CN118964990A - A rock burst early warning method based on KAN neural network - Google Patents
A rock burst early warning method based on KAN neural network Download PDFInfo
- Publication number
- CN118964990A CN118964990A CN202411026021.0A CN202411026021A CN118964990A CN 118964990 A CN118964990 A CN 118964990A CN 202411026021 A CN202411026021 A CN 202411026021A CN 118964990 A CN118964990 A CN 118964990A
- Authority
- CN
- China
- Prior art keywords
- neural network
- rock burst
- early warning
- layer
- kan
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 149
- 239000011435 rock Substances 0.000 title claims abstract description 129
- 238000000034 method Methods 0.000 title claims abstract description 15
- 238000004364 calculation method Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 15
- 238000012544 monitoring process Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 210000002569 neuron Anatomy 0.000 claims description 61
- 238000012545 processing Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 3
- 238000012905 input function Methods 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Neurology (AREA)
- Mathematical Optimization (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Pure & Applied Mathematics (AREA)
- Molecular Biology (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Computing Systems (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Linguistics (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
本发明涉及矿井中冲击地压等危险灾害预警领域，具体为一种基于KAN神经网络的冲击地压预警方法，包括预处理原始微震监测数据；建立冲击地压预警数据集；构建基于KAN神经网络的冲击地压预警算法；训练基于KAN神经网络的冲击地压预警算法；基于KAN神经网络的冲击地压预警算法应用。本发明基于KAN神经网络的冲击地压预警算法中需要训练的参数仅有w和cz，与以往神经网络相比训练参数大大减少；而且由于KAN神经网络中的计算函数非线性描述能力更优，所以在不依靠人为设定的预警指标情况下基于KAN神经网络的冲击地压预警算法仅靠少量参数即可取得比传统神经网络更优的性能。
The present invention relates to the field of early warning of dangerous disasters such as rock burst in mines, and specifically to a rock burst early warning method based on a KAN neural network, including preprocessing original microseismic monitoring data; establishing a rock burst early warning data set; constructing a rock burst early warning algorithm based on a KAN neural network; training the rock burst early warning algorithm based on a KAN neural network; and applying the rock burst early warning algorithm based on a KAN neural network. In the rock burst early warning algorithm based on a KAN neural network of the present invention, the parameters that need to be trained are only w and c z , which greatly reduces the training parameters compared with previous neural networks; and because the calculation function in the KAN neural network has a better nonlinear description ability, the rock burst early warning algorithm based on the KAN neural network can achieve better performance than the traditional neural network with only a small number of parameters without relying on artificially set early warning indicators.
Description
Technical Field
The invention relates to the field of dangerous disaster early warning of rock burst and the like in mines, in particular to a rock burst early warning method based on a KAN neural network.
Background
The rock burst has the characteristics of violent damage and sudden occurrence, and is a typical dynamic disaster in the deep mine exploitation process. The occurrence of rock burst can cause casualties of workers and the damage of roadways, and the current use of mine monitoring data for real-time early warning of the rock burst is a precondition for effectively preventing the occurrence of rock burst.
The microseismic monitoring data is high-quality monitoring data which can reflect mine exploitation conditions. Because of the complexity of rock burst early warning tasks, it is difficult to obtain a satisfactory early warning result by directly utilizing a neural network algorithm to process microseismic monitoring data. In the existing rock burst early warning algorithm (such as patent ZL 202111558332.8) based on the neural network, rock burst early warning indexes are often constructed by utilizing microseismic monitoring data, then weights are given to the rock burst early warning indexes by means of a complex neural network algorithm, and then the impact risk is calculated. In the past rock burst early warning algorithm, due to limited fitting capacity of a network, dominant rock burst early warning indexes are required to be manually determined by means of expert experience, and the early warning indexes can only be used for one mine, and have limited fitting capacity. The traditional rock burst early warning algorithm is usually composed of a convolutional neural network and a cyclic neural network, and the rock burst early warning algorithm of the neural network has huge calculation parameters and has the defects of long training and reasoning time and the like. And because of the complexity of the networks, the networks are always ash box states in the rock burst early warning process, and the meaning corresponding to the node parameters in the networks is difficult to determine.
