Rock burst early warning method based on shuffleNet-informer model

Abstract

The invention discloses a rock burst early warning method based on a shuffleNet-informer model, and the method comprises the steps: collecting ground sound frequency, energy monitoring data and hydraulic prop pressure data which have the same time sequence on site, carrying out the preprocessing of the data, and starting to construct a sample data set; training, testing and verifying the model by using the training set, the verification set and the test set respectively; a shuffleNet-informer model obtained through training, testing and verification is used for processing ground audio frequency and energy monitoring data collected in real time and data of pressure borne by the hydraulic prop, and a predicted value is obtained; and the early warning intensity of the predicted ground pressure impact event is finally determined by referring to grading early warning indexes of mines with similar geological and mining conditions and expert experience scores. According to the method, three kinds of on-site data with different characteristics are adopted for prediction processing, so that rock burst early warning is more accurate; according to the method, convolution operation in an informer model is improved by utilizing shuffleNet, the training efficiency and generalization performance of the model are improved, the calculation complexity of the prediction model is greatly reduced, the real-time performance of prediction can be better ensured, meanwhile, the precision of long-time prediction can be ensured, and early warning evaluation is more reasonable.

Description

Technical Field

[0001] The present invention relates to the field of coal mine safety technology, and in particular to a rock burst warning method based on a shuffleNet-informer model. Background Art

[0002] With the continuous development of the economy and society, the demand for natural resources has also increased, and mineral resources are one of them. It can be said that mineral resources are the basic raw materials for the survival and development of human society, and they play a vital role in the development of the entire national economy. At present, the development and utilization of mineral resources is in the stage of "from shallow to deep, from easy to difficult". Deep resources have huge potential, and the development and utilization of deep resources is inevitable for the development of the mining industry. However, deep mineral mining faces the situation of "three highs and one disturbance", namely "high ground stress, high permeability ground pressure, high ground temperature and strong mining disturbance", which makes the environment of deep mineral mining very harsh, and it is very easy to cause various engineering geological disasters, rock burst is one of them. Therefore, early warning of rock burst is a key step in safe underground mining.

[0003] In the past, most of the early warnings for rock burst were measured by mechanical devices. Although the results were highly accurate, they consumed a lot of manpower and material resources, and could not effectively predict rock burst in real time. In addition, the prediction methods based on transformer and LSTM models have a large amount of calculation and are not very accurate for long-term predictions. Although the informer model overcomes the above shortcomings, it also consumes a lot of computing resources when processing large-scale data sets. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a rock burst warning method and a warning composite evaluation method based on the shuffleNet-informer model to solve the technical problems of rock burst prediction in terms of prediction accuracy, prediction calculation amount, and warning evaluation.

[0005] A rock burst warning method based on a shuffleNet-informer model, comprising:

[0006] I) Collect geosonde frequency, energy monitoring data and hydraulic prop pressure data with the same time series on site. After preprocessing the collected data, the sample data sets of the three data are obtained according to the time series arrangement and combination, and the sample data sets are divided into training set, test set and validation set in a ratio of 8:1:1;

[0007] II) Use the training set, validation set, and test set to train, test, and validate the shuffleNet-informer model respectively;

[0008] Training the shuffleNet-informer model involves:

[0009] 1) First, preprocess the sample data, including missing value processing and normalization. The preprocessed sample data set is embedded and positionally encoded to obtain the initial input data for the encoding part of the shuffleNet-informer model;

[0010] 2) Perform multi-head sparse self-attention and convolution pooling operations on the initial input data of the encoding part of the model to obtain the final feature map of the encoding part.

[0011] 3) The second half of the initial input data of the encoding part obtained in step 1) is used as the first half of the input data of the decoding part of the shuffleNet-informer model to guide the model decoding part to predict the data. The second half of the input data of the decoding part is a 0 matrix, called the mask part. The masked multi-head sparse self-attention operation and multi-head attention operation are performed on the input data of the decoding part to obtain the output of the decoding part of the shuffleNet-informer model.

