A safety production management system based on big data

Abstract

The present invention belongs to the technical field of coal mine production management, and discloses a safety production management system based on big data; the system comprises: collecting coal mine production safety data; pre-processing the coal mine production safety data to obtain perfect production safety data; identifying n potential risk patterns based on the perfect production safety data; building a network flow model based on the n potential risk patterns, and calculating the maximum flow value of the network flow model; calculating the optimal safety production strategy using an artificial bee colony algorithm according to the maximum flow value and the n potential risk patterns; and sending the optimal safety production strategy to an on-site production management terminal, thereby significantly improving the intelligence level, decision-making accuracy and management efficiency of coal mine safety production management, and effectively reducing the risk of safety accidents.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine production management, and more specifically, to a safe production management system based on big data. Background Art

[0002] The patent with application publication number CN114137925A discloses a coal mine production safety intelligent management system, which relates to the field of coal mine management technology, including a terminal server, a remote control platform, a mobile terminal and a monitoring terminal; the terminal server includes a data storage module and a GIS model module; the monitoring terminal includes a data transmission module, a temperature and humidity monitoring module, an oxygen concentration monitoring module, a power consumption monitoring module, a coal mine roof monitoring module, a ventilation monitoring module, a toxic gas monitoring module, a dust monitoring module, a personnel monitoring module, a personnel violation monitoring module, and an attendance record module; the remote control platform includes a data analysis module and an automatic repair module; there are multiple mobile terminals, and the purpose is to provide a coal mine production safety intelligent management system that can run automatically, reduce manpower input, and thereby ensure the production progress of the coal mine and the personal safety of the staff.

[0003] However, the existing coal mine production safety management system still has the problem of insufficient utilization of big data, and is unable to fully and accurately identify potential risks, resulting in some hidden dangers being ignored; secondly, the existing system often adopts a fixed risk assessment model, lacks adaptability to the dynamically changing production environment, and is difficult to cope with emergencies; in actual production, the data processing effect is not good, which may lead to deviations in risk assessment and affect the accuracy of decision-making; in addition, the existing system is difficult to effectively capture and utilize information in safety production data, and ignores the mutual influence and evolution process between risk factors; when formulating safety production strategies, there is a lack of global optimization perspective, and it is difficult to make reasonable trade-offs among multiple risk factors, which may lead to improper resource allocation or deviations in key prevention and control directions; these problems may lead to the accumulation of safety hazards in actual production and increase the risk of accidents.

[0004] In view of this, the present invention proposes a safety production management system based on big data to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: A production safety management system based on big data, comprising: a data acquisition and processing module, for collecting coal mine production safety data; and pre-processing the coal mine production safety data to obtain perfect production safety data;

[0006] The risk identification module identifies n potential risk patterns based on improved production safety data;

[0007] A model building module builds a network traffic model based on n potential risk patterns and calculates the maximum traffic value of the network traffic model;

[0008] The strategy planning module is used to calculate the optimal safety production strategy based on the maximum flow value and n potential risk patterns using the artificial bee colony algorithm; the optimal safety production strategy is sent to the on-site production management terminal, and each module is connected by wired and / or wireless means.

[0009] Furthermore, the coal mine production safety data includes gas concentration, temperature, pressure and displacement;

[0010] The method of preprocessing the coal mine production safety data includes:

[0011] Clean the coal mine production safety data, including removing duplicate data, processing missing values, and identifying and processing abnormal values, to obtain complete production safety data;

[0012] Ways to identify and handle outliers include:

[0013] The same type of data of coal mine production safety data is taken as M data points and mapped to high-dimensional space to obtain high-dimensional data points;

[0014] For all high-dimensional data points, define the number of adaptive nearest neighbors k, calculate the distance vector from each high-dimensional data point to all other high-dimensional data points, select the smallest k distance vectors from all distance vectors, and calculate the mean of these k distance vectors;

[0015] An abnormal threshold is set, and the high-dimensional data points whose mean values ​​of k distance vectors are greater than the abnormal threshold constitute the abnormal value candidate set. For the high-dimensional data points in the abnormal value candidate set, they are marked and retained using the constructed binary features to obtain complete production safety data.

[0016] Furthermore, the mapping to the high-dimensional space includes:

[0017] M data points constitute the data set ,in, It is a data point The vector representation of the kernel matrix K is constructed based on the data set X; the elements in the kernel matrix K are ;

[0018] ;in, For data points The vector representation of is the weight coefficient of the mth feature dimension; for No. feature dimensions, for No. feature dimensions, for and Adaptive scale parameters between ;

[0019] Adaptive scale parameters ; In the formula, for The local density estimate of for The local density estimate of ;

[0020] ;in, For data points Data points in the field, for and The scale parameter between is the distance penalty coefficient; for The weight coefficient of The local density estimate of Consistency;

[0021] The kernel matrix K is centralized to obtain a centralized kernel matrix, the centralized kernel matrix is ​​eigen-decomposed, and the largest m2 eigenvalues ​​and corresponding eigenvectors are selected, and the largest m2 eigenvalues ​​and corresponding eigenvectors are constructed into a new matrix K'; the diagonal of K' is the largest m2 eigenvalues, and each column is the corresponding eigenvector; the original data points are mapped to a high-dimensional space based on the new matrix K'.

