An early warning system and emergency response method for coal mine fires based on BiLSTM-Attention

Abstract

This invention belongs to the field of fire monitoring technology, and specifically relates to an early warning system and emergency response method for externally-caused coal mine fires based on BiLSTM-Attention. The early warning system includes a monitoring module for collecting environmental data from the mine; an early warning module for predicting fires and issuing early warnings using algorithms; and an information transmission module for transmitting the early warning signals to workers. The emergency response method includes the following steps: receiving and processing information: generating a mine roadway topology network map, where personnel positioning stations on the GIS map are nodes in the topology map, and the roadways between nodes are links in the topology map; receiving early warning information from the early warning system, generating evacuation routes based on the location of the monitoring points and the topology network map; and an information synchronization module for sending the evacuation routes to workers through the information transmission module. This invention can predict and warn of fires in advance and generate evacuation routes to guide escape, reducing the risk of externally-caused fires in mines.

Description

Technical Field

[0001] This invention belongs to the field of fire monitoring technology, and in particular relates to an early warning system and emergency response method for coal mine fires caused by external factors based on BiLSTM-Attention. Background Technology

[0002] External fires in coal mines are caused by external heat sources, such as open flames, blasting, methane and coal dust explosions, malfunctioning mechanical equipment, mechanical friction, and short circuits. They can occur anywhere in the mine, but are most common in shafts, shafts, mechanical and electrical chambers, powder magazines, and in roadways or working faces where mechanical and electrical equipment is located. Because external fires have numerous ignition sources and are generally located in areas with good ventilation, they develop very rapidly, causing significant property damage and casualties.

[0003] Traditional monitoring and early warning methods collect a large amount of monitoring data, but have limited data processing and analysis capabilities. They usually issue warnings only after a fire has occurred, making it difficult to detect potential fire risks and anomalies. For rapidly developing external fires in mines, their warning timeliness is low.

[0004] Meanwhile, conventional time-series prediction algorithms suffer from low prediction accuracy and weak generalization, further reducing the accuracy and real-time effectiveness of traditional monitoring and early warning methods. Furthermore, when an external fire warning is issued, it is difficult to effectively notify underground workers. At the same time, most mines lack timely and effective emergency response measures, causing underground workers to miss the optimal escape time, further increasing the risk of fire accidents. Summary of the Invention

[0005] The technical problem to be solved by this invention is to reduce the risk of external fire accidents in mines.

[0006] Therefore, this invention provides an early warning system and emergency response method for coal mine fires caused by external factors based on BiLSTM-Attention.

[0007] The technical solution adopted by this invention to solve its technical problem is: a BiLSTM-Attention coal mine external fire early warning system, comprising: a monitoring module for collecting environmental data related to fire at various monitoring points in the mine; an early warning module for predicting the probability of future fires through algorithms and issuing early warning signals; and an information transmission module for transmitting the early warning signals issued by the early warning module to the staff.

[0008] Through the above technical solution, based on real-time monitoring of underground environmental data, the algorithm in the early warning module predicts the likelihood of a fire occurring in the near future, generates an early warning signal, and transmits the early warning signal to the workers, enabling them to escape in advance before the fire occurs, thereby reducing the risk of external fire accidents in the mine.

[0009] Furthermore, the environmental data collected by the monitoring module mainly includes real-time environmental data of areas prone to external fires, such as underground electromechanical chambers, conveyor belt transport roadways, and auxiliary transport roadways, as well as historical environmental data stored in the InfluxDB time-series database.

[0010] Furthermore, the early warning module includes a BiLSTM-Attention fire prediction model, which comprises an input layer, a bidirectional LSTM layer, an Attention layer, and an output layer. The collected real-time and historical environmental data are processed and pre-processed before being input into the BiLSTM-Attention fire prediction model, which outputs predicted values ​​of environmental parameters for the monitoring area. The early warning module sets threshold values ​​for the parameters of the monitoring points in the external fire monitoring area. If the predicted value of the BiLSTM-Attention fire prediction model exceeds the threshold, an early warning message is generated for the location of the monitoring point.

