Mine personnel trajectory prediction method, system and equipment based on artificial intelligence

Abstract

The invention relates to the technical field of mine safety analysis, and discloses a mine personnel trajectory prediction method, system and device based on artificial intelligence. The method comprises the following steps: collecting mine personnel positioning data, and performing standardization processing; calculating trajectory quality through a neural network; converting the spatio-temporal feature vector to perform image recognition; performing multi-time scale prediction by applying an activation function; matching with a mine space structure, and clustering to identify a group behavior mode; and calculating a safety index, and generating a safety scheduling strategy. According to the method, the future trajectory of the mine personnel can be accurately predicted through the special environment constraint of the mine and the working behavior mode of the miner, and the prediction result is converted into a safe scheduling strategy.

Description

Technical Field

[0001] This application relates to the technical field of mine safety analysis, and particularly to a method, system and device for predicting the trajectory of mine personnel based on artificial intelligence. Background Art

[0002] The mine operation environment is complex and potentially dangerous, and personnel safety management has always been an important task for mining enterprises. Traditional mine personnel management methods mainly rely on manual sign-in, telephone reporting and simple positioning devices, which can only provide the real-time location information of miners and cannot predict future movement trajectories. With the development of Internet of Things technology and positioning technology, mine personnel positioning systems have been widely used. By deploying sensor networks at key positions in the mine and receiving signals from intelligent positioning devices worn by miners, real-time monitoring of miners' positions is achieved. These systems usually adopt technologies such as RFID, Bluetooth, and UWB, and can record the position coordinates of miners and the corresponding timestamps to form historical trajectory data. At the same time, the environmental monitoring system can collect environmental parameters such as temperature, humidity, gas concentration and ventilation conditions in each area of the mine, providing basic data support for safety management.

[0003] However, the existing technologies have obvious deficiencies. First, these systems can only provide the historical and real-time location information of miners, lacking the ability to predict future trajectories and unable to support forward-looking safety management and resource allocation. Second, some existing trajectory prediction methods are designed for open environments such as urban roads and do not consider the physical constraints and safety requirements of the special underground environment of the mine. Direct application will result in prediction results that do not conform to the actual situation. Third, existing methods often ignore the problem of quality heterogeneity of miners' trajectory data. The data quality collected at different times and in different regions varies significantly, and using it without discrimination will reduce the prediction accuracy. Fourth, existing methods usually regard miners as independent individuals for prediction, ignoring the group cooperation relationship in mine operations, resulting in prediction results that do not conform to the actual operation rules. Finally, there is a lack of a mechanism to directly associate the prediction results with safety management decisions, and it is difficult to transform the prediction results into specific safety scheduling measures. Summary of the Invention

[0004] This application provides a method, system and device for predicting the trajectory of mine personnel based on artificial intelligence, which is used to accurately predict the future trajectories of mine personnel through the constraints of the special mine environment and the working behavior patterns of miners, and transform the prediction results into safety scheduling strategies.

[0005] In a first aspect, the present application provides an artificial intelligence-based mine personnel trajectory prediction method. The artificial intelligence-based mine personnel trajectory prediction method includes: collecting mine personnel positioning data and environmental parameter data, performing standardization processing on the mine personnel positioning data to obtain structured trajectory data; performing quality assessment on the structured trajectory data, calculating trajectory quality through constructing an artificial neural network with two hidden layers to obtain trajectory data; converting the trajectory data and the environmental parameter data into spatio-temporal feature vectors, performing image pattern recognition processing through the artificial neural network to obtain trajectory vectors; applying a rectified linear unit activation function and a Sigmoid function to the trajectory vectors for multi-time scale processing, calculating gradients through a loss function weighted by quality weights to obtain trajectory prediction data; performing constraint matching on the trajectory prediction data and the mine space structure, identifying miner group behavior patterns through a density clustering algorithm based on computer vision to obtain target trajectory prediction data; calculating spatial safety distances, time conflict probabilities, and regional load indicators according to the target trajectory prediction data, and generating a mine personnel safety scheduling strategy.

[0006] In a second aspect, the present application provides an artificial intelligence-based mine personnel trajectory prediction system. The artificial intelligence-based mine personnel trajectory prediction system includes:

[0007] A collection module, configured to collect mine personnel positioning data and environmental parameter data, and perform standardization processing on the mine personnel positioning data to obtain structured trajectory data;

[0008] A calculation module, configured to perform quality assessment on the structured trajectory data, calculate trajectory quality through constructing an artificial neural network with two hidden layers to obtain trajectory data;

[0009] A conversion module, configured to convert the trajectory data and the environmental parameter data into spatio-temporal feature vectors, and perform image pattern recognition processing through the artificial neural network to obtain trajectory vectors;

[0010] A processing module, configured to apply a rectified linear unit activation function and a Sigmoid function to the trajectory vectors for multi-time scale processing, calculate gradients through a loss function weighted by quality weights to obtain trajectory prediction data;

[0011] A matching module, configured to perform constraint matching on the trajectory prediction data and the mine space structure, and identify miner group behavior patterns through a density clustering algorithm based on computer vision to obtain target trajectory prediction data;

[0012] A generation module, configured to calculate spatial safety distances, time conflict probabilities, and regional load indicators according to the target trajectory prediction data, and generate a mine personnel safety scheduling strategy.

[0013] In a third aspect, there is provided a mine personnel trajectory prediction device based on artificial intelligence, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the mine personnel trajectory prediction device based on artificial intelligence to execute the above-mentioned mine personnel trajectory prediction method based on artificial intelligence.

[0014] In a fourth aspect, there is provided a computer-readable storage medium, in which instructions are stored, and when it runs on a computer, it enables the computer to execute the above-mentioned mine personnel trajectory prediction method based on artificial intelligence.

[0015] In the technical solution provided by this application, by collecting mine personnel positioning data and environmental parameter data and performing standardized processing, a structured representation of miners' trajectories is achieved, the quality of the structured trajectory data is evaluated, and an artificial neural network with two hidden layers is applied to calculate the trajectory quality, effectively solving the problem of data quality heterogeneity in the mine environment and improving the reliability of the training data for the prediction model. The trajectory data and environmental parameter data are converted into spatio-temporal feature vectors, and image pattern recognition processing is performed through an artificial neural network, making full use of the advantages of deep learning in spatial pattern recognition and effectively extracting the key features of miners' trajectories. The rectified linear unit activation function and the Sigmoid function are applied to the trajectory vectors for multi-time scale processing, realizing trajectory predictions with three different time spans of short-term, medium-term, and long-term, meeting the requirements of different application scenarios, and the design of calculating the gradient of the loss function weighted by quality weights effectively suppresses the interference of low-quality data on model training. The trajectory prediction data is constraint-matched with the mine spatial structure, and the group behavior patterns of miners are identified through a density clustering algorithm based on computer vision, making the prediction results not only conform to the physical constraints of the mine but also fit the actual work rules of miners, greatly improving the prediction accuracy. Finally, based on the target trajectory prediction data, the spatial safety distance, time conflict probability, and regional load index are calculated, and a mine personnel safety scheduling strategy is generated, realizing the direct conversion of the prediction results into safety management decisions. In this specific application field of mines, this solution fully considers the contributions of the characteristics of artificial intelligence algorithms to the solution, especially the multi-level feature extraction ability of artificial neural networks, the non-linear mapping ability of the rectified linear unit and Sigmoid activation functions, the pattern recognition ability of computer vision, and the group behavior modeling ability of density clustering. Compared with traditional methods, the present invention can not only accurately predict the future trajectories of miners but also identify potential safety risks and automatically generate scheduling strategies, significantly improving the accuracy of mine safety management. Description of the Drawings

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a schematic diagram of an embodiment of the method for predicting the trajectory of mine personnel based on artificial intelligence in the embodiments of the present application;

[0018] Figure 2 It is a schematic diagram of an embodiment of the system for predicting the trajectory of mine personnel based on artificial intelligence in the embodiments of the present application;

[0019] Figure 3 It is a structural schematic block diagram of the device for predicting the trajectory of mine personnel based on artificial intelligence in the embodiments of the present invention. Detailed implementation manners

[0020] The embodiments of the present application provide a method, system and device for predicting the trajectory of mine personnel based on artificial intelligence. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "include" or "have" and any deformation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the following describes the specific process of the embodiments of the present application. Please refer to Figure 1 , an embodiment of the method for predicting the trajectory of mine personnel based on artificial intelligence in the embodiments of the present application includes:

[0022] Step S101, collect the positioning data of mine personnel and environmental parameter data, and perform standardization processing on the positioning data of mine personnel to obtain structured trajectory data;

[0023] Step S102, perform quality evaluation on the structured trajectory data, calculate the trajectory quality by constructing an artificial neural network with two hidden layers, and obtain trajectory data;

[0024] Step S103: Convert the trajectory data and environmental parameter data into spatio-temporal feature vectors, and perform image pattern recognition processing through an artificial neural network to obtain trajectory vectors;

[0025] Step S104: Apply the rectified linear unit activation function and the Sigmoid function to the trajectory vectors for multi-time scale processing, calculate the gradient through a loss function weighted by quality weights, and obtain trajectory prediction data;

[0026] Step S105: Constrainedly match the trajectory prediction data with the mine spatial structure, and identify the group behavior patterns of miners through a density clustering algorithm based on computer vision to obtain target trajectory prediction data;

[0027] Step S106: Calculate the spatial safety distance, time conflict probability, and regional load index according to the target trajectory prediction data, and generate a mine personnel safety scheduling strategy.

[0028] It can be understood that the execution subject of this application can be a mine personnel trajectory prediction system based on artificial intelligence, or a terminal or a server. Specifically, it is not limited here. In this embodiment of the application, the server is used as the execution subject for illustration.