[0022] Furthermore, the The weight coefficients are obtained by:

[0023] The data points Each data point in the domain is considered as an independent cluster, and any two independent clusters are calculated. and The comprehensive similarity between ;

[0024] ;in, is a control parameter greater than 0, for and The tightness of the connection between for and The Euclidean distance between the center points, is the distance penalty factor,

[0025] ;in, for and The number of edges between Independent cluster The number of internal data points, Independent cluster The number of internal data points; is the weight allocation factor;

[0026] Merge the two independent clusters with the highest similarity to obtain a new cluster, and recalculate the comprehensive similarity between the new cluster and other independent clusters. Repeat until all data points are clustered into one cluster, and finally calculate the cluster density of all clusters.

[0027] Normalize the obtained cluster density and use it as the weight score ; Map the weight score to the corresponding cluster The weight coefficient ;in is a hyperparameter greater than 0.

[0028] Furthermore, the calculation formula of the cluster density is:

[0029] ;in, Cluster The cluster density, Cluster The volume of is the volume penalty factor, is a small constant, is the shape penalty coefficient, Cluster The degree of abnormality;

[0030] Abnormality ;in, Cluster Any data point within Cluster The geometric center of Cluster The mean of the Euclidean distances of all data points to the geometric center, Cluster The standard deviation of the Euclidean distances of all data points to the geometric center; for and The Euclidean distance between .

[0031] Furthermore, the potential risk pattern identification method includes:

[0032] Arrange the improved production safety data in chronological order to construct a time series database;

[0033] Treat each data sequence in the time series database as a transaction sequence and assign a unique id to the transaction sequence; convert the time series database into vertically formatted data;

[0034] Traverse the vertically formatted data. Initially, take each data element as a candidate pattern and calculate the support of each candidate pattern in the entire time series database.

[0035] A minimum support threshold is preset, and candidate patterns whose support is greater than or equal to the minimum support threshold constitute the first frequent pattern set F_1;

[0036] Using the candidate patterns in the first frequent element set F_1, generate the candidate pattern set H1={C_L} from the bottom up, where C_L is the Lth candidate pattern;

[0037] For each candidate pattern in the candidate pattern set H1, the support of the candidate pattern in the time series database is calculated by intersecting the inverted lists of each data element in the vertically formatted data;

[0038] Preset a frequent threshold, and record the candidate patterns whose support is greater than or equal to the frequent threshold into the qth frequent pattern set F_q;

[0039] Based on the frequent pattern set F_q, generate a candidate pattern set and build the next frequent pattern set F_q+1 until no new candidate pattern set can be generated, and output all the discovered frequent pattern sets;

[0040] For all the frequent patterns in the found frequent pattern set , calculate its risk score ;

[0041] ;in, Frequent mode No. The weight coefficient of each data element; , and All are frequently controlled parameters; For the The time interval between the time when a data element appears and the current time, is the preset cycle parameter;

[0042] Preset risk threshold , for any frequent pattern in all the frequent pattern sets found , if its risk score Greater than or equal to the risk threshold , then the frequent pattern Mark as preliminary potential risk patterns; sort the marked preliminary potential risk patterns in descending order according to the size of support, and extract the first n as the final potential risk patterns.

[0043] Furthermore, the calculation formula of the support is:

[0044] ;in, Is a candidate pattern The number of different transaction sequences, is the total number of transaction sequences in the time series database, is the importance weight parameter, is the time series continuous weight parameter, Candidate Mode The importance scoring function is is the temporal continuity scoring function; Candidate Mode support level;

[0045] Importance scoring function ;in, Candidate Mode Length, Is a candidate mode The data elements, It is The weight coefficient of each data element; is the preset danger threshold, The default value is the planning value;

[0046] Temporal continuity scoring function ;in, Candidate Mode The data elements, is the attenuation parameter, is to adjust the hyperparameters, for and The timing interval between .

[0047] Furthermore, the network traffic model is constructed in the following manner:

[0048] Create a directed graph G(V, E), where V is a set of nodes and E is a set of edges, and add source points and sink points to V. Create a node for each potential risk pattern, recorded as a potential risk pattern node, and add it to V. Connect an edge with a capacity of 1 unit from the source point to each potential risk pattern node, and connect an edge with a capacity of 1 unit from each potential risk pattern node to the sink point; for any two potential risk pattern nodes and , calculate the temporal correlation between them, if the temporal correlation between them is greater than the preset correlation threshold, then and Add an edge with a capacity of time series correlation between them;

[0049] The calculation formula of time series correlation is:

[0050] ;

[0051] in, for and The joint probability of co-occurring in historical data, for The probability of appearing alone in historical data, for The probability of appearing alone in historical data, for The risk weight factor is for The risk weight factor is is a risk weight adjustment parameter greater than 0;

[0052] Based on the directed graph G(V, E), a hierarchical network is constructed, with the source point as the 0th layer, the nodes that can be directly reached from the source point as the 1st layer, and so on; until no new nodes in the next layer can be found, or the sink is reached, the hierarchical network construction is completed, and the hierarchical network is the network traffic model.

