A mining face multi-source heterogeneous sensor data fusion processing method and system

Abstract

The application discloses a kind of mining face multi-source heterogeneous sensing data fusion processing method and system, it is related to data processing technical field, including the following steps: obtaining coal mine mining face sensor group multichannel time series data, and carry out signal propagation path analysis, identify potential failure sensor;Potential failure sensor is constructed based on graph model, and error aggregation evaluation is carried out to node using community discovery algorithm, and error amplification node is obtained;The preset jump radius is set to each error amplification node, and multiple rounds of signal compensation adjustment are carried out in range, and jump neighborhood subgraph set is formed;Signal evaluation is carried out to subgraph set using target optimization algorithm, and the optimal compensation adjustment ratio of each error amplification node is determined;Then, according to the optimal proportion dynamic adjustment graph model, generate first error compensation set;Finally, real-time compensation control is carried out to sensor group, effectively solve the problem that compensation caused by single failure sensor is iterated and superimposed in chain, and then lead to far-end sensor data amplification distortion.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method and system for fusing and processing multi-source heterogeneous sensor data from a mining face. Background Technology

[0002] In complex industrial settings such as coal mine faces, numerous sensors are deployed to monitor key environmental factors and equipment operating status in real time, including gas concentration, temperature, humidity, air velocity, roof pressure, equipment vibration, and electrical parameters. These sensors typically form a distributed, multi-source, heterogeneous sensing network, providing data support for safe production and intelligent management. However, due to the harsh environments of high temperature, high humidity, high dust, and strong electromagnetic interference, sensors are prone to problems such as signal drift, decreased sensitivity, data delay, communication interruption, and even complete failure. Especially in the enclosed and rapidly changing environment of underground coal mines, sensors not only face physical structural damage and contamination but may also be affected by external factors such as the start-up and shutdown of motors, gas release fluctuations, or roadway ventilation disturbances, leading to a significant decrease in data stability and reliability.

[0003] Traditional compensation methods are mostly based on single-point correction or linear interpolation of neighboring nodes. When a sensor fails, the system often transfers its missing or abnormal signal to neighboring nodes through a compensation model. However, in the complex network structure of sensor groups, the compensation effect caused by a failed sensor is not simply transmitted linearly, but rather it is superimposed along the network chain through multiple iterations, causing the error to amplify and spread at each level, thus distorting the data from remote sensors. This phenomenon not only reduces the reliability of the overall monitoring data, but may also lead to misjudgments of critical safety risks such as excessive gas levels and abnormal roof pressure.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for fusing and processing multi-source heterogeneous sensor data from a mining face. By realizing dynamic error compensation and real-time control of the sensor group, the problem of amplified and distorted data from remote sensors is solved when a failed sensor triggers compensation.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for fusing and processing multi-source heterogeneous sensor data from a coal mine face includes the following steps: acquiring multi-channel time-series data of a sensor group at the coal mine face, and performing signal propagation path analysis on the multi-channel time-series data to identify potentially failed sensors; constructing a graph model based on the potentially failed sensors, and using a community detection algorithm to evaluate the error aggregation of nodes in the graph model to obtain error amplification nodes; setting a preset jump radius for each error amplification node, and performing several rounds of signal compensation adjustment within the preset jump radius to obtain a jump neighborhood sub-graph set; using a target optimization algorithm to evaluate the signal of the jump neighborhood sub-graph set, and obtaining the optimal compensation adjustment ratio for each error amplification node based on the evaluation results; dynamically adjusting the graph model according to the optimal compensation adjustment ratio for each error amplification node to obtain a first error compensation set; and performing real-time compensation control on the sensor group at the coal mine face based on the first error compensation set.

[0008] Preferably, the step of performing signal propagation path analysis on multi-channel time-series data to identify potentially failed sensors specifically involves: extracting time-domain and frequency-domain features from the multi-channel time-series data stream to obtain the signal feature vector of each sensor; calculating the signal correlation matrix between each pair of sensors based on the signal feature vector of each sensor; extracting sensor data fluctuation features based on the signal correlation matrix, and calculating the fluctuation feature vector of each sensor based on the data fluctuation features; constructing a first abnormal feature set based on the fluctuation feature vectors of all sensors and performing several rounds of clustering analysis on the first abnormal feature set using a hierarchical clustering algorithm; obtaining the physical layout and communication topology data between sensors and constructing a signal propagation path diagram; and evaluating the propagation path of the signal propagation path diagram based on the results of several rounds of clustering analysis to identify potentially failed sensors.

[0009] Preferably, the step of constructing a graph model based on potential failure sensors and using a community detection algorithm to evaluate the error aggregation of nodes in the graph model to obtain error amplification nodes specifically involves: constructing a graph model with potential failure sensors as nodes and using a community detection algorithm to partition the graph model to obtain several communities; performing a first error evaluation on the nodes in each community and analyzing the multi-step propagation path of the error within the community subset based on the evaluation results to obtain the cumulative characteristics of error propagation, wherein the cumulative characteristics include the path length and influence range of error propagation; calculating the first error aggregation degree of each community based on the cumulative characteristics of error propagation and filtering the communities based on the first error aggregation degree to obtain error-aggregated communities; performing a second error evaluation on the nodes in the error-aggregated communities to obtain the second error aggregation degree of each node; and identifying error amplification nodes based on the second error aggregation degree.

