Underground powerhouse surrounding rock stability prediction method, system, device and medium

Abstract

The present application relates to the technical field of engineering construction, and proposes an underground powerhouse surrounding rock stability prediction method, system, device and medium, the method collects the time series monitoring data of the surrounding rock of the underground powerhouse; the time series monitoring data is modeled by a fractal dimension to obtain a detection fractal dimension function, the scaling factor, correction coefficient and scale factor in the detection fractal dimension function are locally searched according to a hybrid frog algorithm to obtain an optimal scaling factor, an optimal correction coefficient and an optimal scale factor; based on the optimal scaling factor, the optimal correction coefficient, the optimal scale factor and the detection fractal dimension function, a target dimension feature extraction function is constructed, and the target dimension feature value in the time series monitoring data is extracted according to the target dimension feature extraction function; the target dimension feature value and the surrounding rock sampling vector are input into a chaotic neural network to predict the deformation of the surrounding rock, and a surrounding rock deformation prediction result is obtained. The present application helps to improve the timeliness and accuracy of surrounding rock stability monitoring and early warning.

Description

Technical Field

[0001] This invention relates to the field of engineering construction technology, and in particular to a method, system, equipment and medium for predicting the stability of surrounding rock in underground powerhouses. Background Technology

[0002] In the field of geological exploration for water conservancy and hydropower projects, surrounding rock stability assessment has always been a key aspect of the design and construction process. Traditional surrounding rock stability assessments mainly rely on geological exploration data, field test results, and engineers' experience and judgment. While these traditional methods have been widely used throughout history, their limitations have become increasingly apparent as the scale of projects expands and the complexity of geological conditions increases.

[0003] Chinese patent CN116258071A discloses a method for evaluating the effectiveness of surrounding rock monitoring indicators. The method includes: acquiring various monitoring indicator data for a cross-section of a tunnel; dividing the surrounding rock deformation process into several damage stages based on abrupt changes in surrounding rock displacement; establishing a two-dimensional table for each cross-section to store the monitoring data; inputting one or more monitoring indicator columns from the two-dimensional table as data features and damage stage columns as data labels into a machine learning model; classifying the data features according to the data labels; obtaining a prediction accuracy matrix for surrounding rock damage determination; and obtaining the significance score of each data feature to determine the effectiveness of the monitoring indicators. However, although the above scheme uses numerical simulation to simulate the surrounding rock deformation data from the external displacement monitoring data, the accuracy of surrounding rock stability monitoring is low because the external displacement and deformation of the surrounding rock alone cannot determine the internal stress and deformation. Therefore, it is essential to provide a method and system for predicting the stability of surrounding rock in underground powerhouses to improve the timeliness and accuracy of surrounding rock stability monitoring and early warning. Summary of the Invention

[0004] The purpose of this invention is to overcome the aforementioned defects and problems in the existing technology and provide a method, system, equipment, and medium for predicting the stability of surrounding rock in underground powerhouses. By monitoring acoustic emission and recording sampling vectors of the surrounding rock, multi-angle data acquisition of the surrounding rock condition is achieved, and time-series data is used to capture subtle changes in the surrounding rock, thereby improving the timeliness and accuracy of surrounding rock stability monitoring and early warning.

[0005] In a first aspect, the present invention provides a method for predicting the stability of surrounding rock in an underground powerhouse, comprising:

[0006] Collect time-series monitoring data of the surrounding rock of the underground powerhouse, wherein the time-series monitoring data includes acoustic emission signal data of the surrounding rock and sampling vector of the surrounding rock;

[0007] The time-series monitoring data is modeled using fractal dimension to obtain the detection fractal dimension function. The scaling factor, correction coefficient, and scale factor in the detection fractal dimension function are then locally searched using the hybrid leapfrog algorithm to obtain the optimal scaling factor, optimal correction coefficient, and optimal scale factor.

[0008] Based on the optimal scaling factor, the optimal correction coefficient, the optimal scale factor, and the detection fractal dimension function, a target dimension feature extraction function is constructed, and the target dimension feature values ​​in the time-series monitoring data are extracted according to the target dimension feature extraction function.

[0009] The target dimension feature value and the surrounding rock sampling vector are input into a chaotic neural network to predict the surrounding rock deformation, so as to obtain the surrounding rock deformation prediction result.

[0010] Based on the above technical solutions, preferably, the collection of time-series monitoring data of the surrounding rock of the underground powerhouse specifically includes:

[0011] Acquire surrounding rock monitoring data, surrounding rock acoustic emission signal data, and support structure stress monitoring data corresponding to multiple measuring points within a preset interval range in the underground space corresponding to the underground powerhouse. Integrate all support structure stress monitoring data and all surrounding rock monitoring data, and construct multiple surrounding rock sampling vectors corresponding to each measuring point.

[0012] When collecting the surrounding rock monitoring data and the support structure stress monitoring data, each corresponding measuring point simultaneously monitors the horizontal and vertical displacement of the surrounding rock, as well as the radial and tangential stress of the surrounding rock.

[0013] More preferably, the step of performing fractal dimension modeling on the time-series monitoring data to obtain the detection fractal dimension function specifically includes:

[0014] The acoustic emission signal data of the surrounding rock is segmented on the time axis to obtain multiple acoustic emission sequences of the surrounding rock, and the acoustic emission sequences of the surrounding rock are normalized to obtain a standard acoustic emission sequence of the surrounding rock.

[0015] Based on the standard surrounding rock acoustic emission sequence, an initial detection fractal dimension function is constructed;

[0016] The local fractal values ​​at different sampling scales are calculated by using the initial detection fractal dimension function, and the sampling logarithmic relationship between the sampling scale and the corresponding local fractal value is obtained by linear fitting.

[0017] Based on the sampling logarithmic relationship and the initial detection fractal dimension function, the detection fractal dimension function is constructed.

[0018] More preferably, the step of performing a local search for the scaling factor, correction coefficient, and scale factor in the detected fractal dimension function using the hybrid leapfrog algorithm specifically includes:

[0019] Multiple frog individuals are randomly generated and initialized within the feasible region of each frog individual, wherein each frog individual represents an optimized parameter combination including scaling factor, correction coefficient and scale factor;

[0020] Based on the detected fractal dimension function and fractal reference value, a fractal reference function is constructed, and the fitness of each individual frog is calculated according to the fractal reference function and fractal reference value.

[0021] Frog individuals in the population are sorted in ascending order of fitness, and the population is divided into multiple sub-cultures, each containing several frog individuals.

[0022] In each subculture gene, the frog individual with the best fitness and the frog individual with the worst fitness are selected respectively, and the position of the frog individual with the worst fitness is updated based on the position update function to perform local search;

[0023] When each cultural gene has undergone at least one round of local search and reached the maximum number of iterations, the population is recombined and sorted again according to fitness in ascending order. The best frog individual in the population is updated, and the position of the globally best frog individual is recorded.

[0024] More preferably, the step of inputting the target dimension feature value and the surrounding rock sampling vector into a chaotic neural network to predict the surrounding rock deformation, in order to obtain the surrounding rock deformation prediction result, specifically includes:

[0025] The target dimension feature values ​​and the surrounding rock sampling vector are normalized to obtain standard target dimension feature values ​​and standard surrounding rock sampling vectors;

[0026] The standard target dimension feature value and the standard surrounding rock sampling vector are concatenated to obtain the surrounding rock feature vector, and the surrounding rock feature vector is input into the chaotic neural network.