[0011] Furthermore, the construction of the BiLSTM-Attention fire prediction model includes the following process: Constructing an initialization model: This model includes the BiLSTM output weights W o Forgetting weight W f Input weight W i and candidate memory weights W g The input weights, along with the bias value b, control the updating of input information. The forgetting weights control the forgetting of previous memory units. The output weights control the output of information read from memory units. The candidate memory weights are used to generate candidate memory vectors. The output prediction result is generated by inputting the initial values ​​and historical monitoring data values ​​into the BiLSTM-Attention model and finally outputting the prediction result through the output layer. The model parameters are updated by calculating the loss function, using the mean squared error (MSE) loss function to calculate the error and partial derivative of the initial model output, updating the parameters in the initial model, reducing the model error, and thus completing the training of the prediction model and finally establishing the model.

[0012] Furthermore, the bidirectional LSTM layer includes a forward training sequence structure and a backward training sequence structure. The two structures are symmetrical and the information transmission directions are opposite. The forward training sequence structure is used to calculate the current time information forward, and the backward training sequence structure is used to calculate the same sequence backward. The bidirectional LSTM layer can predict all input vectors according to the temporal sequence of input features, integrate past and future information to output the output layer, and finally integrate forward and backward information to output the prediction result.

[0013] Furthermore, the information transmission module includes a sending module, an online personnel management module, and a message recording module. The sending module is divided into an intranet sending module and a public network sending module. The intranet sending module is used to receive messages between the client and the server. The online personnel management module manages the recipients of various types of messages. The message recording module mainly realizes the query of real-time and historical messages, batch confirmation of messages, maintenance of message cache, and configuration of message permissions.

[0014] This invention also provides an emergency response method, comprising the following steps: receiving and processing information: generating an abstract roadway topology network map based on the actual mine roadway GIS map, where personnel positioning stations on the GIS map are nodes of the topology map, roadways between nodes are links of the topology map, and the endpoint node is a disaster refuge chamber or mine entrance; receiving early warning information issued by an early warning system, the early warning information including the name of the measuring point, environmental data of the measuring point, and the location of the measuring point, and generating a disaster refuge route based on the location of the measuring point and the topology network map; information synchronization module: sending the disaster refuge route generated in the receiving and processing step to relevant personnel underground through an information transmission module.

[0015] By adopting the above technical solution, the optimal disaster avoidance route is generated based on the early warning message and the mine GIS topology map and sent to the personnel underground. When a dangerous situation occurs, this method can reduce casualties among the personnel underground and increase the emergency response capability of the mine for external fires.

[0016] Furthermore, in the topology network diagram, the length between each node and the width of the main and secondary alleyways are set, and the weight between links is set according to the ease of passage. The alleyway length is l, the width is d, and the user-defined ease of passage e is used, with the value of e between 0 and 1. Then, the weight of the alleyway is...

[0017] Furthermore, in the step of receiving and processing information, the weight of the link adjacent to the location of the early warning measuring point is set to infinity, and a disaster avoidance route from the adjacent location of the early warning measuring point to the nearest refuge chamber or wellhead is generated. Then the disaster avoidance route generated by the adjacent nodes will avoid the location of the early warning measuring point.

[0018] Furthermore, the generation of the evacuation route is implemented using Dijkstra's algorithm: selecting all routes that are connected to the starting point V of the evacuation route. s The neighboring node with the smallest weight, V k Its weight is W k Add the node to the route; and select all nodes that are related to V. k The neighboring node with the smallest weight, V m Its weight is W km , with V k As the new intermediate node, modify the starting point V of the disaster evacuation route. s To V m weight W m If V s To V m Passing through intermediate node V k Compare starting point V s To V m weight W m With starting point V s To intermediate node V k and intermediate node V k To V m weight W k +W km Modify the starting point to V m The weight is a smaller weight, and it is added to the route.