[0012] 4) The output data of the decoding part of the shuffleNet-informer model is input into a fully connected layer, which outputs a series of prediction results.

[0013] III) using the informer model obtained by training, testing and verification in step II) to process the real-time collected ground sound frequency, energy monitoring data and pressure data of the hydraulic support to obtain a predicted value;

[0014] IV) Referring to the graded warning indicators of mines with similar geological and mining conditions in the past, determine the rating of the predicted data obtained in step III), and finally determine the time and warning intensity of the predicted ground pressure impact event.

[0015] Furthermore, the shuffleNet-informer model is divided into the original data processing layer, the shuffleNet-informer model layer, and the data output layer. The original data processing layer includes data preprocessing, embedding operations, and position coding splicing; the shuffleNet-informer model layer is the main part of the rock burst warning method, which extracts the time series features in the input data through the shuffleNet-informer model and predicts the future time series features; the data output layer is mainly composed of a fully connected layer, which converts the feature map output by the shuffleNet-informer model layer and gives the final prediction result.

[0016] Furthermore, the data preprocessing operation includes: missing value processing and normalization. The missing value processing refers to deleting data containing missing values ​​or interpolating missing values ​​with possible values. When the missing ratio is very small, the data containing missing values ​​can be directly deleted. When the missing data accounts for a considerable proportion, multiple interpolation can be used. The normalization refers to mapping the data to decimals between [0,1] or [-1,1], so that the model can converge to the optimal solution more easily and speed up the solution.

[0017] Furthermore, the shuffleNet-informer model optimizes the convolution operation in the informer model using the core architecture of channel shuffling in the shuffleNet model. The channel shuffle is a variant of grouped convolution. Although grouped convolution can greatly reduce the computational cost, the output of a channel can only come from one group of input channels, which prevents information exchange between channels and weakens the expression ability of the neural network. Channel shuffle is composed of two stacked grouped convolution modules. The data output by the first convolution module is redivided and regrouped, that is, "shuffled", to form new data. When the second convolution module groups the input data, the data of each new group is mixed by the data of different groups of the first convolution module, thereby realizing the association of data between different groups.

[0018] Beneficial effects of the present invention:

[0019] 1. The rock burst warning method based on the shuffleNet-informer model of the present invention adopts three different types of field data for prediction and processing, involving three characteristics of vibration, energy and stress, so that the rock burst warning is more accurate.

[0020] 2. The rock burst warning method based on the shuffleNet-informer model of the present invention adopts the shuffleNet-informer model. Compared with the previous informer prediction model, the training calculation cost consumed by the convolution operation is greatly reduced through the group convolution and channel shuffling mechanism. In addition, the shortcomings of the group convolution that cannot realize the information exchange between channels and weaken the expression ability of the neural network are overcome. Therefore, the present invention can better ensure the accuracy and real-time performance of the prediction, and at the same time can control the consumption of training calculation cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the overall flow chart of the rock burst warning method based on the shuffleNet-informer model.

[0022] Figure 2This is the overall architecture diagram of the shuffleNet-informer model.

[0023] Figure 3 Schematic diagram of the convolution structure of the shuffleNet-informer model. DETAILED DESCRIPTION

[0024] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0025] The rock burst warning method based on the shuffleNet-informer model in this embodiment includes:

[0026] I) Collect geotonic frequency, energy monitoring data and pressure data of hydraulic props with the same time series on site. After preprocessing the collected data, the sample data sets of three kinds of data are obtained according to the time series arrangement and combination, and the sample data sets are divided into training set, test set and validation set in the ratio of 8:1:1; the geotonic frequency, energy monitoring data and pressure data of hydraulic props are predicted respectively. The training set is used to train the shuffleNet-informer model to obtain the training model; the test set is used to evaluate the performance of the training model; the validation set is used to adjust and select the network parameters of the final model.