[0010] Preferably, the step of setting a preset jump radius for each error amplification node and performing several rounds of signal compensation adjustment within the preset jump radius to obtain a jump neighborhood subgraph set specifically involves: extracting the network topology of each error amplification node and dynamically calculating the jump radius based on the second error aggregation degree; using the error amplification node as the root node, traversing all nodes within the jump radius using a breadth-first search algorithm to obtain a first jump neighborhood subgraph; performing multiple rounds of signal compensation adjustment on the first jump neighborhood subgraph, with each round of adjustment based on a weighted average of the current node's signal value and the neighboring node's signal values; and recording the node signal values ​​in each round of signal compensation adjustment to obtain a jump neighborhood subgraph set.

[0011] Preferably, the step of using a target optimization algorithm to evaluate the signal of the skip neighborhood subgraph set and obtaining the optimal compensation adjustment ratio for each error amplification node based on the evaluation results specifically involves: defining a signal evaluation objective function; using a target optimization algorithm to perform multi-objective optimization on the skip neighborhood subgraph set to generate a candidate compensation adjustment ratio set; applying the candidate compensation adjustment ratio set to the corresponding skip neighborhood subgraphs to obtain several first skip neighborhood subgraphs and obtaining the signal adjustment value of each node in the first skip neighborhood subgraphs to obtain a first signal sequence; dividing the first signal sequence according to a preset signal fluctuation range using a threshold to obtain several first signal segments; performing local mean and deviation analysis on the several first signal segments respectively, and constructing an error accumulation curve; performing data analysis on the error accumulation curve, and obtaining the optimal compensation adjustment ratio for each error amplification node based on the analysis results.

[0012] Preferably, the step of dynamically adjusting the graph model according to the optimal compensation adjustment ratio of each error amplification node to obtain the first error compensation set specifically involves: applying the optimal compensation adjustment ratio of each error amplification node to the graph model, correcting the node signal value of the error amplification node, and obtaining the corrected node signal value; recalculating the edge weights of the graph model based on the corrected node signal value to obtain the adjusted graph model; and traversing all nodes of the adjusted graph model to calculate the error compensation value of each node, thereby obtaining the first error compensation set.

[0013] Preferably, the real-time compensation control of the coal mine face sensor group based on the first error compensation set specifically involves: acquiring the coal mine face sensor group data in real time and loading the first error compensation set; dynamically retrieving the error compensation value of the corresponding node in the first error compensation set based on the coal mine face sensor group data; and using the error compensation value to correct the coal mine face sensor group data to generate compensated sensing data.

[0014] A fusion processing system for multi-source heterogeneous sensor data from a mining face is also provided, applied to the aforementioned fusion processing method for multi-source heterogeneous sensor data from a mining face. The system includes a data acquisition module, an error aggregation module, a compensation adjustment module, an optimization evaluation module, a model adjustment module, and a compensation control module.

[0015] The data acquisition module is used to acquire multi-channel time-series data of the sensor group at the coal mine face, and to perform signal propagation path analysis on the multi-channel time-series data to identify potentially failed sensors.

[0016] The error aggregation module is used to construct a graph model based on potential failure sensors and to use a community detection algorithm to evaluate the error aggregation of nodes in the graph model to obtain error amplification nodes.

[0017] The compensation and adjustment module is used to set a preset jump radius for each error amplification node and perform several rounds of signal compensation and adjustment within the preset jump radius range to obtain a jump neighborhood sub-map.

[0018] The optimization evaluation module is used to evaluate the signal of the skip neighborhood sub-graph using the objective optimization algorithm, and obtain the optimal compensation adjustment ratio for each error amplification node based on the evaluation results.

[0019] The model adjustment module is used to dynamically adjust the graph model according to the optimal compensation adjustment ratio of each error amplification node to obtain the first error compensation set.

[0020] The compensation control module is used to perform real-time compensation control on the sensor group at the coal mine face based on the first error compensation set.

[0021] The technical effects and advantages of the present invention regarding the fusion processing method and system for multi-source heterogeneous sensor data from a mining face are as follows:

[0022] This invention analyzes the signal propagation path of multi-channel time-series data to accurately identify potentially faulty sensors. Based on a graph model and community detection algorithm, it performs clustered error assessment and further utilizes skip neighborhood compensation and target optimization algorithms to achieve optimal signal adjustment for error amplification nodes. This dynamically corrects the graph model and generates a first error compensation set, ultimately enabling real-time compensation control of the sensor group at the coal mine face. This method and system offer several technical advantages: they effectively suppress the amplification of errors caused by single or locally faulty sensors through multiple iterations in the network, improving the overall accuracy and reliability of the sensor group's data; they dynamically adapt to the complex environment of the mining face and the characteristics of multi-source heterogeneous sensors, achieving globally coordinated error compensation; simultaneously, through graph modeling and optimization algorithms, they quantify error propagation paths and node importance, enabling refined control; furthermore, the system can operate in real-time, balancing safety monitoring and automated control needs, providing a high-precision and stable data foundation for safe coal mine production and intelligent management. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a method for fusing and processing multi-source heterogeneous sensor data from a mining face according to the present invention.

[0024] Figure 2 This is a schematic diagram of the structure of a fusion processing system for multi-source heterogeneous sensor data from a mining face according to the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1, Figure 1 This invention presents a method for fusing and processing multi-source heterogeneous sensor data from a mining face, comprising the following steps:

[0027] S1: Acquire multi-channel time-series data of the sensor group at the coal mine face, and perform signal propagation path analysis on the multi-channel time-series data to identify potentially failed sensors.

