A feature enhancement prediction method for sensitive features of mine microseismic events

Abstract

The application discloses a kind of feature enhancement prediction methods of mine microseismic event sensitive characteristics, belong to time series prediction and signal processing technical field.Mine microseismic event monitoring data is collected, mine microseismic event target feature sequence is constructed, moving average sequence and first-order difference sequence of the target feature are constructed;Target feature sequence, moving average sequence and first-order difference sequence are input into multi-input joint prediction model DFformer;Target feature sequence is predicted.The application is aimed at the problems of insufficient input information of mine microseismic event sensitive characteristics, timing fluctuation sensitivity, low prediction stability and precision, etc., introduces the moving average sequence and first-order difference sequence of target feature, fully excavates the amplitude, trend and change gradient of target feature sequence, significantly improves the capture ability and prediction robustness of the model to feature extreme value, not only strengthens the correlation constraint between different sequences, but also effectively suppresses noise and mutation interference, improves the prediction accuracy of target sequence.

Description

Technical Field

[0001] This invention belongs to the field of time series prediction and signal processing technology, specifically relating to a feature enhancement prediction method for sensitive features of mine microseismic events. Background Technology

[0002] During mining operations, rock mass fracturing can induce microseismic events. Monitoring these events generates microseismic monitoring data. Characteristic time-series data closely related to rockburst manifestations are extracted from this data and termed microseismic sensitivity features. Microseismic sensitivity features are crucial for assessing rock mass stability and rockburst risk; therefore, accurate prediction based on these features has significant engineering implications.

[0003] However, microseismic sensitive characteristic data exhibit significant nonlinearity, nonstationarity, and abruptness. Their time-series data fluctuate dramatically, and the occurrence of extreme values ​​is highly random, posing considerable challenges to time-series forecasting. Most existing methods directly use a single feature sequence as model input for modeling and forecasting, resulting in typically low prediction accuracy. The main reason is that relying solely on the amplitude information of a single feature makes it difficult to fully exploit the trend changes and rates of change within the feature sequence, leading to slow model responses to abrupt changes and extreme values, and insufficient prediction stability and accuracy. Furthermore, when the feature sequence contains random perturbations, the model is susceptible to noise, resulting in amplified fluctuations or over-smoothing, further affecting the reliability of extreme value predictions.

[0004] Therefore, it is necessary to propose a feature enhancement prediction method for sensitive characteristics of mine microseismic events. This method should, while maintaining the authenticity of the original target prediction sequence, fully explore the trend changes and gradients of the target feature sequence, thereby improving the ability to capture extreme values ​​of the target feature sequence and the stability of the prediction. Summary of the Invention

[0005] To address the problems of insufficient utilization of feature information and low prediction accuracy of single features in existing time series prediction of mine microseismic sensitive features, this invention proposes a feature enhancement prediction method for mine microseismic sensitive features. This method constructs a moving average sequence and a first-order difference sequence based on the target prediction feature sequence. It innovatively introduces a feature enhancement prediction strategy of "sub-feature construction - simultaneous input of multiple features - coupled prediction of multiple features." While preserving the original information of the target feature sequence, it fully extracts the inherent fluctuation patterns of the target feature sequence from the moving average sequence and the first-order difference sequence, significantly improving the accuracy and stability of the target feature sequence prediction.

[0006] The core of this invention lies in constructing a dedicated multi-feature sequence for the target feature sequence, and on this basis, achieving synchronous input and coupled prediction of multiple features. This strategy not only preserves the amplitude characteristics of the target feature sequence but also fully exploits its trend changes and gradient information. Furthermore, the strong coupling constraint mechanism formed by the synchronous joint prediction of multiple features enables mutual constraints among the multiple input feature sequences, effectively suppressing noise and abrupt changes, thereby significantly improving the overall stability and generalization ability of the target feature sequence prediction.

[0007] This invention adopts the following technical solution: a feature enhancement prediction method for sensitive features of mine microseismic events, comprising the following steps:

[0008] Step 1: Extract sensitive feature data sequences of microseismic events based on mine microseismic event monitoring data. , Represented as:

[0009] ,

[0010] in, Represents a timestamp. for The numerical values ​​of the sensitive characteristics of the microseismic event corresponding to the given time;

[0011] Step 2: Construct a moving average sequence of sensitive feature data for microseismic events and first-order difference sequence , and They are represented as follows:

[0012]

[0013]

[0014]

[0015]

[0016] in, for Moving average of sensitive characteristic data of microseismic events at different times. for The first-order difference value of the sensitive feature data of microseismic events at time, where k is the length of the sliding window;

[0017] Step 3: Sequence of sensitive features of microseismic events Moving average sequence of sensitive characteristics of microseismic events and the first-order difference sequence of microseismic event sensitivity characteristics As input, the data is used for training and inference in the prediction model DFformer to obtain a prediction sequence of sensitive features of microseismic events. , Represented as:

[0018] ,

[0019] .

[0020] By adopting the above technical solution, the present invention has at least the following beneficial effects:

[0021] This invention constructs a dedicated multi-feature sequence based on the temporal characteristics of the target feature sequence and introduces a constraint mechanism for synchronous coupling prediction of multiple features. This mechanism, while fully preserving the target feature amplitude, its trend changes, and gradient information, effectively suppresses noise and abrupt changes, significantly improving the overall stability and generalization ability of the prediction. This method features full feature utilization, strong adaptability, high prediction accuracy, and simple implementation, possessing high technical value and promising application prospects. Overall, this invention represents a substantial improvement over existing technologies in terms of data fidelity, algorithm reversibility, computational efficiency, and model compatibility, providing a novel technical solution for feature enhancement prediction of sensitive features of mine microseismic events. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This invention provides a method for enhancing and predicting the daily total energy characteristics of microseismic events in mines.

[0024] Figure 2 This is a time-series data sequence of total daily energy.

[0025] Figure 3 This is a daily total energy moving average sequence;

[0026] Figure 4 The first-order difference sequence of total daily energy;

[0027] Figure 5 Here is a structural diagram of the DFformer model;

[0028] Figure 6 A comparison chart of the predicted total daily energy;

[0029] Figure 7 The waveforms show the maximum daily energy and total daily energy of a certain mine.

[0030] Figure 8 A comparison chart of the predicted and actual daily maximum energy output of a certain mine;

[0031] Figure 9 This is a comparison chart of the predicted and actual daily total energy output of a certain mine. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Using microseismic energy time series data from a certain mine as experimental data, such as Figure 1 As shown, this embodiment of the invention provides a method for enhancing and predicting the daily total energy characteristics of microseismic events in mines, including:

[0034] Step 1: Extract the daily total energy data sequence , Represented as:

[0035] ,

[0036] in, Represents a timestamp. for The daily total energy value corresponding to a given time, such as Figure 2 As shown.

[0037] In this embodiment, the total daily energy of mine microseismic events from 2018 to 2021 was collected.

[0038] Step 2: Construct a moving average series of daily total energy data First-order difference sequence of total daily energy data , and They are represented as follows:

[0039] ,

[0040] ,

[0041] ,

[0042] ,

[0043] in, for The moving average of the daily total energy data at any given time. for The first-order difference value of the total daily energy data at time point, where k is the length of the sliding window.

[0044] In this embodiment, a moving average sequence and a first-order difference sequence of daily total energy data for mine microseismic events from 2018 to 2021 are constructed, as follows: Figure 3 and 4 As shown.

[0045] Table 1 shows some of the total daily energy data, the moving average data of total daily energy, and the first-order difference data of total daily energy.

[0046] Table 1

[0047]

[0048] Step 3: Sequence of daily total energy data Daily total energy data series moving average series First-order difference sequence of total daily energy data The pre-trained prediction model DFformer is input for training and inference to obtain the daily total energy prediction sequence. , Represented as:

[0049] ,

[0050] .

[0051] The DFformer model structure consists of N encoders and decoders connected in series, as shown in the following diagram. Figure 5As shown, the model first performs Dish-TS normalization on the input temporal data to eliminate scale differences between different features; then, it performs feature embedding and superimposes positional encoding to supplement temporal sequence information; next, the processed features are fed into the encoder to mine long dependencies, which are then processed through a double Fourier attention mechanism. The data processed by the double Fourier attention mechanism and the data after positional encoding undergo a first residual processing. The data after the first residual processing is then processed by a feedforward network. The data after the feedforward network and the data after the first residual processing undergo a second residual processing. The data after the second residual processing is then input into subsequent encoders for processing, until the Nth encoder finishes processing and outputs the processed data. The input positional encoding data is processed by a dual Fourier attention mechanism. The data processed by the dual Fourier attention mechanism and the positional encoding data undergo a first residual processing. This first residual processing data is then processed by the data processed by the Nth encoder using the dual Fourier attention mechanism. This second residual processing is performed on the data processed by the dual Fourier attention mechanism and the data processed by the first residual processing. This second residual processing data is then processed by a feedforward network. Finally, the data processed by the feedforward network and the data processed by the second residual processing undergo a third residual processing, completing the feature mapping. Finally, inverse normalization is performed on the model's inference output to restore its dimensions, yielding the final prediction result.

[0052] In terms of data processing, the DFformer model incorporates the Dish-TS normalization method. Traditional static normalization methods rely on fixed global statistics, which cannot adapt to the dynamic changes in time-series distributions, often leading to decreased prediction performance. Dish-TS, by constructing a bilinear mapping network, dynamically predicts two sets of distribution parameters (input and output) from historical sequences, enabling adaptive handling of distribution shifts. Furthermore, microseismic energy data is characterized by strong non-stationarity, large fluctuations, and contains numerous outliers. Therefore, this application adds a dual Fourier attention module to the DFformer model, mapping the time series to the frequency domain via Fourier transform and calculating attention in the frequency domain. First, the location-encoded input data... After linear transformation, we obtain Fourier transforms were performed to obtain ;use Perform dot product and summation operations to obtain the calculation result. Finally, perform an inverse Fourier transform (IFFT) on M to obtain the final result. .

[0053] Finally, the prediction results are compiled and output. Although the model is for the daily total energy sequence... Daily total energy moving average sequence and the first-order difference sequence of total daily energy Both were predicted, but the daily total energy moving average series and the daily total energy first-order difference series only served as constraints during the prediction process. Therefore, only the total energy prediction results were retained in the output. To complete the task of predicting microseismic signals in mines.

[0054] The specific results of the daily total energy prediction from the DFformer model are as follows: Figure 6 As shown in Table 2, the prediction results indicators are as follows: S represents inputting only the total daily energy and outputting only the total daily energy; MS represents inputting the total daily energy, the moving average sequence of the total daily energy, and the first-order difference sequence of the total daily energy, but only predicting the total daily energy; M represents the method of this patent, which inputs the total daily energy, the moving average sequence of the total daily energy, and the first-order difference sequence of the total daily energy, and predicts them simultaneously.

[0055] Table 2

[0056]

[0057] The main coal seam of a certain mine is the No. 3 coal seam, with an average thickness of 8.82m and a burial depth between -800m and -1200m. The No. 3 coal seam and its roof and floor exhibit a weak tendency for rockburst, and the mine has been identified as a rockburst-prone mine. During production, high-energy microseismic events frequently occur, repeatedly inducing rockburst manifestations. For example, on October 1, 2021, a microseismic event of 83.5kJ was detected, with its hypocenter located 0.7m ahead of the 2305 working face and 57.7m above the roof. The ground tremors were clearly felt during the event.

[0058] The 2305 working face in the second mining area of ​​this mine was taken as the research object. This working face is the fifth mining face, with a burial depth ranging from -810m to -980m, and multiple layers of fine sandstone distributed above the roof. Raw microseismic monitoring data from January 2018 to November 2021 were used. The data first underwent a series of preprocessing steps, and finally, microseismic characteristic parameters were extracted in "days," mainly including daily maximum energy and daily total energy. Each characteristic sequence contains a total of 1414 data values. The waveforms of daily maximum energy and daily total energy are shown below. Figure 7 As shown. During the longwall mining operation, a dual-energy threshold-based early warning mechanism was employed: if the maximum daily energy of microseismic events detected exceeded 30 kJ and the total daily energy exceeded 100 kJ, the system automatically triggered a rockburst early warning, prompting on-site intervention measures such as rockburst prevention and pressure relief. Figure 9 Data analysis revealed that within the study period, seven sets of data (January 31, 2021; March 19, 2021; May 26, 2021; June 1, 2021; June 5, 2021; June 20, 2021; and October 10, 2021) simultaneously met the aforementioned dual requirements, thus triggering a warning signal. Figure 8 and Figure 9 It can be seen that the method of this patent predicts four out of six warning points, which can effectively capture most of the early warning information of high-energy anomalies and has a strong disaster sensitivity identification capability.

[0059] The present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A feature enhancement prediction method for sensitive characteristics of microseismic events in mines, characterized in that, Includes the following steps: Step 1: Extract sensitive feature data sequences of microseismic events based on mine microseismic event monitoring data. , Represented as: ， in, Represents a timestamp. for The numerical values ​​of the sensitive characteristics of the microseismic event corresponding to the given time; Step 2: Construct a moving average sequence of sensitive feature data for microseismic events and first-order difference sequence , and They are represented as follows: ， ， ， ， in, for Moving average of sensitive characteristic data of microseismic events at different times. for The first-order difference value of the sensitive feature data of microseismic events at time, where k is the length of the sliding window; Step 3: Sequence of sensitive features of microseismic events Moving average sequence of sensitive characteristics of microseismic events and the first-order difference sequence of microseismic event sensitivity characteristics As input, the data is used for training and inference in the prediction model DFformer to obtain a prediction sequence of sensitive features of microseismic events. , Represented as: ， 。

Images