[0027] II) Use the training set, validation set, and test set to train, test, and validate the shuffleNet-informer model respectively;

[0028] Training the shuffleNet-informer model involves:

[0029] 1) Perform preprocessing operations on the sample data, including missing value processing and normalization. When the missing ratio is very small, the data containing missing values ​​can be directly deleted. When the missing data accounts for a considerable proportion, multiple interpolation can be used. The multiple interpolation method generates an interpolation number pool for missing values. Each number in the number pool can be used as an interpolation for missing values. Therefore, several sets of complete data will be generated. These sets are statistically analyzed, and the best complete data set is selected according to the scoring function. Normalization refers to mapping the data to decimals between [0,1] or [-1,1], so that the model can converge to the optimal solution more easily and speed up the solution.

[0030] 2) The preprocessed sample data set is embedded and positionally encoded to obtain the initial input data of the encoding part of the shuffleNet-informer model;

[0031] 3) The encoding part of the shuffleNet-informer model performs a sparse self-attention operation on the initial input data to obtain an attention matrix, and then performs a self-attention distillation operation on the attention matrix to obtain the final feature map.

[0032] The key components of the sparse self-attention mechanism are the same as those of the self-attention mechanism, including query, key, and value. The query Q represents the current element, which is used to match the key K; the key K represents other elements in the sequence, which are used to match the query Q; the value V is also other elements in the sequence. Once the key K matches the query, the relevant value V will be used to construct the output. The attention weight is calculated by the similarity between the query Q and the key K, which determines the contribution of the value V to the output. The sparse self-attention mechanism focuses on those Qs with high attention weights, recalculates the attention weights of these Qs with respect to all Ks, and the attention weights of the remaining Qs are taken as the average of all attention weights.

[0033] refer to Figure 3 As shown in the figure, the self-attention distillation operation refers to performing several convolution and pooling operations on the attention matrix to prune the input dimension. After each convolution and pooling operation, the length of the input sequence is halved. First, the convolution input data is divided into large groups, and the first convolution operation is performed on each large group of data. The results are divided into small groups, and the data are redistributed to ensure that each large group of data is composed of different small group data. The reorganized data is convolved for the second time to obtain the final output data.

[0034] 4) The second half of the encoder input data after the embedding operation and position encoding splicing in step 2) is used as the first half of the input data of the shuffleNet-informer model decoding part to guide the prediction of the shuffleNet-informer model. The second half of the input data of the shuffleNet-informer model decoding part is a mask matrix, that is, a 0 matrix with the same shape as the predicted target. The shuffleNet-informer model decoding part performs a masked multi-head sparse self-attention operation on the input data, and then performs a multi-head attention operation on the data output by the masked multi-head sparse self-attention operation to obtain the output of the shuffleNet-informer model decoding part; the masked multi-head sparse self-attention operation is basically the same as the sparse self-attention operation of the decoding part, the difference being that the inner product of the masked part in the input data is set to negative infinity to shield the attention of the front part to the masked part, so that it has no effect when the parameters are updated. The multi-head attention operation is also the inner product operation of the three matrices Q, K, and V. After the decoding part, each position to be predicted has a vector.

[0035] 3) The final output of the decoding part of the shuffleNet-informer model is input into a fully connected layer, which outputs a series of prediction results.

[0036] III) using the optimal informer model obtained through training, testing and verification in step II) to process the real-time collected ground sound frequency, energy monitoring data and pressure data of the hydraulic support to obtain a predicted value;

[0037] IV) Determine the rating T1 of the geoacoustic frequency, energy monitoring data and predicted data of the pressure on the hydraulic prop obtained in step III) by referring to the graded warning indicators of mines with similar geological and mining conditions in the past and the expert experience rating, which is specifically divided as follows:

[0038]

[0039] Among them, T1 is 1, 2, 3, and 4, respectively representing no impact ground pressure danger, weak danger, medium danger, and strong danger.

