A method of intelligent early warning while drilling based on multi-source disaster information entropy fusion

Abstract

The invention discloses a method for intelligent early warning while drilling based on multi-source disaster information entropy fusion. A multi-source sensing module is used to obtain multi-source data of different depths during the drilling process, and then the depth segments are divided, and the multi-source data of each depth segment are normalized and dimensionless. The method is used to increase the difference of data in each depth segment, and optimize the subsequent data processing. Then, the processed data is subjected to information entropy fusion data processing, which can further amplify the difference of data of the same category and optimize the data fusion effect. After the processed data is input into the KAN network training, an early warning model is established. Finally, the early warning model is used to input the real-time collected data into the model, so that the information entropy fusion data of a depth segment that has not been detected ahead can be predicted, and after comparing the data with a threshold value, it is finally determined that there is a risk of flood, fire or landslide in a depth segment ahead, and corresponding measures are taken in time according to the risk situation, thereby ensuring the safety of tunnel excavation.

Description

Technical Field

[0001] The invention relates to an intelligent early warning method while drilling based on multi-source disaster information entropy fusion, belonging to the field of advanced disaster prediction in the mining process. Background Art

[0002] Floods, fires, landslides and other disasters are prone to occur during resource exploitation. Although mines have achieved preliminary monitoring and early warning of these disasters, the level of intelligence is low, making it difficult to achieve self-analysis and decision-making capabilities. Disasters can easily cause serious economic losses and even threaten personal safety.

[0003] In the process of resource development, disasters such as water, fire, and landslides are prone to occur. Flood accidents refer to disasters caused by the sudden influx of water sources such as groundwater, aquifer water, fault water, or water accumulation in old tunnels through cracks, faults, or other channels; fire accidents refer to disasters caused by spontaneous combustion, explosives, gas leakage, etc. during coal mining; landslide accidents refer to disasters caused by insufficient roof support strength, water gushing, spontaneous combustion of coal seams, and poor geological conditions.

[0004] Therefore, how to timely discover different disasters in front of the excavation face during resource development and take timely measures to avoid dangers during the excavation process and ultimately effectively ensure the safety of excavation is the research direction required by the present invention. Summary of the invention

[0005] In view of the problems existing in the above-mentioned prior art, the present invention provides an intelligent while-drilling advance warning method based on multi-source disaster information entropy fusion, which can eventually warn of multi-source disasters ahead of the tunnel excavation process by normalizing the collected multi-source data and fusing information entropy, thereby ensuring the safety of tunnel excavation.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is: a method for early warning while drilling based on multi-source disaster information entropy fusion, the specific steps are:

[0007] Step 1: Data collection: Install the multi-source sensor module in the measurement while drilling rod, and drill a hole in the detection area along with the drill rod, and continuously collect multi-source data at different depths during the drilling process;

[0008] Step 2: Data normalization: The multi-source data of different depths obtained in step 1 are first divided into different depth segments according to a set distance, and the multi-source data of each depth segment are normalized and dimensionless;

[0009] Step 3: Information entropy fusion data processing: The multi-source data of each depth segment processed in step 2 are processed by information entropy fusion data respectively, and then the warning thresholds of floods, fires and landslides that may occur during the drilling process are obtained as benchmark data;

[0010] Step 4: Construct a multi-source heterogeneous disaster intelligent early warning model: Use the KAN network to establish an early warning model, and then input the data of each depth segment processed by the information entropy fusion data in step 3 into the established early warning model for training. After the training is completed, the model can output the information entropy fusion data of a depth segment that has not been detected ahead, thereby completing the construction of a multi-source heterogeneous disaster intelligent early warning model;

[0011] Step 5, advanced intelligent warning: re-drilling detection, and the multi-source sensor module collects data in real time during the drilling process, and a group of data is formed for each depth segment drilled. After obtaining multiple groups of data, stop drilling, repeat the processing of steps 2 and 3 for each group of data, and obtain multiple groups of information entropy fusion data. Finally, input the data into the multi-source heterogeneous disaster intelligent warning model established in step 4, and then the model can output the information entropy fusion data of a depth segment that has not been detected in front, and compare the fusion data with the benchmark data in step 3. If it exceeds, it is judged that there is a risk of flood, fire or landslide in the depth segment ahead, and then stop drilling and take corresponding measures; if it does not exceed, it is judged that there is no risk of flood, fire or landslide in the depth segment ahead, and then continue to drill a depth segment, and process the data collected in the depth segment to obtain information entropy fusion data to form a new group of data, and then input this group of data into the model. At this time, the model outputs the information entropy fusion data of a depth segment that has not been detected in front of the current position, and then compares it with the benchmark data to judge whether the disaster situation continues, so as to realize the advanced intelligent warning while drilling.