Disclosure of Invention
In order to solve the problems, the invention provides a rock burst early warning method based on a KAN neural network, which is used for determining the probability of future microseismic events having rock burst risk, and comprises the following steps:
s1: preprocessing original microseismic monitoring data
The original microseismic monitoring data comprises a plurality of original microseismic events which are continuously monitored and recorded, wherein each original microseismic event comprises the time, the X coordinate, the Y coordinate, the Z coordinate, the energy and whether the rock burst happens or not; preprocessing an original microseismic event to obtain a preprocessed microseismic event, wherein the preprocessed microseismic event comprises a time interval, a place X coordinate, a place Y coordinate, a place Z coordinate, energy and whether rock burst is performed or not, and the time interval is the time interval between the microseismic event and the previous microseismic event;
s2: establishing rock burst early warning data set
Extracting continuous M times of preprocessed microseismic events, taking data of the M-th to m+t-th preprocessed microseismic events as a data source of one sample in rock burst early warning data, wherein M is more than or equal to 1 and less than or equal to M-t, and M-t samples can be established at most, wherein each sample comprises a characteristic and a label, the characteristic of each sample comprises a time interval, a place X coordinate, a place Y coordinate, a place Z coordinate and energy of the m+t-th preprocessed microseismic events, and the label of each sample is the rock burst risk of the m+t-th microseismic events; dividing the rock burst early warning data set into a training set and a testing set;
s3: construction of rock burst early warning algorithm based on KAN neural network
The constructed rock burst early warning algorithm based on the KAN neural network has 4 layers of neural networks, each layer of neural network comprises input, calculation functions and output, and the output of the former layer of neural network is the input of the next layer of neural network; the number of neurons of each layer of neural network is n 1、n2、n3 and n 4,n4 =2 in sequence;
the input of the rock burst early warning algorithm is also the input of the first layer neural network, and is the sample characteristic in the rock burst early warning data set Including time interval, location X-coordinate, location Y-coordinate, location Z-coordinate and energy, respectively calculated as X 0,1、x0,2、x0,3、x0,4、x0,5;
The number of the calculation functions of the first layer neural network is equal to the product of the number n 0 =5 of the input functions and the number of the neurons, and the output functions are the number of the neurons of the first layer neural network; the input number of the second layer and the following neural networks is equal to the neuron number of the previous layer of the neural network; the number of the calculation functions of the second layer and the subsequent neural networks is equal to the product of the number of neurons of the previous layer and the number of neurons of the present layer, and the number of the output of the second layer and the subsequent neural networks is equal to the number of neurons of the second layer and the subsequent neural networks;
The ith neuron of the first layer neural network is represented by (l, i l), and the output of the neuron (l, i l) is used A representation; connecting inputs to a layer-1 neural networkAnd its neuron (l, i l) are written asFor the layer i neural network, assuming that there are n l neurons in total, the number of inputs is the number n l-1 of neurons of the previous layer neural network, the inputs are the sample feature x 0,i when being the first layer neural network, the number n l-1 = 5, and the output of a neuron (i, i l) is the calculated sum of all incoming inputs of that neuron:
In the formula, To connect inputA computational function with the ith neuron of the layer i neural network; the output of the ith l-1 th neuron in the layer 1 neural network, which is also the input to the layer 1 neural network, when l=1 Representing a sample feature;
the constructed burst early warning algorithm based on the KAN neural network has 4 layers of neural networks, takes the characteristics of a sample as input, and outputs the constructed burst early warning algorithm based on the KAN neural network are as follows:
Wherein y 1,y2 represents the risk probability of no rock burst and the risk probability of rock burst respectively; i 1,i2,i3 is the ith neuron in the first, second and third layer neural networks respectively; i 0 denotes the ith input of the first layer neural network;
Wherein the function is calculated Is the sum of the basis function b (x) and the spline function:
wherein w is a trainable parameter; spline (x) is a spline function; b (x) is a basis function, x is an input value;
b(x)＝SiLU(x)＝x/(1+e-x) (4)
wherein SiLU denotes Sigmoid Linear Unit activation function; e is a natural constant;
Where spline (x) is a linear spline function:
Wherein, B z (DEG) represents the corresponding Z times in the spline curve, and a cubic spline curve is adopted in the patent, so that the maximum value Z of Z is=3; c z represents a weight value for B z (·) is a trainable parameter;
s4: rock burst early warning algorithm based on KAN neural network
Firstly, giving an initial value of each trainable parameter in a rock burst early warning algorithm based on a KAN (Kan-based neural network), initializing the trainable parameter to 0 aiming at a spline function in a calculation function, and initializing other trainable parameters according to an Xavier initialization rule; inputting a training set sample into a rock burst early-warning algorithm based on a KAN neural network for training, evaluating the difference between an early-warning result of the rock burst early-warning algorithm and a sample label by using a cross entropy loss function, and updating parameters in the rock burst early-warning algorithm to be trained by using a back propagation algorithm;
evaluating the performance of a trained rock burst early warning algorithm by using a test set sample; and (3) storing the rock burst early-warning algorithm parameters with the best effect to obtain a trained rock burst early-warning algorithm based on the KAN neural network.