[0053] Furthermore, the calculation method of the maximum flow value includes:

[0054] Based on the hierarchical network, starting from the source point, a depth-first search is performed. During the depth-first search, it is only allowed to advance from the current layer to the next layer along the direction of the edge. If the sink is reached during the depth-first search, an augmenting path is found, all nodes and edges on this augmenting path are recorded, and the minimum residual capacity CT of all edges on the augmenting path is calculated; along the augmenting path, the minimum residual capacity CT is pushed to each edge; for the forward edge, the residual capacity of the edge is reduced by CT; for the reverse edge, the residual capacity of the edge is increased by CT; the augmenting path is repeatedly searched until no augmenting path can be found. At this time, the total flow received by the sink is the maximum flow value.

[0055] Furthermore, the calculation method of the optimal production safety strategy includes:

[0056] Construct a solution space, taking n potential risk patterns as n4 dimensions in the solution space, and each solution vector in the solution space ,in, is a positive integer greater than 1 and less than n4, Indicates The extent to which a potential risk model is adopted;

[0057] Define the goodness function, the formula of the goodness function is:

[0058] ;in, is the solution vector The sum of all dimensions of According to the solution vector The maximum flow value in the calculated network flow model, is the maximum flow value of the original network traffic model; The solution vector The value of the goodness function of ; According to the solution vector The calculated overall risk value of the strategy; , and is the weight coefficient of superiority;

[0059] A certain number of bee colony individuals are randomly initialized in the solution space to form an initial bee colony. The value of the goodness function of each bee colony individual is calculated, which is recorded as goodness;

[0060] Define the number of bee colony individuals as honey-collecting worker bees and scout bees; for each honey-collecting worker bee, according to its current solution vector , a new solution vector is generated by mutation operation in its neighborhood , calculate the new solution vector The pros and cons of ,like Greater than , then use replace , become a new honey-collecting worker bee, otherwise keep constant;

[0061] Calculate the sum of the merits of all honey-collecting worker bees, sf. For each scout bee, select a honey-collecting worker bee according to the roulette selection strategy and the proportion of the scout bee's fitness value to sf. Generate a new solution vector through mutation operation in the neighborhood of this honey-collecting worker bee, and calculate the merit of this new solution vector. If this merit is greater than the merit of the selected honey-collecting worker bee, then replace the selected honey-collecting worker bee with this new solution vector to become the new honey-collecting worker bee.

[0062] After repeating the preset number of iterations, the swarm individual with the highest current merit is output as the optimal solution vector. ,in, is the optimal solution vector The extent to which a potential risk model is adopted;

[0063] Preset intervention threshold , for each ,like , then the optimal solution vector is adopted Intervention measures for a potential risk model are taken, otherwise the potential risk model is ignored. According to the adopted potential risk model, safety production countermeasures and control measures are formulated to obtain the optimal safety production strategy.

[0064] Technical effects and advantages of a big data-based production safety management system of the present invention:

[0065] Through the comprehensive collection and preprocessing of coal mine production safety data, the system of the present invention lays a solid foundation for subsequent analysis; the algorithm combined with adaptive parameter adjustment and threshold setting is adopted to greatly improve the accuracy and comprehensiveness of potential risk identification; the system constructs a network traffic model, which effectively captures the temporal correlation between potential risk patterns, providing a more comprehensive and in-depth perspective for risk assessment; the system significantly improves the reliability of risk assessment in dealing with outliers and noise data; combined with the artificial bee colony algorithm, the system realizes global optimization among multiple risk factors, making the formulation of safe production strategies more scientific and reasonable; in addition, the system has strong adaptability and can flexibly respond to complex and changeable production environments; in general, it realizes the full process intelligent management from data collection, risk identification to strategy formulation, which significantly improves the intelligence level, decision-making accuracy and management efficiency of coal mine safety production management, and effectively reduces the risk of safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A schematic diagram of a safety production management system based on big data according to the present invention;

[0067] Figure 2 It is a schematic diagram of a safe production management method based on big data of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] Example 1

[0070] See also Figure 1 As shown, the safety production management system based on big data described in this embodiment includes:

[0071] The data collection and processing module is used to collect coal mine production safety data; and pre-process the coal mine production safety data to obtain complete production safety data;

[0072] The risk identification module identifies n potential risk patterns based on improved production safety data;

[0073] A model building module builds a network traffic model based on n potential risk patterns and calculates the maximum traffic value of the network traffic model;

[0074] The strategy planning module is used to calculate the optimal production safety strategy based on the maximum flow value and n potential risk patterns using the artificial bee colony algorithm; the optimal production safety strategy is sent to the on-site production management terminal, thereby realizing intelligent management and optimization of coal mine production safety; each module is connected by wired and / or wireless means to realize data transmission between modules.

[0075] Furthermore, coal mine production safety data includes gas concentration, temperature, pressure and displacement; gas concentration reflects the gas content in the coal mine, and too high a temperature may cause a gas explosion; temperature and pressure reflect the state of the coal seam, and abnormalities may cause roof falls and spalling; displacement reflects the deformation of the coal mining area, and abnormal displacement may cause safety hazards such as underground collapse.