[0028] In this embodiment, signal propagation path analysis is performed on multi-channel time-series data to identify potentially failed sensors, specifically as follows:

[0029] Time-domain and frequency-domain features are extracted from the multi-channel time-series data stream to obtain the signal feature vector of each sensor;

[0030] Based on the signal feature vector of each sensor, calculate the signal correlation matrix between each pair of sensors;

[0031] The fluctuation characteristics of sensor data are extracted based on the signal correlation matrix, and the fluctuation feature vector of each sensor is calculated based on the data fluctuation characteristics.

[0032] Based on the wave feature vectors of all sensors, a first abnormal feature set is constructed and a hierarchical clustering algorithm is used to perform several rounds of clustering analysis on the first abnormal feature set.

[0033] Acquire physical layout and communication topology data between sensors, and construct signal propagation path diagram;

[0034] Based on the results of several rounds of cluster analysis, the signal propagation path is evaluated to identify potential sensor failures.

[0035] It should be noted that the multi-channel time-series data of the coal mine face sensor group includes, but is not limited to, continuous monitoring data from various heterogeneous types of sensors such as gas sensors, wind speed sensors, temperature sensors, humidity sensors, dust sensors, vibration sensors, pressure sensors, and equipment status sensors. Each channel corresponds to the time-series signal of one sensor. The system first preprocesses the raw time-series data of each sensor, including noise reduction, missing value imputation, and standardization. Then, in the time domain, it extracts indicators describing signal stability and trends, such as mean, variance, skewness, kurtosis, and autocorrelation coefficient. Simultaneously, in the frequency domain, it extracts frequency features such as spectral energy distribution, dominant frequency, and bandwidth power ratio through Fast Fourier Transform (FFT) or wavelet transform to characterize the periodicity and energy characteristics of the signal. Finally, the time-domain and frequency-domain features are concatenated to form a multi-dimensional feature vector, which serves as the signal feature vector of the sensor for subsequent correlation and anomaly analysis.

[0036] Secondly, the signal feature vectors of all sensors are normalized to ensure the comparability of features with different dimensions. Then, for any pair of sensors, metrics such as Pearson correlation coefficient, cosine similarity, or mutual information are used to measure their similarity in the feature space. For example, for the Pearson correlation coefficient, if two sensors have a high correlation in both the time and frequency domains, it indicates that their signal change trends are consistent. The correlation analysis results of all sensor pairs are then filled into a symmetric matrix to form a signal correlation matrix, where the matrix elements represent the correlation strength between any two sensors. This matrix provides the basis for subsequent fluctuation feature extraction.

[0037] Then, by analyzing the row vectors of the signal correlation matrix, the correlation distribution between each sensor and other sensors is extracted. Combined with the correlation fluctuation trend within the time window, the fluctuation characteristics of the sensors are obtained. Specifically, the sensor data fluctuation characteristics include: correlation variance (representing the stability of the correlation relationship), correlation mean (representing the overall degree of coordination), correlation change rate (representing the fluctuation intensity), mutation detection index (such as CUSUM, used to detect sudden anomalies), and time-series correlation entropy based on a sliding window (used to measure correlation complexity). These characteristics are integrated to form the fluctuation feature vector of each sensor, which can effectively characterize its dynamic performance in a multi-source signal network.

[0038] Furthermore, the fluctuation feature vectors of all sensors are combined according to a unified dimension to construct the first anomaly feature set, representing the overall dynamic fluctuation characteristics of all sensors under the current mining face environment. To identify potential anomalies, the system employs a hierarchical clustering algorithm (such as Ward's minimum variance method or agglomerated hierarchical clustering based on Euclidean distance) to merge and split this anomaly feature set layer by layer. In each round of clustering, the system gradually refines or merges the sensor set according to different clustering thresholds or round control, thereby obtaining anomaly groups of different granularities. After several rounds of iterative clustering analysis, sensor groups exhibiting significant deviations in fluctuation characteristics can be identified, providing a basis for further identification of potentially failed sensors.

[0039] Furthermore, the physical layout and communication topology data of the sensors are obtained through the construction drawings of the mining face, sensor installation coordinate information, and communication network configuration files. The physical layout includes the actual coordinates and installation height of the sensors in locations such as roadways and working faces, while the communication topology includes the wired / wireless connections between sensors and their attributes such as link bandwidth and latency. Based on this data, the system constructs a signal propagation path graph with sensors as nodes and physical connectivity and communication links as edges. This path graph not only reflects the physical location relationships between sensors but also demonstrates the signal propagation path and potential transmission delays in the communication network, providing graphical model support for subsequent propagation path assessment and anomaly tracing.

[0040] Finally, the abnormal sensor group obtained from hierarchical clustering analysis is mapped onto the signal propagation path graph. The propagation pattern of abnormal features in the graph is evaluated by combining indicators such as propagation path length, node degree centrality, and edge weight. If a sensor acts as an anomalous starting point or key relay node in multiple propagation paths, it indicates that it may be causing the anomalous signal to spread in the network, and the system marks it as a potentially failed sensor. Here, a potentially failed sensor refers to a sensor node that, although not completely failed or disconnected from the network, has already exhibited abnormal data fluctuations, weakened correlation, or caused an abnormal spread trend in the propagation path. Identifying these potentially failed sensors helps to perform compensation and maintenance in advance, preventing an overall decline in system monitoring performance.