[0012] Furthermore, the multi-source sensing module in step one includes a temperature acquisition module, a humidity acquisition module and a strain acquisition module. The temperature acquisition module is used to collect temperature data of the surrounding environment during the drilling process in real time. The humidity acquisition module is used to collect humidity data of the surrounding environment during the drilling process in real time. The strain acquisition module is used to collect strain data of the drill rod during the drilling process in real time.

[0013] Furthermore, it is characterized in that the step 2 specifically comprises: firstly dividing into different depth segments with a set distance of 2m to obtain a total of multiple groups of data; then processing each group of data as follows:

[0014] By min-max normalization:

[0015] Where: x is the original data, min(x) is the minimum value in the data, max(x) is the maximum value in the data, and x' is the normalized data;

[0016] Normalize dimensionlessly by Z-score:

[0017] Where: x is the original data, x' is the dimensionless data, μ is the mean, and σ is the standard deviation.

[0018] Furthermore, the specific process of information entropy fusion data processing in step 3 is as follows:

[0019] Assume that the number of data in a depth segment is n, and the total amount of information contained in the i-th multi-source data in the depth segment is x i , then the total information amount Q of this depth segment is expressed as:

[0020]

[0021] Then the frequency P of the i-th multi-source data in the total information volume is i for:

[0022]

[0023] Where P i The relationship satisfied is:

[0024] x i The amount of self-information I(x i ) is defined as: I(x i )=-log a P i ;

[0025] I(x i ) is a random variable, for all information I(x i ) to find the average value, that is, to calculate the mathematical expectation of all information from the information, which is the information entropy of the signal, denoted as P(x). The specific formula is as follows, where: P i represents the possibility of the occurrence of the i-th multi-source data; lnp i Measure the amount of information contained in the i-th multi-source data;

[0026]

[0027] Through the theory of information entropy fusion, the n groups of multi-source data (i.e., temperature data, humidity data, and strain data) in the depth segment are fused, and the data conditions of the n groups of multi-source data in the safe state and the disaster state are analyzed to determine the warning thresholds for possible floods, fires, and landslides in this depth segment. This step is repeated for the remaining depth segments to obtain the warning thresholds for each depth segment, and finally the average of each warning threshold is used as the benchmark data. Each subsequent time a new depth segment is drilled, the warning threshold for possible floods, fires, and landslides in this depth segment is determined, and the average is combined with the warning thresholds of the previous depth segments as the new benchmark data, and the benchmark data is continuously optimized in this way.

[0028] Further, the step 4 is specifically as follows:

[0029] The KAN network is used to establish the early warning model. The KAN network is a combination of 1 layers, and the representation form is:

[0030]

[0031] Among them, Φ 1 is the B-spline function matrix corresponding to the first KAN layer; x is the input matrix; Indicates the connection between layers and the conformity of functions;

[0032] The simplified KAN network is characterized as follows:

[0033]

[0034] Among them, Φ out is the function matrix corresponding to the output KAN layer; Φ in is the function matrix corresponding to the input KAN layer; x is the input matrix;

[0035] The n groups of information entropy fused data are input into the KAN network model for training. After the training is completed, the model can output the information entropy fused data of a depth segment that has not been detected ahead, thereby completing the construction of a multi-source heterogeneous disaster intelligent early warning model.