[0076] Furthermore, the method of preprocessing the coal mine production safety data includes:

[0077] The coal mine production safety data is cleaned, which includes removing duplicate data, processing missing values ​​(such as linear interpolation, polynomial interpolation, etc.), and identifying and processing outliers.

[0078] Ways to identify and handle outliers include:

[0079] The same type of data of coal mine production safety data is taken as M data points and mapped to high-dimensional space to obtain high-dimensional data points.

[0080] Ways to map to high-dimensional space include:

[0081] M data points constitute the data set ,in, It is a data point The vector representation (dimension is d) constructs the kernel matrix K based on the data set X; the elements in the kernel matrix K are ;

[0082] ;in, For data points (and do not index the same data point at the same time), is the weight coefficient of the mth feature dimension (each attribute field corresponds to a feature dimension), which is used to adjust the importance of this dimension when calculating similarity; for No. feature dimensions, for No. feature dimensions, for and The adaptive scale parameter between them can obtain different similarity measures for different data pairs at different scales, which is more flexible.

[0083] Adaptive scale parameters ; In the formula, for The local density estimate of for The local density estimate of .

[0084] The local density estimate of ;in, For data points Data points within the area (given by the preset area radius), for and The scale parameter between (used to control the scale of the distance between data points), is the distance penalty coefficient, which is used to adjust the influence of the distance term; for The weight coefficient reflects right the importance of The local density estimate of Consistent.

[0085] The weight coefficient Ways to obtain include:

[0086] The data points Each data point in the domain is considered as an independent cluster, and any two independent clusters are calculated. and The comprehensive similarity between .

[0087] ;in, is a control parameter greater than 0, for and The tightness of the connection between for and The Euclidean distance between the center points, is the distance penalty factor.

[0088] ;in, for and The number of edges between them is the number of pairs of data points. satisfy , and and The Euclidean distance between and The maximum diameter of Independent cluster The number of internal data points, Independent cluster The number of internal data points; It is the weight distribution factor, which controls the weight distribution of the density of the two.

[0089] The two independent clusters with the highest similarity are merged to obtain a new cluster, and the comprehensive similarity between the new cluster and other independent clusters is recalculated. This process is repeated until all data points are clustered into one cluster. Finally, the cluster density of all clusters is calculated.

[0090] The cluster density is calculated as:

[0091] ;in, Cluster The cluster density, Cluster The volume of (obtained using the method of minimum enclosing hypersphere volume), is the volume penalty factor, which is adjusted according to the data distribution density. is a small constant, such as 0.001, used to avoid the denominator being 0. is the shape penalty coefficient, Cluster The abnormality of .

[0092] ;in, Cluster Any data point within Cluster The geometric center of Cluster The mean of the Euclidean distances of all data points to the geometric center, Cluster The standard deviation of the Euclidean distances of all data points to the geometric center; for and The Euclidean distance between .

[0093] Normalize the obtained cluster density and use it as the weight score ; Map the weight score to the corresponding cluster The weight coefficient ;in It is a hyperparameter greater than 0, which controls the degree of weight distribution.

[0094] The kernel matrix K is centralized (i.e., each row and column of K is subtracted from the mean of the corresponding row / column) to obtain a centralized kernel matrix. The centralized kernel matrix is ​​eigendecomposed and the largest m2 eigenvalues ​​and corresponding eigenvectors are selected. The largest m2 eigenvalues ​​and corresponding eigenvectors are constructed into a new matrix K'; the diagonal of K' is the largest m2 eigenvalues, and each column is the corresponding eigenvector; the original data points are mapped to a high-dimensional space based on the new matrix K'.

[0095] In this way, the original d-dimensional data is mapped to a high-dimensional space. M2 can calculate the distance in the high-dimensional space according to the adaptive settings, which can better discover the inherent nonlinear structure of the data.

[0096] For all high-dimensional data points, the number of adaptive neighbors k is defined, the distance vector from each high-dimensional data point to all other high-dimensional data points is calculated, the smallest k distance vectors are selected from all distance vectors, and the mean of these k distance vectors is calculated.

[0097] An abnormal threshold is set, and the high-dimensional data points whose mean values ​​of k distance vectors are greater than the abnormal threshold constitute the abnormal value candidate set. For the high-dimensional data points in the abnormal value candidate set, they are marked and retained using the constructed binary features; thus, perfect production safety data is obtained.

[0098] Specifically, for data points, express is a data point in the outlier candidate set; express is a data point that is not in the outlier candidate set. For each data point, a binary feature is constructed. ; That is, when When it is an outlier, ; Otherwise 0; Each data point in the data set All expanded to .

[0099] Outliers are retained while outlier marker features are added, and outliers are used as valuable information instead of simply discarding them, thereby improving the robustness and generalization ability in the presence of outliers.

[0100] Further, potential risk patterns can be identified by:

[0101] The complete production safety data (gas concentration, temperature, pressure, displacement) are arranged in chronological order to construct a time series database (composed of data sequences).

[0102] Each data sequence in the time series database is treated as a transaction sequence, and the transaction sequence is assigned a unique id; the time series database is converted into vertically formatted data, that is, a separate inverted list is established for each different data element in the transaction sequence (such as different gas concentration values) to store the ids of all transaction sequences containing the data element.