[0041] S2, a graph model is constructed based on potential failure sensors, and a community detection algorithm is used to evaluate the error aggregation of nodes in the graph model to obtain error amplification nodes;

[0042] In this embodiment, a graph model is constructed based on the potential failure sensor, and a community detection algorithm is used to evaluate the error aggregation of nodes in the graph model to obtain error amplification nodes, specifically:

[0043] Using potentially failing sensors as nodes, a graph model is constructed, and a community detection algorithm is used to partition the graph model to obtain several communities;

[0044] An initial error assessment is performed on each node within a community, and the multi-step propagation path of the error within the community subset is analyzed based on the assessment results to obtain the cumulative characteristics of error propagation. The cumulative characteristics include the path length and impact range of error propagation.

[0045] The first error clustering degree of each community is calculated based on the cumulative characteristics of error propagation, and the communities are filtered according to the first error clustering degree to obtain error clustered communities.

[0046] A second error assessment is performed on the nodes of the error clustering community to obtain the second error clustering degree of each node;

[0047] The error amplification node is identified based on the second error clustering degree.

[0048] It should be noted that, firstly, the potentially failing sensors identified in the previous step are treated as a set of nodes, and edge weights are defined based on the signal correlation strength, physical layout connectivity, and communication topology information between sensors to construct a weighted undirected graph model. In this graph model, nodes represent potentially failing sensors, and higher edge weights indicate greater data dependency or signal propagation strength between two sensors. Subsequently, the system uses a community detection algorithm to partition the graph model, preferably using the Louvain algorithm or a greedy optimization method based on maximizing modularity. The specific steps include: firstly, initializing each node as an independent community; then, traversing all nodes and temporarily moving them to adjacent communities, calculating whether the overall modularity of the graph model increases after the move, and retaining the move operation if it does; repeating this process until all node moves no longer increase the modularity; finally, performing community compression on the formed initial communities, treating each community as a new super node, reconstructing a simplified graph model, and repeating the above community partitioning operation until the modularity converges, resulting in several stable community partitions. In this way, the graph model is divided into a series of locally highly correlated subgroups, laying the foundation for subsequent error propagation and clustering analysis.

[0049] Secondly, an initial error assessment is performed on each node within the community. This is achieved by comparing the node's real-time signal with its historical average, the average of its neighboring nodes, and the overall environmental benchmark, calculating its error deviation and stability index. Subsequently, the system simulates the multi-step propagation of error within the community subset: using the current node as the error source, the error propagates layer by layer along the edges of the community subgraph. The propagation intensity at each step decays according to the edge weight and accumulates in the error values ​​of subsequent nodes. In this way, the cumulative error value for each propagation path can be obtained. The path length of error propagation refers to the number of edges traversed from the source node to the target node, reflecting the number of steps in error propagation; the scope of influence refers to the size of the set of nodes cumulatively affected by the error during propagation, reflecting the scale of the community that the error may affect. By recording the error accumulation characteristics of each node under multi-step propagation, the error diffusion capability within the community can be comprehensively characterized.

[0050] Furthermore, the first error clustering degree is calculated based on the cumulative error propagation characteristics of all nodes in the community. Specifically, a weighted average of the error propagation path lengths of all nodes in the community is calculated, and this is combined with the normalized value of the influence range to construct an error clustering degree index function. Subsequently, the first error clustering degrees of all communities are sorted or thresholds are set to select communities with error clustering degrees significantly higher than the average level, defined as error clustering communities. These communities are key areas for error diffusion and will be the focus of further in-depth analysis.

[0051] Finally, a more refined second error assessment is performed on nodes within the selected error clustering communities. This process not only calculates the local error deviation value of each node but also introduces dynamic indicators such as the node's betweenness centrality in the propagation path, the average error of its neighbors, and the error accumulation rate to construct a second error clustering degree. The second error clustering degree reflects a node's ability to amplify errors during propagation within the community, i.e., whether it significantly increases the propagation and accumulation of errors within the community as a relay or key node. By comparing the second error clustering degree values ​​of each node, the system identifies nodes with clustering degrees significantly higher than the threshold or located in the top several percentiles, marking them as error amplification nodes. These nodes are usually the key sources that cause the continuous amplification and spread of local network errors and are the core targets for subsequent compensation and adjustment.

[0052] S3, set a preset jump radius for each error amplification node, and perform several rounds of signal compensation adjustment within the preset jump radius range to obtain the jump neighborhood sub-map set;

[0053] In this embodiment, a preset jump radius is set for each error amplification node, and several rounds of signal compensation adjustment are performed within the preset jump radius range to obtain a jump neighborhood sub-map set, specifically:

[0054] Extract the network topology of each error amplification node and dynamically calculate the jump radius based on the second error clustering degree;

[0055] Using the error amplification node as the root node, a breadth-first search algorithm is used to traverse all nodes within the jump radius to obtain the first jump neighborhood subgraph.

[0056] On the first-hop neighborhood subgraph, multiple rounds of signal compensation adjustment are performed, with each round of adjustment based on a weighted average of the current node's signal value and the neighboring node's signal value.

[0057] Record the node signal values ​​for each round of signal compensation adjustment to obtain the skip neighborhood sub-graphet.