[0019] The beneficial effects of this invention are as follows: First, it improves upon traditional time-series prediction methods. By training a time-series model based on prediction results from key underground monitoring points, the model can calculate environmental data for a future timeframe using real-time and historical environmental data. The calculated environmental data is then compared with predetermined environmental thresholds to generate timely and effective underground fire early warning information. Next, a message center module is designed to promptly send early warning messages to underground workers via the mine's intranet and public network, allowing workers to escape before a fire occurs and reducing the risk of externally caused fires in the mine. Simultaneously, based on the early warning messages and the mine's GIS topology map, an improved Dijkstra algorithm is used to generate optimal evacuation routes and send them to underground personnel. In the event of a fire, this method can reduce casualties among underground personnel and increase the mine's emergency response capabilities against externally caused fires. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Figure 1 This is a schematic diagram of the early warning process of an early warning system for coal mine external fires based on BiLSTM-Attention, according to the present invention.

[0022] Figure 2This is a flowchart of the training process for the BiLSTM-Attention coal mine prediction model based on this invention.

[0023] Figure 3 This is a structural diagram of the message center module of this invention.

[0024] Figure 4 This is a node topology diagram shown in an embodiment of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0026] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, features defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0027] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] Example 1

[0029] Combination Figures 1-3 A BiLSTM-Attention early warning system for coal mine fires caused by external factors includes a main and backup monitoring unit and a communication mechanism for signal transmission. The communication mechanism is mainly an underground ring network. The main and backup monitoring units are equipped with a monitoring module, an early warning module, an information transmission module, and an emergency program.

[0030] The monitoring module is used to collect mine environmental data. The monitoring module includes multiple monitoring sensors installed at measuring points in areas prone to fires caused by external factors, such as underground electromechanical chambers, conveyor belt roadways, and auxiliary transport roadways. The monitoring sensors include, but are not limited to, various sensor devices such as dust sensors, CO sensors, O2 sensors, CH4 sensors, and temperature sensors. The monitoring sensors will upload the real-time environmental detection data measured at each measuring point in the mine to the main and backup monitoring machines through a communication mechanism, and save it to the InfluxDB time-series database in the main and backup monitoring machines. The InfluxDB time-series database in the main and backup monitoring machines stores the real-time environmental data and historical environmental data of the measuring points.

[0031] Reference Figure 2 The early warning module includes a data preprocessing group, a prediction group, and an early warning group. The data preprocessing group is used to organize the real-time and historical environmental data of the measurement points stored in the InfluxDB time-series database. The data preprocessing group cleans the data, and its data preprocessing method includes the following steps:

[0032] Step 101: Clean the data and fill in any missing data points using Newton's interpolation method, where the interpolation polynomial is expressed as... Where f[x0,x1,…,x] i ] represents i+1 data points (x0, y0), (x1, y1), ..., (x i ,y i The difference quotient. Definition:

[0033]

[0034] Where, f(x) i )=y i That is, f(x) passes through the known point (x). i ,y i The fitted interpolation function;

[0035] Step 102: Remove data noise by using rule constraints to eliminate obvious outliers;

[0036] Step 103: Downsampling data, similar to the pooling method, uses a sliding window to slide across the historical monitoring data with a window size of 16 data points, and selects the peak value of the data within the window as the flag data to reduce the amount of data and computational complexity;

[0037] Step 104: Data standardization, based on the formula The data is processed using the minimax standardization method to eliminate the dimensional differences between different data.

[0038] Step 105: Split the dataset into training set, validation set and test set in chronological order.

[0039] The prediction group includes a BiLSTM-Attention fire prediction model, which comprises an input layer, a bidirectional LSTM layer, an attention layer, and an output layer. The input layer uses real-time and historical environmental monitoring data from various monitoring points, processed and organized by the data preprocessing group. The bidirectional LSTM layer adds a backpropagation layer to the LSTM in the Bidirectional Long Short-Term Memory Neural Network (Bi-LSTM). The forward and backward training sequences are two independent LSTM structures. These structures are symmetrical, with opposite information transmission directions. The forward training sequence calculates the current time-series information, and the backward training sequence calculates the same sequence backward. The Bi-LSTM predicts all input vectors based on the temporal sequence of the input features. The hidden layer integrates past and future information and outputs it to the output layer. Finally, the forward and backward information are integrated to output the prediction result. The gradients of the prediction model with respect to the input and parameters are calculated in the bidirectional LSTM layer and the attention layer. The output layer is a fully connected layer.