[0103] Traverse the vertically formatted data. Initially, each data element is taken as a candidate pattern and the support of each candidate pattern in the entire time series database is calculated.

[0104] The calculation formula for support is:

[0105]

[0106] in, Is a candidate pattern The number of different transaction sequences, is the total number of transaction sequences in the time series database, is the importance weight parameter, is the time series continuous weight parameter, Candidate Mode The importance scoring function is is the temporal continuity scoring function; Candidate Mode support.

[0107] Importance scoring function ;

[0108] in, Candidate Mode The length (number of data elements), Is a candidate mode The data elements, It is The weight coefficient of each data element reflects the importance of the data element to safe production; is the preset danger threshold, The default value is a planning value used to scale the score to the interval [0, 1].

[0109] Based on historical data, different weight coefficients are set for different types of data elements (such as gas concentration, temperature, pressure, etc.) to reflect their importance to safe production.

[0110] Temporal continuity scoring function ;in, Candidate Mode The data elements, is the attenuation parameter, which controls the influence of the time interval on the function. is a hyperparameter used to adjust the intensity of the interval penalty. for and The timing interval between .

[0111] The temporal continuity scoring function tends to discover patterns with better temporal continuity, and continuous anomalies are often more worthy of attention than scattered anomalies.

[0112] A minimum support threshold is preset, and candidate patterns whose support is greater than or equal to the minimum support threshold constitute the first frequent pattern set F_1.

[0113] Using the candidate patterns in the first frequent element set F_1, a longer candidate pattern set H1={C_L} is generated from the bottom up, where C_L is the Lth candidate pattern; for each candidate pattern in the candidate pattern set H1, the support of the candidate pattern in the time series database is calculated by intersecting the inverted lists of each data element in the vertically formatted data.

[0114] A frequent threshold is preset, and candidate patterns whose support is greater than or equal to the frequent threshold are recorded into the qth frequent pattern set F_q of the corresponding length.

[0115] Based on the frequent pattern set F_q, generate a longer candidate pattern set and construct the next frequent pattern set F_q+1 until no new candidate pattern set can be generated, and output all discovered frequent pattern sets.

[0116] For all the frequent patterns in the found frequent pattern set , calculate its risk score ;

[0117] ;in, Frequent mode No. The weight coefficient of each data element reflects the importance of the element to safe production; , and They are all adjustable frequent control parameters, and the optimal values ​​are determined through algorithm optimization and cross-validation according to the time series distribution characteristics of the actual data; For the The time interval between the time when a data element appears and the current time, It is a preset period parameter (such as 24 hours, one week, etc.).

[0118] Preset risk threshold , for any frequent pattern in all the frequent pattern sets found , if its risk score Greater than or equal to the risk threshold , then the frequent pattern Mark as preliminary potential risk patterns; sort the marked preliminary potential risk patterns in descending order according to the size of support, and extract the first n as the final potential risk patterns.

[0119] Furthermore, the network traffic model is constructed in the following ways:

[0120] Create a directed graph G(V, E), where V is a set of nodes and E is a set of edges, and add source points and sink points to V. Create a node for each potential risk pattern, recorded as a potential risk pattern node, and add it to V. Connect an edge with a capacity of 1 unit from the source point to each potential risk pattern node, and connect an edge with a capacity of 1 unit from each potential risk pattern node to the sink point; for any two potential risk pattern nodes and , calculate the temporal correlation between them, if the temporal correlation between them is greater than the preset correlation threshold, then and Add an edge with capacity equal to the temporal association between them.

[0121] It should be noted that a capacity of 1 means that each potential risk mode node can only receive a maximum of 1 unit of traffic from the source point.

[0122] The calculation formula of time series correlation is:

[0123] ;

[0124] in, for and The joint probability of co-occurring in historical data, for The probability of appearing alone in historical data, for The probability of appearing alone in historical data, for The risk weight factor is for The risk weight factor is is a risk weight adjustment parameter greater than 0; the historical data is the coal mine production safety monitoring data for a period of time that has been acquired before building the network traffic model, which is used to analyze the temporal correlation between different risk modes.

[0125] Based on the directed graph G(V, E), a hierarchical network is constructed, with the source point as the 0th layer, the nodes that can be directly reached from the source point as the 1st layer, and so on; until no new nodes in the next layer can be found, or the sink is reached, the hierarchical network construction is completed, and the hierarchical network is the network traffic model; the maximum traffic value is obtained based on the hierarchical network planning.

[0126] Based on the hierarchical network, starting from the source point, a depth-first search is performed. During the depth-first search, only the direction of the edge is allowed to advance from the current layer to the next layer. If the sink is reached during the depth-first search, an augmenting path is found, all nodes and edges on the augmenting path are recorded, and the minimum residual capacity CT of all edges on the augmenting path is calculated (traversing all edges on the augmenting path, finding the minimum value of the residual capacity of each edge, that is, obtaining the value of CT); along the augmenting path, the minimum residual capacity CT is pushed to each edge; for the forward edge (from the layer closer to the source point to the layer farther from the source point), the residual capacity of the edge is reduced by CT; for the reverse edge (from the layer farther from the source point to the layer closer to the source point), the residual capacity of the edge is increased by CT; the augmenting path is repeatedly searched until no augmenting path can be found. At this time, the total flow received by the sink is the maximum flow value, which reflects the maximum sum of temporal associations between the given n risk patterns.