[0040] V) Based on the ground frequency, energy monitoring data and predicted data of the pressure on the hydraulic support obtained in step III), these predicted data are linearly fitted to simplify them into a simple curve, and the curve is secondarily derived to study the speed of change of its slope, that is, the trend change of the predicted data. According to, there are three general trends: (1) basically unchanged; (2) slowly rising; (3) rapidly rising. These three trends are assigned -1, 0, and 1 values ​​of the correction warning level T2 in turn, as shown below:

[0041]

[0042] VI) Add the obtained T1 and T2 values ​​to get the final warning level T. Therefore, the value range of T is between 0-5, and it is divided into 4 danger levels, and corresponding measures are taken. The details are as follows: when T is 0, there is no possibility of rock burst, and normal production is carried out; when T is 1 or 2, there is no possibility of rock burst, and the monitoring of rock burst at the data collection site needs to be strengthened; when T is 3, weak rock burst is possible, and rock burst management measures need to be taken to reduce the intensity of rock burst and the size of the harm caused in the future; when T is 4 or 5, production should be stopped and wait for the rock burst event to occur and subside.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A rock burst warning method based on the shuffleNet-informer model, characterized in that: include: I) Collecting geosonde frequency, energy monitoring data and pressure data of hydraulic props with the same time series on site, preprocessing the collected data, and arranging and combining them according to the time series to obtain sample data sets of the three types of data, and dividing the sample data sets into training set, test set and validation set in a ratio of 8:1:1; II) Use the training set, validation set, and test set to train, test, and validate the shuffleNet-informer model respectively; III) using the shuffleNet-informer model obtained by training, testing and verification in step II) to predict the real-time collected ground sound frequency, energy monitoring data and pressure data of the hydraulic support to obtain a predicted value; IV) referring to the graded warning indicators of mines with similar geological and mining conditions in the past and the expert experience scores, analyzing the predicted data obtained in step III), and finally determining the time and warning intensity of the predicted ground pressure impact event.

2. The rock burst warning method based on the shuffleNet-informer model according to claim 1 is characterized in that: The data preprocessing operation includes: missing value processing and normalization.

3. The rock burst warning method based on the shuffleNet-informer model according to claim 1 is characterized in that: The shuffleNet-informer model as a whole is divided into an original data processing layer, a shuffleNet-informer model layer, and a data output layer. The original data processing layer includes data preprocessing, embedding operations, and position coding splicing; the shuffleNet-informer model layer is the main part of the rock burst warning method. The shuffleNet-informer model is used to extract the time series features in the input data and predict the future time series features. The data output layer is mainly composed of a fully connected layer, which converts the feature map output by the shuffleNet-informer model layer and gives the final prediction result.

4. The rock burst warning method based on the shuffleNet-informer model according to claim 2 is characterized in that: The missing value processing refers to deleting the data containing missing values ​​or interpolating the missing values ​​with possible values. When the missing ratio is very small, the data containing missing values ​​can be directly deleted. When the missing data accounts for a considerable proportion, multiple interpolation methods can be used; The normalization refers to mapping the data to decimals between [0, 1] or [-1, 1], so that the model can converge to the optimal solution more easily and the solution speed can be accelerated.

5. The rock burst warning method based on the shuffleNet-informer model according to claim 3 is characterized in that: The shuffleNet-informer model optimizes the convolution operation in the informer model by using the core architecture of channel shuffling in the shuffleNet model. The channel shuffle is a variant of grouped convolution. Although grouped convolution can greatly reduce the computational cost, the output of a channel can only come from one group of input channels, which prevents information exchange between channels and weakens the expression ability of the neural network. The channel shuffle is composed of two stacked grouped convolution modules, which redistribute and regroup the data output by the first convolution module, that is, shuffle, to form new data. When the second convolution module groups the input data, the data of each new group is mixed by the data of different groups of the first convolution module, thereby realizing the association of data between different groups.

Images