[0058] It should be noted that, firstly, from the global signal propagation diagram of the sensor group at the mining face, each error amplification node and its direct neighbor relationships are extracted to obtain the local network topology of that node, including the edge weights connected to it, node degree centrality, and adjacency matrix information. To determine the compensation range, the system dynamically calculates the jump radius based on the node's second error clustering degree: specifically, the jump radius is set to be proportional to the second error clustering degree. When the node's second error clustering degree is high, it indicates a large influence range on error propagation, so the jump radius is appropriately increased; when the clustering degree is low, the jump radius is decreased. The specific partitioning steps are as follows: first, a basic radius value is set; then, the node's second error clustering degree value is normalized to the [0,1] interval, and the dynamic radius is calculated; finally, the radius value is rounded and limited to the preset minimum and maximum radius interval to ensure stable and reasonable calculation results. In this way, the jump radius can be adaptively determined for each error amplification node.

[0059] Furthermore, using the error amplification node as the root node, a breadth-first search (BFS) algorithm is employed for hierarchical traversal, starting from its local network topology. The traversal begins at the root node and visits its neighboring nodes layer by layer, expanding one layer at a time until the jump radius limit is reached. During the traversal, all visited nodes and their edge weights with the root node are recorded, and a corresponding adjacency matrix is ​​constructed. Finally, all nodes within the jump radius and their connecting edges form the first jump neighborhood subgraph. This subgraph preserves the local propagation relationships between the error amplification node and its surrounding nodes while shielding irrelevant nodes outside the radius, thus providing a localized processing environment for subsequent compensation adjustments.

[0060] Furthermore, multiple rounds of signal compensation and adjustment are performed in the first-hop neighborhood subgraph. Specifically, in each iteration, the system obtains the current signal value of each node in the subgraph and the signal values ​​of all its neighboring nodes, and calculates a weighted average based on the edge weights. This ensures that the signal of each node not only retains its original information but also allows for correction based on the states of its neighboring nodes, thereby gradually reducing the impact of abnormal signals on the local network. The entire process typically involves several iterations until the node signals stabilize or the preset number of iterations is reached.

[0061] Finally, during the multi-round signal compensation adjustment process, a snapshot of the signal values ​​of all nodes in the subgraph is taken after each round, forming the compensation state subgraph for that round. As the iteration progresses, the system generates multiple subgraphs with different states round by round, and all these subgraphs constitute a skip neighborhood subgraph set. This subgraph set can reflect the signal evolution of the error amplification node and its neighborhood during the dynamic compensation process, providing a data foundation for subsequent signal evaluation and determination of the optimal compensation ratio using objective optimization algorithms. Ultimately, the skip neighborhood subgraph set is stored as a set of time-series evolution subgraphs, becoming a complete description of the error amplification node compensation process.

[0062] S4. The target optimization algorithm is used to evaluate the signal of the skip neighborhood sub-graph and the optimal compensation adjustment ratio of each error amplification node is obtained based on the evaluation results.

[0063] In this embodiment, a target optimization algorithm is used to evaluate the signal of the skip neighborhood sub-graph, and the optimal compensation adjustment ratio for each error amplification node is obtained based on the evaluation results, specifically:

[0064] Define a signal evaluation objective function, and use an objective optimization algorithm to perform multi-objective optimization on the skip neighborhood sub-graph to generate a candidate compensation adjustment ratio set;

[0065] The candidate compensation adjustment ratio set is applied to the corresponding jump neighborhood subgraph to obtain several first jump neighborhood subgraphs and the signal adjustment value of each node in the first jump neighborhood subgraph is obtained to obtain the first signal sequence.

[0066] The first signal sequence is divided into several first signal segments according to a preset signal fluctuation range by a threshold.

[0067] Local mean and deviation analyses were performed on several first signal segments, and error accumulation curves were constructed.

[0068] Data analysis was performed on the error accumulation curve, and the optimal compensation adjustment ratio for each error amplification node was obtained from the analysis results.

[0069] It should be noted that for the skip neighborhood subgraph, a multi-objective optimization algorithm is used to generate a candidate compensation adjustment ratio set. First, a signal evaluation objective function is set, which comprehensively considers the following factors: 1. Signal stationarity objective: minimizing the variance of the compensated signal; 2. Signal accuracy objective: minimizing the mean square error between the compensated signal and the reference signal or historical normal signal; 3. Local consistency objective: minimizing the difference between adjacent node signals to ensure local coordination; 4. Energy preservation objective: maintaining the mean of the compensated signal within a certain range to avoid over-adjustment. By optimizing the objective function, several candidate compensation adjustment ratios are obtained, thus generating a candidate compensation adjustment ratio set.

[0070] Furthermore, each set of proportions in the candidate compensation adjustment set is sequentially applied to the skip neighborhood subgraph. Specifically, the signal values ​​of the error amplification node and its neighboring nodes are compensated and adjusted according to the candidate proportions, generating different versions of the subgraph, each version being called the first skip neighborhood subgraph. In each subgraph, the system calculates the signal adjustment values ​​of all nodes at that proportion and arranges these adjusted signal values ​​according to node number or traversal order to obtain a complete first signal sequence. Since each set of candidate proportions generates a corresponding signal sequence, the system ultimately obtains multiple different first signal sequences, providing basic data for subsequent threshold segmentation and error analysis.