[0036] Compared with the prior art, the present invention adopts a multi-source sensing module to obtain multi-source data of different depths during the drilling process, and then divides the depth segments, and normalizes and dimensionlessly processes the multi-source data of each depth segment; its purpose is to increase the difference in data of each depth segment, thereby optimizing the subsequent data processing, and then performing information entropy fusion data processing on the processed data, which can further amplify the difference between data of the same category and optimize the data fusion effect; the processed data is input into the KAN network training to establish an early warning model; finally, the early warning model is used to input the real-time collected data into the model, which can predict the information entropy fusion data of a depth segment that has not been detected ahead, and after comparing the data with the threshold, it is finally determined that there is a risk of flood, fire or landslide in a depth segment ahead, and corresponding measures are taken in time according to the risk situation, thereby ensuring the safety of tunnel excavation. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the process of the present invention;

[0038] Figure 2 It is a schematic diagram of the layout of the multi-source sensor module in the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below.

[0040] like Figure 1 As shown, the specific steps of the present invention are:

[0041] Step 1: Data collection: Install the multi-source sensor module in the measurement while drilling rod, and drill a hole in the detection area along with the drill rod, and continuously collect multi-source data at different depths during the drilling process; Figure 2 As shown, the multi-source sensing module in step one includes a temperature acquisition module, a humidity acquisition module and a strain acquisition module. The temperature acquisition module is used to collect the temperature data of the surrounding environment during the drilling process in real time. The humidity acquisition module is used to collect the humidity data of the surrounding environment during the drilling process in real time. The strain acquisition module is used to collect the strain data of the drill rod during the drilling process in real time.

[0042] Step 2: Data normalization: The multi-source data of different depths obtained in step 1 are first divided into different depth segments according to the set distance, and the multi-source data of each depth segment are normalized and dimensionless. The purpose is to optimize the effect of data processing and accelerate the convergence speed of the KAN intelligent algorithm. Specifically, the data is first divided into different depth segments with a set distance of 2m, namely 0-2m, 2-4m, 4-6m, 6-8m, 8-10m, and a total of multiple groups of data are obtained; then each group of data is processed as follows:

[0043] By min-max normalization:

[0044] Where: x is the original data, min(x) is the minimum value in the data, max(x) is the maximum value in the data, and x' is the normalized data;

[0045] Normalize dimensionlessly by Z-score:

[0046] Where: x is the original data, x' is the dimensionless data, μ is the mean, and σ is the standard deviation.

[0047] Step 3: Information entropy fusion data processing: Perform information entropy fusion data processing on the multi-source data of each depth segment after processing in step 2. The technical purpose of information entropy fusion is to improve the system's ability to handle uncertainty and complex data. Shannon proposed the concept of information entropy to represent the degree of order of information. The more disordered the information is, the greater the information entropy is; the more ordered the information is, the smaller the information entropy is.

[0048] If a disaster such as flood, fire or landslide occurs somewhere during resource exploitation, the temperature, humidity and stress will change slightly. The more information needs to be transmitted, the faster the change will be, which will increase the information entropy value. Information entropy is a quantitative indicator used to evaluate the amount of information contained in the overall signal. A depth segment of every two meters of drilling detection depth is selected, and the temperature, humidity and stress acquisition module data obtained are processed by information entropy fusion data. The purpose is to amplify the mutation position and optimize the data processing effect; because the data segmentation differences of temperature and humidity during the whole drilling process are too small, it is still difficult to see significant differences after dimensionless normalization, so the information entropy fusion theory is introduced to process the measured data; this method can amplify the differences between data of the same category and optimize the data fusion effect; the temperature and humidity data after information entropy fusion processing can effectively distinguish the differences between data of different depth segments, and the subsequent early warning model established by the KAN network plays an optimization role in intelligent recognition. The specific process is as follows:

[0049] Assume that the number of data in a depth segment is n, and the total amount of information contained in the i-th multi-source data in the depth segment is x i , then the total information amount Q of this depth segment is expressed as:

[0050]

[0051] Then the frequency P of the i-th multi-source data in the total information volume is i for:

[0052]

[0053] Where P i The relationship satisfied is:

[0054] x i The amount of self-information I(x i ) is defined as: I(x i )=-log a P i ;

[0055] I(x i ) is a random variable, for all information I(x i ) to find the average value, that is, to calculate the mathematical expectation of all information from the information, which is the information entropy of the signal, denoted as P(x). The specific formula is as follows, where: P i represents the possibility of the occurrence of the i-th multi-source data; lnp i Measure the amount of information contained in the i-th multi-source data;

[0056]

[0057] Through the theory of information entropy fusion, the n groups of multi-source data (i.e., temperature data, humidity data, and strain data) in the depth segment are fused, and the data conditions of the n groups of multi-source data in the safe state and the disaster state are analyzed to determine the warning threshold for possible floods, fires, and landslides in this depth segment. This step is repeated for the remaining depth segments to obtain the warning thresholds for each depth segment, and finally the average of each warning threshold is used as the benchmark data. Each subsequent time a new depth segment is drilled, the warning threshold for possible floods, fires, and landslides in this depth segment is determined, and the average is combined with the warning thresholds of the previous depth segments as the new benchmark data, and the benchmark data is continuously updated in this way.