[0027] Based on the internal state function and output state function of the chaotic neural network, the deformation prediction results of the surrounding rock are obtained.

[0028] More preferably, the expression for the target dimension feature extraction function is:

[0029]

[0030] d WS =T(δ;α) * ,β * m * );

[0031] Where T() represents the detection fractal dimension function; δ represents the sampling scale; x n denoted as the amplitude of the nth sampling point after segmentation of the acoustic emission signal data in the surrounding rock; N represents the minimum number of scales required to completely cover the signal at the sampling scale δ; α represents the scaling factor corresponding to the cumulative length of the signal; β represents the correction coefficient corresponding to the mapping value of signal change and scale relationship; m represents the scale factor corresponding to the scale window used in fractal dimension estimation; MSE represents the mean square error; K represents the total number of acoustic emission sequences in the surrounding rock acoustic emission signal data; T k This represents the fractal dimension value of the acoustic emission sequence of the k-th segment of the surrounding rock; This represents the fractal reference value under the acoustic emission sequence of the surrounding rock in the k-th segment; d WS α represents the feature extraction function for the target dimension. * β represents the optimal scaling factor corresponding to the signal accumulation length. * This represents the optimal correction coefficient corresponding to the mapping value between signal change and scale; m * This represents the optimal scale factor corresponding to the scale window used in fractal dimension estimation.

[0032] More preferably, the expression for the chaotic neural network is:

[0033] F(t)=[d WS (t),σ(t),ε(t),P(t),....];

[0034]

[0035] yi(t+1)=f(x i (t+1));

[0036] Where F(t) represents the surrounding rock feature vector input to the chaotic neural network at time t; d WS (t) represents the target dimension feature value at time t; ε(t) represents the strain value of the surrounding rock at time t; σ(t) represents the compressive strength of the rock mass material monitored at time t; P(t) represents the real-time surrounding rock pressure monitored at time t; x i (t) represents the internal state function of the i-th neuron at time t; g() represents the return mapping function; J represents the total number of quantized neurons; w ij The connection weights represent the contribution of the j-th quantized neuron to the state update of the i-th neuron; H represents the number of dimensions of the surrounding rock sampling vector; v ih F represents the connection weights that represent the contribution of the surrounding rock features in the h-th dimension to the state update of the i-th neuron; h (t) represents the surrounding rock feature in the h-th dimension of the input chaotic neural network at time t; θi The bias parameter of the i-th neuron is represented by ; c represents the perturbation strength parameter; z(t) represents the chaotic sequence generated by the chaotic mapping; y i (t+1) represents the output state function of the i-th neuron at time t+1; f() represents the nonlinear activation function.

[0037] Secondly, the present invention also provides a system for predicting the stability of surrounding rock in underground powerhouses, which is applied to the method described above, the system comprising:

[0038] A time-series monitoring data acquisition module is used to acquire time-series monitoring data of the surrounding rock of an underground powerhouse, wherein the time-series monitoring data includes acoustic emission signal data of the surrounding rock and sampling vectors of the surrounding rock;

[0039] The target dimension feature value acquisition module is used to perform fractal dimension modeling on the time-series monitoring data to obtain the detection fractal dimension function, and to perform local search on the scaling factor, correction coefficient, and scale factor in the detection fractal dimension function according to the hybrid leapfrog algorithm to obtain the optimal scaling factor, optimal correction coefficient, and optimal scale factor; based on the optimal scaling factor, the optimal correction coefficient, the optimal scale factor, and the detection fractal dimension function, a target dimension feature extraction function is constructed, and the target dimension feature value in the time-series monitoring data is extracted according to the target dimension feature extraction function;

[0040] The surrounding rock deformation prediction module is used to input the target dimension feature value and the surrounding rock sampling vector into a chaotic neural network to predict the surrounding rock deformation and obtain the surrounding rock deformation prediction result.

[0041] Thirdly, the present invention also provides an underground powerhouse surrounding rock stability prediction device, including a memory and a processor;

[0042] The memory is used to store computer program code and transmit the computer program code to the processor;

[0043] The processor is configured to execute the method described above according to instructions in the computer program code.

[0044] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0046] 1. This invention discloses a method, system, equipment, and medium for predicting the stability of surrounding rock in an underground powerhouse. By monitoring acoustic emission and recording sampling vectors of the surrounding rock, multi-angle data acquisition of the surrounding rock state is achieved, ensuring higher information content and completeness of the monitoring data. Furthermore, time-series data is used to capture subtle changes in the surrounding rock, improving the timeliness and accuracy of surrounding rock stability monitoring and early warning. A fractal dimension model is used to model the monitoring data, revealing the complex nonlinear behavior during the deformation process of the surrounding rock. By detecting the fractal dimension function, the inherent structural characteristics of the data are captured. A hybrid leapfrog algorithm is used for local search, and by adjusting the scaling factor, correction coefficient, and scale factor, the optimal parameter combination can be obtained, significantly improving the accuracy and robustness of the fractal model in capturing the deformation characteristics of the surrounding rock. Simultaneously, a chaotic neural network is used to input and process the target features and sampling vectors, better handling the randomness, nonlinearity, and complexity issues present in the system, achieving efficient prediction of the surrounding rock deformation process.

[0047] 2. In the method, system, equipment, and medium for predicting the stability of surrounding rock in an underground powerhouse, the present invention simultaneously collects surrounding rock monitoring data, surrounding rock acoustic emission signal data, and support structure stress monitoring data at multiple measuring points within a preset interval distance in the underground space. This ensures comprehensive coverage and spatial distribution integrity of monitoring information, helps to capture the state changes of surrounding rock and support structure at different locations, and monitors the displacement and stress state of the surrounding rock, which can comprehensively describe the stress state and changing trend of the surrounding rock and support structure from both deformation and stress perspectives. Integrating all support structure stress monitoring data and surrounding rock monitoring data can fully utilize multi-source data from various measuring points, achieve cross-validation of information, and enhance the effectiveness of signals. At the same time, multiple surrounding rock sampling vectors are constructed for each measuring point, so that subsequent data analysis, feature extraction, and model construction can better reflect the local surrounding rock characteristics of each measuring point, improve the model's ability to perceive local anomalies, and utilize surrounding rock displacement and stress data. This not only allows for timely acquisition of minute changes in the surrounding rock in different directions but also provides multi-dimensional input for hybrid data models, significantly improving the accuracy of stability prediction. Attached Figure Description

[0048] 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.

[0049] Figure 1 This is a flowchart of a method for predicting the stability of surrounding rock in an underground powerhouse, provided by the present invention.

[0050] Figure 2 This is a fitting curve of the size of the acoustic emission source in the surrounding rock and the released energy provided by the present invention.

[0051] Figure 3 This is a structural block diagram of an underground powerhouse surrounding rock stability prediction system provided by the present invention.

[0052] Figure 4 This is a structural block diagram of an underground powerhouse surrounding rock stability prediction device provided by the present invention. Detailed Implementation

[0053] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0054] See Figure 1 This invention provides a method for predicting the stability of surrounding rock in underground powerhouses, the method comprising steps S1 to S4.