[0029] Specifically, it is completed through a sensor network deployed at preset positions in the mine. These sensors receive the position coordinate data and timestamp data uploaded by the intelligent positioning devices worn by miners. At the same time, the collection of environmental parameter data includes information such as temperature, humidity, gas concentration, and ventilation conditions. The collected original trajectory data is cleaned, mainly targeting three types of abnormal points: coordinate points beyond the physical boundary of the mine, abnormal timestamp points, and coordinate jump points. There are often missing points in the cleaned trajectory data. At this time, a spatio-temporal interpolation algorithm is used for filling. This algorithm considers the continuous movement characteristics of miners and calculates the reasonable coordinates of the missing points according to the spatio-temporal relationship between the front and back position points. The complete trajectory sequence is divided into multiple trajectory segments according to a preset time window. Spatial features and time features are extracted from these trajectory segments to form featureized trajectory data. These data are associated with the functional attributes of the mine area, and semantic annotation is performed on the trajectory points to generate structured trajectory data.

[0030] Quality assessment of structured trajectory data first obtains initial quality parameters through multi-dimensional scoring, including integrity index, accuracy index, regularity index, and semantic consistency index. These initial quality parameters are weighted and combined through a trajectory quality scoring function to generate a normalized quality score. Then, an artificial neural network is constructed. This network has two hidden layers. The input layer has two nodes to receive the basic features of the trajectory data. Each of the two hidden layers has 8 nodes to process feature transformation. The output layer has 1 node to output the quality score. After the structured trajectory data is converted into a two-dimensional spatio-temporal image, it is input into this neural network. Local features are extracted through convolutional operations, and then global features are extracted through pooling operations and fully connected operations to obtain the trajectory quality score. To optimize the scoring accuracy, a trajectory contrast learning task is constructed. Similar scores are generated for trajectories of the same miner and close in time, and different scores are generated for trajectories with significant differences in regularity and periodicity. The network parameters are optimized by minimizing the contrast loss function to obtain a calibrated quality score. Based on this score, quality weights are assigned to the structured trajectory data to generate trajectory data.

[0031] In the process of converting trajectory data and environmental parameter data into spatio-temporal feature vectors, the trajectory data is converted into a time-sequence position lattice in the mine three-dimensional coordinate system, and at the same time, the environmental parameter data is mapped to the attribute values of the corresponding position points. A multi-channel mine trajectory activity heat map is generated through spatial rasterization processing to form a trajectory image matrix. The contrast of this matrix is enhanced to highlight the trajectory features in key areas such as roadways, working faces, and refuge chambers. The enhanced trajectory image is decomposed into four channels: spatial channel, velocity channel, acceleration channel, and environmental parameter channel. After compression and dimensionality reduction processing, a spatio-temporal feature vector is generated. This vector is input into the first hidden layer of the artificial neural network. 8 nodes use 3×3 convolutional kernels to extract local features, identify behavioral features such as the turning pattern, stop point distribution, and speed change of the miner, and form a behavioral feature map. After this map undergoes a non-linear transformation through the rectified linear unit activation function, it is input into the second hidden layer of the neural network. 8 nodes use 5×5 convolutional kernels to extract context information with a larger receptive field, identify the operation rules and path selection patterns of the miner in different mine areas, and obtain context feature representation data. Finally, the mine topology structure constraint is applied to this data, and the three-dimensional information of space, time, and environment is integrated through a fully connected layer to generate a trajectory vector.

[0032] When performing multi-time scale processing on the trajectory vector, the prediction task is divided into three scales: short-term prediction, medium-term prediction, and long-term prediction. The trajectory vector is divided in the time dimension to obtain three sets of prediction input data. The short-term prediction input data is processed by applying the rectified linear unit activation function to capture the local continuity and short-term change patterns of the trajectory. The medium-term prediction input data models the long-distance dependence relationship by calculating the correlation weight matrix at each time point in the trajectory sequence. The long-term prediction input data processes the output of the hidden state through the Sigmoid function to model the long-term uncertainty of the trajectory. Based on the quality weight of the trajectory data, a weighted loss function is constructed to calculate the prediction error, and the gradient is calculated through the backpropagation algorithm and the network parameters are updated to optimize the prediction results. Finally, the prediction results of the three time scales are fused to obtain the trajectory prediction data.

[0033] When performing constraint matching between the trajectory prediction data and the mine spatial structure, the prediction data is compared with the three-dimensional mine structure diagram in terms of spatial coordinates, and the prediction points that do not meet the physical accessibility are removed, including the points located inside the rock formation, the points that cross the safety isolation area, and the points that exceed the working area range, to obtain the spatially constrained trajectory data. Then, the historical trajectories of multiple miners are grouped and sorted according to the work team, work nature, and activity area to form a grouped trajectory set. Density clustering processing based on computer vision is applied to this set, and the trajectory points that are close in space and time are grouped together. The center point and activity range of each group are extracted to generate a behavior density distribution map. Based on this map, the activity patterns of miners in different areas are identified, including the stay pattern at the work point, the inspection route pattern, the material transportation pattern, and the emergency evacuation pattern, to form the behavior pattern data. The spatially constrained trajectory data is matched with the behavior pattern data, and the direction and rhythm of the predicted trajectory are adjusted according to the current miner's position and task type to make it conform to the characteristics of the corresponding behavior pattern, generating the behavior-optimized trajectory. Based on the collaborative relationship between miners in the same work group, the behavior-optimized trajectory is adjusted for group consistency to ensure the spatio-temporal coordination of the predicted trajectories of collaborative miners, obtaining the target trajectory prediction data.

[0034] When generating a safety scheduling strategy based on target trajectory prediction data, the data is grouped and sorted by miner ID and time series, and the position coordinate sequence of each miner in the future time period is calculated to form a personnel distribution prediction map. Based on this map, the minimum spatial distance between any two miners is calculated through the Euclidean distance formula. If it is less than the preset safety threshold, it is marked as a potential dangerous contact point to obtain the spatial safety distance index. Time probability analysis is carried out on the potential dangerous contact points, the probability value of two miners appearing in the same area and the same time period is calculated, and the spatial overlap degree is calculated through the Gaussian distribution model of miner position coordinates to obtain the time conflict probability index. The target trajectory prediction data is mapped onto the mine area division map, the number density of miners in each area at different time periods is counted, and the load rate is calculated through the ratio of the area capacity to the actual number of people to obtain the area load index. According to these three indexes, a multi-objective optimization model is constructed, and they are combined into a safety risk score through the linear weighted method, the risk levels of each area and time period are divided, and a risk distribution map is generated. Based on the risk distribution map, a mine personnel safety scheduling strategy including personnel diversion suggestions, working time adjustment plans and emergency evacuation routes is generated.

[0035] In the embodiments of the present application, by collecting mine personnel positioning data and environmental parameter data and performing standardized processing, a structured representation of miners' trajectories is achieved, the quality of the structured trajectory data is evaluated, and an artificial neural network with two hidden layers is applied to calculate the trajectory quality, effectively solving the problem of data quality heterogeneity in the mine environment and improving the reliability of the training data for the prediction model. The trajectory data and environmental parameter data are converted into spatio-temporal feature vectors, and image pattern recognition processing is performed through an artificial neural network, making full use of the advantages of deep learning in spatial pattern recognition and effectively extracting the key features of miners' trajectories. The rectified linear unit activation function and the Sigmoid function are applied to the trajectory vectors for multi-time scale processing, achieving trajectory predictions for three different time spans: short-term, medium-term, and long-term, meeting the requirements of different application scenarios. The design of calculating the gradient of the loss function weighted by quality weights effectively suppresses the interference of low-quality data on model training. The trajectory prediction data is constraint-matched with the mine spatial structure, and the group behavior patterns of miners are identified through a density clustering algorithm based on computer vision, making the prediction results not only conform to the physical constraints of the mine but also fit the actual working rules of miners, greatly improving the prediction accuracy. Finally, based on the target trajectory prediction data, the spatial safety distance, time conflict probability, and regional load index are calculated, and a mine personnel safety scheduling strategy is generated, realizing the direct conversion of the prediction results into safety management decisions. In this specific application field of mines, this solution fully considers the contributions of artificial intelligence algorithm features to the solution, especially the multi-level feature extraction ability of the artificial neural network, the non-linear mapping ability of the rectified linear unit and the Sigmoid activation function, the pattern recognition ability of computer vision, and the group behavior modeling ability of density clustering. Compared with traditional methods, the present invention can not only accurately predict the future trajectories of miners, but also identify potential safety risks and automatically generate scheduling strategies, significantly improving the accuracy of mine safety management.

[0036] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0037] Collect the position coordinate data and timestamp data uploaded by the intelligent positioning devices worn by miners through the sensor network deployed at preset positions in the mine to obtain the original trajectory data. At the same time, collect the temperature, humidity, gas concentration, and ventilation condition data in the mine area to obtain the environmental parameter data;

[0038] Perform cleaning processing on the original trajectory data to remove coordinate points beyond the physical boundaries of the mine, abnormal timestamp points, and coordinate jump points, obtaining the cleaned trajectory data;

[0039] For the missing data points in the cleaned trajectory data, use a spatio-temporal interpolation algorithm for filling processing to obtain a trajectory sequence;

[0040] The trajectory sequence is segmented according to a preset time window to obtain multiple trajectory segments;

[0041] Spatial features and temporal features are extracted from the multiple trajectory segments, including displacement vectors, velocity, acceleration, steering angle, residence time, and movement frequency, to obtain featureized trajectory data;

[0042] The featureized trajectory data is associated with the functional attributes of the mine area, and semantic annotation is performed on the trajectory points to obtain structured trajectory data.

[0043] Specifically, data is collected through a sensor network deployed at preset positions in the mine. These sensor networks mainly include positioning signal receivers distributed at key positions such as roadway intersections, working faces, and refuge chambers, which are used to receive signals emitted by intelligent positioning devices worn by miners. The intelligent positioning device is a small device carried by miners and usually integrates an RFID tag, Bluetooth, or UWB (Ultra-Wideband) positioning module, capable of broadcasting the location information of miners in real time. When miners move in the mine with this device, the receivers at preset positions will capture the signals and record the location coordinate data (including three-dimensional coordinates of X, Y, and Z) and the corresponding timestamp data, forming the original trajectory data. At the same time, environmental monitoring sensors arranged in the mine will collect data on temperature, humidity, gas concentration, and ventilation conditions in each area.