[0071] Furthermore, a reasonable fluctuation range for the sensor signal is predefined, such as the upper and lower limits obtained based on historical normal operating condition statistics or a threshold range set by industry standards. For each first signal sequence, the relationship between the signal value and the threshold is checked point by point: if the signal value is continuously within the normal range, it is divided into an independent signal segment; if the signal value continuously exceeds the normal range, it is divided into another signal segment. Through this process, the entire signal sequence is divided into several first signal segments with different fluctuation characteristics. By segmenting, the normal and abnormal intervals in the signal sequence can be separated, thus facilitating subsequent local mean and deviation analysis.

[0072] Furthermore, statistical analysis is performed on each segmented first signal segment to calculate its local mean and standard deviation, which characterize the stability and fluctuation intensity of that interval. Subsequently, the system accumulates the deviation results of each signal segment in chronological order to obtain an error accumulation sequence, which is then plotted as an error accumulation curve. The vertical axis of the error accumulation curve represents the amount of accumulated error, and the horizontal axis represents the sequence or time progression of the signal segments. This curve can intuitively reflect the impact of the compensation ratio on the overall signal stability. For example, if the curve shows a rapid rise, it indicates that the error is still amplified during the compensation process; if the curve tends to flatten, it indicates that the compensation effect is good.

[0073] Finally, by analyzing the error accumulation curve, the system comprehensively considers the curve's trend, maximum value, slope, and convergence characteristics to evaluate the merits of each candidate compensation adjustment ratio. Specifically, the system extracts indicators such as the mean value of the curve's stable interval, peak value, and convergence speed to calculate a comprehensive score. The lower the score, the better the ratio is at reducing errors and stabilizing the signal. Ultimately, the system selects the group with the best comprehensive score from all candidate ratios as the optimal compensation adjustment ratio for that error amplification node. This ratio not only ensures the signal's stability globally but also effectively suppresses further error amplification in the local network, thus providing a basis for subsequent dynamic model adjustments.

[0074] S5. The graphical model is dynamically adjusted according to the optimal compensation adjustment ratio of each error amplification node to obtain the first error compensation set.

[0075] In this embodiment, the graph model is dynamically adjusted according to the optimal compensation adjustment ratio of each error amplification node to obtain the first error compensation set, specifically:

[0076] The optimal compensation adjustment ratio of each error amplification node is applied to the graphical model to correct the node signal value of the error amplification node, thus obtaining the corrected node signal value.

[0077] Based on the corrected node signal values, the edge weights of the graph model are recalculated to obtain the adjusted graph model.

[0078] Traverse all nodes of the adjusted graph model, calculate the error compensation value of each node, and obtain the first error compensation set.

[0079] It should be noted that, firstly, the optimal compensation adjustment ratio determined for each error amplification node in the preceding steps is read and applied to the original signal value of the corresponding node. Through calculation, the original monitoring information of the node is preserved while introducing the compensation effect of the neighborhood, thereby effectively reducing the degree of deviation of the abnormal signal. This correction operation is performed on all error amplification nodes one by one to obtain the corrected node signal value, providing input data for the subsequent overall adjustment of the graphical model.

[0080] Furthermore, based on the corrected node signal values, the edge weights in the graph model are updated. Specifically, for any two nodes, if the updated edge weight is lower than a preset threshold, the edge connection is considered unreliable and is deleted to ensure the sparsity and robustness of the graph structure. Finally, after all edge weights are updated, an adjusted graph model is formed, which can more realistically reflect the compensated network signal relationships.

[0081] Finally, the adjusted graph model is traversed, and an error compensation value is calculated for each node. The error compensation value is defined as the difference between the node's corrected signal and its original signal, representing the compensation magnitude applied to that node by the system to eliminate the error. By calculating the difference for all nodes and recording their corresponding node identifiers, timestamps, and confidence parameters, the system generates a first error compensation set. The first error compensation set is a centralized compensation parameter library used to describe the compensation information required by each node globally at a given moment. It includes not only correction data for error-amplified nodes but also indirect compensation reference values ​​for other nodes. This compensation set will be called in real time in subsequent steps for dynamic compensation control of the sensor group at the coal mine face.

[0082] S6, Real-time compensation control of the sensor group at the coal mine face is performed based on the first error compensation set.

[0083] In this embodiment, real-time compensation control is performed on the sensor group at the coal mine face based on the first error compensation set, specifically as follows:

[0084] Real-time acquisition of sensor data from the coal mine face and loading of the first error compensation set;

[0085] The error compensation value of the corresponding node in the first error compensation set is dynamically retrieved based on the data from the sensor group at the coal mine face.

[0086] Error compensation values ​​are used to correct the data of the sensor group at the coal mine face, generating compensated sensor data.

[0087] It should be noted that a multi-source heterogeneous sensor array deployed underground in the coal mine collects multi-channel data on the mining face environment and equipment in real time, including sensor output signals such as gas concentration, wind speed, wind pressure, temperature, humidity, dust concentration, roof pressure, and equipment motor current. The data acquisition module and sensors maintain a stable connection via a wired industrial bus or wireless communication network. The acquisition cycle can be set to seconds or minutes to ensure the timeliness of the monitoring data. All collected data undergoes preprocessing (such as timestamp alignment, packet loss retransmission, and preliminary filtering) through an edge gateway before being uploaded to the ground monitoring center or local server to form a real-time updated data stream.