[0127] It should be noted that in the network flow model, each edge is assigned a capacity value, which represents the maximum allowed flow on the edge. When the flow starts to be pushed, the flow value on the edge will gradually increase; at this time, the residual capacity on the edge refers to the additional flow that the edge can currently accommodate.

[0128] The calculated maximum flow value can evaluate the overall risk intensity and correlation degree of a given set of risk patterns.

[0129] Furthermore, the calculation method of the optimal safety production strategy includes:

[0130] Construct the solution space, take n potential risk patterns as n4 dimensions in the solution space, and the value range of each dimension is [0, 1], which indicates the degree of adoption of the corresponding risk pattern. Each solution vector in the solution space ,in, is a positive integer greater than 1 and less than n4. The solution vector represents a safety production strategy. Indicates The extent to which a potential risk model is adopted.

[0131] Define the goodness function, the formula of the goodness function is:

[0132] ;in, is the solution vector The sum of all dimensions of represents the overall intervention degree of the strategy, According to the solution vector The maximum flow value in the calculated network traffic model (based on Dynamically adjust the edge capacity in the network traffic model, thereby affecting the maximum flow value of the model). is the maximum flow value of the original network traffic model; The solution vector The value of the goodness function.

[0133] According to the solution vector The calculated overall risk value of the strategy (a weighted value of the aforementioned risk scores); , and It is the weight coefficient of the quality, which can be adjusted according to actual needs.

[0134] A certain number of swarm individuals (solution vectors) are randomly initialized in the solution space to form an initial swarm. The value of the quality function of each swarm individual is calculated, which is recorded as the quality.

[0135] Define the number of bee colony individuals as honey-collecting worker bees and scout bees; for each honey-collecting worker bee, according to its current solution vector , a new solution vector is generated by mutation operation within its neighborhood (defined domain radius acquisition) , calculate the new solution vector The pros and cons of ,like Greater than , then use replace , become a new honey bee, otherwise keep constant.

[0136] Calculate the sum of the merits of all honey-collecting worker bees sf. For each scout bee, select a honey-collecting worker bee according to the roulette selection strategy and the proportion of the scout bee's fitness value to sf. Generate a new solution vector through mutation operation in the neighborhood of this honey-collecting worker bee, and calculate the merit of this new solution vector. If this merit is greater than the merit of the selected honey-collecting worker bee, then replace the selected honey-collecting worker bee with this new solution vector to become the new honey-collecting worker bee. After repeating the preset number of iterations, output the colony individual with the highest merit as the optimal solution vector ,in, is the optimal solution vector The extent to which a potential risk model is adopted.

[0137] Preset intervention threshold , for each ,like , then the optimal solution vector is adopted Intervention measures for a potential risk pattern are taken, otherwise the potential risk pattern is ignored. According to the adopted potential risk pattern, specific production safety countermeasures and control measures are formulated. For example, a database containing potential risk patterns and their corresponding production safety countermeasures and control measures is established in advance. Each risk pattern may correspond to multiple countermeasures and measures, which are sorted according to effectiveness and feasibility to obtain the optimal production safety strategy.

[0138] In this embodiment, through the comprehensive collection and preprocessing of coal mine production safety data, the system lays a solid foundation for subsequent analysis; the algorithm is adopted, combined with adaptive parameter adjustment and threshold setting, which greatly improves the accuracy and comprehensiveness of potential risk identification; the system constructs a network traffic model, effectively captures the temporal correlation between potential risk patterns, and provides a more comprehensive and in-depth perspective for risk assessment; the system significantly improves the reliability of risk assessment in dealing with outliers and noise data; combined with the artificial bee colony algorithm, the system realizes global optimization among multiple risk factors, making the formulation of safe production strategies more scientific and reasonable; in addition, the system has strong adaptive capabilities and can flexibly respond to complex and changing production environments; in general, it realizes full-process intelligent management from data collection, risk identification to strategy formulation, significantly improving the intelligence level, decision-making accuracy and management efficiency of coal mine safety production management, and effectively reducing the risk of safety accidents.

[0139] Example 2

[0140] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a safety production management method based on big data, including:

[0141] S1. Collect coal mine production safety data; and pre-process the coal mine production safety data to obtain complete production safety data;

[0142] S2. Based on the improved production safety data, the time series pattern mining algorithm is used to identify n potential risk patterns;

[0143] S3, constructing a network traffic model based on n potential risk patterns, and calculating the maximum traffic value of the network traffic model;

[0144] S4. According to the maximum flow value and n potential risk patterns, the optimal production safety strategy is calculated using the artificial bee colony algorithm; the optimal production safety strategy is sent to the on-site production management terminal.

[0145] Example 3

[0146] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned big data-based safe production management method is implemented.