[0044] The cleaning process will detect and delete coordinate points that exceed the physical boundaries of the mine. These points are usually caused by signal reflection or interference resulting in positioning errors. The specific operation is to compare each coordinate point with the three-dimensional structure diagram of the mine. If the coordinates of a point fall within the solid rock formation or outside the mine's mining area, it is regarded as an abnormal point and deleted. Secondly, the cleaning process will identify abnormal timestamp points, including cases of time reversal (the timestamp of the latter point is earlier than that of the former point) and time jumps (the time interval between adjacent points is abnormally large). Finally, the cleaning process will detect coordinate jump points, that is, the situation where miners move an unreasonably large distance in a very short time. Such points are usually caused by instantaneous interference of the positioning signal or equipment failure. By calculating the movement speed between adjacent points, if it exceeds the normal walking speed threshold of humans, set to 2 - 3 meters per second, it is determined as an abnormal point and deleted. Although the abnormal points are removed from the trajectory data obtained after the cleaning process, there may be a problem of data loss.

[0045] The missing data points in the cleaned trajectory data are completed using the space-time interpolation algorithm. The space-time interpolation algorithm comprehensively considers spatial continuity and temporal coherence, and infers the position of the missing point based on the known points before and after the missing point. Common interpolation methods include linear interpolation, spline interpolation, etc. In the mine environment, considering the characteristics of the miner's movement path being restricted by the mine structure, space-time interpolation must also be combined with the physical constraints of the mine passage. The specific operation is: determine the time position of the missing point, and then calculate the most likely position coordinates of the miner at that time point based on the coordinates and timestamps of the known points before and after. If the missing time period is long, complex interpolation must be performed in combination with the miner's historical movement pattern and mine path planning. After the interpolation is completed, a continuous and complete trajectory sequence is formed to ensure the temporal continuity of the trajectory data.

[0046] The completed trajectory sequence is segmented according to the preset time window. The size of the time window is usually selected based on the miner's work characteristics and analysis requirements. Common time windows include 10 minutes, 30 minutes or 1 hour. The segmentation process is to divide the complete trajectory into fixed-length segments in chronological order. Each segment contains all the location point information of the miner within the time window. For example, if 30 minutes is selected as the time window, the trajectory of a miner's 8-hour working time will be segmented into 16 trajectory segments. The window boundary problem needs to be handled during segmentation. Spatial features and temporal features are extracted from the multiple trajectory segments obtained by segmentation. The spatial features include displacement vector (direction and distance between two adjacent points), speed (displacement divided by time interval), acceleration (rate of change of speed) and steering angle (angle of change in direction of movement). Temporal features include residence time (duration of speed close to zero in a specific area) and movement frequency (number of moves per unit time). The feature extraction process is actually to convert the original position-time data into a high-level feature set that describes the miner's behavior pattern. For example, for a trajectory segment, the displacement vector between every two adjacent points is calculated, and then the average speed and acceleration per unit time are calculated. At the same time, the time period when the speed is close to zero is detected to identify the stop behavior, and the number of direction changes greater than a specific angle is counted to identify the turning behavior. These features together constitute the characterized trajectory data.

[0047] The purpose of associating the characterized trajectory data with the functional attributes of the mine area and semantically annotating the trajectory points is to increase the semantic understanding of the trajectory data. The functional attributes of the mine area refer to the uses and characteristics of different areas, such as mining working faces, transportation tunnels, safe shelters, equipment maintenance areas, etc. The association process first matches the spatial coordinates of each trajectory point with the mine functional area map to determine the functional area where the point is located. Then, according to the functional attributes of the area and the behavioral characteristics of the miners in the area, semantic labels are added to the trajectory points, such as "working at the working face", "moving along the transportation tunnel", "resting in the shelter", etc. This semantic annotation makes the trajectory data no longer a simple position sequence, but an activity record with business meaning, forming structured trajectory data.

[0048] For example, after the worker entered the mine wearing the smart positioning device, during the process from the locker room to the working face, the sensor network recorded his movement trajectory, including a location coordinate point and corresponding timestamp every 5 seconds, and the environmental sensor recorded the environmental parameters of each area along the way. In the data cleaning stage, the system detected some abnormal points, such as wall penetration points and speed abnormal points (such as moving from the main lane to the refuge room instantly), and removed these points. After cleaning, it was found that the worker had about 1 minute of data missing when passing through a section with weak signals. The system used the spatiotemporal interpolation algorithm, based on the last position point before the missing and the first position point after the missing, combined with the mine channel structure, to calculate the possible path points during the missing period and complete the data. The completed complete trajectory is divided into multiple segments in units of 30 minutes, and features are extracted in each segment, such as calculating the average speed of workers in different areas (0.8 m / s in the main lane and 0.3 m / s in the working face), stop behavior (staying at the working face for 25 minutes continuously), etc. Finally, according to the mine functional area map, semantic annotations are added to the trajectory points to clearly identify workers' movements in the main tunnels, coal mining operations on the working face, short stops in the rest area, and other activities, forming structured trajectory data containing rich semantic information.

[0049] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0050] The structured trajectory data is scored in multiple dimensions to obtain initial quality parameters. The initial quality parameters are weighted and combined according to the importance of each indicator through the preset trajectory quality scoring function to obtain a normalized quality score.

[0051] Construct and train an artificial neural network with two hidden layers, wherein the input layer has two nodes, the two hidden layers each have eight nodes, and the output layer has one node, to obtain an artificial neural network;

[0052] Convert the structured trajectory data into a two-dimensional spatio-temporal image, input it into an artificial neural network, extract local features through convolutional operations, extract global features through pooling operations and fully connected operations, and obtain a trajectory quality score;

[0053] Construct a trajectory contrast learning task, generate similar scores for trajectories of the same miner and close in time, generate different scores for trajectories with significant differences in regularity and periodicity, and optimize the network parameters by minimizing the contrast loss function to obtain a calibrated quality score;

[0054] Based on the calibrated quality score, assign quality weights to the structured trajectory data to obtain the trajectory data.

[0055] Specifically, perform multi-dimensional scoring on the structured trajectory data. The process includes evaluating the trajectory data from four dimensions: integrity, accuracy, regularity, and semantic consistency. The integrity index measures the coverage ratio of the trajectory data in time, which is obtained by calculating the ratio of the number of valid data points to the total working time. Complete trajectory data should cover the entire working process of the miner without large segments missing. The accuracy index measures the position accuracy level of the trajectory data and is evaluated by analyzing the deviation degree between the positioning points and the known reference points in the mine. Accurate trajectory data should accurately reflect the actual position of the miner. The regularity index measures the periodic intensity reflected in the trajectory data and is evaluated by analyzing the repetitive patterns and time regularities of the trajectory. Highly regular trajectories usually reflect the stable work behavior patterns of the miner. The semantic consistency index measures the matching degree between the trajectory data and the miner's work tasks and is evaluated by comparing the degree of conformity between the trajectory activity area and the pre-determined work plan. High semantic consistency indicates that the trajectory data can accurately reflect the actual work content of the miner.

[0056] The measurement results of these four types of indicators form the initial quality parameters, and then these parameters are weighted and combined through a preset trajectory quality scoring function. The weighting process considers the importance weights of each indicator. Usually, the weights of integrity and accuracy are relatively high because these are basic quality features; the weights of regularity and semantic consistency are relatively low, but their weights may be increased in specific analysis tasks. The result after weighted combination is normalized to map the quality score to the range of 0 to 1, where 1 represents the highest quality and 0 represents the lowest quality. The normalization process ensures the comparability of the quality scores of different trajectories.

[0057] The structure of the artificial neural network includes an input layer, two hidden layers, and an output layer. The input layer has two nodes that respectively receive the basic features of the trajectory, such as the number of trajectory points and the coverage time length. Each of the two hidden layers has 8 nodes, which are responsible for feature transformation and non-linear mapping. The output layer has 1 node that outputs the final quality score value. The network is trained using the supervised learning method, using existing high-quality trajectory samples as positive examples and low-quality trajectory samples as negative examples. The network parameters are optimized through the backpropagation algorithm, enabling the network to accurately distinguish trajectory data of different quality levels. The artificial neural network after training can perform quality scoring on new trajectory data without manually calculating various indicators.

[0058] Converting structured trajectory data into a two-dimensional spatio-temporal image is to facilitate neural network processing. The conversion process maps the three-dimensional spatial trajectory and one-dimensional time information onto a two-dimensional plane to form a spatio-temporal image. Specifically, the time is taken as the horizontal axis and the spatial position as the vertical axis, and the trajectory points are mapped to pixel points on the image, and the continuous trajectory forms lines or curves in the image. Different colors or brightness levels can represent different motion states or environmental parameter values. This image representation makes the spatial pattern and time pattern visible simultaneously, facilitating feature extraction by the convolutional neural network. The converted two-dimensional spatio-temporal image is input into the artificial neural network. First, local features, such as movement patterns and turning features in a short period of time, are extracted through the convolutional layer. The convolutional operation uses multiple different filters to slide on the image to identify pattern features such as edges and textures in the image. Then, the feature map is dimension-reduced through the pooling layer, reducing the computational amount while retaining key information. Global features are extracted through the fully connected layer, integrating the relationships between local features to form a comprehensive evaluation of the overall trajectory quality and outputting the trajectory quality score.

[0059] Constructing a trajectory contrast learning task is an unsupervised learning method for evaluating trajectory quality in the absence of explicit quality labels. The core idea of contrastive learning is that trajectories of the same miner under similar conditions should have similar quality characteristics, while trajectories of different miners or under significantly different conditions may have different quality characteristics. In specific implementation, first define the trajectory similarity criterion: Trajectory pairs of the same miner and close in time (such as different time periods of the same day) are regarded as positive sample pairs, and they should obtain similar quality scores; Trajectory pairs with significant differences in regularity and periodicity (such as trajectories on normal working days and equipment failure days) are regarded as negative sample pairs, and they should obtain different quality scores. For each pair of trajectories, calculate their feature representations and quantify the differences between them through a contrastive loss function. The design goal of the contrastive loss function is to minimize the distance between positive sample pairs and maximize the distance between negative sample pairs. By minimizing this loss function, continuously optimize the neural network parameters so that the network can learn a more accurate trajectory quality evaluation criterion and obtain calibrated quality scores. Assigning quality weights to structured trajectory data based on the calibrated quality scores is a key step in applying the quality evaluation results to actual data processing. The quality weight assignment uses a non-linear mapping function to convert the calibrated quality scores into weighting coefficients. A commonly used mapping function is the exponential function, which makes high-quality trajectories obtain exponentially higher weights and greatly reduces the impact of low-quality trajectories. This weighting strategy ensures that subsequent trajectory analysis and prediction are mainly based on high-quality data and reduces the interference of low-quality data. The structured trajectory data after assigning quality weights forms trajectory data.