[0040] The construction of the BiLSTM-Attention fire prediction model includes the following process:

[0041] Step 201: Construct the initialization model, which includes the BiLSTM output weights W0 and the forgetting weights W f Input weights W i and candidate memory weights W g The input weight matrix, along with the bias value b, controls the updating of input information. The forgetting weight matrix controls the forgetting of previous memory units. The output weight matrix controls the output of information read from memory units. The candidate memory weight matrix generates candidate memory vectors. Specifically, based on the bidirectional LSTM layer...

[0042] Input Gate:

[0043] Forgotten Gate:

[0044] Output gate:

[0045] Memory unit update:

[0046] In the formula, σ is the sigmoid function, ⊙ is element-wise multiplication, W and b are the weights and biases, xt is the t-th element in the input sequence, and h t-1 i is the hidden state of the previous time step. t f to t These represent the outputs of the input gate, forget gate, and output gate, respectively. t This is the output of the memory unit.

[0047] In BiLSTM, forward and backward LSTM units compute the forward and backward hidden states, respectively. The final output is the concatenation of the hidden states from both directions.

[0048] Step 202: Output the prediction result. Input the initial values ​​and historical monitoring data values ​​into the BiLSTM-Attention model, and finally output the prediction result through the output layer. The fully connected output layer vector undergoes a linear transformation and a non-linear activation function y = f(W). x +b), where W∈R n×m is the weight matrix, b is the bias vector, and the activation function is the sigmoid function.

[0049] Step 203: Update model parameters and calculate the loss function. Calculate the output error and partial derivatives of the initial model using the mean squared error (MSE) loss function. Proceed the output error to the fully connected layer, calculate the gradient of the intermediate vector in the fully connected layer, and pass it to the previous layer. Pass the gradient of the intermediate vector to the BiLSTM and Attention layers, and calculate their gradients with respect to the input and parameters. Use algorithms such as stochastic gradient descent (SGD) to update the value of each parameter based on the gradient of the loss function with respect to the parameters. Monitor the loss function values ​​on the training and validation sets to determine if the model has converged. If the loss function value has converged to a small value, training can be stopped to avoid overfitting, thus completing the training of the prediction model and ultimately building the model.

[0050] The model predicts the data for the next 10 minutes based on the monitoring values ​​of key monitoring points in areas affected by external fires. Historical monitoring data from monitoring points for 1 hour is collected, preprocessed, and input into the model to obtain standardized predicted values. These are then de-standardized to obtain the final predicted values.

[0051] In the early warning group, threshold values ​​for monitoring points in the external fire monitoring area are set. If the predicted value of the BiLSTM-Attention fire prediction model exceeds the threshold, an early warning message for the location of the monitoring point is generated and sent to the information transmission module through the communication mechanism. The early warning message includes the monitoring point name, environmental data of the monitoring point, and location of the monitoring point, etc., based on the location of the monitoring point.

[0052] Reference Figure 3The information transmission module includes a sending module, an online personnel management module, and a message recording module. The sending module retrieves the ID numbers of all personnel within the warning area through the online personnel management module. The online personnel management module mainly manages the recipients of various types of messages to ensure successful message delivery. The message recording module mainly enables real-time and historical message queries, batch message confirmation, maintenance of message cache, and message permission configuration.

[0053] The system sends information to personnel in the warning area, and simultaneously activates the mine's audible and visual alarms to notify them. The sending module is divided into an intranet sending module and a public network sending module. For intranet message sending, when only an intranet is available, messages are sent via WebSocket. The system selects the user based on message type matching; if the user is online, the message is sent directly; if the user is offline, the message is added to a pending message queue for caching, and sent to the user when they come online. Message records are also saved to a database in the information recording module for easy real-time and historical message retrieval. The public network sending module supports message reception on both web and mobile devices. Public network sending utilizes cloud push services provided by JPush and Alibaba Cloud to send and receive various public network messages. First, the public network sending source is configured, and the Push function is added to the App. The platform then calls the public network sending interface to send the generated information to the message center module.