Preferably, step S1 further includes: and carrying out standardization treatment on the time interval, the location X coordinate, the location Y coordinate, the location Z coordinate and the energy of the preprocessed microseismic events.
Preferably, in step S2, the label is subjected to One-Hot encoding, and the label after the encoding is 1 for the rock burst, and the label after the encoding is not 0 for the rock burst.
Preferably, the method further comprises step S5: KAN neural network-based rock burst early warning algorithm application
And (3) acquiring the latest monitored continuous t original microseismic events, performing data processing according to the steps S1-S2 to obtain a sample with only features and no tag, inputting the features of the sample into the rock burst early warning algorithm trained in the step S4 and based on the KAN neural network to obtain the probability that the future microseismic event has rock burst risk.
The invention has the beneficial effects that: the parameters to be trained in the rock burst early warning algorithm based on the KAN neural network only comprise w and c z, and compared with the conventional neural network, the training parameters are greatly reduced; and because the nonlinear description capability of the calculation function in the KAN neural network is better, the rock burst early warning algorithm based on the KAN neural network can obtain better performance than the traditional neural network by only relying on a small number of parameters under the condition of not relying on the artificially set early warning index. Meanwhile, as the calculation function and the learnable parameters in the algorithm are clear, the formula of the algorithm per se can be expressed (see formula 2) for the rock burst early warning task, and the algorithm has a certain interpretation.
Drawings
FIG. 1 is a flowchart of a method for early warning of rock burst based on a KAN neural network;
fig. 2 is a frame diagram of a rock burst early warning algorithm based on a KAN neural network.
Detailed Description
The technical scheme of the invention is described in more detail below with reference to the accompanying drawings.
A rock burst early warning method based on a KAN neural network is used for early warning the probability of rock burst risk (rock burst) of a future microseismic event under the condition of no artificial setting of early warning indexes, and comprises the following steps:
s1: preprocessing original microseismic monitoring data
The original microseismic monitoring data comprises a plurality of original microseismic events which are continuously monitored and recorded, wherein each original microseismic event comprises the time, the X coordinate, the Y coordinate, the Z coordinate, the energy and whether the rock burst happens or not; preprocessing an original microseismic event to obtain a preprocessed microseismic event, wherein the preprocessed microseismic event comprises a time interval, a place X coordinate, a place Y coordinate, a place Z coordinate, energy and whether rock burst is generated, the time interval is the time interval between the microseismic event and the previous microseismic event, and the time interval, the place X coordinate, the place Y coordinate, the place Z coordinate and the energy are subjected to standardized processing; the standardized processing is a data processing method commonly used in the art, and is not described herein;
s2: establishing rock burst early warning data set
Extracting continuous M times of preprocessed microseismic events, taking data of the M-th to m+t-th preprocessed microseismic events as a data source of One sample in rock burst early warning data, wherein M is more than or equal to 1 and less than or equal to M-t, and M-t samples can be established at most, wherein each sample comprises a feature and a label, the feature of each sample comprises a time interval, a place X coordinate, a place Y coordinate, a place Z coordinate and energy of the M-th to m+t-1-th preprocessed microseismic events, the label of each sample is the rock burst risk of a future microseismic monitoring event, namely the rock burst risk (whether the rock burst is the rock burst) of the m+t-th microseismic event, the label is required to be subjected to One-Hot encoding processing, and the label of the rock burst is 1 instead of the label of the rock burst is 0 after processing; dividing the rock burst early warning data set into a training set and a testing set according to the proportion of 7:3;