[0060] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0061] Convert the trajectory data into a time-sequence position lattice in the mine three-dimensional coordinate system, map the environmental parameter data to the attribute values of the corresponding position points, and generate a multi-channel mine trajectory activity heat map through spatial rasterization processing to obtain a trajectory image matrix;

[0062] Perform contrast enhancement on the trajectory image matrix to obtain an enhanced trajectory image. The contrast enhancement is used to enhance the trajectory features of roadways, working faces, and emergency shelters;

[0063] Decompose the enhanced trajectory image into spatial channels, velocity channels, acceleration channels, and environmental parameter channels, perform compression and dimensionality reduction processing to generate spatio-temporal feature vectors;

[0064] Apply a 3×3 convolutional kernel to the spatio-temporal feature vectors through 8 nodes of the first hidden layer of the artificial neural network for local feature extraction, and identify miner behavior features including turning patterns, stop point distributions, and speed changes to obtain a behavior feature map;

[0065] The behavioral feature mapping is non-linearly transformed through the rectified linear unit activation function and input into the second hidden layer of the artificial neural network. The context information with a larger receptive field is extracted by 8 nodes using a 5×5 convolutional kernel to identify the operation rules and path selection patterns of miners in different mine areas, and the context feature representation data is obtained;

[0066] Apply the mine topological structure constraint to the context feature representation data, and integrate the three-dimensional information of space, time, and environment through the fully connected layer to obtain the trajectory vector.

[0067] Specifically, the trajectory data contains the position information of miners at different time points, usually represented by the quadruple (x, y, z, t), where (x, y, z) are the coordinates of the miners in the three-dimensional space of the mine, and t is the corresponding timestamp. The time-series position lattice refers to arranging these discrete position points in chronological order to form a spatio-temporal associated point set. At the same time, the environmental parameter data (including temperature, humidity, gas concentration, and ventilation conditions) needs to be mapped to the corresponding position points as the attribute values of the position points. This mapping establishes the association between the position and the environmental conditions, so that each position point carries not only the position information but also the environmental parameter information. Through spatial grid processing, the continuous three-dimensional space is discretized into regular three-dimensional grids, and each grid cell corresponds to a small area in the mine space. This process is similar to dividing the mine space into three-dimensional "pixels". When a position point falls into a certain grid cell, the value of the cell will increase, and multiple position points falling into the same cell will accumulate the value of the cell to form a heat value. Different types of data (such as position, speed, various environmental parameters) are rasterized on different "channels" respectively, and finally a multi-channel mine trajectory activity heat map, that is, a trajectory image matrix, is formed. This matrix is a three-dimensional array, where the first two dimensions represent the spatial coordinates, and the third dimension represents different data channels.

[0068] The contrast enhancement of the trajectory image matrix is to highlight the trajectory features of key areas. Contrast enhancement is an image processing technique that makes important features in the image more obvious by adjusting the distribution of pixel values. In mine trajectory analysis, special attention is paid to the trajectory features of key areas such as roadways, working faces, and refuge chambers, because these areas are the main places where miners are active and contain important behavioral information. Specific methods of contrast enhancement include histogram equalization, contrast stretching, or adaptive contrast enhancement, etc. Histogram equalization makes the image have a uniformly distributed gray histogram by redistributing the gray levels of the image pixel values; contrast stretching expands the range of pixel values through linear or non-linear mapping; adaptive contrast enhancement adopts different enhancement strategies according to the characteristics of different regions of the image. Through contrast enhancement processing, the trajectory features of the roadway, working face, and refuge chamber areas become more distinct, forming an enhanced trajectory image. Decomposing the enhanced trajectory image into multiple channels is to process different types of information separately. The spatial channel records the position distribution of miners, showing the activity range and hot spots of miners in the mine; the speed channel records the moving speed of miners, reflecting the activity intensity in different regions; the acceleration channel records the speed change, which can capture key behaviors such as starting, stopping, and turning; the environmental parameter channel records the environmental conditions at each position, correlating the relationship between activities and the environment. These high-dimensional data usually contain a large amount of redundant information and need to reduce the data dimension through compression and dimensionality reduction processing. Commonly used dimensionality reduction methods include principal component analysis (PCA), linear discriminant analysis (LDA), or autoencoders, etc. PCA projects high-dimensional data into a low-dimensional space by finding the main directions of data variation (principal components); LDA takes into account class information and finds the projection direction that can best distinguish different classes; autoencoders learn the compressed representation of data through neural networks. After dimensionality reduction, the resulting low-dimensional representation is called a spatio-temporal feature vector, which retains the main information of the original data while greatly reducing the amount of data.

[0069] Local feature extraction of the spatio-temporal feature vector is performed through the first hidden layer of the artificial neural network. The first hidden layer contains 8 nodes, and each node corresponds to a 3×3 convolutional kernel. The convolutional kernel is a small 3×3 matrix that can detect local patterns by sliding over the input data and calculating the convolution operation. The 3×3 size enables the convolutional kernel to capture feature changes in a small range, such as local structures like corners and intersections. The convolution operation is to perform element-wise multiplication of the convolutional kernel with a sub-region of the input data and sum them to obtain the value at the corresponding position in the output feature map. Eight different convolutional kernels can detect eight different local patterns, such as vertical edges, horizontal edges, diagonals, etc. In this way, the neural network can identify behavioral features such as the turning patterns of miners (such as turning behaviors at roadway intersections), the distribution of stopping points (such as long stays at the working face), and speed changes (such as deceleration behaviors near the refuge chamber). These local features together constitute the behavioral feature map, which is a new multi-channel feature map, and each channel corresponds to a detected local behavioral pattern.

[0070] The non-linear transformation of the behavioral feature map through the rectified linear unit activation function is to enhance the expression ability of the network. The rectified linear unit (ReLU) is an activation function defined as f(x)=max(0,x), that is, it returns zero for negative inputs and its own value for positive inputs. The main role of ReLU is to introduce non-linearity, enabling the network to learn more complex patterns; at the same time, it is computationally simple and has stable gradients, which is beneficial for the training of deep networks. The feature map after ReLU processing retains the positive value features (indicating detected patterns) while suppressing the negative value features (indicating irrelevant patterns). These processed features are input into the second hidden layer of the artificial neural network, which also contains 8 nodes but uses larger 5×5 convolutional kernels. The receptive field of the 5×5 convolutional kernel is larger and can capture a wider range of context information. The receptive field refers to the size of the region in the input feature map corresponding to a point in the output feature map. A larger receptive field means that more information about surrounding pixels can be considered. Through these larger convolutional kernels, the second hidden layer can identify more advanced patterns, such as the operation rules of miners in different mine areas (such as the cyclic operation mode at the working face) and path selection patterns (such as the preferred choice among multiple paths). These advanced features form the context feature representation data, which not only contains local behavioral information but also the associations of these behaviors in a broader spatial and temporal context.

[0071] Applying mine topological structure constraints to the context feature representation data is to ensure that the generated trajectory vectors conform to the limitations of the mine physical space. The mine topological structure refers to the spatial layout of the mine, including the connection relationships of roadways, impassable areas, and the distribution of each functional area. Taking this topological structure as a constraint condition can exclude physically impossible trajectory predictions (such as passing through walls, crossing obstacles, etc.). In terms of specific implementation, the mine structure is usually represented as a Navigation Graph, where nodes represent key positions (such as intersections, functional areas), and edges represent passable paths. The context features are combined with the navigation graph through a fully connected layer to achieve further integration and refinement of the features. A fully connected layer is a type of layer in a neural network where each input is connected to each output, and it can learn the global combination of input features. In this way, the system can integrate spatial information (where the miner is), temporal information (when to arrive), and environmental information (what the surrounding environment is like) to generate a trajectory vector that comprehensively considers these three dimensions. This trajectory vector is a highly abstract and compressed representation that contains the key features and behavior patterns of the miner's trajectory.

[0072] For example: The movement of a worker in the mine from the main shaft to each equipment maintenance point within a day forms the original trajectory data. These trajectory data are converted into a time-series position lattice in the mine three-dimensional coordinate system, and at the same time, parameters such as temperature and methane concentration measured by environmental sensors along the way are mapped to the corresponding position points. Then, through spatial rasterization processing, the mine space is divided into small cubes of 10 cm × 10 cm × 10 cm, and a heat value is assigned to each cube according to the frequency of the worker passing by, forming a multi-channel heat map matrix. Next, contrast enhancement is performed on this heat map matrix to make the trajectory features in key areas such as equipment maintenance points and main channels more prominent. The enhanced heat map is decomposed into a spatial channel (reflecting the position distribution), a velocity channel (reflecting the moving speed), an acceleration channel (reflecting the speed change), and an environmental parameter channel (reflecting the environmental conditions at each point). Through principal component analysis for dimensionality reduction, a 100-dimensional spatio-temporal feature vector is obtained. This vector is input into the first hidden layer of an artificial neural network, and 8 3×3 convolutional kernels respectively detect different local features, such as the staying behavior of the worker at the equipment point, the turning behavior at the roadway intersection, etc., to form a behavior feature map. After being processed by the ReLU activation function, this map is input into the second hidden layer, and 8 5×5 convolutional kernels further extract more advanced features, such as the inspection route pattern of the worker, the operation rules in different areas at different times, etc., to obtain the context feature representation. Finally, these features are combined with the mine navigation graph, and all information is integrated through a fully connected layer to generate a trajectory vector that can accurately describe the behavior rules of this maintenance worker.