[0054] Example 2

[0055] Combination Figure 4 An emergency response method for a BiLSTM-Attention-based early warning system for coal mine fires caused by external factors includes the following steps:

[0056] Step 1: Receive and process information. The main and backup monitoring units are equipped with emergency programs. These programs generate an abstract tunnel topology network map based on the actual mine tunnel GIS map. Real-time personnel location stations on the GIS map are nodes in the topology, and the tunnels between stations are links. The endpoint nodes are refuge chambers or mine entrances, with mine entrances being high-priority endpoint nodes. Weights between links are set based on length, width of the main and auxiliary tunnels, and ease of passage. For example, assuming a tunnel length of l, width of d, and a user-defined ease of passage e (ranging from 0 to 1), the weights of the links for that tunnel are... For example, assuming node V0 is the starting point of the route, find the minimum weight between V0 and all other nodes, and save the minimum weight between other nodes and the starting node as a set W, for example W{V0,0;V1,∞;V2,∞;V3,∞;V4,∞;} means that in the initial stage, the minimum weight between other nodes and the starting node is assumed to be infinite, and the weight between the starting node and itself is 0. All nodes in the set W are marked as unvisited.

[0057] After the early warning team issues an early warning message, based on the type, location, and monitoring values ​​of the monitoring points in the early warning message, where the location of the monitoring points is the area covered by the early warning message, the emergency procedure obtains the personnel location substation information for that area and generates evacuation routes based on the topology network diagram. The generation of evacuation routes includes the following steps:

[0058] Determine the starting point and ending point: Set the topology node corresponding to the measurement point location in the early warning information as the starting point of the disaster avoidance route, and set the location of its adjacent node as the starting point of the disaster avoidance route in the same way; the ending node is the disaster avoidance chamber or wellhead, and set the link weight adjacent to the location of the early warning measurement point to infinity;

[0059] Route planning: The evacuation route is generated using Dijkstra's algorithm: the endpoint of the evacuation route is set to V. e The starting point of the disaster evacuation route is set at V. s ① Based on the calculation of the weights between nodes, first select the node V with the smallest weight that is adjacent to the starting point of the disaster evacuation route. k It is related to the starting node V s The weight between them is W k ① Add the node to the route; ② Select all nodes related to V. k The neighboring node with the smallest weight, V m Its weight is W km , with V k Modify the starting point V to become the new intermediate node. s To V m weight W m If V s To V m Passing through intermediate node V k Compare starting point V s To V m weight W m With starting point V s To intermediate node V k and intermediate node V k To V m weight W k +W km Modify the starting point to V m The weight is set to a smaller weight and added to the route. ③ Repeat steps ② and ③ above until the starting point V is reached. sTo the destination V e All nodes are included in the route, and the route then passes through the nodes with the lowest weight.

[0060] Combination Figure 4 Specifically, based on the number of nodes n, n sets P(n) are created. Set P(n) represents all nodes with links to node n and their weight information. For example, the information in set P(0) is P(0){V1, 3; V2, 7; V4, 13}, which represents the weight information of three nodes with links to node V0. The information in set P(1) is P(1){V0, 3; V2, 3; V4, 8}, which represents the weight information of three nodes with links to node V0.

[0061] Then, find the node with the smallest current weight in the weight set W between other nodes and the starting node, which is unvisited and not equal to ∞. For example, in the route weight set W{V0,0;V1,∞;V2,∞;V3,∞;V4,∞;}, the node with the smallest current weight is V0. Then, find the established node set P(0){V1,3;V2,7;V4,13} adjacent to V0. P(0) represents all nodes with links to node V0. Update the newly established minimum weight set W{V0,0;V1,∞;V2,∞;V3,∞;V4,∞;} between other nodes and the starting node to the route weight set W'{V0,0;V1,3;V2,7;V3,∞;V4,13;} with V0 as the starting point, and record the path. Mark the V0 node in W' as visited.

[0062] Find the node with the smallest current weight among the weight sets W of other nodes and the starting node, which is unvisited and not equal to ∞: For example, in the updated set W'{V0,0;V1,3;V2,7;V3,∞;V4,13;}, the node with the smallest current weight is V1. Find the set of nodes adjacent to V1 P(1){V0,3;V2,3;V4,8}, P(1) contains three nodes V0,V2,V4.