[0147] Since the electronic device introduced in this embodiment is an electronic device used to implement a safety production management method based on big data in the embodiment of this application, based on the safety production management method based on big data introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the safety production management method based on big data in the embodiment of this application, it belongs to the scope of protection of this application.

[0148] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0149] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A safety production management system based on big data, characterized in that: include: Data collection and processing module, used to collect coal mine production safety data; And pre-process the coal mine production safety data to obtain perfect production safety data; The risk identification module identifies n potential risk patterns based on improved production safety data; A model building module builds a network traffic model based on n potential risk patterns and calculates the maximum traffic value of the network traffic model; The strategy planning module is used to calculate the optimal safety production strategy based on the maximum flow value and n potential risk modes using the artificial bee colony algorithm; the optimal safety production strategy is sent to the on-site production management terminal, and each module is connected by wired and / or wireless means; The potential risk patterns may be identified by: Arrange the improved production safety data in chronological order to construct a time series database; Treat each data sequence in the time series database as a transaction sequence and assign a unique id to the transaction sequence; convert the time series database into vertically formatted data; Traverse the vertically formatted data. Initially, take each data element as a candidate pattern and calculate the support of each candidate pattern in the entire time series database. A minimum support threshold is preset, and candidate patterns whose support is greater than or equal to the minimum support threshold constitute the first frequent pattern set F_1; Using the candidate patterns in the first frequent element set F_1, generate the candidate pattern set H1={C_L} from the bottom up, where C_L is the Lth candidate pattern; For each candidate pattern in the candidate pattern set H1, the support of the candidate pattern in the time series database is calculated by intersecting the inverted lists of each data element in the vertically formatted data; Preset a frequent threshold, and record the candidate patterns whose support is greater than or equal to the frequent threshold into the qth frequent pattern set F_q; Based on the frequent pattern set F_q, generate a candidate pattern set and build the next frequent pattern set F_q+1 until no new candidate pattern set can be generated, and output all the discovered frequent pattern sets; For all the frequent patterns in the found frequent pattern set , calculate its risk score ; ;in, Frequent mode No. The weight coefficient of each data element; , and All are frequently controlled parameters; For the The time interval between the time when a data element appears and the current time, is the preset cycle parameter; Preset risk threshold , for any frequent pattern in all the frequent pattern sets found , if its risk score Greater than or equal to the risk threshold , then the frequent pattern Mark as preliminary potential risk patterns; sort the marked preliminary potential risk patterns in descending order according to the support degree, and extract the first n as the final potential risk patterns; The network traffic model is constructed in the following manner: Create a directed graph G(V, E), where V is a set of nodes and E is a set of edges, and add source points and sink points to V. Create a node for each potential risk pattern, recorded as a potential risk pattern node, and add it to V. Connect an edge with a capacity of 1 unit from the source point to each potential risk pattern node, and connect an edge with a capacity of 1 unit from each potential risk pattern node to the sink point; for any two potential risk pattern nodes and , calculate the temporal correlation between them, if the temporal correlation between them is greater than the preset correlation threshold, then and Add an edge with a capacity of time series correlation between them; The calculation formula of time series correlation is: ； in, for and The joint probability of co-occurring in historical data, for The probability of appearing alone in historical data, for The probability of appearing alone in historical data, for The risk weight factor is for The risk weight factor is is a risk weight adjustment parameter greater than 0; Based on the directed graph G(V, E), a hierarchical network is constructed, with the source point as the 0th layer, the nodes that can be directly reached from the source point as the 1st layer, and so on; until no new nodes in the next layer can be found, or the sink point is reached, the hierarchical network construction is completed, and the hierarchical network is the network traffic model; The calculation method of the maximum flow value includes: Based on the hierarchical network, starting from the source point, a depth-first search is performed. During the depth-first search, only the direction of the edge is allowed to advance from the current layer to the next layer. If the sink is reached during the depth-first search, an augmenting path is found, all nodes and edges on the augmenting path are recorded, and the minimum residual capacity CT of all edges on the augmenting path is calculated; along the augmenting path, the minimum residual capacity CT is pushed to each edge; for the forward edge, the residual capacity of the edge is reduced by CT; for the reverse edge, the residual capacity of the edge is increased by CT; the augmenting path is searched repeatedly until no augmenting path can be found. At this time, the total flow received by the sink is the maximum flow value; The calculation method of the optimal production safety strategy includes: Construct a solution space, taking n potential risk patterns as n4 dimensions in the solution space, and each solution vector in the solution space ,in, is a positive integer greater than 1 and less than n4, Indicates The extent to which a potential risk model is adopted; Define the goodness function, the formula of the goodness function is: ;in, is the solution vector The sum of all dimensions of According to the solution vector The maximum flow value in the calculated network traffic model, is the maximum flow value of the original network traffic model; The solution vector The value of the goodness function of ; According to the solution vector The calculated overall risk value of the strategy; , and is the weight coefficient of superiority; A certain number of bee colony individuals are randomly initialized in the solution space to form an initial bee colony. The value of the goodness function of each bee colony individual is calculated, which is recorded as goodness; Define the number of bee colony individuals as honey-collecting worker bees and scout bees; for each honey-collecting worker bee, according to its current solution vector , a new solution vector is generated by mutation operation in its neighborhood , calculate the new solution vector The pros and cons of ,like Greater than , then use replace , become a new honey bee, otherwise keep constant; Calculate the sum of the merits of all honey-collecting worker bees, sf. For each scout bee, select a honey-collecting worker bee according to the roulette selection strategy and the proportion of the scout bee's fitness value to sf. Generate a new solution vector through mutation operation in the neighborhood of this honey-collecting worker bee, and calculate the merit of this new solution vector. If this merit is greater than the merit of the selected honey-collecting worker bee, then replace the selected honey-collecting worker bee with this new solution vector to become the new honey-collecting worker bee. After repeating the preset number of iterations, the swarm individual with the highest current merit is output as the optimal solution vector. ,in, is the optimal solution vector The extent to which a potential risk model is adopted; Preset intervention threshold , for each ,like , then the optimal solution vector is adopted Intervention measures for a potential risk model are taken, otherwise the potential risk model is ignored. According to the adopted potential risk model, safety production countermeasures and control measures are formulated to obtain the optimal safety production strategy.