[0073] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0074] The prediction time is divided into three scales: short-term prediction, medium-term prediction, and long-term prediction. The trajectory vector is divided in the time dimension to obtain three sets of prediction input data;

[0075] The short-term prediction input data is input into the temporal convolutional network structure. By applying the rectified linear unit activation function to process the output of each convolutional layer, the local continuity and short-term change patterns of the trajectory are captured to obtain the short-term trajectory prediction result;

[0076] The medium-term prediction input data is input into the self-attention mechanism network structure. By calculating the correlation weight matrix of each time point in the trajectory sequence, the long-distance dependence relationship is modeled to obtain the medium-term trajectory prediction result;

[0077] The long-term prediction input data is input into the recurrent neural network structure. By processing the output of the hidden state through the Sigmoid function, the long-term uncertainty of the trajectory is modeled to obtain the long-term trajectory prediction result;

[0078] Based on the quality weight of the trajectory data, a weighted loss function is constructed. The prediction errors are calculated for the short-term trajectory prediction result, medium-term trajectory prediction result, and long-term trajectory prediction result respectively. By calculating the gradient and updating the network parameters, the target prediction result is obtained;

[0079] The target prediction result is fused, and the prediction points that do not meet the physical reachability are filtered through the mine space constraint to obtain the trajectory prediction data.

[0080] Specifically, dividing the prediction time into three scales of short-term prediction, medium-term prediction, and long-term prediction is an important strategy in the mine personnel trajectory prediction method. This multi-scale prediction framework can provide a dedicated processing mechanism for prediction requirements with different time spans. Specifically, short-term prediction usually refers to the trajectory prediction for the next 5 - 15 minutes, mainly focusing on the position where the miner is about to arrive; medium-term prediction refers to the trajectory prediction for the next 15 - 60 minutes, focusing on the movement path of the miner within a shift; long-term prediction refers to the trajectory prediction for the next 1 - 4 hours, focusing on the activity pattern of the miner throughout the work shift. Dividing the trajectory vector in the time dimension means splitting the trajectory vector into three different sets of prediction input data according to the time length of the historical data and the time span of the prediction target. For example, for short-term prediction, the trajectory data of the last 10 minutes may be used as input; for medium-term prediction, the data of the last 30 minutes may be used; for long-term prediction, the data of the last 2 hours may be used. This division takes into account the differences in the dependence on historical information for prediction tasks at different time scales. Short-term prediction is more dependent on recent behaviors, while long-term prediction requires longer historical information to identify patterns.

[0081] Feeding the short-term prediction input data into the Temporal Convolutional Network (TCN) structure is the core step of short-term trajectory prediction. The Temporal Convolutional Network (TCN) is a variant of convolutional neural network specialized for processing temporal data. Its feature is to use one-dimensional convolutional operations to process time series and expand the receptive field through causal convolution and dilated convolution techniques. Causal convolution ensures that the prediction only depends on the current and past inputs and is not affected by future data; dilated convolution enables the network to efficiently increase the receptive field and capture longer-range temporal dependencies by inserting holes in the convolutional kernel. In the prediction of mine personnel trajectories, the Temporal Convolutional Network receives the short-term prediction input data and extracts the temporal features of the trajectory through multiple convolutional operations. The output of each convolutional layer is processed by the Rectified Linear Unit (ReLU) activation function, which can retain useful positive features and set irrelevant negative features to zero, enhancing the expressive power of the network. In this way, the Temporal Convolutional Network can effectively capture the local continuity of the trajectory (such as continuous movement along the same roadway) and short-term change patterns (such as round-trip movement at the working face), and finally output the short-term trajectory prediction results, that is, the possible position sequence of miners within the next 5 - 15 minutes. Feeding the medium-term prediction input data into the self-attention mechanism network structure is a processing method for longer-term range prediction. The self-attention mechanism is a technique that can calculate the relationships between elements within a sequence. It generates a weight matrix representing these relationships by learning the association strength between each element in the sequence and all other elements. In the prediction of mine personnel trajectories, the self-attention mechanism allows the model to consider the long-range dependencies between different time points in the trajectory sequence, which is crucial for capturing the movement patterns of miners over a longer period. The specific operation is to feed the medium-term prediction input data (such as the trajectory of the past 30 minutes) into the self-attention network. The network first calculates the query vector (Query), key vector (Key), and value vector (Value) for each time point in the trajectory sequence, then calculates the correlation score through the dot product of the query vector and the key vector, converts the score into weights through the softmax function, and finally performs weighted summation on the value vector with these weights to obtain a representation considering the long-range dependencies. This mechanism can identify trajectory points that are far apart in time but have strong correlations, such as the behavior pattern of miners regularly returning to a certain area to check equipment, thus generating more accurate medium-term trajectory prediction results.

[0082] Feeding long-term prediction input data into a recurrent neural network structure is a method for handling the longest time range prediction. A recurrent neural network (RNN) is a neural network specialized for processing sequential data, characterized by its ability to transmit information through internal states (memory) to handle sequences of arbitrary length. In mine personnel trajectory prediction, long-term prediction needs to consider the activity patterns of miners over an entire work shift or even multiple shifts. Traditional RNNs are difficult to handle such long sequence dependencies, so improved RNN variants such as long short-term memory networks (LSTMs) or gated recurrent units (GRUs) are usually adopted. These variants can selectively remember important information and forget irrelevant information by introducing a gating mechanism, thus effectively handling long sequences. The long-term prediction input data (such as the trajectories of the past 2 hours) is fed into the recurrent neural network, and the network continuously updates its internal state as time steps progress, accumulating historical information. Finally, the hidden state of the network is processed through the Sigmoid function to compress the output value between 0 and 1. This processing method helps to model the long-term uncertainty of the trajectory because the uncertainty of long-term prediction is usually higher and needs to be expressed in a probabilistic way. In this way, the recurrent neural network can generate trajectory prediction results that take into account long-term behavior patterns.

[0083] Each trajectory is assigned a quality weight that reflects the reliability of its data quality. These quality weights are now used to construct a weighted loss function, imposing different learning importance on trajectory samples of different qualities. Specifically, for the three prediction results of short-term, medium-term, and long-term, the errors between the predicted values and the true values are calculated respectively. Commonly used error metrics include the mean absolute error (MAE), mean squared error (MSE), or mean absolute percentage error (MAPE), etc. Then, these errors are weighted by the quality weights, making the errors from high-quality trajectories contribute more to the total loss, thus guiding the model to learn more about the patterns in high-quality samples. The weighted loss function calculates the gradients through optimization algorithms such as gradient descent and updates the network parameters, continuously adjusting the model to reduce the weighted loss, and finally obtaining the optimized target prediction result based on high-quality data. Short-term prediction, medium-term prediction, and long-term prediction each have their advantages: short-term prediction is more accurate for recent predictions, medium-term prediction can capture longer dependencies, and long-term prediction is better at grasping the overall patterns. The fusion process is to comprehensively consider these three prediction results to generate the final predicted trajectory. Commonly used fusion methods include weighted average, ensemble learning, or multi-task learning, etc. In the weighted average method, the three prediction results are combined according to pre-set weights or weights dynamically calculated based on the current prediction reliability; in the ensemble learning method, the three predictions are regarded as different "expert opinions" and integrated through voting or averaging; in the multi-task learning method, the three prediction tasks share the underlying features but have their own prediction heads and achieve complementarity through joint optimization. The predicted results obtained by fusion also need to be filtered through the mine space constraints to eliminate the prediction points that do not meet physical accessibility. These constraints come from the three-dimensional structure diagram of the mine, ensuring that the predicted trajectories do not show physically impossible situations such as passing through walls, crossing obstacles, or having unreasonable speeds. After being filtered by the constraints, the final trajectory prediction data that conforms to physical laws is obtained.

[0084] For example: The worker performs cyclic operations at the coal mining face every day, and the working mode is relatively fixed. First, trajectory vectors are extracted from historical trajectory data and then divided into three groups according to the prediction time scale: the short-term prediction uses the trajectory vectors of the most recent 15 minutes, the medium-term prediction uses the trajectory vectors of the most recent 45 minutes, and the long-term prediction uses the trajectory vectors of the most recent 2 hours. For short-term prediction, the trajectory vectors of the most recent 15 minutes are input into a temporal convolutional network, which processes through multiple layers of convolution and ReLU activation functions to capture the local movement patterns of the worker on the coal mining face, such as the round-trip movement from one end to the other, and predicts the possible position changes of the worker within the next 10 minutes. For medium-term prediction, the trajectory vectors of the most recent 45 minutes are input into a self-attention network, which calculates the correlation weight matrix between different time points to identify the pattern that the worker returns to the support control area for adjustment every about 30 minutes, and predicts the activity trajectory in the next 30 minutes. For long-term prediction, the trajectory vectors of the most recent 2 hours are input into an LSTM network, which accumulates historical information through memory units to identify the changes in the working rhythm of the worker during the entire shift, such as the frequent activities in the first half of the shift and the slightly slower pace in the second half, and predicts the approximate activity range for the remaining working time. These three prediction results are trained and optimized by constructing a loss function based on the trajectory quality weight. For example, since the signal at the coal mining face is good and the data quality is high, its trajectory obtains a higher quality weight and accounts for a greater proportion in the loss calculation. Finally, the three optimized prediction results are fused. The short-term prediction contributes the most to the positions of recent points, the medium-term prediction contributes the most to the positions around 30 minutes, and the long-term prediction provides guidance for the overall trend. The fused trajectory also needs to pass through the mine structure constraints to ensure that the predicted path only follows the actual existing roadways and working faces and does not pass through walls or cross obstacles, obtaining trajectory prediction data that conforms to physical laws.

[0085] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0086] Compare the spatial coordinates of the trajectory prediction data with the three-dimensional mine structure diagram, and eliminate the prediction points that do not conform to the mine channels, including the points inside the rock formation, the points crossing the safety isolation area, and the points exceeding the working area range, to obtain spatially constrained trajectory data;

[0087] Classify and organize the historical trajectories of multiple miners in the spatially constrained trajectory data, and group them according to work teams, work natures, and activity areas to obtain a grouped trajectory set;

[0088] Apply density clustering processing based on computer vision to the grouped trajectory set, cluster the trajectory points with similar time and space into one group, and extract the center point and activity range of each group to obtain a behavior density distribution map;

[0089] Based on the behavior density distribution map, identify the activity patterns of miners in different areas, including the stay patterns at working points, inspection route patterns, material transportation patterns, and emergency evacuation patterns, to obtain behavior pattern data;

[0090] Match the spatial constraint trajectory data with the behavior pattern data, and adjust the direction and rhythm of the predicted trajectory according to the current miner's position and task type to make it conform to the characteristics of the corresponding behavior pattern, obtaining an optimized behavior trajectory;

[0091] Based on the collaborative relationship among miners in the same work group, make a group consistency adjustment to the optimized behavior trajectory to ensure the spatio-temporal coordination of the predicted trajectories of collaborative miners, obtaining target trajectory prediction data.