[0063] If the value of W{V0} in set W has not been visited, the value of W{V0} is not equal to ∞, and W{V0} < W{V1} + P(1){V0}, then update the value of {V0} in set W and record the path. Since V0 has been visited, it is not updated; if the value of W{V2} in set W has not been visited, the value of W{V2} is not equal to ∞, and W{V2} < W{V1} + P(1){V2}, the value of W{V2} is 7, the value of W{V1} is 3, and the value of P(1){V2} is 3, which meets the conditions, then update the value of {V2} in set W, and update W{V0,0; V1,3; V2,7; V3,∞; V4,13;} to W{V0,0; V1,3; V2,6; V3,∞; V4,13;}, and record the path; if the value of W{V4} in set W has not been visited, the value of W{V4} is not equal to ∞, and W{V4} < W{V1} + P(1){V4}, the value of W{V4} is 13, the value of W{V1} is 3, and the value of P(1){V4} is 8, which meets the conditions, then update the value of {V4} in set W, and update W{V0,0; V1,3; V2,6; V3,∞; V4,13;} to W{V0,0; V1,3; V2,6; V3,∞; V4,11;}, and record the path.

[0064] 至此P(1)一共包含三个节点V0,V2,V4全部访问完毕,将W中的V1节点标记为已被访问状态。将W’{V0,0;V1,3;V2,6;V3,∞;V4,13;}更新为W”{V0,0;V1,3;V2,6;V3,∞;V4,11;} At this point, all three nodes V0, V2, and V4 in P(1) have been visited, and the V1 node in W is marked as visited. Update W’{V0,0; V1,3; V2,6; V3,∞; V4,13;} to W”{V0,0; V1,3; V2,6; V3,∞; V4,11;}

[0065] 重复上述步骤直到权重集合W中所有节点都被标记为已访问状态,则起始节点V0到其余所有节点的最小权重均已被找到,其最小权重路径也已被记录。从而最终生成预警测点相邻位置到最近避难硐室或井口(高优先级)避灾路线,则相邻节点生成的避灾路线会避开预警测点所在位置。 Repeat the above steps until all nodes in the weight set W are marked as visited. Then, the minimum weights from the starting node V0 to all other nodes have been found, and their minimum weight paths have also been recorded. Thus, an evacuation route from the adjacent position of the warning measurement point to the nearest refuge chamber or the wellhead (high priority) is finally generated, and the evacuation routes generated by adjacent nodes will avoid the location of the warning measurement point.

[0066] 信息同步模块:将步骤接收并处理信息中生成的避灾路线通过信息传递模块通过内网发送模块和公网发送模块发送至井下相关人员。 Information synchronization module: Send the evacuation routes generated in the step of receiving and processing information to relevant personnel underground through the information transmission module, the intranet sending module, and the public network sending module.

[0067] In summary, this method first improves upon traditional time-series forecasting methods by training a time-series model based on forecasts from key underground monitoring points. This model then calculates future environmental data using real-time and historical data. By comparing the calculated environmental data with predetermined thresholds, timely and effective underground fire early warning information is generated. Next, a message center module is designed to promptly send early warning messages to underground workers via the mine's intranet and public network, enabling them to escape before a fire occurs and reducing the risk of external fire accidents in the mine. Simultaneously, based on the early warning messages and the mine's GIS topology map, an improved Dijkstra algorithm is used to generate optimal evacuation routes and send them to underground personnel. In the event of an emergency, this method can reduce casualties among underground personnel and increase the mine's emergency response capabilities against external fires.

[0068] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined by the scope of the claims.