[0058] Step 4. Construct a multi-source heterogeneous disaster intelligent early warning model: Use the KAN network to establish the early warning model. The KAN network based on the KAN representation theorem can effectively alleviate the performance degradation problem caused by the increase in the number of input features (i.e., the increase in temperature acquisition module, humidity acquisition module, strain acquisition module, etc.) in a short period of time; the advantage of the KAN network is that its powerful pruning ability can automatically remove unimportant detection and monitoring data within the safe time of the entire drilling process to simplify the network structure, which can not only reduce the complexity of the model, but also improve the training speed and generalization ability of the model.

[0059] Construct the mapping relationship from input to output, that is, construct the transformation (matrix) from the input layer to the output layer:

[0060]

[0061] The KAN network is a combination of 1 layers, represented as:

[0062]

[0063] Among them, Φ 1 is the B-spline function matrix corresponding to the first KAN layer; x is the input matrix; Indicates the connection between layers and the conformity of functions;

[0064] The simplified KAN network is characterized as:

[0065]

[0066] Among them, Φ out is the function matrix corresponding to the output KAN layer; Φ in is the function matrix corresponding to the input KAN layer; x is the input matrix;

[0067] The n groups of data after information entropy fusion processing of drilling depths of 0-2m, 2-4m, 4-6m, 6-8m, and 8-10m are input into the KAN network model for training. After the training, the model can output the information entropy fusion data of a depth segment that has not been detected in front, thereby completing the construction of a multi-source heterogeneous disaster intelligent early warning model; the purpose of this step is to predict whether the temperature data, humidity data, and strain data within the range of 2m in front of the drilling position exceed the set safety threshold, and then achieve the purpose of advanced intelligent early warning.

[0068] Step 5, advanced intelligent warning: re-drilling detection, and the multi-source sensor module collects data in real time during the drilling process, and a group of data is formed for each depth segment drilled. After obtaining multiple groups of data, stop drilling, repeat the processing of steps 2 and 3 for each group of data, and obtain multiple groups of information entropy fusion data. Finally, input the data into the multi-source heterogeneous disaster intelligent warning model established in step 4, and then the model can output the information entropy fusion data of a depth segment that has not been detected in front, and compare the fusion data with the benchmark data in step 3. If it exceeds, it is judged that there is a risk of flood, fire or landslide in the depth segment ahead, and then stop drilling and take corresponding measures; if it does not exceed, it is judged that there is no risk of flood, fire or landslide in the depth segment ahead, and then continue to drill a depth segment, and process the data collected in the depth segment to obtain information entropy fusion data to form a new group of data, and then input this group of data into the model. At this time, the model outputs the information entropy fusion data of a depth segment that has not been detected in front of the current position, and then compares it with the benchmark data to judge whether the disaster situation continues, so as to realize the advanced intelligent warning while drilling.