[0088] Furthermore, a first error compensation set is maintained in the data center. This set is dynamically adjusted from the graph model described above and stores the error compensation values ​​for each node under different time windows and their applicable conditions. When the real-time compensation control module receives a new batch of sensor data, it synchronously calls and loads the currently valid first error compensation set. The loading method is as follows: based on the current timestamp and data batch number, the corresponding compensation parameter file is retrieved from the compensation set database, cached in memory, and a mapping table between node numbers and compensation values ​​is established. This ensures that the compensation values ​​can be quickly matched to the real-time acquired sensor data, achieving millisecond-level response.

[0089] Secondly, the real-time sensor data is processed point by point, with each data point containing a unique sensor identifier. Based on this identifier, a quick lookup is performed in the compensation set mapping table to dynamically retrieve the corresponding error compensation value. If the node has a definite compensation parameter within the current time window, that value is returned directly; otherwise, if the node is not included in the compensation set, its error compensation value is automatically set to zero or the average compensation value of neighboring nodes is used as a substitute. Through this dynamic retrieval mechanism, the system ensures that each sensor data point can find a corresponding compensation reference, thereby avoiding uncorrected omissions.

[0090] Finally, the retrieved error compensation value is superimposed and corrected with the real-time sensor raw signal. For example, in a coal mine working face, the temperature sensor numbered T101, due to sensor unit aging, consistently outputs a value about 2°C lower than the actual value. The system configures an error compensation value of (+2.0) for this node in the compensation set. When the real-time acquired raw temperature signal of node T101 is 24.5°C, the system calculates the compensated signal as 26.5°C, which is closer to the actual environmental value. As another example, the gas sensor numbered G305 is identified as having a slight drift in the neighborhood correlation assessment, and its compensation value is set to −0.05. When its raw acquisition value is 0.82%, the compensated value is 0.77%, which is within the actual concentration range. In this way, a set of globally compensated sensor data can be dynamically generated and provided to the safety monitoring and dispatching system, achieving accurate perception of the coal mine working face.

[0091] Example 2, Figure 2 This invention presents a fusion processing system for multi-source heterogeneous sensor data from a mining face, comprising a data acquisition module, an error aggregation module, a compensation and adjustment module, an optimization and evaluation module, a model adjustment module, and a compensation and control module.

[0092] The data acquisition module is used to acquire multi-channel time-series data of the sensor group at the coal mine face, and to perform signal propagation path analysis on the multi-channel time-series data to identify potentially failed sensors.

[0093] The error aggregation module is used to construct a graph model based on potential failure sensors and to use a community detection algorithm to evaluate the error aggregation of nodes in the graph model to obtain error amplification nodes.

[0094] The compensation and adjustment module is used to set a preset jump radius for each error amplification node and perform several rounds of signal compensation and adjustment within the preset jump radius range to obtain a jump neighborhood sub-map.

[0095] The optimization evaluation module is used to evaluate the signal of the skip neighborhood sub-graph using the objective optimization algorithm, and obtain the optimal compensation adjustment ratio for each error amplification node based on the evaluation results.

[0096] The model adjustment module is used to dynamically adjust the graph model according to the optimal compensation adjustment ratio of each error amplification node to obtain the first error compensation set.

[0097] The compensation control module is used to perform real-time compensation control on the sensor group at the coal mine face based on the first error compensation set.

[0098] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0099] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0100] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0101] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0102] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0103] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for fusing and processing multi-source heterogeneous sensor data from a mining face, characterized in that, Includes the following steps: Acquire multi-channel time-series data of sensor groups at the coal mine face, and perform signal propagation path analysis on the multi-channel time-series data to identify potentially failed sensors; A graph model is constructed based on potential failure sensors, and a community detection algorithm is used to evaluate the error aggregation of nodes in the graph model to obtain error amplification nodes, specifically: Using potentially failing sensors as nodes, a graph model is constructed and a community detection algorithm is used to partition the graph model to obtain several communities. The first error assessment is performed on the nodes in each community, and the multi-step propagation path of the error in the community subset is analyzed based on the assessment results to obtain the cumulative characteristics of error propagation. The cumulative characteristics include the path length and the range of influence of error propagation. The first error clustering degree of each community is calculated based on the cumulative characteristics of error propagation, and the communities are filtered according to the first error clustering degree to obtain error clustered communities. A second error assessment is performed on the nodes of the error clustering community to obtain the second error clustering degree of each node; Identify error amplification nodes based on the second error clustering degree; A preset jump radius is set for each error amplification node, and several rounds of signal compensation adjustment are performed within the preset jump radius range to obtain a jump neighborhood sub-map. The target optimization algorithm is used to evaluate the signal of the skip neighborhood sub-graph and the optimal compensation adjustment ratio for each error amplification node is obtained based on the evaluation results. The graphical model is dynamically adjusted according to the optimal compensation adjustment ratio of each error amplification node to obtain the first error compensation set; Real-time compensation control is performed on the sensor group at the coal mine face based on the first error compensation set.

2. The method for fusing and processing multi-source heterogeneous sensor data from a mining face according to claim 1, characterized in that, The step of performing signal propagation path analysis on multi-channel time-series data to identify potentially faulty sensors specifically involves: Time-domain and frequency-domain features are extracted from the multi-channel time-series data stream to obtain the signal feature vector of each sensor; Based on the signal feature vector of each sensor, calculate the signal correlation matrix between each pair of sensors; The fluctuation characteristics of sensor data are extracted based on the signal correlation matrix, and the fluctuation feature vector of each sensor is calculated based on the data fluctuation characteristics. Based on the wave feature vectors of all sensors, a first abnormal feature set is constructed and a hierarchical clustering algorithm is used to perform several rounds of clustering analysis on the first abnormal feature set. Acquire physical layout and communication topology data between sensors, and construct signal propagation path diagram; Based on the results of several rounds of cluster analysis, the signal propagation path is evaluated to identify potential sensor failures.