s3: construction of rock burst early warning algorithm based on KAN neural network
1-2, The constructed burst pressure early warning algorithm based on the KAN neural network has 4 layers of neural networks, and the number of neurons of each layer of neural network is n 1、n2、n3 and n 4 in sequence; in this embodiment, n 1、n2、n3 and n 4 are 8, 16, 8 and 2 respectively, where the number n 4 of neurons in the last layer of neural network is related to the type of rock burst risk, and in this patent, the rock burst early warning result only has two cases of rock burst risk and rock burst-free risk, so the number n 4 of neurons in the last layer of neural network is 2; each layer of neural network comprises an input, a calculation function and an output, wherein the number of the calculation functions is an integer multiple of neurons;
The input of the rock burst early warning algorithm is also the input of a first layer of neural network, and is sample characteristics X 0,i0 in the rock burst early warning data set, including time interval, location X coordinate, location Y coordinate, location Z coordinate and energy, which are respectively calculated as X 0,1、x0,2、x0,3、x0,4、x0,5; the output of the rock burst early warning algorithm is also the output of a4 th layer neural network, and corresponds to the label of a sample, wherein the probability of rock burst (with rock burst risk) and the probability of rock burst (without rock burst risk) are the probability of a future microseismic event; the output of the previous layer of neural network is the input of the next layer of neural network;
The number of the calculation functions of the first layer neural network is equal to the product of the number 5 of the input functions and the number of the neurons, 40 are used in the embodiment, and the output functions are the number of the neurons of the first layer neural network; the input number of the second layer and the following neural networks is equal to the neuron number of the previous layer of the neural network; the number of the calculation functions of the second layer and the subsequent neural networks is equal to the product of the number of the neurons of the previous layer (the number of the neurons of the previous layer is also the number of the neurons of the present layer which are the number of 128, 128 and 16 respectively in the embodiment; the output number of the second layer and the following neural networks is equal to the number of the neurons;
Representing the ith neuron of the first layer neural network in a form (l, i l) similar to coordinates, and using the output of the neuron (l, i l) Representation at the same timeIs also an input to the layer 1 neural network; for example, neuron (2, 6) represents the 6 th neuron of the layer 2 neural network, x 2,6 represents the output of neuron (2, 6) and is also the input to the layer 3 neural network; connecting inputs to a layer-1 neural networkAnd its neuron (l, i l) are written asFor example, the 5 th neuron in the layer 2 neural network outputs x 2,5 and is also the input to the layer 3 neural network, and the computational function connecting its input x 2,5 and its 6 th neuron in the layer 3 neural network is noted asFor the layer i neural network, assuming that it has n l neurons in total, its input number is the number n l-1 of neurons of the previous layer neural network, its input is the sample feature x 0,i when it is the first neural network, the input number n l-1 = 5, the output of a neuron (i, i l) is the calculated sum of all incoming inputs of that neuron:
In the formula, To connect inputA computational function with the ith neuron of the layer i neural network; the output of the ith l-1 th neuron in the layer 1 neural network, which is also the input to the layer 1 neural network, when l=1 Representing a sample feature; The output of the ith l th neuron in the first layer neural network is also the input of the first+1 layer neural network;
The constructed burst early warning algorithm based on the KAN neural network has 4 layers of neural networks, takes the characteristics of a sample as an input vector, and outputs the constructed burst early warning algorithm based on the KAN neural network are as follows:
