loss values; a space-state-multipath matrix is ​​constructed using a deep learning model to generate a multidimensional feature map of the equipment; abnormal equipment is screened based on health scores and anomaly scores are obtained through analysis; a three-level dynamic response is achieved by comparing with preset thresholds: a normal mode and an enhanced recognition mode; for high-concurrency scenarios, a sliding window statistical and priority flow limiting mechanism is adopted. By evaluating recognition accuracy, communication stability, and resource consumption, the threshold is adaptively adjusted and the multipath compensation strategy is optimized; this method significantly improves the accuracy of downhole equipment status recognition and communication reliability, while ensuring resource utilization efficiency.

[0082] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0083] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for recognizing the state of a fully mechanized mining equipment in a mine based on deep learning and transmitting data using Wi-Fi 6, characterized in that, include: By using cameras and various sensors deployed underground, key operating parameters and environmental vibration data of the fully mechanized mining equipment are collected in real time, forming a spatiotemporal-vibration joint database. After multipath effect analysis, vibration multipath path loss values ​​are obtained. The spatiotemporal-vibration joint database data is input into a preset deep learning model, which outputs equipment state baselines and constructs a spatial-state-multipath matrix that integrates vibration multipath characteristics, extracting a set of baseline feature vectors. The process of forming the spatiotemporal-vibration joint database and obtaining vibration multipath path loss values ​​through multipath effect analysis includes: aligning key operating parameters and environmental vibration data by timestamp and equipment spatial location, constructing a spatiotemporal index for each data point, with the timestamp as the primary key and the equipment ID and spatial coordinates as auxiliary indexes, and obtaining spatiotemporal-vibration joint features through alignment and integration. A time-series database is used to store sensor data, and a relational database is used to manage device metadata. A spatiotemporal-vibration joint database is obtained by linking multi-source data through a unified timestamp. Based on a spatiotemporal-vibration joint database, the vibration acceleration time series of each device at different time points and spatial locations is extracted; according to the geometric dimensions of the downhole environment and the relative positional relationship between the sensor and the device, the number and distribution of typical multipath propagation paths are determined; using a pre-set Rayleigh fading statistical model, the envelope amplitude distribution of each path is calculated; combined with the actual measured small displacement changes caused by vibration, the relative phase and amplitude attenuation factors of each path are adjusted, and Monte Carlo simulation is performed on the instantaneous synthesized signal of multiple paths; through statistical analysis of the simulation results, the additional loss distribution characteristics of the multipath are obtained, and thus the vibration multipath loss value is obtained. Based on the extracted baseline feature vector set, key features are activated through weighted nonlinear transformation and ReLU function; a convolutional neural network is introduced for iterative training to generate a multidimensional feature map of the equipment; based on the equipment status index, vibration multipath path loss value and preset retransmission limit, the equipment health score is obtained; and the multidimensional feature map of the equipment is filtered to obtain the multidimensional feature map of abnormal equipment. The multi-dimensional feature map of abnormal devices is mapped to the abnormal score and compared with the preset initial switching threshold to realize the abnormal classification. The corresponding identification mode is matched according to the abnormal classification. It also includes the identification and control of concurrent abnormal reporting under the enhanced identification mode. The actual recognition effect and multipath compensation effect under different recognition modes are comprehensively evaluated, including the status recognition accuracy, wireless communication stability and resource consumption of each abnormal device. Based on the evaluation results, the initial switching threshold is automatically adjusted, and the device adaptability of the abnormal recognition and compensation strategy is intelligently optimized according to the multipath compensation performance under different modes.

2. The method for identifying the status of fully mechanized mining equipment and transmitting Wi-Fi 6 based on deep learning according to claim 1, characterized in that, Constructing a space-state-multipath matrix that integrates vibration multipath characteristics includes: inputting raw multi-parameter data from a spatiotemporal-vibration joint database into a pre-trained deep learning model, and obtaining the state of each device through the device detection branch of the deep learning model. Where 0 indicates that the equipment is in normal condition and 1 indicates that the equipment is in abnormal condition; The baseline feature vector regression branch in the deep learning model reads the actual coordinates of the device in the downhole three-dimensional space from the spatiotemporal-vibration joint database. Based on the vibration multipath loss value corresponding to the equipment. Predicting the maximum number of pre-set retransmissions For each device, maintain a small-dimensional vector. A fixed-size sparse 3D mesh is built based on pre-stored images of the entire mine operation area. Each mesh cell corresponds to a certain spatial voxel. For the i-th device, its coordinates are... Mapped to the nearest voxel unit, serving as the position index dimension of the matrix; based on the voxel unit, a state level dimension and a multipath influence dimension are assigned to each device, forming a three-dimensional index at the corresponding position in the three-dimensional matrix. In the blank, fill in the corresponding multi-radial quantity for this device. This yields the space-state-multipath matrix.

3. The deep learning-based mine fully mechanized mining equipment state recognition and Wi-Fi6 transmission method according to claim 2, characterized in that, Based on the space-state-multipath matrix, a set of baseline feature vectors is extracted, including: expanding the three-dimensional space-state-multipath matrix into a one-dimensional vector to obtain the set of baseline feature vectors at the current time t. ;in, This represents the device state corresponding to device i. This represents the vibration multipath loss value corresponding to device i. This represents the preset maximum number of retransmissions for device i. Let T be the coding rate adjustment factor for device i, and T be the transpose symbol.

4. The deep learning-based mine fully mechanized mining equipment state recognition and Wi-Fi6 transmission method according to claim 1, characterized in that, Generate a multidimensional feature map of the devices, including: the set of baseline feature vectors for each device i at the current time obtained in step one, and the spatial coordinates of the devices. With type identification; the eigenvectors in the baseline eigenvector set are multiplied by a preset first set of weight matrices through a weighted nonlinear affine transformation, and the corresponding preset bias vectors are added. The ReLU activation function is then applied element-wise to the affine transformation result. The output of the previous activation step is used as the input to the next affine transformation, and the ReLU activation operation is repeated. The output of the final mapping layer is denoted as the intermediate eigenvector. ;in, Indicates the number of calculations currently in progress. Layer and L represents the total number of layers in the weighted nonlinear affine transformation design. Indicates the first Layer output; Intermediate representation of all devices According to spatial coordinates The type identifier is mapped onto a sparse 3D mesh, and the meshes are then stitched together to obtain the initial feature map. ;by As input, the system is continuously updated through a convolutional neural network. In each iteration, the corresponding vibration multipath loss value of each device is extracted. Generate channel-level gain , representing the weight vector that the i-th device needs to dynamically adjust on each feature channel, in the initial feature map. Multiply by the channel The final device multidimensional feature map was obtained. .

5. The deep learning-based mine fully mechanized mining equipment state recognition and Wi-Fi6 transmission method according to claim 4, characterized in that, The selection principle for the multidimensional feature map of abnormal equipment is as follows: In the multidimensional feature map of equipment, for each equipment i, extract the equipment status index from its corresponding position. Vibration multipath loss value and preset retransmission limit The device health score for each device is obtained by weighted summation. And assign a health score to each device. The device health score will be compared with the preset device health threshold. Device index i that is below the preset device health threshold is collected into the abnormal device set. For each In the device multidimensional feature map Position the spatial coordinates of the device Centered on the spatial coordinates, small tensors formed by extracting their neighborhood are used as the multidimensional feature map of the device's anomalous features. Where j is the index of the faulty device and , where m is the number of abnormal devices selected from the device multidimensional feature map.

6. The deep learning-based mine fully mechanized mining equipment state recognition and Wi-Fi6 transmission method according to claim 1, characterized in that, According to the abnormal classification situation, the corresponding recognition mode is matched, including: for each abnormal device, an abnormal device multi-dimensional feature map is obtained , and through a global average pooling operation, it is compressed into a one-dimensional feature vector ; One-dimensional feature vector The input is fed into a separate fully connected layer, and a scalar anomaly score is generated through linear transformation and a nonlinear activation function. The linear transformation is based on the formula: ;in, This is the intermediate output of the linear transformation. and The fully connected layer has independent preset weights and preset biases; the nonlinear activation is based on the formula: Where sigma is the Sigmoid function, which limits the score to the interval [0,1]. comparing the score of each abnormal device with a preset initial switching threshold value and and performing comparison; When a low level anomaly is flagged, the regular recognition mode is maintained; when When this occurs, it is marked as a medium-level anomaly, and a minor compensation is performed while maintaining the normal identification pattern. When a high-level anomaly is marked, the automatic switching to the enhanced recognition mode is performed.

7. The deep learning-based mine fully mechanized mining equipment state recognition and Wi-Fi6 transmission method according to claim 6, characterized in that, The identification criteria for concurrent anomaly reporting in the enhanced identification mode include: a sliding time window of fixed length. The total number of devices identified as high-level anomalies within this window is recorded as the sliding window device statistics value. According to the formula: ;in, Indicates the time of the j-th device Abnormal scores, Represents all moments Backtracking from the current time t to... All states within this interval are included in the statistics; Sliding window device statistics values are compared to a preset concurrent reporting threshold The preset concurrent reporting threshold represents the maximum number of high-level abnormal devices allowed to enter the reinforcement mode within any window When sliding window device statistics Meets the preset concurrent reporting threshold If the scenario is identified as a high-concurrency abnormal reporting scenario, the system will enter a high-concurrency rate limiting state from then until the end of the next sliding window.

8. The deep learning-based mine fully mechanized mining equipment state recognition and Wi-Fi6 transmission method according to claim 7, characterized in that, Enhanced control over concurrent anomaly reporting in the identification mode, including: anomaly scoring based on abnormal devices. and equipment health score The priority of each abnormal device is obtained by weighted summation. Based on the priority of each abnormal device. Sort all faulty devices in descending order and generate a priority queue; Based on the generated priority queue, take the front The abnormal device, denoted as the reinforcement identification device set of the current sliding window, enters the reinforcement identification mode, and the remaining devices are postponed to the next sliding window for reevaluation.

9. The deep learning-based mine fully mechanized mining equipment state recognition and Wi-Fi6 transmission method according to claim 1, characterized in that, The initial switching threshold is automatically adjusted, including: calculating the status recognition accuracy, wireless communication stability and resource overhead scores for each abnormal device j within a preset period; for each abnormal device, a comprehensive performance score is obtained by weighting the recognition accuracy and subtracting the resource consumption and stability penalty in the corresponding recognition mode; the scores of the regular recognition mode and the enhanced recognition mode of all devices are averaged to obtain two global averages, and the difference between the two is compared. When the global average score in the reinforced recognition mode is greater than that in the regular recognition mode, it is determined that the reinforced recognition mode is excellent this time; and a preset downshift value is used to automatically reduce the switching threshold value previously used for the hierarchical switching and ; When the global average score in the reinforced recognition mode is less than or equal to the global average score in the normal recognition mode, the switching threshold is automatically increased by a preset up-regulation value and , reducing the triggering range of the reinforced recognition mode.

Images