3. The method for fusing and processing multi-source heterogeneous sensor data from a mining face according to claim 2, characterized in that, The process of setting a preset jump radius for each error amplification node and performing several rounds of signal compensation adjustment within the preset jump radius to obtain a jump neighborhood sub-map is as follows: Extract the network topology of each error amplification node and dynamically calculate the jump radius based on the second error clustering degree; Using the error amplification node as the root node, a breadth-first search algorithm is used to traverse all nodes within the jump radius to obtain the first jump neighborhood subgraph. On the first-hop neighborhood subgraph, multiple rounds of signal compensation adjustment are performed, with each round of adjustment based on a weighted average of the current node's signal value and the neighboring node's signal value. Record the node signal values ​​for each round of signal compensation adjustment to obtain the skip neighborhood sub-graphet.

4. The method for fusing and processing multi-source heterogeneous sensor data from a mining face according to claim 3, characterized in that, The objective optimization algorithm is used to evaluate the signal of the skip neighborhood sub-graph, and the optimal compensation adjustment ratio for each error amplification node is obtained based on the evaluation results. Specifically: Define a signal evaluation objective function, and use an objective optimization algorithm to perform multi-objective optimization on the skip neighborhood sub-graph to generate a candidate compensation adjustment ratio set; The candidate compensation adjustment ratio set is applied to the corresponding jump neighborhood subgraph to obtain several first jump neighborhood subgraphs and the signal adjustment value of each node in the first jump neighborhood subgraph is obtained to obtain the first signal sequence. The first signal sequence is divided into several first signal segments according to a preset signal fluctuation range by a threshold. Local mean and deviation analyses were performed on several first signal segments, and error accumulation curves were constructed. Data analysis was performed on the error accumulation curve, and the optimal compensation adjustment ratio for each error amplification node was obtained from the analysis results.

5. The method for fusing and processing multi-source heterogeneous sensor data from a mining face according to claim 4, characterized in that, The first error compensation set is obtained by dynamically adjusting the graphical model according to the optimal compensation adjustment ratio of each error amplification node. Specifically: The optimal compensation adjustment ratio of each error amplification node is applied to the graphical model to correct the node signal value of the error amplification node, thus obtaining the corrected node signal value. Based on the corrected node signal values, the edge weights of the graph model are recalculated to obtain the adjusted graph model. Traverse all nodes of the adjusted graph model, calculate the error compensation value of each node, and obtain the first error compensation set.

6. The method for fusing and processing multi-source heterogeneous sensor data from a mining face according to claim 5, characterized in that, The real-time compensation control of the sensor group at the coal mine face based on the first error compensation set is specifically as follows: Real-time acquisition of sensor data from the coal mine face and loading of the first error compensation set; The error compensation value of the corresponding node in the first error compensation set is dynamically retrieved based on the data from the sensor group at the coal mine face. Error compensation values ​​are used to correct the data of the sensor group at the coal mine face, generating compensated sensor data.

7. A fusion processing system for multi-source heterogeneous sensor data from a mining face, applied to the fusion processing method for multi-source heterogeneous sensor data from a mining face as described in any one of claims 1-6, characterized in that, It includes a data acquisition module, an error aggregation module, a compensation and adjustment module, an optimization and evaluation module, a model adjustment module, and a compensation and control module. The data acquisition module is used to acquire multi-channel time-series data of the sensor group at the coal mine face, and to perform signal propagation path analysis on the multi-channel time-series data to identify potentially failed sensors. The error aggregation module is used to construct a graph model based on potentially failing sensors and employs a community detection algorithm to evaluate the error aggregation of nodes in the graph model, thereby obtaining error amplification nodes. Specifically: Using potentially failing sensors as nodes, a graph model is constructed and a community detection algorithm is used to partition the graph model to obtain several communities. The first error assessment is performed on the nodes in each community, and the multi-step propagation path of the error in the community subset is analyzed based on the assessment results to obtain the cumulative characteristics of error propagation. The cumulative characteristics include the path length and the range of influence of error propagation. The first error clustering degree of each community is calculated based on the cumulative characteristics of error propagation, and the communities are filtered according to the first error clustering degree to obtain error clustered communities. A second error assessment is performed on the nodes of the error clustering community to obtain the second error clustering degree of each node; Identify error amplification nodes based on the second error clustering degree; The compensation and adjustment module is used to set a preset jump radius for each error amplification node and perform several rounds of signal compensation and adjustment within the preset jump radius range to obtain a jump neighborhood sub-map. The optimization evaluation module is used to evaluate the signal of the skip neighborhood sub-graph using the objective optimization algorithm, and obtain the optimal compensation adjustment ratio for each error amplification node based on the evaluation results. The model adjustment module is used to dynamically adjust the graph model according to the optimal compensation adjustment ratio of each error amplification node to obtain the first error compensation set. The compensation control module is used to perform real-time compensation control on the sensor group at the coal mine face based on the first error compensation set.

Images