Wherein y 1,y2 represents the risk probability of no rock burst and the risk probability of rock burst respectively; n 0 denotes the input vector, i.e. the sample characteristics Five parameters of a common time interval, a location X coordinate, a location Y coordinate, a location Z coordinate and energy are x0,1、x0,2、x0,3、x0,4、x0,5,n0＝5;i1,i2,i3 respectively, namely an ith neuron in the first layer of neural network, the second layer of neural network and the third layer of neural network; i 0 denotes the ith input of the first layer neural network;
Wherein the function is calculated Is the sum of the basis function b (x) and the spline function:
wherein w is a trainable parameter; spline (x) is a spline function; b (x) is a basis function, x is an input value;
b(x)＝SiLU(x)＝x/(1+e-x) (4)
wherein SiLU denotes Sigmoid Linear Unit activation function; e is a natural constant;
in this embodiment, spline (x) is a linear spline function:
Wherein, B z (DEG) represents the corresponding Z times in the spline curve, and a cubic spline curve is adopted in the patent, so that the maximum value Z of Z is=3; c z represents a weight value for B z (·) is a trainable parameter;
In the conventional convolutional neural network and the conventional cyclic neural network, firstly, an input value is calculated to be an output value by using a calculation function, then the output value is activated by using an activation function, and the structures of the neural networks are difficult to express by formulas due to the structural reasons of an algorithm; the activation function of the KAN neural network constructed by the invention has stronger nonlinear description capability after being matched with the learnable parameters in the calculation function, and the structure of the activation function can be represented by a formula (2);
s4: rock burst early warning algorithm based on KAN neural network
Firstly, giving an initial value of each trainable parameter in a rock burst early warning algorithm based on a KAN (Kan-based neural network), initializing the trainable parameter to 0 aiming at a spline function in a calculation function, and initializing other trainable parameters according to an Xavier initialization rule; randomly disturbing the training set samples, inputting a rock burst early-warning algorithm based on a KAN neural network for training, evaluating the difference between the early-warning result of the rock burst early-warning algorithm and a sample label by using a cross entropy loss function, and updating parameters in the rock burst early-warning algorithm to be trained by using a back propagation algorithm;
Evaluating the performance of a trained rock burst early warning algorithm by using a test set sample; after 200 training cycles, the rock burst early warning algorithm parameters with the best effect are stored, and the trained rock burst early warning algorithm based on the KAN neural network is obtained.
Further comprising step S5: KAN neural network-based rock burst early warning algorithm application
And (3) obtaining the latest monitored continuous t original microseismic events, performing data processing according to the steps S1-S2 to obtain a sample with only features and no tag, inputting the features of the sample into the rock burst early warning algorithm based on the KAN neural network trained in the step S4, and obtaining the probability that the future one microseismic event has rock burst risk (is rock burst).
The parameters to be trained in the rock burst early warning algorithm based on the KAN neural network only comprise w and c z, and compared with the conventional neural network, the training parameters are greatly reduced; and because the nonlinear description capability of the calculation function in the KAN neural network is better, the rock burst early warning algorithm based on the KAN neural network can obtain better performance than the traditional neural network by only relying on a small number of parameters under the condition of not relying on the artificially set early warning index. Meanwhile, as the calculation function and the learnable parameters in the algorithm are clear, the formula of the algorithm per se can be expressed (see formula 2) for the rock burst early warning task, and the algorithm has a certain interpretation.