[0055] Step S1: Collect time-series monitoring data of the surrounding rock of the underground powerhouse. The time-series monitoring data includes acoustic emission signal data of the surrounding rock and sampling vector of the surrounding rock.

[0056] In this embodiment, acoustic emission sensors, displacement sensors, and strain sensors can be used to monitor time-series data. The acoustic emission sensors can be high-sensitivity piezoelectric sensors or capacitive sensors to capture minute acoustic signals, effectively covering the high-frequency signals generated by the common propagation of microcracks in rock masses. The displacement sensors and strain sensors can also be equipped with strain gauges, displacement meters, etc., as needed to record physical quantities such as deformation and stress changes in the surrounding rock.

[0057] The deployment of sensors in key areas should be based on the geological structure and engineering risk points of the surrounding rock of the underground powerhouse, selecting critical areas prone to cracking, damage, or deformation. For example, areas prone to cracking or stress concentration should be prioritized for detection. To improve data representativeness, sensor networks or arrays are typically used. This not only obtains data from multiple angles and directions but also allows for the assessment of the overall behavior of the surrounding rock through spatial data correlation analysis. All sensors should be connected to the data acquisition device via standardized interfaces. A unified time reference (e.g., GPS synchronization or network clock synchronization) must be used between all sensors to ensure temporal consistency of multi-sensor data, which is beneficial for subsequent data correlation analysis.

[0058] Furthermore, acoustic emission signals typically contain high-frequency components, thus requiring a sufficiently high sampling frequency (e.g., several kHz to tens of kHz) to capture signal details. The specific value needs to be determined based on the signal bandwidth and the expected monitoring target. For low-frequency signals such as displacement and strain, an appropriate sampling frequency should be selected based on the rate of change to ensure data continuity without generating redundant data. Related sensors and data acquisition systems should have high resolution (e.g., 16-bit or higher) to capture subtle changes in surrounding rock deformation and avoid data loss due to quantization. For acoustic emission signals, a trigger threshold needs to be preset to ensure that data is recorded only when the signal strength exceeds the ambient noise level, thereby improving data quality and saving storage space. Initial setup can be done through trial sampling based on site noise levels to determine a reasonable threshold before formal implementation. A mode combining continuous sampling and event triggering can be set. When an abnormal signal exceeding the threshold is detected, data for that period is automatically saved; simultaneously, low-frequency sampling is also performed during low-risk periods for subsequent comparison and reference. Ensuring all sensors share the same time base is crucial for multi-dimensional analysis of time-series data. GPS synchronization, NTP (Network Time Protocol), or local clock calibration can be used to prevent data alignment problems caused by clock drift.

[0059] In one example, monitoring data of the surrounding rock, acoustic emission signal data of the surrounding rock, and stress monitoring data of the support structure corresponding to multiple measuring points within a preset interval range in the underground space corresponding to the underground powerhouse are acquired. All stress monitoring data of the support structure and all monitoring data of the surrounding rock are integrated, and multiple surrounding rock sampling vectors corresponding to each measuring point are constructed. When collecting the monitoring data of the surrounding rock and the stress monitoring data of the support structure, each corresponding measuring point simultaneously monitors the horizontal and vertical displacement of the surrounding rock, as well as the radial and tangential stress of the surrounding rock.

[0060] Based on the geological profile, fault strike, and joint distribution of the surrounding rock of the underground powerhouse, areas with potential weak points and concentrated fissures are selected as key monitoring areas. For critical components such as the support structure, roof slab, and arch columns, where the risk of concentrated stress and displacement is high, monitoring points should be deployed nearby.

[0061] For areas with relatively homogeneous geological conditions, a uniformly spaced monitoring point layout can be adopted. For example, a monitoring point can be set every 5 to 10 meters to ensure data collection covers the entire area. In potentially hazardous areas or where support structures are concentrated, the spacing can be reduced to 2 to 3 meters to ensure that more subtle changes can be captured. On the same horizontal plane, multiple monitoring locations can be selected based on the size of the surrounding rock, construction design, and construction process requirements. For surrounding rock with obvious stratification or depth changes, multiple vertical monitoring locations can be selected to obtain response data at different depths. A grid can be constructed on the horizontal plane, with a monitoring point at the center of each grid. The grid size can be designed to be 5m × 5m or denser, depending on the site size and monitoring requirements. Based on the volumetric distribution and structural characteristics of the surrounding rock, vertically layered data can be superimposed on the horizontal plane. For example, monitoring points can be set up on the roof, vault, and floor of the underground powerhouse to achieve three-dimensional displacement and stress monitoring.

[0062] In this embodiment, by pre-setting intervals within the underground space, monitoring data of the surrounding rock, acoustic emission signal data of the surrounding rock, and stress monitoring data of the support structure are simultaneously collected at multiple measuring points. This ensures comprehensive coverage and complete spatial distribution of monitoring information, which helps to capture the state changes of the surrounding rock and support structure at different locations. Furthermore, monitoring the displacement and stress state of the surrounding rock can comprehensively describe the stress state and changing trend of the surrounding rock and support structure from both deformation and stress perspectives. Integrating all stress monitoring data of the support structure and monitoring data of the surrounding rock can fully utilize the multi-source data from each measuring point, achieve cross-validation of information, and enhance the effectiveness of the signal. At the same time, multiple surrounding rock sampling vectors are constructed for each measuring point, so that subsequent data analysis, feature extraction, and model construction can better reflect the local surrounding rock characteristics of each measuring point, improve the model's ability to perceive local abnormal changes, and utilize the surrounding rock displacement and stress data. This not only allows for timely acquisition of minute changes in the surrounding rock in different directions but also provides multi-dimensional input for the hybrid data model, significantly improving the accuracy of stability prediction.

[0063] Furthermore, high-precision digital displacement gauges or fiber optic sensors are used for displacement sensing, with a resolution of 0.01 mm or higher typically recommended. Two sensors are installed at each measuring point to collect lateral (horizontal) and vertical displacement data, respectively. Sufficient bonding area must be maintained during installation to ensure tight coupling between the sensor and the surrounding rock or support structure. For surrounding rock stress monitoring, pressure gauges or strain gauge sensors are commonly used, measuring radial and tangential stress. The stress sensor should have high linear response and repeatability at peak stress. Depending on the geographical location of the measuring point, the sensor should be calibrated to ensure its measurement direction aligns with the actual stress direction of the surrounding rock. Acoustic emission sensors: Piezoelectric high-sensitivity acoustic emission sensors are used, with a wide frequency response range (typically above 10 kHz to hundreds of kHz), capable of capturing fast transient signals. They are fixed in areas prone to micro-damage and crack propagation, using dedicated mounting brackets and high-quality coupling agents to improve signal transmission efficiency. A multi-channel high-speed acquisition system is configured to simultaneously receive signals from multiple sensors. For displacement and stress monitoring, the sampling frequency can be set between 1 and 10 Hz; for acoustic emission monitoring, the sampling frequency is 10 kHz or even higher. The module must have a high-resolution data converter (such as a 16-bit or 24-bit ADC) to ensure the accuracy of the acquired data meets monitoring requirements. A unified time synchronization scheme, such as GPS clock, NTP network clock synchronization, or a dedicated synchronization module, should be adopted to ensure that all data records have a unified timestamp. Real-time data upload should be achieved using a local area network (WLAN or wired network) to ensure centralized data management at the monitoring center. The data acquisition system should be designed to resist electromagnetic interference, using shielded cables and filters to ensure high-quality data acquisition even in the low-interference, complex environment of an underground plant.

[0064] In this embodiment, the pre-set monitoring point scheme ensures both wide-area monitoring in areas with homogeneous geological conditions and high-density point distribution in key areas prone to weakening or deformation (such as fault zones, fracture concentration areas, and critical parts of the support structure). This allows for both understanding the overall state of the surrounding rock and capturing subtle local changes. By constructing a grid on the horizontal plane and layering points vertically, multi-dimensional data acquisition of the underground space is achieved, which helps to accurately reconstruct the deformation and stress conditions of the surrounding rock at different depths and directions, providing comprehensive information for subsequent evaluation. High-precision displacement and stress sensors are used to effectively capture minute horizontal and vertical displacements of the surrounding rock, as well as radial and tangential stress changes. The sampled data not only has high temporal resolution but also high spatial resolution, ensuring that the data reflects the true state of the surrounding rock. All support structure stress monitoring data and all surrounding rock monitoring data are integrated, and through spatiotemporal data alignment, signal denoising, and fusion processing, a unified data format is formed, providing stable and high-quality data samples for model input. For each measuring point, a sampling vector is constructed based on the collected multi-dimensional data such as displacement, stress, and acoustic emission. This enables subsequent data-driven analysis (such as fractal feature extraction and intelligent prediction) to more accurately reflect the actual engineering condition and improve the robustness and accuracy of the prediction model.

[0065] like Figure 2 As shown, Figure 2 The figure shows the fitted curves of the acoustic emission source size D and the released energy En in the surrounding rock, divided into two subplots (a) and (b), corresponding to different fitting conditions. The horizontal axis D represents the size of the acoustic emission source, ranging from 1.1 to 1.9. The vertical axis En represents the energy released by the acoustic emission, normalized to [0,1]. The figure contains two curves, corresponding to the first group of samples (dashed line) and the second group of samples (solid line), respectively.

[0066] exist Figure 2 In (a) of the diagram, as the size D of the acoustic emission source increases, the released energy En shows a rapid decreasing trend, followed by a stabilizing trend. The curves of the first and second groups of samples show some differences when D is small (approximately 1.1 to 1.3), but gradually converge after D > 1.5, indicating that the energy release characteristics of the two groups of samples are similar over a larger size range. Figure 2Figure (b) also illustrates the relationship between D and En, but the curve decreases at a faster rate. The first and second groups of samples show significant differences when D is small (approximately 1.1 to 1.3), with the energy release of the first group slightly higher than that of the second. As D increases, the two curves gradually converge after D > 1.4, eventually becoming consistent. Clearly, there is a non-linear relationship between the acoustic emission source size and the released energy En; energy release decreases rapidly with increasing size and tends to stabilize over a larger size range. The first and second groups of samples exhibit some differences in energy release characteristics within a small size range, but show consistency over a large size range, reflecting the acoustic emission characteristics of the surrounding rock under different conditions.

[0067] Step S2 involves performing fractal dimension modeling on the time-series monitoring data to obtain the detection fractal dimension function, and then using the hybrid leapfrog algorithm to perform a local search on the scaling factor, correction coefficient, and scale factor in the detection fractal dimension function to obtain the optimal scaling factor, optimal correction coefficient, and optimal scale factor.

[0068] In this embodiment, the acoustic emission signal data of the surrounding rock is segmented on the time axis to obtain multiple acoustic emission sequences of the surrounding rock. The acoustic emission sequences of the surrounding rock are then normalized to obtain a standard acoustic emission sequence of the surrounding rock. Based on the standard acoustic emission sequence of the surrounding rock, an initial detection fractal dimension function is constructed. The local fractal values ​​at different sampling scales are calculated using the initial detection fractal dimension function, and the sampling logarithmic relationship between the sampling scale and the corresponding local fractal value is obtained based on linear fitting. Based on the sampling logarithmic relationship and the initial detection fractal dimension function, a detection fractal dimension function is constructed.

[0069] One or more amplitude thresholds are set based on actual monitoring data and signal noise levels. Generally, when the signal amplitude exceeds the preset threshold, it marks the start of an acoustic emission event; when the signal drops to near the background noise level, it marks the end of the event. The entire time series is scanned. When the signal rises from below the threshold to above the threshold, this moment is recorded as the start of the acoustic emission event. Then, the point where the signal drops from a high amplitude state to below the threshold or remains below the threshold for a certain period is found and recorded as the end of the event. For two events that are close in time and may occur consecutively, a minimum event interval can be preset to avoid incorrectly segmenting individual consecutive events; simultaneously, a minimum duration is set to filter out short noise signals, ensuring that only valid events are captured. Alternatively, the entire monitoring time can be divided into fixed periods (e.g., every 1 second or every few seconds), and energy calculations or other statistical analyses can be performed on the signal within each window. The window length can be selected appropriately based on the duration of the acoustic emission event and the signal sampling frequency; the power, mean, standard deviation, and other information within each window are calculated, and the threshold is used to determine whether the window contains an acoustic emission event. If the conditions are met, the signal within that window can be considered as an event segment; otherwise, it is discarded or merged with adjacent windows. After extracting each segment, multiple acoustic emission sequences are formed. Each sequence is normalized to ensure that the amplitudes of different events fall within the same standard range, facilitating subsequent fractal dimension modeling. Normalization typically involves mapping the amplitude to the 0-1 interval or standardizing it based on statistics such as the mean and standard deviation. Each acoustic emission sequence should include its start and end times, amplitude information, and its position within the entire acquired data window, facilitating subsequent temporal window reconstruction and feature extraction.

[0070] In this embodiment, the expression for the target dimension feature extraction function is:

[0071]

[0072] d WS =T(δ;α) * ,β * m * );

[0073] Where T() represents the detection fractal dimension function; δ represents the sampling scale, which is an integer multiple of the minimum sampling interval; x n denoted as the amplitude of the nth sampling point after segmentation of the acoustic emission signal data in the surrounding rock; N represents the minimum number of scales required to completely cover the signal at the sampling scale δ; α represents the scaling factor corresponding to the cumulative length of the signal; β represents the correction coefficient corresponding to the mapping value of signal change and scale relationship; m represents the scale factor corresponding to the scale window used in fractal dimension estimation; MSE represents the mean square error; K represents the total number of acoustic emission sequences in the surrounding rock acoustic emission signal data; T kThis represents the fractal dimension value of the acoustic emission sequence of the k-th segment of the surrounding rock; This represents the fractal reference value under the acoustic emission sequence of the surrounding rock in the k-th segment; d WS α represents the feature extraction function for the target dimension. * β represents the optimal scaling factor corresponding to the signal accumulation length. * This represents the optimal correction coefficient corresponding to the mapping value between signal change and scale; m * This represents the optimal scale factor corresponding to the scale window used in fractal dimension estimation.

[0074] For the standard acoustic emission sequence x n Calculate the signal increment |x| at the sampling scale δ. n+δ -x n |, multiply the signal increment by the scaling factor α, and divide the signal increment |x n+δ -x n Adjust β exponentially, sum over all sampling points n, and obtain the cumulative value. Perform a logarithmic transformation on the cumulative value and the scale factor, and calculate the numerator and denominator respectively: Dividing the numerator and denominator yields the initial detection fractal dimension function.

[0075] Understandably, since T(δ; α, β, m) is a fractal dimension function used to describe the fractal characteristics of acoustic emission signal data from surrounding rock at different sampling scales δ, it modifies the signal increment |x n+δ -x n The cumulative value of | is logarithmically transformed, and combined with the logarithmic relationship of the sampling scale, to calculate the fractal dimension. The parameters α, β, and m are optimized by minimizing the mean squared error (MSE) so that the fitted fractal dimension function T(δ; α, β, m) closely approximates the fractal reference value. The error is minimized.

[0076] This step also includes steps S21 to S25.

[0077] Step S21: Randomly generate multiple frog individuals and initialize them within the feasible region of each frog individual. Each frog individual represents an optimization parameter combination including scaling factor, correction coefficient, and scale factor.

[0078] In this step, each individual frog represents an optimization parameter combination containing a scaling factor α, a correction coefficient β, and a scale factor m. The scaling factor α can be defined as a positive number, and its value typically ranges from [α...]. min α max For example, [0.1, 10]; the correction coefficient β is defined as a positive number, and its value usually ranges from [β]. min ,β maxFor example, [0.5, 2]; the scaling factor m is defined as a positive number, and its value usually ranges from [m...]. min m max For example, [1, 5]. Set the frog population size to q (i.e., generate q individual frogs). For each individual frog, randomly generate its parameter combination (α, β, m), with a scaling factor α ranging from [α...]. min α max A value is randomly generated within the range [ ]. That is, α = α min +(α min -α min )rand(0,1), and similarly, the correction coefficient β and the scaling factor m can be randomly generated in the same way.

[0079] The position of each frog individual is represented by its parameter combination (α, β, m). The generated (α, β, m) is used as the initial position of the frog individual and stored in the population. This random generation process is repeated q times to obtain a population containing q frog individuals, i.e., population = {(α1, β1, m1), (α2, β2, m2), ..., (α...} q ,β q m a Ensure that the parameters (α, β, m) for each individual frog are within the feasible region. If a parameter is out of range, it can be regenerated or truncated to the boundary value, for example, if α > α. max Let α = α max .

[0080] Step S22: Based on the detected fractal dimension function and fractal reference value, construct a fractal reference function, and calculate the fitness of each individual frog according to the fractal reference function and fractal reference value.

[0081] In this step, through the fractal reference function Calculate the fitness of each individual frog.

[0082] Step S23: Sort the frog individuals in the population in ascending order of fitness, and divide the population into multiple sub-culture gene bodies, wherein each culture gene body contains several frog individuals.

[0083] In this step, the number of subculture genes, M, is set, typically a factor of the population size, q. Each subculture gene contains q / M of individual frogs. If q is not divisible by M, the remaining individuals can be distributed among the first few genes.

[0084] Following the ordered sequence of individual frogs, the population is sequentially distributed into M subculture gene groups. For example, subculture gene group 1 contains frogs from the 1st to the q / Mth ordered frogs; subculture gene group 2 contains frogs from the q / M+1th to the 2qMth ordered frogs. This process continues until all frogs have been distributed.

[0085] Step S24: Select the frog individual with the best fitness and the frog individual with the worst fitness in each subculture gene body, and update the position of the frog individual with the worst fitness based on the position update function to perform local search.

[0086] In this step, each subculture genera has been sorted by fitness from best to worst. The best frog individual represents the individual with the lowest fitness among the subculture genera, denoted as Best, and its position can be represented as BestPosition = (α best ,β best m best The worst frog individual represents the individual with the highest fitness in the subculture gene pool, denoted as Worst, and its position can be represented as WorstPosition=(α worst ,β worst m worst The position update formula can be expressed as NewPosition = WorstPosition + r(BestPosition - WorstPosition), where r is a random number, usually uniformly distributed in the range [0,1], used to control the update step size.

[0087] Step S25: When each cultural gene has undergone at least one round of local search and reached the maximum number of iterations, the population is recombined and sorted again according to fitness in ascending order. The best frog individual in the population is updated, and the position of the globally best frog individual is recorded.

[0088] In this embodiment, multiple frog individuals are randomly generated and initialized within the feasible region to ensure population diversity and global search capability, providing a good starting point for subsequent optimization. Fitness is calculated based on the detection fractal dimension function and fractal reference value, accurately assessing individual performance and improving optimization accuracy. Hierarchical optimization and parallel search are achieved through fitness-increasing sorting and population division into multiple sub-culture gene bodies, maintaining population diversity while improving the efficiency of combining global and local searches. In the local search, the worst-performing frog individual is optimized based on the position update function, bringing it closer to the optimal individual. Simultaneously, random step size and boundary checking mechanisms prevent getting trapped in local optima, and a global search supplementary mechanism enhances the algorithm's robustness. Finally, by recombining the population and updating the global optimum, the algorithm gradually approaches the global optimum, dynamically adjusting the population structure to balance local and global search capabilities. The overall technical performance is characterized by an efficient combination of global and local search, a dynamic balance between diversity and convergence, strong adaptability, high optimization efficiency, and strong robustness. It is suitable for complex nonlinear, multimodal, high-dimensional, and dynamic optimization problems, and can quickly find high-quality solutions and gradually approach the global optimum.

[0089] Step S3: Based on the optimal scaling factor, optimal correction coefficient, optimal scale factor, and detection fractal dimension function, construct the target dimension feature extraction function, and extract the target dimension feature values ​​from the time series monitoring data according to the target dimension feature extraction function.

[0090] In this step, based on the probe fractal dimension function T(δ; α, β, m) and the optimized parameters, the function is written as d WS =T(δ;α) * ,β * m * ), where δ represents the sampling scale. The parameters in the function, after optimization, can better reflect the fractal characteristics or complexity of time-series data signals.

[0091] Furthermore, the target dimension feature extraction function d WS Incorporating parameter α * β * and m * The function for detecting fractal dimension can be further constructed into a function of the following form: d WS (δ)=α * Δln(|x n+δ -x n |)+β * ×δ m *Because time-series monitoring data often exhibits multi-scale characteristics, meaning the data may display different patterns of change at different scales, preprocessing of the raw data is necessary to extract effective fractal features. This includes calculating local differences or fluctuations at different sampling scales or time delays. This generates a quantity describing local changes in the data at each scale, allowing for the acquisition of multiple sets of fractal features from the same dataset. WS (δ) value. For each sampling scale δ, the corresponding d is obtained through the above target dimension feature extraction function. WS (δ) value. If a uniform target dimension feature is required, d values ​​at multiple scales can be used. WS (δ) values ​​are statistically summarized, for example, to calculate d at different scales. WS The average value of (δ) is used as a global feature description.

[0092] After processing the data obtained in the above steps, one or more target dimension feature values ​​are output. These values ​​reflect the fractal properties of time-series monitoring data at multiple scales and the complexity of the signals, providing a quantitative basis for subsequent data monitoring, anomaly detection, and prediction tasks.

[0093] Step S4: Input the target dimension feature value and the surrounding rock sampling vector into the chaotic neural network to predict the surrounding rock deformation and obtain the surrounding rock deformation prediction result.

[0094] This step also includes steps S41 to S43.

[0095] Step S41: Normalize the target dimension feature values ​​and the surrounding rock sampling vector to obtain standard target dimension feature values ​​and standard surrounding rock sampling vectors.

[0096] In this step, the minimum, maximum, or mean and standard deviation of the target dimension feature vector and the surrounding rock sampling vector are calculated respectively, and then normalized feature by feature. This ensures that the values ​​of the two sets of data are at similar scales in their respective dimensions. If the distribution characteristics of the two sets of data differ significantly, an appropriate normalization method can be selected according to the specific circumstances to preserve data information as much as possible.

[0097] Step S42: Concatenate the standard target dimension feature value and the standard surrounding rock sampling vector to obtain the surrounding rock feature vector, and input the surrounding rock feature vector into the chaotic neural network.

[0098] The expression for a chaotic neural network is:

[0099] F(t)=[d WS (t),σ(t),ε(t),P(t),....];

[0100]

[0101] y i (t+1)=f(x i (t+1));

[0102] Where F(t) represents the surrounding rock feature vector input to the chaotic neural network at time t; d WS (t) represents the target dimension feature value at time t; ε(t) represents the strain value of the surrounding rock at time t; σ(t) represents the compressive strength of the rock mass material monitored at time t; P(t) represents the real-time surrounding rock pressure monitored at time t; x i (t) represents the internal state function of the i-th neuron at time t; g() represents the return mapping function; J represents the total number of quantized neurons; w ij The connection weights represent the contribution of the j-th quantized neuron to the state update of the i-th neuron; H represents the number of dimensions of the surrounding rock sampling vector; v ih F represents the connection weights that represent the contribution of the surrounding rock features in the h-th dimension to the state update of the i-th neuron; h (t) represents the surrounding rock feature in the h-th dimension of the input chaotic neural network at time t; θ i The bias parameter of the i-th neuron is represented by ; c represents the perturbation strength parameter; z(t) represents the chaotic sequence generated by the chaotic mapping; y i (t+1) represents the output state function of the i-th neuron at time t+1; f() represents the nonlinear activation function.

[0103] Step S43: Based on the internal state function and output state function of the chaotic neural network, obtain the surrounding rock deformation prediction results.

[0104] In this embodiment, by normalizing the target dimension feature values ​​and the surrounding rock sampling vectors respectively, the differences in dimensions and numerical ranges between different data are eliminated, ensuring that the two sets of data are on the same numerical scale. This not only improves the consistency and comparability of the data but also provides a reliable foundation for subsequent data fusion, reducing problems caused by numerical instability during neural network training. The normalized standard target dimension feature values ​​and standard surrounding rock sampling vectors are concatenated to form a unified surrounding rock feature vector. This fusion method can fully retain the key information in both sets of data and construct a comprehensive description containing multi-dimensional features, providing richer and more comprehensive input information for the chaotic neural network, which is beneficial for capturing the complex nonlinear characteristics in the surrounding rock deformation process. Based on the constructed surrounding rock feature vector, the chaotic neural network uses its internal state function and output state function for dynamic modeling and prediction, effectively learning and expressing the complex dynamic behavior of the system.

[0105] In one example, assume the output layer of the chaotic neural network has only one neuron, and the output value is denoted as y(t), ranging from [0,1], representing a comprehensive index of the "degree of instability of the surrounding rock". The closer y(t) is to 1, the closer the model judges the surrounding rock to an unstable state; the closer y(t) is to 0, the closer the model judges the surrounding rock to a stable state. Based on past engineering monitoring data and field experience, to facilitate rapid decision-making, two thresholds T1 and T2 (where 0≤T1<T2<1) can be set to divide the output value into three risk levels. For example: stable zone: 0≤y(t)<T1; critical zone: T1≤y(t)<T2; unstable zone: T2≤y(t)≤1.

[0106] Specific thresholds can be set based on historical rock fracturing and instability data, through statistical analysis or expert experience. For example, T1 = 0.4, T2 = 0.7. If the network output y(t) < 0.4, the surrounding rock is considered to be in a safe or relatively stable range; if the network output y(t) is between 0.4 and 0.7, it may mean that a certain degree of micro-crack propagation has occurred inside the surrounding rock, requiring strengthened monitoring and early warning; if the network output y(t) ≥ 0.7, it indicates a significantly increased risk of surrounding rock instability, and safety measures such as support, pressure relief, or evacuation should be taken immediately.

[0107] Furthermore, in practical engineering, the state of the surrounding rock evolves continuously over time, and a single output only represents the situation at a certain moment. To make a more stable and reliable judgment, the following methods can be used to perform trend analysis on y(t): take the average or weighted average of y(t) within a time window (such as the most recent 5 or 10 monitoring times) to smooth out the influence of noise; if the output y(t) exceeds a certain threshold in multiple consecutive acquisitions, then a warning or alarm is triggered; evaluate the growth (or decline) rate of y(t) over time. If a rapid increase occurs in a short period of time, it indicates that the surrounding rock condition is deteriorating and should be given high attention.

[0108] In this embodiment, by monitoring acoustic emission and recording sampling vectors of the surrounding rock of the underground powerhouse, multi-angle data acquisition of the surrounding rock state is achieved, ensuring that the monitoring data has higher information content and completeness. Furthermore, time-series data is used to capture subtle changes in the surrounding rock, improving the timeliness and accuracy of monitoring and early warning. A fractal dimension model is used to model the monitoring data, which can reveal the complex nonlinear behavior in the deformation process of the surrounding rock. By detecting the fractal dimension function, the inherent structural characteristics of the data are captured. A hybrid leapfrog algorithm is used for local search. By adjusting the scaling factor, correction coefficient, and scale factor, the optimal parameter combination can be obtained, which significantly improves the accuracy and robustness of the fractal model in capturing the deformation characteristics of the surrounding rock. At the same time, a chaotic neural network is used to input and process the target features and sampling vectors, which can better handle the randomness, nonlinearity, and complexity problems in the system, and achieve efficient prediction of the deformation process of the surrounding rock.

[0109] See Figure 3 The present invention also provides a system for predicting the stability of surrounding rock in underground powerhouses, comprising:

[0110] A time-series monitoring data acquisition module is used to acquire time-series monitoring data of the surrounding rock of an underground powerhouse, wherein the time-series monitoring data includes acoustic emission signal data of the surrounding rock and sampling vectors of the surrounding rock;

[0111] The target dimension feature value acquisition module is used to perform fractal dimension modeling on the time-series monitoring data to obtain the detection fractal dimension function, and to perform local search on the scaling factor, correction coefficient, and scale factor in the detection fractal dimension function according to the hybrid leapfrog algorithm to obtain the optimal scaling factor, optimal correction coefficient, and optimal scale factor; based on the optimal scaling factor, the optimal correction coefficient, the optimal scale factor, and the detection fractal dimension function, a target dimension feature extraction function is constructed, and the target dimension feature value in the time-series monitoring data is extracted according to the target dimension feature extraction function;

[0112] The surrounding rock deformation prediction module is used to input the target dimension feature value and the surrounding rock sampling vector into a chaotic neural network to predict the surrounding rock deformation and obtain the surrounding rock deformation prediction result.

[0113] In one example, the time-series monitoring data acquisition module is used to acquire surrounding rock monitoring data, surrounding rock acoustic emission signal data, and support structure stress monitoring data corresponding to multiple measuring points within a preset interval range in the underground space corresponding to the underground plant. It integrates all support structure stress monitoring data and all surrounding rock monitoring data and constructs multiple surrounding rock sampling vectors corresponding to each measuring point.

[0114] In one example, when the time-series monitoring data acquisition module is used to collect surrounding rock monitoring data and support structure stress monitoring data, each corresponding measuring point simultaneously monitors the horizontal and vertical displacement of the surrounding rock, as well as the radial and tangential stress of the surrounding rock.

[0115] In one example, the target dimension feature value acquisition module is used to segment the surrounding rock acoustic emission signal data on the time axis, obtain multiple segments of surrounding rock acoustic emission sequences, and perform amplitude normalization processing on the surrounding rock acoustic emission sequences to obtain a standard surrounding rock acoustic emission sequence; based on the standard surrounding rock acoustic emission sequence, an initial detection fractal dimension function is constructed; the initial detection fractal dimension function is used to calculate the local fractal values ​​at different sampling scales, and the sampling logarithmic relationship between the sampling scale and the corresponding local fractal values ​​is obtained based on linear fitting; based on the sampling logarithmic relationship and the initial detection fractal dimension function, the detection fractal dimension function is constructed.

[0116] In one example, the target dimension feature value acquisition module randomly generates multiple frog individuals and initializes them within the feasible region of each frog individual. Each frog individual represents an optimized parameter combination including scaling factor, correction coefficient, and scale factor. Based on the probe fractal dimension function and fractal reference value, a fractal reference function is constructed, and the fitness of each frog individual is calculated according to the fractal reference function and fractal reference value. The frog individuals in the population are sorted in ascending order of fitness, and the population is divided into multiple sub-culture gene bodies, where each culture gene body contains several frog individuals. In each sub-culture gene body, the frog individual with the best fitness and the worst fitness are selected respectively, and the position of the worst fitness frog individual is updated based on the position update function to perform local search. When at least one round of local search has been performed in each culture gene body and the maximum number of iterations has been reached, the population is recombined and sorted again in ascending order of fitness, the best frog individual in the population is updated, and the position of the globally best frog individual is recorded.

[0117] In one example, the surrounding rock deformation prediction module is used to normalize the target dimension feature values ​​and the surrounding rock sampling vector to obtain standard target dimension feature values ​​and standard surrounding rock sampling vectors; the standard target dimension feature values ​​and standard surrounding rock sampling vectors are concatenated to obtain the surrounding rock feature vector, and the surrounding rock feature vector is input into a chaotic neural network; the surrounding rock deformation prediction result is obtained based on the internal state function and output state function of the chaotic neural network.

[0118] In one example, the expression for the target dimension feature extraction function is:

[0119]

[0120] d WS =T(δ;α) * ,β * m* );

[0121] Where T() represents the detection fractal dimension function; δ represents the sampling scale, which is an integer multiple of the minimum sampling interval; x n denoted as the amplitude of the nth sampling point after segmentation of the acoustic emission signal data in the surrounding rock; N represents the minimum number of scales required to completely cover the signal at the sampling scale δ; α represents the scaling factor corresponding to the cumulative length of the signal; β represents the correction coefficient corresponding to the mapping value of signal change and scale relationship; m represents the scale factor corresponding to the scale window used in fractal dimension estimation; MSE represents the mean square error; K represents the total number of acoustic emission sequences in the surrounding rock acoustic emission signal data; T k This represents the fractal dimension value of the acoustic emission sequence of the k-th segment of the surrounding rock; This represents the fractal reference value under the acoustic emission sequence of the surrounding rock in the k-th segment; d WS α represents the feature extraction function for the target dimension. * β represents the optimal scaling factor corresponding to the signal accumulation length. * This represents the optimal correction coefficient corresponding to the mapping value between signal change and scale; m * This represents the optimal scale factor corresponding to the scale window used in fractal dimension estimation.

[0122] In one example, the expression for a chaotic neural network is:

[0123] F(t)=[d WS (t),σ(t),ε(t),P(t),....];

[0124]

[0125] y i (t+1)=f(x i (t+1));

[0126] Where F(t) represents the surrounding rock feature vector input to the chaotic neural network at time t; d WS (t) represents the target dimension feature value at time t; ε(t) represents the strain value of the surrounding rock at time t; σ(t) represents the compressive strength of the rock mass material monitored at time t; P(t) represents the real-time surrounding rock pressure monitored at time t; x i (t) represents the internal state function of the i-th neuron at time t; g() represents the return mapping function; J represents the total number of quantized neurons; w ij The connection weights represent the contribution of the j-th quantized neuron to the state update of the i-th neuron; H represents the number of dimensions of the surrounding rock sampling vector; v ih F represents the connection weights that represent the contribution of the surrounding rock features in the h-th dimension to the state update of the i-th neuron; h(t) represents the surrounding rock feature in the h-th dimension of the input chaotic neural network at time t; θ i The bias parameter of the i-th neuron is represented by ; c represents the perturbation strength parameter; z(t) represents the chaotic sequence generated by the chaotic mapping; y i (t+1) represents the output state function of the i-th neuron at time t+1; f() represents the nonlinear activation function.

[0127] See Figure 4 The present invention also provides an underground powerhouse surrounding rock stability prediction device, including a memory and a processor;

[0128] The memory is used to store computer program code and transmit the computer program code to the processor;

[0129] The processor is used to execute the above-described method for predicting the stability of surrounding rock in an underground powerhouse according to instructions in the computer program code.

[0130] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for predicting the stability of surrounding rock in an underground powerhouse.

[0131] Generally, the computer instructions for implementing the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself, which is temporarily propagating.

[0132] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EKROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0133] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks such as TensorFlow and PyTorch can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer or to an external computer (e.g., via the Internet using an Internet service provider) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0134] The aforementioned equipment and non-transitory computer-readable storage media can be found in the detailed description of a method for predicting the stability of surrounding rock in an underground powerhouse and its beneficial effects, which will not be repeated here.

[0135] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting the stability of surrounding rock in an underground powerhouse, characterized in that, include: Collect time-series monitoring data of the surrounding rock of the underground powerhouse, wherein the time-series monitoring data includes acoustic emission signal data of the surrounding rock and sampling vector of the surrounding rock; The time-series monitoring data is modeled using fractal dimension to obtain the detection fractal dimension function. The scaling factor, correction coefficient, and scale factor in the detection fractal dimension function are then locally searched using the hybrid leapfrog algorithm to obtain the optimal scaling factor, optimal correction coefficient, and optimal scale factor. Based on the optimal scaling factor, the optimal correction coefficient, the optimal scale factor, and the detection fractal dimension function, a target dimension feature extraction function is constructed, and the target dimension feature values ​​in the time-series monitoring data are extracted according to the target dimension feature extraction function. The target dimension feature value and the surrounding rock sampling vector are input into a chaotic neural network to predict the surrounding rock deformation, so as to obtain the surrounding rock deformation prediction result. The expression for the target dimension feature extraction function is: ； ； ； in, This represents the function for detecting fractal dimension. Indicates the sampling scale; This indicates that the acoustic emission signal data of the surrounding rock has been processed in segments. The amplitude of each sampling point; Indicates the sampling scale The minimum number of scales required to fully cover the signal; This represents the scaling factor corresponding to the cumulative length of the signal; This represents the correction coefficient corresponding to the mapping value between signal change and scale. This represents the scale factor corresponding to the scale window used in fractal dimension estimation. Indicates mean square error; This indicates the total number of acoustic emission sequences in the surrounding rock acoustic emission signal data; Indicates the first Fractal dimension values ​​of acoustic emission sequences in surrounding rock segments; Indicates the first Fractal reference values ​​under acoustic emission sequences of surrounding rock segments; This represents the feature extraction function for the target dimension. This represents the optimal scaling factor corresponding to the signal accumulation length; This represents the optimal correction coefficient corresponding to the mapping value between signal change and scale. This represents the optimal scale factor corresponding to the scale window used in fractal dimension estimation.

2. The method for predicting the stability of surrounding rock in an underground powerhouse according to claim 1, characterized in that, The time-series monitoring data collected from the surrounding rock of the underground powerhouse specifically includes: Acquire surrounding rock monitoring data, surrounding rock acoustic emission signal data, and support structure stress monitoring data corresponding to multiple measuring points within a preset interval range in the underground space corresponding to the underground powerhouse. Integrate all support structure stress monitoring data and all surrounding rock monitoring data, and construct multiple surrounding rock sampling vectors corresponding to each measuring point. When collecting the surrounding rock monitoring data and the support structure stress monitoring data, each corresponding measuring point simultaneously monitors the horizontal and vertical displacement of the surrounding rock, as well as the radial and tangential stress of the surrounding rock.

3. The method for predicting the stability of surrounding rock in an underground powerhouse according to claim 1, characterized in that, The step of performing fractal dimension modeling on the time-series monitoring data to obtain the fractal dimension function for detection specifically includes: The acoustic emission signal data of the surrounding rock is segmented on the time axis to obtain multiple acoustic emission sequences of the surrounding rock, and the acoustic emission sequences of the surrounding rock are normalized to obtain a standard acoustic emission sequence of the surrounding rock. Based on the standard surrounding rock acoustic emission sequence, an initial detection fractal dimension function is constructed; The local fractal values ​​at different sampling scales are calculated by using the initial detection fractal dimension function, and the sampling logarithmic relationship between the sampling scale and the corresponding local fractal value is obtained by linear fitting. Based on the sampling logarithmic relationship and the initial detection fractal dimension function, the detection fractal dimension function is constructed.

4. The method for predicting the stability of surrounding rock in an underground powerhouse according to claim 3, characterized in that, The local search for scaling factors, correction coefficients, and scale factors in the detected fractal dimension function based on the hybrid leapfrog algorithm specifically includes: Multiple frog individuals are randomly generated and initialized within the feasible region of each frog individual, wherein each frog individual represents an optimized parameter combination including scaling factor, correction coefficient and scale factor; Based on the detected fractal dimension function and fractal reference value, a fractal reference function is constructed, and the fitness of each individual frog is calculated according to the fractal reference function and fractal reference value. Frog individuals in the population are sorted in ascending order of fitness, and the population is divided into multiple sub-cultures, each containing several frog individuals. In each subculture gene, the frog individual with the best fitness and the frog individual with the worst fitness are selected respectively, and the position of the frog individual with the worst fitness is updated based on the position update function to perform local search; When each cultural gene has undergone at least one round of local search and reached the maximum number of iterations, the population is recombined and sorted again according to fitness in ascending order. The best frog individual in the population is updated, and the position of the globally best frog individual is recorded.

5. The method for predicting the stability of surrounding rock in an underground powerhouse according to claim 1, characterized in that, The step of inputting the target dimension feature value and the surrounding rock sampling vector into a chaotic neural network to predict the surrounding rock deformation, in order to obtain the surrounding rock deformation prediction result, specifically includes: The target dimension feature values ​​and the surrounding rock sampling vector are normalized to obtain standard target dimension feature values ​​and standard surrounding rock sampling vectors; The standard target dimension feature value and the standard surrounding rock sampling vector are concatenated to obtain the surrounding rock feature vector, and the surrounding rock feature vector is input into the chaotic neural network. Based on the internal state function and output state function of the chaotic neural network, the deformation prediction results of the surrounding rock are obtained.

6. The method for predicting the stability of surrounding rock in an underground powerhouse according to claim 5, characterized in that, The expression for the chaotic neural network is: ； ； ； in, Indicates the first The surrounding rock feature vector is input into the chaotic neural network at any time; Indicates the first Time-bound target dimension feature value; Indicates the first The strain value of the surrounding rock at any given time; Indicates the first The compressive strength of the rock mass material is monitored at all times; express Real-time surrounding rock pressure monitored at all times; Indicates the first One neuron in The internal state function at time t; Represents the return mapping function; This indicates the total number of quantified neurons; Indicates the first The quantization neuron pairs with the first The contribution connection weights for the state update of each neuron; Indicates the number of dimensions of the surrounding rock sampling vector; Indicates the first The characteristics of the surrounding rock under the first dimension The contribution connection weights for the state update of each neuron; Indicates the first The input at time t in the chaotic neural network Rock characteristics in multiple dimensions; Indicates the first Bias parameters of each neuron; Indicates the disturbance intensity parameter; This represents the chaotic sequence generated by the chaotic mapping; Indicates the first One neuron in Output state function at time t; This represents a non-linear activation function.

7. A system for predicting the stability of surrounding rock in an underground powerhouse, characterized in that, The system is applied to the method according to any one of claims 1-6, the system comprising: A time-series monitoring data acquisition module is used to acquire time-series monitoring data of the surrounding rock of an underground powerhouse, wherein the time-series monitoring data includes acoustic emission signal data of the surrounding rock and sampling vectors of the surrounding rock; The target dimension feature value acquisition module is used to perform fractal dimension modeling on the time-series monitoring data to obtain the detection fractal dimension function, and to perform local search on the scaling factor, correction coefficient, and scale factor in the detection fractal dimension function according to the hybrid leapfrog algorithm to obtain the optimal scaling factor, optimal correction coefficient, and optimal scale factor; based on the optimal scaling factor, the optimal correction coefficient, the optimal scale factor, and the detection fractal dimension function, a target dimension feature extraction function is constructed, and the target dimension feature value in the time-series monitoring data is extracted according to the target dimension feature extraction function; The surrounding rock deformation prediction module is used to input the target dimension feature value and the surrounding rock sampling vector into a chaotic neural network to predict the surrounding rock deformation and obtain the surrounding rock deformation prediction result.

8. A device for predicting the stability of surrounding rock in an underground powerhouse, characterized in that, Including memory and processor; The memory is used to store computer program code and transmit the computer program code to the processor; The processor is configured to execute the method as described in any one of claims 1 to 6 according to instructions in the computer program code.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images