[0069] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for early warning while drilling based on multi-source disaster information entropy fusion, characterized in that: The specific steps are: Step 1: Data collection: Install the multi-source sensor module in the measurement while drilling rod, and drill a hole in the detection area along with the drill rod, and continuously collect multi-source data at different depths during the drilling process; Step 2: Data normalization: The multi-source data of different depths obtained in step 1 are first divided into different depth segments according to a set distance, and the multi-source data of each depth segment are normalized and dimensionless; Step 3: Information entropy fusion data processing: The multi-source data of each depth segment processed in step 2 are processed by information entropy fusion data respectively, and then the warning thresholds of floods, fires and landslides that may occur during the drilling process are obtained as benchmark data; Step 4: Construct a multi-source heterogeneous disaster intelligent early warning model: Use the KAN network to establish an early warning model. The KAN network is a combination of 1 layers, and the representation form is: Among them, Φ1 is the B-spline function matrix corresponding to the first KAN layer; x is the input matrix; ° represents the sign of the inter-layer connection function; The simplified KAN network is characterized as follows: Among them, Φ out is the function matrix corresponding to the output KAN layer; Φ in is the function matrix corresponding to the input KAN layer; x is the input matrix; Input n groups of information entropy fusion processed data into the KAN network model for training. After training, the model can output information entropy fusion data of a depth segment that has not been detected ahead, thereby completing the construction of a multi-source heterogeneous disaster intelligent early warning model. Step 5, advanced intelligent warning: re-drilling detection, and the multi-source sensor module collects data in real time during the drilling process, and a group of data is formed for each depth segment drilled. After obtaining multiple groups of data, stop drilling, repeat the processing of steps 2 and 3 for each group of data, and obtain multiple groups of information entropy fusion data. Finally, input the data into the multi-source heterogeneous disaster intelligent warning model established in step 4, and then the model can output the information entropy fusion data of a depth segment that has not been detected in front, and compare the fusion data with the benchmark data in step 3. If it exceeds, it is judged that there is a risk of flood, fire or landslide in the depth segment ahead, and then stop drilling and take corresponding measures; if it does not exceed, it is judged that there is no risk of flood, fire or landslide in the depth segment ahead, and then continue to drill a depth segment, and process the data collected in the depth segment to obtain information entropy fusion data to form a new group of data, and then input this group of data into the model. At this time, the model outputs the information entropy fusion data of a depth segment that has not been detected in front of the current position, and then compares it with the benchmark data to judge whether the disaster situation continues, so as to realize advanced intelligent warning while drilling.

2. According to claim 1, the method for early warning while drilling based on multi-source disaster information entropy fusion is characterized in that: The multi-source sensing module in step one includes a temperature acquisition module, a humidity acquisition module and a strain acquisition module. The temperature acquisition module is used to collect temperature data of the surrounding environment during the drilling process in real time. The humidity acquisition module is used to collect humidity data of the surrounding environment during the drilling process in real time. The strain acquisition module is used to collect strain data of the drill rod during the drilling process in real time.

3. According to claim 2, the method for early warning while drilling based on multi-source disaster information entropy fusion is characterized in that: The specific step 2 is: firstly divide the data into different depth segments with a set distance of 2m, and obtain multiple groups of data; then process each group of data as follows: By min-max normalization: Where: x is the original data, min(x) is the minimum value in the data, max(x) is the maximum value in the data, and x' is the normalized data; Normalize dimensionlessly by Z-score: Where: x is the original data, x' is the dimensionless data, μ is the mean, and σ is the standard deviation.

4. According to claim 1, the method for early warning while drilling based on multi-source disaster information entropy fusion is characterized in that: The specific process of information entropy fusion data processing in step 3 is as follows: Assume that the number of data in a depth segment is n, and the total amount of information contained in the i-th multi-source data in the depth segment is x i , then the total information amount Q of this depth segment is expressed as: Then the frequency P of the i-th multi-source data in the total information volume is i for: Where P i The relationship satisfied is: x i The amount of self-information I(x i ) is defined as: I(x i )=-log a P i ; I(x i ) is a random variable, for all information I(x i ) is used to find the average value, that is, to calculate the mathematical expectation of all information, which is the information entropy of the depth segment, denoted as P(x). The specific formula is as follows: i represents the possibility of the occurrence of the i-th multi-source data; lnp i Measure the amount of information contained in the i-th multi-source data; Through the theory of information entropy fusion, the n groups of multi-source data in the depth segment are fused, and the data conditions of the n groups of multi-source data in the safe state and the disaster state are analyzed to determine the warning thresholds for possible floods, fires, and landslides in the depth segment. This step is repeated for the remaining depth segments to obtain the warning thresholds for each depth segment, and finally the average of each warning threshold is used as the benchmark data. Each subsequent time a new depth segment is drilled, the warning threshold for possible floods, fires, and landslides in the depth segment is determined, and the average is combined with the warning thresholds of the previous depth segments as the new benchmark data, and the benchmark data is continuously optimized in this way.

Images