Claims (4)
1. A rock burst early warning method based on a KAN neural network is used for determining the probability of future microseismic events having rock burst dangers, and is characterized by comprising the following steps:
s1: preprocessing original microseismic monitoring data
The original microseismic monitoring data comprises a plurality of original microseismic events which are continuously monitored and recorded, wherein each original microseismic event comprises the time, the X coordinate, the Y coordinate, the Z coordinate, the energy and whether the rock burst happens or not; preprocessing an original microseismic event to obtain a preprocessed microseismic event, wherein the preprocessed microseismic event comprises a time interval, a place X coordinate, a place Y coordinate, a place Z coordinate, energy and whether rock burst is performed or not, and the time interval is the time interval between the microseismic event and the previous microseismic event;
s2: establishing rock burst early warning data set
Extracting continuous M times of preprocessed microseismic events, taking data of the M-th to m+t-th preprocessed microseismic events as a data source of one sample in rock burst early warning data, wherein M is more than or equal to 1 and less than or equal to M-t, and M-t samples can be established at most, wherein each sample comprises a characteristic and a label, the characteristic of each sample comprises a time interval, a place X coordinate, a place Y coordinate, a place Z coordinate and energy of the m+t-th preprocessed microseismic events, and the label of each sample is the rock burst risk of the m+t-th microseismic events; dividing the rock burst early warning data set into a training set and a testing set;
s3: construction of rock burst early warning algorithm based on KAN neural network
The constructed rock burst early warning algorithm based on the KAN neural network has 4 layers of neural networks, each layer of neural network comprises input, calculation functions and output, and the output of the former layer of neural network is the input of the next layer of neural network; the number of neurons of each layer of neural network is n 1、n2、n3 and n 4,n4 =2 in sequence;
the input of the rock burst early warning algorithm is also the input of the first layer neural network, and is the sample characteristic in the rock burst early warning data set Including time interval, location X-coordinate, location Y-coordinate, location Z-coordinate and energy, respectively calculated as X 0,1、x0,2、x0,3、x0,4、x0,5;
The number of the calculation functions of the first layer neural network is equal to the product of the number n 0 =5 of the input functions and the number of the neurons, and the output functions are the number of the neurons of the first layer neural network; the input number of the second layer and the following neural networks is equal to the neuron number of the previous layer of the neural network; the number of the calculation functions of the second layer and the subsequent neural networks is equal to the product of the number of neurons of the previous layer and the number of neurons of the present layer, and the number of the output of the second layer and the subsequent neural networks is equal to the number of neurons of the second layer and the subsequent neural networks;
The ith neuron of the first layer neural network is denoted by (l, i l), and the output of neuron (l, i l) is denoted by x l,il; the computational function connecting its input x l-1,il-1 and its neurons (l, i l) in the layer-1 neural network is noted as For the layer i neural network, assuming that there are n l neurons in total, the number of inputs is the number n l-1 of neurons of the previous layer neural network, the inputs are the sample feature x 0,i when being the first layer neural network, the number n l-1 = 5, and the output of a neuron (i, i l) is the calculated sum of all incoming inputs of that neuron:
In the formula, To connect inputA computational function with the ith neuron of the layer i neural network; the output of the ith l-1 th neuron in the layer 1 neural network, which is also the input to the layer 1 neural network, when l=1 Representing a sample feature;
the constructed burst early warning algorithm based on the KAN neural network has 4 layers of neural networks, takes the characteristics of a sample as input, and outputs the constructed burst early warning algorithm based on the KAN neural network are as follows:
Wherein y 1,y2 represents the risk probability of no rock burst and the risk probability of rock burst respectively; i 1,i2,i3 is the ith neuron in the first, second and third layer neural networks respectively; i 0 denotes the ith input of the first layer neural network;
Wherein the function is calculated Is the sum of the basis function b (x) and the spline function:
wherein w is a trainable parameter; spline (x) is a spline function; b (x) is a basis function, x is an input value;
b(x)＝SiLU(x)＝x/(1+e-x) (4)
wherein SiLU denotes Sigmoid Linear Unit activation function; e is a natural constant;
Where spline (x) is a linear spline function:
Wherein, B z (DEG) represents the corresponding Z times in the spline curve, and a cubic spline curve is adopted in the patent, so that the maximum value Z of Z is=3; c z represents a weight value for B z (·) is a trainable parameter;
s4: rock burst early warning algorithm based on KAN neural network
Firstly, giving an initial value of each trainable parameter in a rock burst early warning algorithm based on a KAN (Kan-based neural network), initializing the trainable parameter to 0 aiming at a spline function in a calculation function, and initializing other trainable parameters according to an Xavier initialization rule; inputting a training set sample into a rock burst early-warning algorithm based on a KAN neural network for training, evaluating the difference between an early-warning result of the rock burst early-warning algorithm and a sample label by using a cross entropy loss function, and updating parameters in the rock burst early-warning algorithm to be trained by using a back propagation algorithm;
evaluating the performance of a trained rock burst early warning algorithm by using a test set sample; and (3) storing the rock burst early-warning algorithm parameters with the best effect to obtain a trained rock burst early-warning algorithm based on the KAN neural network.