[0092] Specifically, the three-dimensional mine structure diagram is a digital representation of the mine space, including the spatial layout and geometric shapes of various functional areas such as roadways, working faces, and refuge chambers, as well as the location information of non-passable areas such as rock strata and safety isolation zones. In the spatial coordinate comparison process, the three-dimensional coordinates (x, y, z) of each predicted point are first compared with the channel positions in the mine structure diagram to check whether the point falls within a legal passage area. For predicted points located inside the rock strata, that is, points whose coordinates fall within the solid rock area, these points clearly violate the physical laws because miners cannot pass through walls and must move along the excavated roadways and working faces. For predicted points that cross the safety isolation area, that is, points whose coordinates pass through the restricted access areas set for safety reasons (such as high gas concentration, unstable roof), these points are safety violations and miners should not enter these areas under actual circumstances. For predicted points that exceed the working area range, that is, points whose coordinates exceed the normal working activity range of miners, these points may be unreasonable predictions caused by model extrapolation. By removing these three types of predicted points that do not conform to the physical constraints of the mine, the remaining predicted points form spatial constraint trajectory data, which represent physically feasible miner movement paths. Classifying and organizing the historical trajectories of multiple miners in the spatial constraint trajectory data is to identify the behavior patterns of different types of miners. First, group according to work teams, aggregating the trajectories of miners in the same shift (such as the morning shift, middle shift, night shift) together because miners in the same shift usually have similar work schedules and collaborative relationships. Second, group according to work nature, classifying the trajectories of miners with the same job type (such as coal miners, support workers, mechanical and electrical maintenance workers) into one category because miners with the same job type perform similar work tasks and exhibit similar movement patterns. Finally, group according to activity areas, grouping the trajectories of miners mainly active in the same area (such as a certain working face, a certain section of the transportation roadway) into one group because the regional characteristics will affect the movement behavior of miners. The multi-dimensional classification and organization form a grouped trajectory set.

[0093] Applying computer vision-based density clustering processing to a set of grouped trajectories is a method for identifying trajectory patterns using visual analysis techniques. The core of density clustering processing is the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm, which groups data points based on the density distribution of points, can discover clusters of arbitrary shapes, and can identify noise points. In trajectory analysis, spatio-temporal trajectory points are first mapped into a two-dimensional or three-dimensional space, and points that are spatio-temporally close appear as aggregation regions in the space. The DBSCAN algorithm is based on two parameters: ε (neighborhood radius) and MinPts (minimum number of points to form a cluster). The algorithm traverses each point. If the ε-neighborhood of a point contains at least MinPts points, it is regarded as a core point; if a point is within the ε-neighborhood of a certain core point but is not a core point itself, it is regarded as a border point; a point that is neither a core point nor a border point is regarded as a noise point. By connecting core points that are density-reachable, different cluster groups are formed. For each cluster group, its center point (the average coordinate of all points) and activity range (the spatial distribution range of points) are calculated to form a behavior density distribution map. This distribution map visually shows the concentration and activity boundaries of miners' activities in different areas of the mine. By analyzing the spatial position, density change, and time series characteristics of the cluster distribution, typical behavior patterns can be identified. The stay pattern at the work point is manifested as a high-density aggregation area formed at a specific location (such as the operation position of the shearer, the support operation area). The trajectory points in these areas have a high density and a long time span, indicating that miners have carried out fixed operations here for a long time. The inspection route pattern is manifested as a continuous linear trajectory distribution with a moderate density, covering multiple devices or areas, reflecting the behavior of miners inspecting and monitoring equipment along a predetermined route. The material transportation pattern is manifested as a round-trip trajectory formed between the material storage point and the usage point, usually showing a regular cyclic pattern, reflecting the behavior of material handling and distribution. The emergency evacuation pattern is manifested as a fast-moving trajectory from the working area to the safety exit or the refuge chamber, with a relatively fast speed and a relatively direct path, reflecting the evacuation behavior in case of emergency. By identifying and classifying these patterns, behavior pattern data is formed, including the feature descriptions and parameter settings of different types of behaviors.

[0094] Based on the location information and task information of the current miner (obtained from the shift scheduling system or work task sheet), determine the most likely behavior pattern that the miner is currently executing. For example, if the miner is located at the coal mining face and the task is to operate the coal shearer, then their behavior pattern may be the "staying pattern at the coal mining operation point". Then, compare the predicted trajectory with the characteristics of this behavior pattern. If the predicted trajectory deviates from the typical characteristics of the behavior pattern (such as the predicted trajectory does not show the expected staying time at the operation point), adjustments are made. The adjustments include correcting the direction of the predicted path to conform to the typical path selection of the behavior pattern; adjusting the moving speed and rhythm to conform to the typical activity characteristics of the behavior pattern; increasing or decreasing the staying time at key points to conform to the operation rules. Through this matching and adjustment process, the predicted trajectory is made to better conform to the actual work behavior characteristics of the miner, forming an optimized behavior trajectory. Mine operations are usually team work, and there are collaborative relationships among miners in the same work group. For example, the timbering workers and coal miners need to coordinate and cooperate, and the haulage workers and loading workers need to be temporally connected. The group consistency adjustment first constructs the relationship network of miners within the work group, defines the types of collaboration (such as front-back connection, simultaneous cooperation, mutual support, etc.) and constraint conditions (such as time difference, distance requirements, etc.) among members. Then, check whether the optimized behavior trajectory satisfies these collaborative relationships. For example, if the predicted trajectories of two collaborating miners show that they will arrive at the location where they should work simultaneously at different times, then the time arrangement needs to be adjusted to make them synchronous; if the predicted trajectory shows that the distance between the collaborating miners exceeds the effective collaboration range, then the position needs to be adjusted to maintain an appropriate working distance. Through this group consistency adjustment, ensure that the predicted trajectories of each miner in the team are coordinated with the trajectories of other collaborating members in terms of time and space, and finally form the target trajectory prediction data that conforms to the actual operation rules.

[0095] For example: The workgroup consists of 4 miners, namely shearer operator A, support workers B and C, and safety monitor D. First, the preliminary trajectory prediction data generated by the model is compared with the mine three-dimensional structure diagram, and it is found that there are problems with some prediction points: 3 prediction points of miner A fall inside the coal wall beside the roadway, 2 prediction points of support worker B pass through the temporarily sealed return airway, and several prediction points of safety monitor D exceed the scope of its responsible area. After removing these unreasonable points, spatial constraint trajectory data that conforms to physical constraints is formed. Then, these trajectory data are classified and sorted according to work shift (all belonging to the early shift), work nature (coal mining, support, and monitoring respectively), and activity area (all in the same working face but responsible for different sections), forming a grouped trajectory set. Then, the DBSCAN algorithm is applied for density clustering processing, with ε set to 5 meters (considering the width of the mine roadway) and MinPts set to 10 (based on the data sampling frequency). The algorithm identifies multiple clusters: a high-density area is formed near the shearer, a strip-shaped distribution is formed in the support operation area, and a regular dot-shaped distribution is formed near the ventilation equipment, which together constitute the behavior density distribution map. Based on this distribution map, several typical behavior patterns are further identified: miner A shows a "reciprocating operation pattern", operating the shearer back and forth along the working face; support workers B and C show a "following support pattern", carrying out support operations at a certain distance behind the shearer; safety monitor D shows a "patrol and monitoring pattern", regularly checking each monitoring point. Matching the spatial constraint trajectory data with these behavior patterns, it is found that some predictions of miner A do not show the proper reciprocating rhythm, and the predicted trajectory of support worker B does not match the coal mining progress and needs to be adjusted. After adjustment, considering the collaborative relationship within the workgroup: miner A and support workers B and C need to maintain a safe distance and the support work needs to follow the coal mining progress, and safety monitor D needs to communicate with the three people regularly. Through group consistency adjustment, it is ensured that the predicted trajectories of the four people are coordinated in time and space, forming the target trajectory prediction data.

[0096] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0097] Group and sort the target trajectory prediction data according to miner ID and time series, calculate the position coordinate sequence of each miner in the future time period, and obtain the personnel distribution prediction map;

[0098] Based on the personnel distribution prediction map, calculate the minimum spatial distance between any two miners through the Euclidean distance formula. If the minimum spatial distance is less than the preset safety threshold, it is marked as a potential dangerous contact point, and the spatial safety distance index is obtained;

[0099] Conduct a time probability analysis on the potential dangerous contact points, calculate the probability value of two miners appearing in the same area and the same time period, and obtain the time conflict probability index;

[0100] Map the target trajectory prediction data onto the mine area division map, count the miner number density in each area during different time periods, calculate the load rate by the ratio of the area capacity to the actual number of people, and obtain the area load index;

[0101] Construct a multi-objective optimization model based on the spatial safety distance index, time conflict probability index, and area load index. Combine the three indexes into a safety risk score through the linear weighted method, divide the risk levels for each area and time period, and obtain the risk distribution map;

[0102] Based on the risk distribution map, generate a mine personnel safety scheduling strategy including personnel diversion suggestions, working time adjustment plans, and emergency evacuation routes. The diversion suggestions are calculated by minimizing the area load variance, the time adjustment plan is generated by the peak-shifting algorithm, and the emergency evacuation route is determined by the shortest path algorithm.

[0103] Specifically, grouping and organizing the target trajectory prediction data according to the miner ID and time series is to organize the prediction results in an orderly manner in terms of individuals and time dimensions. The specific operations include grouping the prediction trajectory data of all miners according to the unique identifier (miner ID), sorting each group in ascending order of the timestamp, and forming a continuous position coordinate sequence of each miner in the future time period (such as the next 2 - 4 hours). These sequences are summarized to form a personnel distribution prediction map, which is a spatio-temporal data structure that records the predicted positions of all miners at each future time point.

[0104] Calculating the minimum spatial distance between any two miners based on the personnel distribution prediction map is an important part of safety management. This process uses the Euclidean distance formula, and the calculation method is as follows:

[0105]

[0106] Among them, represents the minimum spatial distance between miner and miner ; T represents the set of prediction times; , , respectively represent the three-dimensional coordinate values of miner at time t; , , respectively represent the three-dimensional coordinate values of miner Three-dimensional coordinate values at time t. During the calculation process, each time point within the predicted time range is traversed, and the Euclidean distance between the two miners at that time point is calculated. The minimum value among them is taken as the minimum spatial distance between the two. If this minimum distance is less than the preset safety threshold (such as 1.5 meters, set based on epidemic prevention or safety operation requirements), then this spatio-temporal point is marked as a potential dangerous contact point, forming a spatial safety distance indicator. This indicator not only includes the distance value but also the time point and spatial position information where the minimum distance occurs.

[0107] Performing a time probability analysis on potential dangerous contact points is a step to further evaluate the risk. This process calculates the probability value of the two miners appearing in the same area during the same time period, and the calculation method is as follows:

[0108]

[0109] where, represents miner and miner having a conflict probability within the time period in area R; represents the distance between the two at time t; represents the safety distance threshold; and respectively represent the positions of the two at time t; is an indicator function, taking the value 1 when the condition is true and 0 otherwise; represents the length of the time period. This calculation takes two aspects into account: one is whether the distance between the two is less than the safety threshold, and the other is whether the two are in the same area at the same time. By counting the time points that meet the conditions within the predicted time period and dividing by the total time length, a time conflict probability indicator is obtained. This indicator reflects the possibility of the two miners having a close contact within a specific area and time period.

[0110] Mapping the target trajectory prediction data to the mine area division map is the starting point for analyzing the regional load. The mine area division map is a map that divides the mine space into multiple regions according to functional or physical boundaries, such as mining faces, transportation roadways, refuge chambers, etc. The mapping process is to determine the region to which the predicted position of each miner belongs at each time point. Then, the number density of miners in each region within different time periods is counted, that is, the number of people per unit area. Next, the load rate is calculated by taking the ratio of the actual number of people to the regional capacity (the maximum allowable number of people determined according to the regional area and safety regulations). For example, if the capacity of a certain working face is 10 people, and it is predicted that 7 people will be present at the same time during a certain period, then the load rate is 70%. This calculation needs to consider the division granularity of the time period, usually divided by hours or half - hours, to obtain the load rate of each region in each time period, forming the regional load index. Constructing a multi - objective optimization model based on the spatial safety distance index, time conflict probability index, and regional load index is the core of safety risk assessment. A multi - objective optimization model refers to an optimization problem that simultaneously considers multiple objective functions. In this context, the three indexes each represent a kind of safety risk factor. The linear weighted method is a method of linearly combining multiple objective functions into a single objective function. Specifically, different weights are assigned to the three indexes according to their importance, and the weighted sum is used to obtain the comprehensive safety risk score. For example, the weight of the spatial safety distance index is 0.4, the weight of the time conflict probability index is 0.3, and the weight of the regional load index is 0.3, and the sum of the weights is 1. In this way, each region can obtain a comprehensive safety risk score at each time period. According to these scores, and in accordance with the preset risk level division criteria (such as low risk, medium risk, high risk), the risk levels of each region in each time period are divided, generating an intuitive risk distribution map, showing the risk levels of different regions at different times.

[0111] Generating a mine personnel safety scheduling strategy based on the risk distribution map is the final application link. The safety scheduling strategy includes three main components: personnel diversion suggestions, working time adjustment plans, and emergency evacuation routes. Personnel diversion suggestions refer to proposing a personnel re - allocation plan for areas with excessive load to relieve congestion. This suggestion is calculated by minimizing the variance of the regional load, that is, trying to balance the load of each region as much as possible to avoid some regions being too crowded while others being too idle. The working time adjustment plan is for the situation of high time conflicts, and it proposes suggestions for staggering the working times. This plan is generated by a peak - shifting algorithm, which reduces the personnel aggregation within the same time period by adjusting the working times of different work types or teams. Emergency evacuation routes are prepared for emergencies and are used to guide miners to evacuate quickly and safely. These routes are determined by the shortest path algorithm (such as Dijkstra's algorithm), which can find the shortest path from any position to the nearest safety exit, taking into account the physical structure of the mine and the distribution of possible dangerous areas.

[0112] The above describes the method for predicting the trajectory of mine personnel based on artificial intelligence in the embodiments of the present application. Next, the system for predicting the trajectory of mine personnel based on artificial intelligence in the embodiments of the present application will be described. Please refer to Figure 2 An embodiment of the system for predicting the trajectory of mine personnel based on artificial intelligence in the embodiments of the present application includes:

[0113] An acquisition module 201, configured to acquire mine personnel positioning data and environmental parameter data, and perform standardization processing on the mine personnel positioning data to obtain structured trajectory data;

[0114] A calculation module 202, configured to perform quality evaluation on the structured trajectory data, calculate the trajectory quality through constructing an artificial neural network with two hidden layers, and obtain trajectory data;

[0115] A conversion module 203, configured to convert the trajectory data and the environmental parameter data into spatio-temporal feature vectors, perform image pattern recognition processing through the artificial neural network, and obtain trajectory vectors;

[0116] A processing module 204, configured to perform multi-time scale processing on the trajectory vectors by applying a rectified linear unit activation function and a Sigmoid function, calculate the gradient through a loss function weighted by quality weights, and obtain trajectory prediction data;

[0117] A matching module 205, configured to perform constraint matching on the trajectory prediction data and the mine space structure, identify the behavior patterns of miner groups through a density clustering algorithm based on computer vision, and obtain target trajectory prediction data;

[0118] A generation module 206, configured to calculate the spatial safety distance, the time conflict probability, and the regional load index according to the target trajectory prediction data, and generate a safety scheduling strategy for mine personnel.

[0119] Through the collaborative cooperation of the above-mentioned various components, by collecting mine personnel location data and environmental parameter data and performing standardized processing, a structured representation of miners' trajectories is achieved, the quality of the structured trajectory data is evaluated, and an artificial neural network with two hidden layers is applied to calculate the trajectory quality, effectively solving the problem of data quality heterogeneity in the mine environment and improving the reliability of the training data for the prediction model. The trajectory data and environmental parameter data are converted into spatio-temporal feature vectors, and image pattern recognition processing is performed through an artificial neural network, making full use of the advantages of deep learning in spatial pattern recognition and effectively extracting the key features of miners' trajectories. The rectified linear unit activation function and the Sigmoid function are applied to the trajectory vectors for multi-time scale processing, achieving trajectory predictions with three different time spans: short-term, medium-term, and long-term, meeting the requirements of different application scenarios. The design of the loss function for calculating the gradient with quality weights effectively suppresses the interference of low-quality data on model training. The trajectory prediction data is constraint-matched with the mine spatial structure, and the group behavior patterns of miners are identified through a density clustering algorithm based on computer vision, making the prediction results not only conform to the physical constraints of the mine but also fit the actual working rules of miners, greatly improving the prediction accuracy. Finally, based on the target trajectory prediction data, the spatial safety distance, time conflict probability, and regional load index are calculated, and a mine personnel safety scheduling strategy is generated, realizing the direct conversion of the prediction results into safety management decisions. In this specific application field of the mine, this solution fully considers the contributions of artificial intelligence algorithm features to the solution, especially the multi-level feature extraction ability of the artificial neural network, the non-linear mapping ability of the rectified linear unit and Sigmoid activation functions, the pattern recognition ability of computer vision, and the group behavior modeling ability of density clustering. Compared with traditional methods, the present invention can not only accurately predict the future trajectories of miners but also identify potential safety risks and automatically generate scheduling strategies, significantly improving the accuracy of mine safety management.

[0120] Above Figure 2 From the perspective of modular functional entities, the artificial intelligence-based mine personnel trajectory prediction system in the embodiments of the present invention is described in detail. Next, the artificial intelligence-based mine personnel trajectory prediction device in the embodiments of the present invention is described in detail from the perspective of hardware processing.

[0121] Figure 3FIG. 0 is a schematic structural diagram of a mine personnel trajectory prediction device based on artificial intelligence provided by an embodiment of the present invention. The mine personnel trajectory prediction device 300 based on artificial intelligence may vary greatly due to configuration or performance, and may include one or more processors (central processing units, CPUs) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 for storing application programs 333 or data 332 (for example, one or more mass storage device terminals). Among them, the memory 320 and the storage media 330 may be transient storage or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the mine personnel trajectory prediction device 300 based on artificial intelligence. Further, the processor 310 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the mine personnel trajectory prediction device 300 to implement the steps of the above-mentioned mine personnel trajectory prediction method based on artificial intelligence.

[0122] The mine personnel trajectory prediction device 300 based on artificial intelligence may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that Figure 3 The shown structural diagram of the mine personnel trajectory prediction device based on artificial intelligence does not limit the mine personnel trajectory prediction device provided by the present invention, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0123] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are run on a computer, the computer is caused to execute the steps of the mine personnel trajectory prediction method based on artificial intelligence.

[0124] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, systems, and units may refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0125] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing an artificial intelligence-based mine personnel trajectory prediction device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0126] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A mine personnel trajectory prediction method based on artificial intelligence, characterized in that: The method comprises: Collecting mine personnel positioning data and environmental parameter data, and performing standardization processing on the mine personnel positioning data to obtain structured trajectory data; Performing quality assessment on the structured trajectory data, calculating trajectory quality by constructing an artificial neural network with two hidden layers to obtain trajectory data; The trajectory data and the environmental parameter data are converted into a spatiotemporal feature vector, and image pattern recognition processing is performed through the artificial neural network to obtain a trajectory vector; Applying a rectified linear unit activation function and a sigmoid function to the trajectory vector for multi-time scale processing, calculating the gradient through a loss function weighted by a quality weight, and obtaining trajectory prediction data; The trajectory prediction data is constrained and matched with the spatial structure of the mine, and the behavior pattern of the miner group is identified by a density clustering algorithm based on computer vision to obtain the target trajectory prediction data; According to the target trajectory prediction data, the spatial safety distance, time conflict probability and regional load index are calculated to generate a mine personnel safety scheduling strategy.

2. The method for predicting mine personnel trajectories based on artificial intelligence according to claim 1, characterized in that: The collecting of mine personnel positioning data and environmental parameter data, and standardization of the mine personnel positioning data to obtain structured trajectory data include: The sensor network deployed at the preset location of the mine collects the location coordinate data and timestamp data uploaded by the smart positioning devices worn by the miners to obtain the original trajectory data. At the same time, the temperature, humidity, gas concentration and ventilation condition data of the mine area are collected to obtain the environmental parameter data. The original trajectory data is cleaned to remove coordinate points, abnormal time stamp points and coordinate jump points beyond the physical boundary of the mine, so as to obtain cleaned trajectory data; The missing data points in the cleaned trajectory data are supplemented by a spatiotemporal interpolation algorithm to obtain a trajectory sequence; Segmenting the trajectory sequence according to a preset time window to obtain a plurality of trajectory segments; Extracting spatial features and temporal features from the plurality of trajectory segments, including displacement vector, velocity, acceleration, steering angle, dwell time, and movement frequency, to obtain characterized trajectory data; The characterized trajectory data is associated with the functional attributes of the mine area, and the trajectory points are semantically annotated to obtain the structured trajectory data.

3. The method for predicting mine personnel trajectory based on artificial intelligence according to claim 1, characterized in that: The quality assessment of the structured trajectory data is performed by constructing an artificial neural network with two hidden layers to calculate the trajectory quality, thereby obtaining the trajectory data, including: The structured trajectory data is scored in multiple dimensions to obtain initial quality parameters, and the initial quality parameters are weighted and combined according to the importance of each indicator through a preset trajectory quality scoring function to obtain a normalized quality score; Construct and train an artificial neural network with two hidden layers, wherein the input layer has two nodes, the two hidden layers each have eight nodes, and the output layer has one node, to obtain an artificial neural network; The structured trajectory data is converted into a two-dimensional spatiotemporal image, input into the artificial neural network, local features are extracted through convolution operation, global features are extracted through pooling operation and full connection operation, and trajectory quality score is obtained; Construct a trajectory contrast learning task to generate similar scores for the trajectories of the same miner and close in time, and generate different scores for trajectories with significant differences in regularity and periodicity. Optimize network parameters by minimizing the contrast loss function to obtain a calibration quality score. Based on the calibration quality score, a quality weight is assigned to the structured trajectory data to obtain the trajectory data.

4. The method for predicting mine personnel trajectories based on artificial intelligence according to claim 1, characterized in that: The step of converting the trajectory data and the environmental parameter data into a spatiotemporal feature vector and performing image pattern recognition processing through the artificial neural network to obtain a trajectory vector includes: The trajectory data is converted into a time-series position matrix in a three-dimensional coordinate system of a mine, the environmental parameter data is mapped into attribute values ​​of corresponding position points, and a multi-channel mine trajectory activity heat map is generated through spatial rasterization processing to obtain a trajectory image matrix; Performing contrast enhancement on the trajectory image matrix to obtain an enhanced trajectory image, wherein the contrast enhancement is used to enhance trajectory features of the tunnel, working surface and refuge chamber; Decomposing the enhanced trajectory image into a spatial channel, a velocity channel, an acceleration channel and an environmental parameter channel, performing compression and dimension reduction processing, and generating a spatiotemporal feature vector; Applying a 3×3 convolution kernel to the spatiotemporal feature vector through 8 nodes of the first hidden layer of the artificial neural network to perform local feature extraction, identify the miner's behavior features including turning patterns, stop point distribution and speed changes, and obtain a behavior feature map; The behavior feature map is nonlinearly transformed through a modified linear unit activation function and input into the second hidden layer of the artificial neural network. The context information with a larger receptive field is extracted through 8 nodes using a 5×5 convolution kernel to identify the working rules and path selection patterns of miners in different mine areas, and obtain context feature representation data; The mine topology constraint is applied to the context feature representation data, and the three-dimensional information of space, time and environment is integrated through a fully connected layer to obtain the trajectory vector.

5. The method for predicting mine personnel trajectory based on artificial intelligence according to claim 1, characterized in that: The step of applying a modified linear unit activation function and a Sigmoid function to the trajectory vector for multi-time scale processing, calculating a gradient through a loss function weighted by a quality weight, and obtaining trajectory prediction data includes: The prediction time is divided into three scales: short-term prediction, medium-term prediction and long-term prediction, and the trajectory vector is divided into three time dimensions to obtain three sets of prediction input data; The short-term prediction input data is input into the temporal convolutional network structure, and the output of each convolutional layer is processed by applying the rectified linear unit activation function to capture the local continuity and short-term change pattern of the trajectory, and obtain the short-term trajectory prediction result; The mid-term prediction input data is input into the self-attention mechanism network structure, and the long-distance dependency relationship is modeled by calculating the correlation weight matrix of each time point in the trajectory sequence to obtain the mid-term trajectory prediction result; The long-term prediction input data is input into the recurrent neural network structure, and the output of the hidden state is processed by the Sigmoid function to model the long-term uncertainty of the trajectory and obtain the long-term trajectory prediction result; Based on the quality weight of the trajectory data, a weighted loss function is constructed, prediction errors are calculated for the short-term trajectory prediction result, the medium-term trajectory prediction result, and the long-term trajectory prediction result, respectively, and a target prediction result is obtained by calculating the gradient and updating the network parameters; The target prediction results are fused and the prediction points that do not meet the physical accessibility are filtered out through the mine space constraints to obtain the trajectory prediction data.

6. The method for predicting the trajectory of mine personnel based on artificial intelligence according to claim 1, characterized in that: The constraint matching of the trajectory prediction data with the spatial structure of the mine is performed, and the behavior pattern of the miner group is identified by a density clustering algorithm based on computer vision to obtain the target trajectory prediction data, including: Compare the predicted trajectory data with the three-dimensional structure diagram of the mine in spatial coordinates, remove predicted points that do not match the mine passage, including points inside the rock formation, points that cross the safety isolation area, and points that exceed the working range, to obtain spatially constrained trajectory data; Classifying and arranging the historical trajectories of multiple miners in the spatial constraint trajectory data, grouping them according to work teams, work nature and activity areas, and obtaining a grouped trajectory set; Applying density clustering processing based on computer vision to the grouped trajectory set, clustering trajectory points that are similar in time and space into one group, and extracting the center point and activity range of each group to obtain a behavior density distribution map; Based on the behavior density distribution map, the activity patterns of miners in different areas are identified, including the stay pattern at the work point, the inspection route pattern, the material transportation pattern, and the emergency evacuation pattern, to obtain behavior pattern data; Matching the spatial constraint trajectory data with the behavior pattern data, adjusting the direction and rhythm of the predicted trajectory according to the current miner position and task type so that it conforms to the characteristics of the corresponding behavior pattern, and obtaining a behavior optimization trajectory; Based on the collaborative relationship between miners in the same working group, the behavior optimization trajectory is adjusted for group consistency to ensure the spatiotemporal coordination of the predicted trajectory of the collaborative miners, and obtain the target trajectory prediction data.

7. The method for predicting mine personnel trajectory based on artificial intelligence according to claim 1, characterized in that: The method of calculating the spatial safety distance, time conflict probability and regional load index according to the target trajectory prediction data and generating a mine personnel safety dispatch strategy includes: The target trajectory prediction data is grouped and sorted according to the miner ID and time series, and the position coordinate sequence of each miner in the future time period is calculated to obtain a personnel distribution prediction map; Based on the personnel distribution prediction map, the minimum spatial distance between any two miners is calculated using the Euclidean distance formula. If the minimum spatial distance is less than a preset safety threshold, it is marked as a potential dangerous contact point to obtain a spatial safety distance index; Performing a time probability analysis on the potential dangerous contact points, calculating the probability of two miners appearing in the same area and at the same time period, and obtaining a time conflict probability index; Mapping the target trajectory prediction data onto a mine area division map, counting the density of miners in each area in different time periods, calculating the load rate by the ratio of regional capacity to the actual number of people, and obtaining a regional load index; According to the spatial safety distance index, time conflict probability index and regional load index, a multi-objective optimization model is constructed, the three indicators are combined into a safety risk score through a linear weighted method, and the risk level of each area and time period is divided to obtain a risk distribution map; Based on the risk distribution map, a mine personnel safety scheduling strategy is generated, which includes personnel diversion suggestions, work time adjustment plans and emergency evacuation routes. The diversion suggestions are calculated by minimizing the regional load variance, the time adjustment plan is generated by a peak-shifting algorithm, and the emergency evacuation route is determined by a shortest path algorithm.

8. A mine personnel trajectory prediction system based on artificial intelligence, characterized in that: Used to implement the method for predicting the trajectory of mine personnel based on artificial intelligence as described in any one of claims 1 to 7, the system for predicting the trajectory of mine personnel based on artificial intelligence comprises: A collection module is used to collect mine personnel positioning data and environmental parameter data, and perform standardized processing on the mine personnel positioning data to obtain structured trajectory data; A calculation module, used for performing quality assessment on the structured trajectory data, and performing trajectory quality calculation by constructing an artificial neural network with two hidden layers to obtain trajectory data; A conversion module, used for converting the trajectory data and the environmental parameter data into a spatiotemporal feature vector, performing image pattern recognition processing through the artificial neural network to obtain a trajectory vector; A processing module, used for applying a rectified linear unit activation function and a sigmoid function to the trajectory vector to perform multi-time scale processing, and calculating a gradient through a loss function weighted by a quality weight to obtain trajectory prediction data; A matching module is used to perform constraint matching between the trajectory prediction data and the spatial structure of the mine, identify the behavior pattern of the miner group through a density clustering algorithm based on computer vision, and obtain the target trajectory prediction data; The generation module is used to calculate the spatial safety distance, time conflict probability and regional load index according to the target trajectory prediction data, and generate a mine personnel safety scheduling strategy.

9. A mine personnel trajectory prediction device based on artificial intelligence, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for predicting mine personnel trajectories based on artificial intelligence described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor executes the method for predicting mine personnel trajectories based on artificial intelligence as claimed in any one of claims 1 to 7.

Images