2. A big data-based production safety management system according to claim 1, characterized in that: The coal mine production safety data include gas concentration, temperature, pressure and displacement; The method of preprocessing the coal mine production safety data includes: Clean the coal mine production safety data, including removing duplicate data, processing missing values, and identifying and processing abnormal values, to obtain complete production safety data; Ways to identify and handle outliers include: The same type of data of coal mine production safety data is taken as M data points and mapped to high-dimensional space to obtain high-dimensional data points; For all high-dimensional data points, define the number of adaptive nearest neighbors k, calculate the distance vector from each high-dimensional data point to all other high-dimensional data points, select the smallest k distance vectors from all distance vectors, and calculate the mean of these k distance vectors; An abnormal threshold is set, and the high-dimensional data points whose mean values ​​of k distance vectors are greater than the abnormal threshold constitute the abnormal value candidate set. For the high-dimensional data points in the abnormal value candidate set, they are marked and retained using the constructed binary features to obtain complete production safety data.

3. A big data-based production safety management system according to claim 2, characterized in that: The mapping to the high-dimensional space includes: M data points constitute the data set ,in, It is a data point The vector representation of the kernel matrix K is constructed based on the data set X; the elements in the kernel matrix K are ; ;in, For data points The vector representation of is the weight coefficient of the mth feature dimension; for No. feature dimensions, for No. feature dimensions, for and Adaptive scale parameters between ; Adaptive scale parameters ; In the formula, for The local density estimate of for The local density estimate of ; ;in, For data points Data points in the field, for and The scale parameter between is the distance penalty coefficient; for The weight coefficient of The local density estimate of Consistency; The kernel matrix K is centralized to obtain a centralized kernel matrix, the centralized kernel matrix is ​​eigen-decomposed, and the largest m2 eigenvalues ​​and corresponding eigenvectors are selected, and the largest m2 eigenvalues ​​and corresponding eigenvectors are constructed into a new matrix K'; the diagonal of K' is the largest m2 eigenvalues, and each column is the corresponding eigenvector; the original data points are mapped to a high-dimensional space based on the new matrix K'.

4. A big data-based production safety management system according to claim 3, characterized in that: Said The weight coefficients are obtained by: The data points Each data point in the domain is considered as an independent cluster, and any two independent clusters are calculated. and The comprehensive similarity between ; ;in, is a control parameter greater than 0, for and The tightness of the connection between for and The Euclidean distance between the center points, is the distance penalty factor, ;in, for and The number of edges between Independent cluster The number of internal data points, Independent cluster The number of internal data points; is the weight allocation factor; Merge the two independent clusters with the highest similarity to obtain a new cluster, and recalculate the comprehensive similarity between the new cluster and other independent clusters. Repeat until all data points are clustered into one cluster, and finally calculate the cluster density of all clusters. Normalize the obtained cluster density and use it as the weight score ; Map the weight score to the corresponding cluster The weight coefficient ;in is a hyperparameter greater than 0.

5. A big data-based production safety management system according to claim 4, characterized in that: The calculation formula of the cluster density is: ;in, Cluster The cluster density, Cluster The volume of is the volume penalty factor, is a small constant, is the shape penalty coefficient, Cluster The degree of abnormality; Abnormality ;in, Cluster Any data point within Cluster The geometric center of Cluster The mean of the Euclidean distances of all data points to the geometric center, Cluster The standard deviation of the Euclidean distances of all data points to the geometric center; for and The Euclidean distance between .

6. A big data-based production safety management system according to claim 5, characterized in that: The calculation formula of the support is: ;in, Is a candidate pattern The number of different transaction sequences, is the total number of transaction sequences in the time series database, is the importance weight parameter, is the time series continuous weight parameter, Candidate Mode The importance scoring function is is the temporal continuity scoring function; Candidate Mode support level; Importance scoring function ;in, Candidate Mode Length, Is a candidate mode The data elements, It is The weight coefficient of each data element; is the preset danger threshold, The default value is the planning value; Temporal continuity scoring function ;in, Candidate Mode The data elements, is the attenuation parameter, is to adjust the hyperparameters, for and The timing interval between .

Images