Claims

1. An early warning system for coal mine fires based on BiLSTM-Attention, characterized in that: include, Monitoring module: Used to collect fire-related environmental data from various monitoring points underground; Early warning module: Predicts the likelihood of future fires using algorithms and issues early warning signals; The early warning module includes a BiLSTM-Attention fire prediction model, which comprises an input layer, a bidirectional LSTM layer, an attention layer, and an output layer. Real-time and historical environmental data are collected, processed, and pre-processed before being input into the BiLSTM-Attention fire prediction model, which outputs predicted values ​​of environmental parameters for the monitoring area. The early warning module sets threshold values ​​for the parameters of the monitoring points in the external fire monitoring area. If the predicted value of the BiLSTM-Attention fire prediction model exceeds the threshold, an early warning message for the location of the monitoring point is generated. The construction of the BiLSTM-Attention fire prediction model includes the following process: Constructing the initialization model: This initialization model includes the BiLSTM output weights. Forgetting weight Input weights and candidate memory weights The input weights, along with the bias value b, are used to control the updating of input information, the forget weights are used to control the forgetting of previous memory units, the output weights are used to control the output of information read from memory units, and the candidate memory gate weights are used to generate candidate memory vectors. Output prediction results: Input the initial values ​​and historical monitoring data values ​​into the BiLSTM-Attention model, and finally output the prediction results through the output layer; Update model parameters: Calculate the loss function. Using the mean squared error (MSE) loss function, calculate the error and partial derivative of the initial model output, update the parameters in the initial model, reduce the model error, and thus complete the training of the prediction model and finally complete the model establishment. The bidirectional LSTM layer includes a forward training sequence structure and a backward training sequence structure. The two structures are symmetrical and the information transmission directions are opposite. The forward training sequence structure is used to calculate the information at the current time step forward, and the backward training sequence structure is used to calculate the same sequence backward. The bidirectional LSTM layer can predict all input vectors according to the temporal sequence of input features, integrate past and future information to output the output layer, and finally integrate forward and backward information to output the prediction result. Information transmission module: Transmits the warning signals issued by the warning module to the staff.

2. The early warning system for coal mine external fires based on BiLSTM-Attention as described in claim 1, characterized in that: The environmental data collected by the monitoring module mainly includes real-time environmental data of areas prone to external fires, such as underground electromechanical chambers, conveyor belt transport roadways, and auxiliary transport roadways, as well as historical environmental data stored in the InfluxDB time-series database.

3. The early warning system for coal mine external fires based on BiLSTM-Attention as described in claim 1, characterized in that: The information transmission module includes a sending module, an online personnel management module, and a message recording module. The sending module is divided into an intranet sending module and a public network sending module. The intranet sending module is used to receive messages between the client and the server. The online personnel management module manages the personnel who are responsible for receiving various types of messages. The message recording module mainly realizes the query of real-time and historical messages, batch confirmation of messages, maintenance of message cache, and configuration of message permissions.

4. An emergency response method for use in an early warning system based on BiLSTM-Attention for coal mine external fires as described in any one of claims 1-3, characterized in that: Includes the following steps: Receive and process information: Generate an abstract roadway topology network map based on the actual mine roadway GIS map. Personnel positioning substations on the GIS map are nodes in the topology map, roadways between nodes are links in the topology map, and the endpoint node is the disaster refuge chamber or mine entrance. Upon receiving an early warning message from the early warning system, the early warning message includes the name of the measuring point, the environmental data of the measuring point, and the location of the measuring point. Based on the location of the measuring point, a disaster avoidance route is generated according to the topology network diagram. Information synchronization module: The disaster avoidance route generated from the information received and processed in the steps is sent to relevant personnel underground through the information transmission module.

5. An emergency response method according to claim 4, characterized in that: The topology network diagram sets the length between nodes, the width of main and secondary alleyways, and the weight between links based on the ease of passage. The alleyway length is l, the width is d, and the user-defined ease of passage is e, with a value between 0 and 1. The weight of that alleyway is... .

6. An emergency response method according to claim 5, characterized in that: In the step of receiving and processing information, the weight of the link adjacent to the location of the early warning measuring point is set to infinity, and a disaster avoidance route from the adjacent location of the early warning measuring point to the nearest refuge chamber or wellhead is generated. Then the disaster avoidance route generated by the adjacent node will avoid the location of the early warning measuring point.

7. An emergency response method according to claim 6, characterized in that: The disaster evacuation route is generated using Dijkstra's algorithm: select all routes that originate from the disaster evacuation route. s The neighboring node with the smallest weight, V k Its weight is W k Add the node to the route; And select all that are related to V k The neighboring node with the smallest weight, V m Its weight is W km , with V k As the new intermediate node, modify the starting point V of the disaster evacuation route. s To V m weight W m If V s To V m Passing through intermediate node V k Compare starting point V s To V m weight W m With starting point V s To intermediate node V k and intermediate node Vk to V m weight W k +W km Modify the starting point to V m The weight is a smaller weight, and it is added to the route.

Images