2. The rock burst warning method according to claim 1, wherein step S1 further comprises: and carrying out standardization treatment on the time interval, the location X coordinate, the location Y coordinate, the location Z coordinate and the energy of the preprocessed microseismic events.
3. The method of claim 1, wherein in step S2, the label is subjected to One-Hot encoding, and the label of the rock burst is 1 after the encoding, and the label of the rock burst is not 0.
4. A rock burst warning method according to any one of claims 1 to 3, further comprising step S5: KAN neural network-based rock burst early warning algorithm application
And (3) acquiring the latest monitored continuous t original microseismic events, performing data processing according to the steps S1-S2 to obtain a sample with only features and no tag, inputting the features of the sample into the rock burst early warning algorithm trained in the step S4 and based on the KAN neural network to obtain the probability that the future microseismic event has rock burst risk.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411026021.0A CN118964990A (en) | 2024-07-30 | 2024-07-30 | A rock burst early warning method based on KAN neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411026021.0A CN118964990A (en) | 2024-07-30 | 2024-07-30 | A rock burst early warning method based on KAN neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118964990A true CN118964990A (en) | 2024-11-15 |
Family
ID=93404857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411026021.0A Pending CN118964990A (en) | 2024-07-30 | 2024-07-30 | A rock burst early warning method based on KAN neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118964990A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119296241A (en) * | 2024-12-10 | 2025-01-10 | 西南科技大学 | A forest fire detection method based on KAN network and multi-sensor fusion |
-
2024
- 2024-07-30 CN CN202411026021.0A patent/CN118964990A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119296241A (en) * | 2024-12-10 | 2025-01-10 | 西南科技大学 | A forest fire detection method based on KAN network and multi-sensor fusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111562996B (en) | Method and system for detecting time sequence abnormality of key performance index data | |
| CN110473592B (en) | A Multi-view Human Synergistic Lethal Gene Prediction Method | |
| TW202139045A (en) | Privacy protection-based target service model determination | |
| CN111027686B (en) | Landslide displacement prediction method, device and equipment | |
| CN113538125A (en) | Risk rating method for optimizing Hopfield neural network based on firefly algorithm | |
| CN107392312B (en) | A Dynamic Adjustment Method Based on DCGAN Performance | |
| CN102622515B (en) | A kind of weather prediction method | |
| WO2018107414A1 (en) | Apparatus, equipment and method for compressing/decompressing neural network model | |
| CN112001110A (en) | Structural damage identification monitoring method based on vibration signal space real-time recursive graph convolutional neural network | |
| CN110046379A (en) | A kind of structure entirety damnification recognition method based on space-frequency information | |
| CN109190834A (en) | Stock price trend prediction method and system neural network based | |
| CN110751108A (en) | A Method for Determining Similarity of Subway Distributed Vibration Signals | |
| CN118964990A (en) | A rock burst early warning method based on KAN neural network | |
| EP4303780A1 (en) | Method and system for solving qubo problems with hybrid classical-quantum solvers | |
| CN116007937B (en) | Intelligent fault diagnosis method and device for transmission parts of mechanical equipment | |
| CN116227365A (en) | Landslide displacement prediction method based on improved VMD-TCN | |
| CN114358192A (en) | Multi-source heterogeneous landslide data monitoring and fusing method | |
| CN106529185A (en) | Historic building displacement combined prediction method and system | |
| CN116738302A (en) | Methods for detecting anomalies in time series data generated by infrastructure equipment in the network | |
| CN102708684A (en) | Short-term traffic flow Volterra-DFP self-adaption prediction method | |
| CN114897277B (en) | Burst-type landslide displacement prediction method based on LSTM | |
| CN116681159A (en) | Short-term power load prediction method based on whale optimization algorithm and DRESN | |
| CN118246510B (en) | Post-training pruning method of neural network | |
| CN113537638A (en) | Short-term wind pressure prediction method and abnormal data completion method and device for high-rise building | |
| CN112348656A (en) | BA-WNN-based personal loan credit scoring method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |