Vertical shaft material impact energy monitoring method and system based on machine learning

Abstract

The embodiment of the invention provides a vertical shaft material impact energy monitoring method and system based on machine learning. An impact characteristic signal set of a vertical shaft buffer structure contact interface is continuously collected. And performing dynamic feature extraction processing on the impact feature signal set to generate an energy dynamic feature sequence reflecting the impact energy change process. And calling a pre-trained impact energy analysis model to carry out energy state analysis processing on the energy dynamic characteristic sequence, and generating an impact energy dynamic change curve related to the initial impact energy characteristic, the buffer dissipation energy characteristic and the residual energy characteristic obtained through analysis. And calling a pre-trained well wall damage prediction model, inputting the impact energy dynamic change curve and preset well wall material characteristic parameters into the well wall damage prediction model, generating a damage risk index of the well wall structure, and executing graded early warning regulation and control operation according to the damage risk index. Therefore, the energy detection accuracy of vertical shaft material impact and the safety of vertical shaft construction are improved.

Description

Technical Field

[0001] The present application relates to the technical field of shaft construction, and in particular to a method and system for monitoring the impact energy of shaft materials based on machine learning. Background Art

[0002] During the construction and implementation of vertical shaft projects, material impact is a common and critical factor. As a key structural form of underground engineering, vertical shafts are widely used in numerous fields, including mining, tunneling, and water conservancy projects. During shaft construction, materials (such as ore, rock blocks, and concrete blocks) frequently fall from heights, violently impacting the shaft's buffer structures (such as the shaft lining and buffer devices). This material impact behavior is highly uncertain and complex. Parameters such as the height, velocity, shape, mass, and angle of the falling material vary, leading to significant variations in the energy and mode of action of each impact. Furthermore, the geological environment in which the shaft is located is complex and variable, and the mechanical properties of the shaft lining material vary depending on factors such as geological conditions and construction techniques, further complicating the impact process.

[0003] The importance of monitoring the impact of materials in shafts is self-evident. Accurate impact energy monitoring can grasp the effects of material impact on the shaft structure in real time, and provide key data support for evaluating the safety of the shaft structure. If the impact energy is too large and is not monitored and effectively controlled in a timely manner, it may cause damage, deformation, or even destruction of the shaft wall structure, leading to serious accidents such as shaft collapse, which will not only cause huge economic losses, but also seriously threaten the lives of construction workers. In addition, by monitoring and analyzing the impact energy, the design and construction parameters of the shaft buffer structure can be optimized, the overall impact resistance of the shaft can be improved, and the service life of the shaft can be extended. However, the existing technology has the problems of low accuracy in energy detection of shaft material impact and low safety in shaft construction.

[0004] Therefore, how to improve the energy detection accuracy of shaft material impact and the safety of shaft construction is an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a shaft material impact energy monitoring method and system based on machine learning, so as to improve the energy detection accuracy of shaft material impact and the safety of shaft construction.

[0006] In a first aspect, the present application provides a method for monitoring the impact energy of shaft materials based on machine learning, comprising: Continuously collecting a set of impact characteristic signals of the contact interface of the shaft buffer structure, wherein the set of impact characteristic signals includes pressure sensing signals and vibration sensing signals of different impact action surfaces; Performing dynamic feature extraction processing on the impact characteristic signal set to generate an energy dynamic feature sequence reflecting the impact energy change process, wherein the energy dynamic feature sequence includes an impact duration feature, a signal amplitude change feature, and a frequency distribution feature; Calling a pre-trained impact energy analysis model to perform energy state analysis on the energy dynamic feature sequence, and generating an impact energy dynamic change curve related to the initial impact energy feature, buffer dissipation energy feature, and residual energy feature obtained by analysis; Calling a pre-trained well wall damage prediction model, inputting the impact energy dynamic change curve and preset well wall material characteristic parameters into the well wall damage prediction model, and generating a damage risk index for the well wall structure; A graded warning and control operation is performed according to the damage risk index.

[0007] In the second aspect, the present application provides a vertical shaft material impact energy monitoring system based on machine learning, wherein the vertical shaft material impact energy monitoring system based on machine learning includes a machine-readable storage medium and a processor, wherein the machine-readable storage medium stores machine-executable instructions, and when the processor executes the machine-executable instructions, the vertical shaft material impact energy monitoring system based on machine learning implements the aforementioned vertical shaft material impact energy monitoring method based on machine learning.

[0008] The machine learning-based method and system for monitoring the impact energy of vertical shaft materials, provided herein, continuously collects pressure and vibration sensor signals from different impact surfaces of the contact interface of the vertical shaft buffer structure to construct a multi-source heterogeneous impact characteristic signal set. Dynamic feature extraction technology is then used to analyze and generate an energy dynamic feature sequence that includes impact duration, signal amplitude change, and frequency distribution characteristics. This comprehensively characterizes the complete evolution of impact energy from generation to attenuation. A pre-trained impact energy analysis model is used to deeply analyze the energy dynamic feature sequence, accurately extracting the initial impact energy, buffer dissipated energy, and residual energy characteristics, and generating a dynamic change curve, thereby achieving quantitative characterization and dynamic tracking of the impact energy state. Furthermore, a pre-trained wellbore damage prediction model is used to generate a damage risk index based on the characteristic parameters of the wellbore material, establishing a closed-loop prediction system from energy monitoring to structural damage assessment. Finally, a graded early warning and control mechanism is used to achieve a graded risk response. By integrating multi-dimensional data fusion with intelligent models, this system overcomes the limitations of single-parameter analysis in traditional monitoring methods and significantly improves the accuracy, real-time performance, and reliability of vertical shaft material impact energy monitoring, thereby effectively reducing the risk of damage to the wellbore structure caused by impact energy accumulation and ensuring the safe and stable operation of the vertical shaft project. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0010] Figure 1 A flow chart of a method for monitoring the impact energy of shaft materials based on machine learning provided in an embodiment of the present application; Figure 2 A structural schematic diagram of a shaft material impact energy monitoring system based on machine learning provided in an embodiment of the present application.

[0011] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0012] To help those skilled in the art better understand the present invention, the following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. It is apparent that the described embodiments are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by those skilled in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0013] The terms "first," "second," and so on, in the specification and claims of the present invention and the accompanying drawings are used to distinguish between different items, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or end comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed therein, or may optionally include other steps or elements inherent to such process, method, product, or end.

[0014] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0015] Figure 1The present invention provides a flow chart of a method for monitoring the impact energy of shaft materials based on machine learning. It should be understood that in other embodiments, the order of some steps in the method for monitoring the impact energy of shaft materials based on machine learning in this embodiment can be shared with each other according to actual needs, or some steps can be omitted or maintained. Figure 1 As shown, the method may include the following steps: Step S110: continuously collecting a set of impact characteristic signals of the contact interface of the shaft buffer structure, wherein the set of impact characteristic signals includes pressure sensing signals and vibration sensing signals of different impact action surfaces.

[0016] In this embodiment, this step is the basis of the entire monitoring method, which aims to continuously obtain various signals generated during the material impact process by deploying sensing equipment at the contact interface of the shaft buffer structure, and provide raw data for subsequent analysis and processing.

[0017] The vertical shaft buffer structure involved in this embodiment is provided with functionally complementary graded buffer areas along the axial direction, specifically including an upper elastic deformation area and a lower plastic energy dissipation area. A continuous and smooth guide surface is provided at the junction of the two areas. The guide surface is covered with a wear-resistant coating. Its function is to guide the impact material flow and impact energy to diffuse smoothly and evenly from the elastic deformation area to the active surface of the entire plastic energy dissipation area, effectively avoiding energy concentration impacting the local area of ​​the lower section, significantly improving the overall buffering efficiency and reducing the risk of stress concentration.

[0018] The upper elastic deformation zone is composed of a metal frame tightly fixed to the well wall with pre-embedded anchors, and its interior is filled with high-elastic damping material. Its main function is to absorb the initial impact energy through recoverable elastic deformation and to initially disperse and guide the remaining impact force.

[0019] The lower plastic energy dissipation zone is rigidly fixed to the bottom of the elastic zone frame by high-strength bolts or connectors. It is composed of high-strength composite material plates designed to withstand the main impact energy. Its core function is to maximize the dissipation of impact kinetic energy through irreversible processes such as plastic deformation of the material, such as controlled crushing or friction energy dissipation, to protect the main structure of the well wall.

[0020] To comprehensively collect impact signature signals, a pressure sensor array can be deployed in a circular pattern at the interface between the elastic deformation zone and the plastic energy dissipation zone. The number of sensors is dynamically configured based on the wellbore diameter, with the spacing between sensors within each ring not exceeding a specific distance. The spacing between nodes is calibrated using a laser rangefinder. The sensors are waterproofly packaged and connected to the data acquisition box via shielded cables. Sealant is applied to the cable joints to prevent water seepage. Furthermore, vibration sensor nodes are evenly distributed across the contact interface of the buffer layer, using a distributed topology to cover the entire impact surface. These sensors are connected to the data acquisition terminal via waterproof cables.

[0021] Through these pressure sensors and vibration sensors, the pressure sensing signals and vibration sensing signals generated by different impact surfaces when the material impacts can be collected respectively. These signals are continuously collected and together constitute a set of impact characteristic signals.

[0022] Step S111: generating an original sensing signal stream with a timestamp based on the pressure sensing signal and vibration sensing signal collected from the contact interface of the shaft buffer structure, wherein the sampling frequency of the original sensing signal stream matches the characteristic frequency of the impact event.

[0023] After collecting pressure and vibration sensor signals, these signals need to be initially processed to generate the original sensor signal stream. The data acquisition terminal is equipped with a multi-channel synchronous acquisition module and a multi-channel signal conditioning module. After the sensors collect the signals, the data acquisition terminal will process them.

[0024] First, a timestamp is added to each collected pressure and vibration sensor signal. The timestamp accurately records the moment the signal is generated, which is crucial for subsequent analysis of the temporal correlation between signals and the timing characteristics of the impact event. Second, a dynamic sampling frequency is set to match the characteristic frequency of the impact event. Only when the sampling frequency matches the characteristic frequency of the impact event can the signal characteristics of the impact event be fully captured, avoiding signal distortion or information loss caused by inappropriate sampling frequency.

[0025] The data acquisition terminal converts the analog sensor output signals into digital signals in real time through a multi-channel synchronous acquisition module and a signal conditioning module. During the conversion process, the correspondence between the timestamp and the signal is maintained, thereby generating a raw sensor signal stream with a timestamp. This raw sensor signal stream is then transmitted to the central processing unit via optical fiber or shielded transmission lines. The anti-interference properties of optical fiber and shielded transmission lines ensure that the signal is not disturbed during transmission, thus ensuring the accuracy of the raw data.

[0026] Step S112: Adopting an adaptive filtering algorithm to perform noise elimination processing on the original sensor signal stream to obtain an effective signal component containing impact energy characteristics.

[0027] During the acquisition and transmission process, the original sensor signal stream will inevitably be interfered with by various noises, such as the electronic noise of the sensor itself, the vibration noise of the surrounding environment of the shaft, electromagnetic interference, etc. These noises will mask the impact energy characteristics and affect the accuracy of subsequent analysis. Therefore, noise elimination processing must be performed.

[0028] In this step, an adaptive filtering algorithm can be used to process the original sensor signal stream. The advantage of the adaptive filtering algorithm is that it can automatically adjust the filter parameters according to the statistical characteristics of the input signal, thereby effectively filtering out noise.

[0029] Specifically, the algorithm continuously analyzes the spectral characteristics and amplitude variations of the raw sensor signal stream, identifies noise components, and then suppresses these noise components by adjusting the filter coefficients. For example, for high-frequency noise in the raw sensor signal stream, the algorithm identifies its frequency range and specifically reduces the signal strength within that frequency range; for irregular impulse noise, the algorithm uses smoothing to reduce its impact.

[0030] After processing by the adaptive filtering algorithm, the noise components in the original sensor signal stream are largely eliminated, and the retained signal components can accurately reflect the energy changes during the material impact process. These components are the effective signal components that contain the impact energy characteristics.

[0031] Step S113: After the valid signal components pass the signal integrity check, the checked valid signal components are grouped based on the spatial position of the impact surface to generate the impact characteristic signal set containing multi-channel signals, and the number of channels of the multi-channel signals is consistent with the number of partitions of the impact surface.

[0032] Obtaining valid signal components containing impact energy signatures does not necessarily mean these signals can be used directly for subsequent analysis. Signal integrity verification is required to ensure signal reliability and integrity. This includes checking the valid signal components for missing data, abnormal transitions, or distortion.

[0033] For example, if a valid signal component experiences data interruptions between consecutive sampling moments, or if the signal amplitude suddenly changes significantly beyond the normal range and this change does not conform to the general pattern of an impact event, the signal will be deemed to have failed integrity verification. For signals that fail verification, the corresponding time period must be re-acquired and processed to ensure data integrity.

[0034] After all valid signal components have passed the signal integrity check, they are grouped according to the spatial location of the impact surface. Since the impact surface is divided into multiple different partitions, each partition corresponds to a specific spatial location. The valid signal components belonging to the same partition are grouped together. This allows the signal of each group to correspond to a specific impact surface area, facilitating subsequent analysis of the impact characteristics of different areas. The impact characteristic signal set generated after grouping contains multi-channel signals, with each channel corresponding to a partition of the impact surface. Therefore, the number of channels in the multi-channel signal is consistent with the number of partitions of the impact surface, ensuring that the impact characteristic information of each partition can be clearly distinguished and processed.

[0035] Step S120: performing dynamic feature extraction processing on the impact feature signal set to generate an energy dynamic feature sequence reflecting the impact energy change process, wherein the energy dynamic feature sequence includes impact duration feature, signal amplitude change feature and frequency distribution feature.

[0036] After obtaining the impact characteristic signal set, it needs to be subjected to dynamic feature extraction processing. The purpose of this step is to extract key features that can reflect the impact energy change process from the complex signal, and then generate an energy dynamic feature sequence to provide input data for the subsequent impact energy analysis model.

[0037] The impact characteristic signal set contains the effective components of the processed pressure sensor signals and vibration sensor signals of different impact surfaces. These signals contain rich information about the impact energy. By performing time domain and frequency domain analysis on these signals, the impact duration characteristics, signal amplitude change characteristics and frequency distribution characteristics can be extracted. These characteristics reflect the changes in impact energy from different angles. Combining them can form an energy dynamic characteristic sequence that comprehensively reflects the impact energy change process.

[0038] Step S121: performing time domain waveform analysis on the pressure sensing signal in the shock characteristic signal set, extracting the signal rising edge slope and peak interval parameters of the pressure sensing signal, and generating a shock duration feature, which is used to describe the time span of a single shock event.

[0039] Time domain waveform analysis is performed on the pressure sensor signal in the impact characteristic signal set. Time domain waveform analysis mainly studies the waveform shape of the pressure sensor signal that changes over time. Through this analysis, characteristic parameters related to the impact duration can be obtained.

[0040] During the analysis process, the pressure sensor signal's rising edge slope is first extracted. The rising edge of the signal refers to the stage where the pressure signal rises from its initial value to its peak value. The rising edge slope is the amount of signal change per unit time during this stage. It is calculated by taking the ratio of the signal difference between two adjacent sampling moments during the rising edge stage to the corresponding time difference, and then averaging these ratios to obtain the rising edge slope. The peak interval parameter is then extracted. The peak interval parameter refers to the time interval between two adjacent pressure signal peaks. This peak interval parameter is obtained by identifying each peak point in the pressure signal waveform and calculating the time difference between the two adjacent peak points.

[0041] Based on the extracted signal rising edge slope and peak interval parameters, the time span from the beginning to the complete end of a single impact event can be comprehensively determined, thereby generating an impact duration feature. This feature can clearly describe the duration of a single impact event and is of great significance for analyzing the duration of impact energy.

[0042] Step S122: performing amplitude spectrum analysis on the vibration sensor signal in the impact characteristic signal set, calculating the ratio of the signal peak amplitude to the root mean square value of the vibration sensor signal, and generating a signal amplitude variation feature, which is used to characterize the intensity fluctuation of the impact energy.

[0043] Amplitude spectrum analysis is performed on the vibration sensor signals in the impact characteristic signal set. Amplitude spectrum analysis can reflect the distribution of vibration sensor signals at different amplitudes. Through this analysis, characteristic parameters related to the fluctuation of impact energy intensity can be obtained.

[0044] During the analysis process, it's necessary to calculate the ratio of the vibration sensor signal's peak amplitude to its RMS value. Peak amplitude refers to the maximum amplitude value in the vibration signal's waveform and reflects the signal's maximum intensity. The RMS value is calculated by squaring the value at each sampling point in the vibration signal, averaging these squared values, and then taking the square root of this average. The resulting RMS value reflects the average energy level of the vibration signal.

[0045] The ratio of the signal's peak amplitude to its RMS value is calculated. This ratio reflects the relative change in the vibration signal's amplitude. A large ratio indicates more dramatic amplitude fluctuations, and correspondingly, greater fluctuations in the impact energy intensity. A small ratio indicates more gradual amplitude fluctuations, and correspondingly, less fluctuations in the impact energy intensity. This ratio is used to generate a signal amplitude variation signature, which effectively characterizes fluctuations in impact energy intensity.

[0046] Step S123: performing frequency domain decomposition on the vibration sensing signal by fast Fourier transform, extracting the power spectrum density distribution of the main frequency components, and generating frequency distribution features, which are used to reflect the distribution of the impact energy in different frequency bands.

[0047] To determine the distribution of impact energy across different frequency bands, the vibration sensor signal must be decomposed in the frequency domain using a fast Fourier transform (FFT). FFT is a mathematical tool that converts signals from the time domain to the frequency domain. This transform decomposes the vibration sensor signal into sine and cosine components of different frequencies, thereby revealing the signal's distribution characteristics in the frequency domain.

[0048] After performing frequency domain decomposition on the vibration sensing signal, it is necessary to extract the power spectral density distribution of the main frequency components. The power spectral density distribution represents the power per unit frequency and reflects the amount of energy carried by different frequency components. By analyzing the distribution of the power spectral density in different frequency bands, the energy contribution of each frequency band can be determined, that is, the proportion of the energy in each frequency band to the total energy of the vibration signal. Based on these energy contributions, a frequency distribution feature is generated. This feature can clearly reflect the distribution of impact energy in different frequency bands. The distribution of impact energy in different frequency bands is closely related to factors such as the intensity of the impact, the properties of the material, and the response of the buffer structure. Therefore, this feature is of great value for in-depth analysis of impact energy characteristics.

[0049] Step S124: splicing the impact duration feature, the signal amplitude change feature, and the frequency distribution feature in chronological order to form the energy dynamic feature sequence with time sequence correlation.

[0050] After extracting the impact duration, signal amplitude variation, and frequency distribution features, these features need to be spliced ​​together in chronological order to form a dynamic energy feature sequence. Since these features are extracted at different time points or time periods during the impact process, they correspond to the characteristics of the impact energy at different moments. Therefore, splicing them together in chronological order can form a continuous, time-correlated sequence.

[0051] For example, the impact duration, signal amplitude variation, and frequency distribution features extracted at the initial stage of an impact event are arranged in chronological order and concatenated with corresponding features extracted during the development and final stages of the impact event to form a complete sequence. This resulting energy dynamic feature sequence comprehensively reflects the entire dynamic process of impact energy, from generation to change and disappearance, encompassing all variations in impact energy over time. This provides complete and valid input data for subsequent energy state analysis using a pre-trained impact energy analysis model.

[0052] Step S130: calling a pre-trained impact energy analysis model to perform energy state analysis on the energy dynamic feature sequence, and generating an impact energy dynamic change curve related to the initial impact energy feature, buffer dissipation energy feature, and residual energy feature obtained by analysis.

[0053] After obtaining the energy dynamic feature sequence, a pre-trained impact energy analysis model is used to analyze its energy state for in-depth analysis of the impact energy state changes. The impact energy analysis model is trained using a large amount of historical impact data and can accurately analyze various characteristics related to impact energy from the energy dynamic feature sequence.

[0054] The impact energy analysis model's analytical process primarily involves in-depth exploration and analysis of the energy dynamics sequence, identifying characteristic information reflecting the initial impact energy, energy dissipated by the buffer structure, and residual energy transmitted to the wellbore. This information then yields the initial impact energy signature, the buffer dissipated energy signature, and the residual energy signature. Based on these analytically derived energy signatures, an impact energy dynamics curve is generated. This curve intuitively displays the dynamic changes in impact energy throughout the entire process, including the magnitude of the initial impact energy, the energy dissipation process of the buffer structure, and the trend of residual energy changes. This provides an important basis for evaluating the effectiveness of the buffer structure and the safety status of the wellbore.

[0055] Step S131: input the energy dynamic feature sequence into the input layer of the impact energy analysis model, and convert the energy dynamic feature sequence into a high-dimensional feature vector through the embedding layer of the impact energy analysis model, wherein the dimension of the high-dimensional feature vector matches the number of hidden layer neurons of the impact energy analysis model.

[0056] First, the energy dynamic feature sequence is input into the input layer of the impact energy analysis model. The function of the input layer is to receive external input data and pass it to the next layer of the model for processing.

[0057] After the energy dynamic feature sequence enters the input layer as input data, it undergoes preliminary organization and transmission to ensure it can be correctly received by the embedding layer. Next, the energy dynamic feature sequence enters the embedding layer of the impact energy analysis model. The embedding layer's main function is to transform the input feature sequence, mapping it to a high-dimensional space to generate a high-dimensional feature vector. This transformation aims to convert the low-dimensional, potentially redundant feature sequence into a high-dimensional, more representative feature vector, thereby better capturing the complex relationships between features.

[0058] The dimension of the high-dimensional feature vector needs to match the number of hidden layer neurons in the impact energy analysis model. This is because the neurons in the hidden layer need to process the high-dimensional feature vector. Dimension matching can ensure that the high-dimensional feature vector can be smoothly input into the hidden layer and enable the hidden layer to fully utilize the information in the high-dimensional feature vector for subsequent calculations and analysis.

[0059] Step S132: performing temporal dependency modeling on the high-dimensional feature vector through the bidirectional long short-term memory network layer of the impact energy analysis model, extracting feature correlation relationships under different time steps, and generating a temporal correlation feature matrix.

[0060] After the high-dimensional feature vector is generated, it is input into the bidirectional long short-term memory (LSTM) network layer of the impact energy analysis model. The LSTM layer is a neural network structure capable of processing sequential data. Its ability to remember past information and predict future information makes it ideal for processing data with time-series characteristics. This network layer models the temporal dependencies of the high-dimensional feature vectors, analyzing the dependencies between the features of the high-dimensional feature vectors at different time steps.

[0061] During the modeling process, the bidirectional long short-term memory network layer processes the high-dimensional feature vector simultaneously from two directions: one from the past to the future, and the other from the future to the past. This allows for a more comprehensive capture of the correlations between features at different time steps. This process extracts correlations between features at different time steps within the high-dimensional feature vector. These correlations include causal relationships and covariation between features. Based on these correlations, a temporal correlation feature matrix is ​​generated. Each element in this matrix represents the strength of the correlation between two features at different time steps, comprehensively reflecting the interactions and dependencies between features across the temporal dimension.

[0062] Step S133: Utilize the attention mechanism module of the impact energy analysis model to perform dynamic weighting processing on the temporal correlation feature matrix, strengthen the feature contribution of key time nodes, and generate an attention weighted feature vector.

[0063] After the temporal correlation feature matrix is ​​generated, it needs to be dynamically weighted using the attention mechanism module of the impact energy analysis model. The core idea of ​​the attention mechanism module is to give higher weight to important information and lower weight to less important information when processing data, thereby highlighting the role of key information.

[0064] In this step, the attention mechanism module analyzes the temporal correlation feature matrix and identifies features of key time nodes that are more important for analyzing the energy state. These key time nodes are typically moments when the impact energy changes significantly, such as the onset of the impact, peak energy, or sharp energy decay. The features at these moments are crucial for accurately analyzing the energy state. The attention mechanism module assigns higher weights to features of these key time nodes and lower weights to features of non-critical time nodes. Through this dynamic weighting process, the contribution of features from key time nodes is strengthened.

[0065] After weighted processing, an attention-weighted feature vector is generated, which can highlight the role of key features in energy state analysis and improve the accuracy and efficiency of subsequent processing.

[0066] Step S134: performing nonlinear mapping processing on the attention weighted feature vector through the fully connected layer of the impact energy analysis model, and outputting the initial impact energy feature, the buffer dissipation energy feature, and the residual energy feature.

[0067] After the attention-weighted feature vector is generated, it is input into the fully connected layer of the impact energy analysis model. A fully connected layer is a network layer composed of multiple neurons, each of which is connected to all neurons in the previous layer. This layer can perform complex nonlinear mapping on the input features.

[0068] In the fully connected layer, the attention-weighted feature vector undergoes a series of operations and transformations through the interaction of weight parameters and activation functions between neurons, mapping it to different feature spaces. This nonlinear mapping process can extract more abstract and essential feature information from the attention-weighted feature vector, ultimately outputting the initial impact energy characteristics, buffer dissipation energy characteristics, and residual energy characteristics. The initial impact energy characteristics reflect the energy size and characteristics of the material at the initial moment of impact and are the starting point of the impact energy. The buffer dissipation energy characteristics reflect the energy dissipated by the buffer structure during the impact process through elastic and plastic deformation, including characteristics such as the energy size and rate of dissipation. The residual energy characteristics reflect the energy ultimately transmitted to the wellbore wall after dissipation by the buffer structure.

[0069] Step S135: generating the impact energy dynamic change curve based on the peak amplitude of the initial impact energy characteristic, the attenuation coefficient of the buffer dissipation energy characteristic, and the frequency spectrum distribution parameter of the residual energy characteristic.

[0070] In this embodiment, the generation of the dynamic change curve of the impact energy needs to be based on the principle of conservation of energy, integrating the key parameters of the initial impact energy, the buffer dissipation process, and the residual energy distribution. The peak amplitude of the initial impact energy characteristic represents the maximum energy intensity when the impact occurs, and is the basis for the starting point of the curve; the attenuation coefficient of the buffer dissipation energy characteristic reflects the efficiency of energy dissipation in the buffer structure over time, and determines the overall attenuation trend of the curve; the spectrum distribution parameter of the residual energy characteristic reflects the distribution of the remaining energy in different frequency bands after buffering, and provides support for the superposition of multi-frequency fluctuation characteristics of the curve. The three are interrelated in the time dimension, and through coordinated integration, a dynamic change curve that can fully reflect the entire process of impact energy from generation to dissipation to residual is formed. This curve will serve as the core quantitative basis for subsequent wellbore damage assessment.

[0071] Step S1351: extracting the peak amplitude of the initial impact energy characteristic to determine the initial value of the energy intensity at the initial stage of the impact energy action.

[0072] The initial impact energy signature is a sequence of energy records at multiple time points, documenting the instantaneous change in energy after the impact begins. To extract the peak amplitude, this sequence must be thoroughly sorted, with the energy values ​​at each time point compared one by one. The node with the largest energy value is selected, and the energy value corresponding to this node is the peak amplitude of the initial impact energy signature. Because this peak occurs at the beginning of the impact and represents the maximum energy at the moment of impact, it can be directly determined as the initial value of the energy intensity at the beginning of the impact energy phase. This initial value not only serves as the starting point for the dynamic change curve of the impact energy but also forms an important foundation for subsequent analysis of the total kinetic energy of the impact.

[0073] Step S1352: Obtain the attenuation coefficient of the buffer dissipation energy characteristic, analyze the attenuation law of the attenuation coefficient over time, and generate energy attenuation trend data.

[0074] The energy dissipation characteristics of a buffer consist of attenuation coefficients at different time points. These coefficients, calculated based on the momentum theorem, reflect the efficiency of the buffer structure in dissipating energy at different moments. Once the attenuation coefficients are obtained, they are subjected to time series analysis to observe how their values ​​change over time. For example, in the initial impact, due to the rapid response of the elastic deformation zone, the attenuation coefficient may show an upward trend, indicating an accelerated rate of energy dissipation. As the impact continues, the plastic energy dissipation zone gradually assumes the primary role in energy dissipation, and the attenuation coefficient may stabilize, reflecting the persistence of energy dissipated by plastic deformation. By summarizing these changing patterns, the decay patterns of the attenuation coefficients over time can be summarized, thereby generating energy attenuation trend data. This data, presented as a multidimensional sequence containing attenuation rate characteristics at different time periods, provides a dynamic basis for generating the impact energy attenuation master curve, ensuring that the master curve accurately reflects the actual energy dissipation process of the buffer structure.

[0075] Step S1353: Analyze the spectrum distribution parameters of the residual energy feature, identify the energy proportions corresponding to different frequency intervals, and generate a spectrum energy distribution feature.

[0076] The spectrum distribution parameter of the residual energy feature is the result obtained after performing frequency domain analysis on the residual energy, and includes the distribution information of the residual energy in different frequency bands.

[0077] To analyze the spectral distribution parameters, the frequency range must be divided into several continuous intervals. The energy percentage of each interval is then calculated. Specifically, the power spectral density area within each frequency interval can be integrated and compared with the total power spectral density area across the entire frequency range to determine the energy percentage for that interval.

[0078] For example, the energy percentage in the low-frequency range may be related to the primary energy component of the material impact, while the energy percentage in the high-frequency range may reflect the vibration characteristics of the buffer structure. Arranging the energy percentages of these different frequency ranges in frequency order generates a spectral energy distribution feature. This feature clearly demonstrates the distribution of residual energy in the frequency domain, providing a quantitative basis for the subsequent superposition of frequency fluctuation components, ensuring that the fluctuation superposition accurately reflects the frequency characteristics of the residual energy.

[0079] Step S1354: Using the initial value of the energy intensity as the starting point of the curve and combining it with the energy attenuation trend data, a main impact energy attenuation curve is generated.

[0080] The energy intensity initial value is the energy value of the curve at time zero, and this is used as the starting point to construct the impact energy attenuation master curve. During the construction process, it is necessary to combine the attenuation patterns of different time periods in the energy attenuation trend data to calculate the energy value at each time node.

[0081] For example, in the first time interval, the energy decay from the initial value is calculated based on the rate of change of the attenuation coefficient in that interval, resulting in the energy value at the end of that interval. In subsequent intervals, the energy values ​​at each node are continuously calculated based on the previous energy values ​​and the attenuation trend of the corresponding interval. These time nodes and their corresponding energy values ​​are marked in the coordinate system and connected with a smooth curve to form the main impact energy decay curve.

[0082] The main curve reflects the overall attenuation trend of impact energy under the action of the buffer structure. Its shape is jointly affected by the elastic deformation zone and the plastic energy dissipation zone. The elastic deformation stage may show a faster energy drop, while the plastic energy dissipation stage may show a gentle attenuation.

[0083] Step S1355: Based on the frequency spectrum energy distribution characteristics, the frequency energy fluctuation component superposition processing is performed on the impact energy attenuation main curve to generate the impact energy dynamic change curve containing multi-frequency energy fluctuation characteristics.

[0084] The spectral energy distribution characteristics provide a basis for superimposing the frequency fluctuations of the shock energy attenuation master curve. The fluctuation amplitude weight corresponding to each frequency interval is determined based on the energy contribution of each frequency interval. The higher the energy contribution of the interval, the greater the impact of its fluctuation components when superimposed. Subsequently, for each frequency interval, a fluctuation component is generated that matches the frequency characteristics of that interval. The frequency of these components aligns with the center frequency of the corresponding interval, and the amplitude is determined by the energy contribution of the interval and the energy scale of the master curve.

[0085] During the superposition process, the energy values ​​of each frequency fluctuation component and the master curve at the same time point are synthesized. This ensures that the master curve not only retains the overall attenuation trend but also incorporates the energy fluctuation characteristics of multiple frequencies. This process generates a dynamic impact energy curve that more realistically reflects the complex fluctuations of impact energy during transmission. Taking into account the non-stationary nature of material impact and the vibration response of the buffer structure, it provides more comprehensive energy data for accurately assessing the impact of residual energy on the wellbore.

[0086] Step S13551: extracting the energy fluctuation amplitude and fluctuation period corresponding to different frequency intervals in the spectrum energy distribution characteristics.

[0087] When extracting the energy fluctuation amplitude and fluctuation period for each frequency interval from the spectral energy distribution characteristics, the energy fluctuation amplitude must be determined in conjunction with the energy contribution of that interval and the energy range of the impact energy attenuation master curve. Generally, the higher the energy contribution, the greater the fluctuation amplitude. This can be calculated by correlating the energy contribution with the energy range of the master curve. Dynamic adjustments must also be made based on the attenuation trend of the master curve to ensure that the fluctuation amplitude at different time points matches the energy level of the master curve. The fluctuation period is directly related to the frequency interval. For each frequency interval, the period corresponding to its center frequency is taken as the fluctuation period of that interval to ensure that the period matches the frequency characteristics.

[0088] The energy fluctuation amplitude and fluctuation period extracted in this way can accurately reflect the energy fluctuation characteristics of each frequency interval, laying the foundation for generating frequency energy fluctuation components, so that the fluctuation components can truly reflect the frequency characteristics of the residual energy.

[0089] Step S13552: Generate frequency energy fluctuation components corresponding to each frequency interval based on the energy fluctuation amplitude and the fluctuation period.

[0090] For each frequency interval, the corresponding frequency energy fluctuation component is generated using the extracted energy fluctuation amplitude and fluctuation period as parameters. The fluctuation component is generated in the form of a periodic function, whose amplitude is the energy fluctuation amplitude in the interval and the period is the corresponding fluctuation period. The function form can be selected as sine or cosine to reflect the fluctuation characteristics of the energy at that frequency. Each generated frequency energy fluctuation component is a function of time and can reflect the instantaneous fluctuation of energy at different frequencies. For example, the fluctuation component in the low-frequency interval may show slower periodic changes, while the fluctuation component in the high-frequency interval may show faster periodic changes. These fluctuation components provide specific fluctuation data for subsequent superposition processing, ensuring that the superimposed curve can reflect the fluctuation characteristics of energy at different frequencies.

[0091] Step S13553: Analyze the energy proportion of each frequency interval and assign a corresponding fluctuation weight coefficient to the frequency energy fluctuation component of the corresponding frequency interval.

[0092] The energy contribution of each frequency interval in the spectral energy distribution characteristics is analyzed. The higher the energy contribution, the greater the potential impact of the residual energy in that interval on the wellbore. Therefore, it should be given a higher weight when superimposing the fluctuation component. The allocation of fluctuation weight coefficients is based on the energy contribution. Through normalization, the sum of the weight coefficients of each interval is ensured to be 1, so that the weights can accurately reflect the relative importance of the energy in each interval. This weight allocation method makes the fluctuation components of frequency intervals with high energy contributions more significant after superposition. It can highlight the energy fluctuations of the main frequency components, make the generated impact energy dynamic change curve more consistent with the actual energy transfer characteristics, and avoid excessive interference of secondary frequency components on the curve shape.

[0093] Step S13554: weighting the frequency energy fluctuation component of each frequency interval with the corresponding fluctuation weight coefficient to generate a weighted frequency energy fluctuation component.

[0094] The weighted frequency energy fluctuation component is obtained by multiplying the frequency energy fluctuation component of each frequency interval by the fluctuation weight coefficient corresponding to that interval. This process actually adjusts the fluctuation component according to the energy importance of each interval. The fluctuation component with a high weight coefficient is amplified, while the fluctuation component with a low weight coefficient is reduced, so that the influence of each fluctuation component matches its energy proportion.

[0095] Through weighted processing, we ensure that the superimposed curve can accurately reflect the actual contribution of different frequency energies, highlight the influence of the main frequency components, and suppress the interference of the secondary frequency components, so that the curve can better reflect the key fluctuation characteristics of the residual energy.

[0096] Step S13555: Synchronously superimpose the weighted frequency energy fluctuation components of all the frequency intervals onto the impact energy attenuation main curve to generate the impact energy dynamic change curve containing multi-frequency energy fluctuation characteristics.

[0097] On the time axis, the weighted frequency energy fluctuation components of each frequency interval are superimposed with the energy value of the impact energy attenuation master curve at the same time point. This superimposed energy value is the sum of the energy value of the master curve at a given time point and all weighted fluctuation components at that time point. The superimposed energy values ​​at all time points are then connected to form a dynamic impact energy change curve that incorporates multi-frequency energy fluctuation characteristics. This curve retains the overall attenuation trend of the master curve while reflecting the fluctuation characteristics of energy at different frequencies through superimposed fluctuation components. This curve provides a more comprehensive description of the dynamic changes in impact energy. Taking into account the complexity of material impact and the dynamic response of the buffer structure, it provides more accurate energy input data for subsequent wellbore damage prediction.

[0098] Step S140: calling a pre-trained well wall damage prediction model, inputting the impact energy dynamic change curve and preset well wall material characteristic parameters into the well wall damage prediction model, and generating a well wall structure damage risk index.

[0099] In this embodiment, the wellbore damage prediction model is constructed through machine learning algorithm training based on historical impact event data and corresponding wellbore damage conditions.

[0100] The core function of this model is to establish the relationship between the dynamic changes of impact energy and the properties of the wellbore material, and then quantify the damage risk of the wellbore structure.

[0101] When the model is invoked, the energy time series characteristics contained in the impact energy dynamic change curve and the preset wellbore material characteristic parameters are required as input. The impact energy dynamic change curve reflects the dynamic characteristics of the impact energy, such as intensity, attenuation rate, and frequency distribution. The wellbore material characteristic parameters include physical and mechanical properties such as elastic modulus, Poisson's ratio, compressive strength, and tensile strength.

[0102] By processing these input data, the model analyzes the interaction between impact energy and wellbore material, and ultimately outputs a damage risk index that can quantify the risk of damage to the wellbore structure. This index provides a decision-making basis for subsequent graded early warning and control.

[0103] Step S141: Obtain the peak energy characteristic, attenuation rate characteristic and spectrum distribution characteristic of the impact energy dynamic change curve by analyzing the impact energy dynamic change curve, wherein the peak energy characteristic corresponds to the peak amplitude of the initial impact energy characteristic, the attenuation rate characteristic corresponds to the attenuation coefficient of the buffer dissipation energy characteristic, and the spectrum distribution characteristic corresponds to the spectrum distribution density of the residual energy characteristic.

[0104] When analyzing the dynamic change curve of impact energy, the first step is to extract the peak energy feature. This feature is the maximum energy value in the curve, corresponding to the peak amplitude of the initial impact energy feature, reflecting the maximum intensity of the impact energy. Secondly, the attenuation rate feature is extracted. This feature is obtained by analyzing the rate of change of the energy value in the curve over time. This feature corresponds to the attenuation coefficient of the buffer dissipation energy feature and reflects the speed of energy decay. Finally, the spectrum distribution feature is extracted by performing frequency domain analysis on the dynamic change curve of impact energy. This feature corresponds to the spectrum distribution density of the residual energy feature and reflects the distribution of energy in different frequency bands. These three features describe the dynamic characteristics of impact energy from different perspectives, providing a foundation for subsequent fusion processing with the characteristic parameters of the wellbore material, ensuring that the model can fully capture the key characteristics of impact energy.

[0105] Step S142: obtaining preset characteristic parameters of the well wall material, wherein the characteristic parameters of the well wall material include elastic modulus parameters, Poisson's ratio parameters, compressive strength parameters, and tensile strength parameters.

[0106] The characteristic parameters of the wellbore material are pre-determined through material testing and inspection, and include several key parameters reflecting the performance of the wellbore material. To obtain these parameters, the corresponding parameter values ​​for the current wellbore structure are retrieved from a pre-set database. These parameters include: the elastic modulus, which reflects the material's ability to resist elastic deformation; the Poisson's ratio, which reflects the ratio of lateral to longitudinal strain under load; the compressive strength, which reflects the maximum compressive stress the material can withstand; and the tensile strength, which reflects the maximum tensile stress the material can withstand. Together, these parameters constitute the wellbore material characteristic parameters, providing a material performance foundation for analyzing the damage risk of the wellbore structure under impact energy. Their accuracy directly impacts the model's prediction accuracy of wellbore damage risk.

[0107] Step S143: performing feature fusion processing on the peak energy feature, the attenuation rate feature, the spectrum distribution feature and the wellbore material characteristic parameters to generate a multi-dimensional input feature vector.

[0108] During feature fusion processing, the peak energy features, attenuation rate features, spectrum distribution features and wellbore material characteristic parameters are first standardized, and the parameters of different dimensions are converted into dimensionless standardized features to ensure that they can be fused in the same dimension.

[0109] Normalization involves converting each eigenvalue to its relative position in the feature sequence, eliminating the effects of dimensional differences. All normalized features are then concatenated in a specific order to form a multidimensional vector, known as the multidimensional input feature vector. This vector incorporates comprehensive information about the impact energy signature and the wellbore material properties, serving as input data for the wellbore damage prediction model, ensuring that the model simultaneously considers the impact of both energy and material factors on wellbore damage.

[0110] Step S144: inputting the multidimensional input feature vector into the feature interaction layer of the wellbore damage prediction model, performing nonlinear correlation modeling on the multidimensional input feature vector through a feature cross network, and generating an interaction feature matrix, wherein the interaction feature matrix represents the interaction characteristics between wellbore material and energy.

[0111] The multidimensional input feature vector is fed into the feature interaction layer of the wellbore damage prediction model. The feature cross-network in this layer, composed of multiple neural network layers, is capable of performing nonlinear transformations and cross-combinations on the input features. For example, the peak energy feature is cross-operated with the elastic modulus parameter to obtain a feature reflecting the interaction between the maximum impact energy intensity and the material's elasticity; the decay rate feature is cross-operated with the compressive strength parameter to obtain a feature reflecting the interaction between the energy decay rate and the material's compressive resistance. In this way, all input features are comprehensively cross-processed to generate an interaction feature matrix. Each element in this matrix represents the interaction strength between two or more features, clearly characterizing the complex interaction characteristics between the wellbore material and energy, providing rich interaction information for subsequent attention mechanism processing.

[0112] Step S145: performing spatial attention weight allocation on the interaction feature matrix through the convolutional attention module of the well wall damage prediction model, strengthening the feature components corresponding to the weak areas of the well wall material, and generating a weighted interaction feature map.

[0113] The convolutional attention module consists of a convolutional layer and an attention mechanism layer. The convolutional layer first performs a convolution operation on the interaction feature matrix, extracting local feature information from the matrix and generating a feature map. The attention mechanism layer then analyzes this feature map and, combined with the characteristic parameters of the wellbore material, identifies characteristic components associated with weak areas in the wellbore material. These weak areas typically correspond to areas of low material strength or concentrated structural stress. These characteristic components are assigned higher attention weights, while characteristic components corresponding to non-weak areas are assigned lower weights. This spatial attention weighting enhances the role of characteristic components corresponding to weak areas in subsequent processing, generating a weighted interaction feature map. This feature map highlights the interaction between weak areas and impact energy, helping to improve the model's accuracy in predicting wellbore damage risk and enabling the model to focus more on damage risks in critical areas.

[0114] Step S146: performing long-short term memory modeling on the weighted interactive feature map to capture the cumulative damage effect of the dynamic change curve of the impact energy on the wellbore structure and generate a time series damage feature vector.

[0115] The weighted interaction feature map is input into the long-short-term memory network layer of the wellbore damage prediction model, which can process sequence data and capture the temporal dependencies therein. In this step, the long-short-term memory network layer analyzes the weighted interaction feature map, tracks the impact of the dynamic change curve of the impact energy on the wellbore structure at different time points, and identifies the cumulative damage process. For example, by analyzing the characteristics of multiple time points, the trend of gradual accumulation of damage in weak areas of the wellbore material with the continued action of the impact energy is captured. This cumulative effect may cause the material performance to gradually degrade and eventually cause significant damage. Through this modeling method, a time-series damage feature vector is generated. This vector contains the damage feature information of the wellbore structure at different time points, which can reflect the cumulative damage effect of the impact energy on the wellbore structure and provide time-series dimension information for the final damage risk index generation.

[0116] Step S147: performing probability conversion on the time series damage feature vector through the output layer of the wellbore damage prediction model to generate the damage risk index.

[0117] The time-series damage feature vector is input into the output layer of the wellbore damage prediction model, which consists of a fully connected layer and an activation function. The fully connected layer performs a linear transformation on the time-series damage feature vector, mapping it into a one-dimensional feature space. The activation function, using a function suitable for probabilistic output, converts the transformed feature value into a probability value within a specific range. This probability value is the damage risk index.

[0118] The damage risk index reflects the likelihood of damage to the shaft wall structure. A higher index indicates a greater risk of damage to the shaft wall structure under the current impact energy, while a lower index indicates a lower risk. This index provides a quantitative basis for subsequent tiered early warning and control measures, enabling relevant personnel to take appropriate measures based on the index.

[0119] Step S150: performing a graded warning control operation according to the damage risk index.

[0120] After obtaining the damage risk index, the corresponding graded warning and control operations are executed based on the index's magnitude. These graded warning and control operations are determined based on the comparison of the damage risk index with the preset warning threshold. Different warning levels correspond to different control measures. These measures can promptly address potential damage risks to the shaft wall structure and ensure safe operation of the shaft. This process combines monitoring, analysis, and control to form a closed-loop safety management system, ensuring timely response to risks posed by impact energy.

[0121] Step S151: Compare the injury risk index with a preset warning threshold range to determine the warning level of the current injury risk index, where the warning levels include a first warning level, a second warning level, and a third warning level.

[0122] In this embodiment, the preset warning threshold ranges are based on industry safety standards, combined with actual shaft operation conditions and historical damage data. These threshold ranges form continuous, non-overlapping intervals, each corresponding to a specific warning level. When performing a comparison, the specific values ​​of the damage risk index must be matched against these intervals one by one.

[0123] For example, if the damage risk index falls within the lowest threshold range, it is determined to be the first warning level; if it falls within the middle threshold range, it is determined to be the second warning level; if it falls within the highest threshold range, it is determined to be the third warning level. This grading method can clearly define the degree of damage risk faced by the wellbore structure and provide clear guidance for subsequent targeted control measures. In the process of determining the warning level, it is necessary to ensure that the division of the threshold range is reasonable and operational, accurately reflecting the differences in different risk levels and facilitating the system's automatic identification and determination.

[0124] Step S152: According to the warning level, executing the warning control operation corresponding to the warning level, the warning control operation includes at least one of outputting warning prompt information, triggering sound and light alarms, restricting construction equipment construction behavior, and starting the emergency support system.

[0125] According to the determined warning level, the corresponding warning control operations can be automatically executed to deal with different degrees of damage risks.

[0126] When it is at the first warning level, warning prompt information can be output. This information can be displayed on the display screen of the monitoring terminal and sent to the mobile devices of relevant managers to remind them to pay attention to changes in the status of the well wall structure and strengthen inspections of on-site conditions.

[0127] When in the second warning level, in addition to outputting warning prompt information, the sound and light alarm device can also be triggered. The on-site alarm emits continuous sound signals and flashing light signals to warn on-site construction personnel to pay attention to safety; at the same time, the construction behavior of construction equipment is restricted, such as reducing the operating speed of material lifting equipment and reducing the frequency of material delivery, thereby reducing the continuous impact on the buffer structure and the well wall.

[0128] When it reaches the third warning level, the emergency support system is immediately activated to quickly deploy the pre-deployed temporary support structure to reinforce the well wall; at the same time, the power supply of the construction equipment is cut off to prevent the equipment from continuing to operate and causing greater impact and damage.

[0129] The execution of these early warning and control operations is progressive. As the warning level increases, the intensity and urgency of the control measures also increase accordingly to ensure that risks can be effectively controlled and the safety of shaft construction can be guaranteed.

[0130] Step S153: adjusting at least one of the pre-tightening force of the composite plate included in the buffer structure and the angle of the guide surface of the buffer structure according to the damage risk index, so as to dynamically optimize the buffering efficiency of the buffer structure.

[0131] The buffering efficiency of the buffer structure directly affects the extent of the impact energy on the well wall. Therefore, adjusting the parameters of the buffer structure according to the damage risk index is the key to dynamically optimizing the buffering efficiency.

[0132] When the damage risk index is high, it indicates that the energy dissipation capacity of the current buffer structure is insufficient to effectively mitigate the impact energy. In this case, the preload of the composite plate can be increased. The preload of the composite plate is adjusted through a hydraulic device at the joint. Increasing the preload can tighten the bond between the plates, improve their ability to resist impact and deformation, and thus enhance the energy efficiency of the plastic energy dissipation zone.

[0133] At the same time, the guide surface angle of the buffer structure can also be adjusted. The adjustment of the guide surface angle is completed by the angle adjustment mechanism installed at its bottom. Appropriately increasing the inclination angle of the guide surface can more effectively guide the impact material flow and energy to a larger range of the plastic energy consumption zone, avoid energy concentration in local areas, and further enhance the overall buffering effect of the buffer structure.

[0134] When the damage risk index is low, the preload force of the composite panel can be appropriately reduced, or the guide surface angle can be adjusted to a flatter state to reduce unnecessary energy consumption and equipment loss.

[0135] Through this dynamic adjustment, the buffer structure can always maintain the best buffering performance according to the actual impact energy situation, forming an adaptive protection mechanism.

[0136] Step S160: extracting the damage sensitivity features of each region of the wellbore output by the middle layer of the wellbore damage prediction model.

[0137] When processing input data, the wellbore damage prediction model's intermediate layer outputs a large number of features reflecting the damage sensitivity of various wellbore regions. These features are known as the damage sensitivity features of each wellbore region. Extracting these features requires determining the outputs of specific neurons in the model's intermediate layer that are relevant to wellbore damage. These neuron outputs can reflect the sensitivity of different regions to impact energy and the susceptibility to damage. For example, neurons with high output values ​​may correspond to areas of the wellbore that are prone to damage from impact energy, while neurons with low output values ​​may correspond to areas with low damage sensitivity. During the extraction process, the neuron position and output value corresponding to each damage sensitivity feature are recorded to form a multidimensional set of damage sensitivity features. This set encompasses damage sensitivity information for each wellbore region and provides the data foundation for subsequent spatial mapping.

[0138] Step S161: Correlation mapping is performed between the damage sensitive features of each area of ​​the well wall and the spatial position information of the well wall structure, so as to establish a corresponding relationship between the damage features and the spatial coordinates.

[0139] The spatial position of the wellbore structure is determined by a pre-established three-dimensional coordinate system, with each region having unique spatial coordinates. When mapping the extracted damage-sensitive features of each wellbore region to these spatial coordinates, each damage-sensitive feature is assigned to a specific spatial coordinate based on the corresponding relationship between the neuronal output of the feature and the wellbore region.

[0140] For example, by using the correspondence between neurons and wellbore regions recorded during model training, the wellbore region corresponding to a damage-sensitive feature can be determined. The spatial coordinates of this region can then be obtained, establishing a one-to-one correspondence between the damage-sensitive feature and the spatial coordinates. This associative mapping gives the abstract damage-sensitive feature a specific spatial location attribute, laying the foundation for the subsequent construction of a mesh model of the wellbore structure and the generation of a 3D damage distribution map.

[0141] Step S162: Based on the correspondence between the damage characteristics and the spatial coordinates, a grid model of the shaft wall structure is constructed in a three-dimensional spatial coordinate system.

[0142] The three-dimensional spatial coordinate system is centered at the shaft's axis, with the Z axis as the axial direction and the X and Y axes as the radial directions. This coordinate system is divided into several grid cells based on the actual size and shape of the shaft wall structure. Each grid cell represents a tiny area of ​​the shaft wall, and its size is determined by the shaft wall's dimensions and accuracy requirements. Then, based on the correspondence between damage features and spatial coordinates, each grid cell is associated with a corresponding damage-sensitive feature. This means that each grid cell contains damage-sensitive information about its location.

[0143] In this way, a grid model of the wellbore structure is constructed. This model is a three-dimensional grid collection, in which each grid cell has clear spatial coordinates and corresponding damage sensitivity characteristics, which can accurately reflect the damage sensitivity distribution of each area of ​​the wellbore.

[0144] Step S163: color-code grid cells of different damage risk levels according to the distribution of the damage risk index in each grid cell in the grid model.

[0145] The distribution of the damage risk index in the grid model is determined by combining the damage sensitivity characteristics corresponding to each grid unit and the overall damage risk index. That is, each grid unit is divided into a corresponding damage risk level according to the size of its damage sensitivity characteristics and the level of the overall damage risk index.

[0146] Color coding uses a gradient color system. For example, grid cells with low damage risk levels are represented by green, those with medium risk levels by yellow, and those with high risk levels by red. During the coding process, a clear color range and corresponding RGB value must be set for each damage risk level to ensure color differentiation and intuitiveness. Through color coding, the damage risk level of each grid cell in the grid model can be visually presented, facilitating the rapid identification of high-risk areas.

[0147] Step S164: Rendering the color-coded grid model to generate a three-dimensional damage distribution map of the wellbore structure, and marking potential high-risk areas with higher damage risk indexes in the three-dimensional damage distribution map.

[0148] The color-coded mesh model is rendered using 3D rendering technology to enhance the model's three-dimensionality and visual quality, making the colors of each mesh cell more vivid and the boundaries clearer. During the rendering process, the model's lighting and transparency can also be adjusted to better illustrate the distribution of damage risks within the wellbore.

[0149] The 3D damage distribution map of the wellbore structure generated in this step is an intuitive 3D visualization model that comprehensively reflects the damage risk status of each area of ​​the wellbore. Within this map, potentially high-risk areas with a high damage risk index are marked with specific markers, such as a flashing red border or a prominent symbol. A text description can also be included alongside the map, indicating the area's spatial coordinates and risk level.

[0150] The generation of three-dimensional damage distribution maps enables managers to intuitively understand the damage risk distribution of the well wall, providing an important reference for formulating targeted maintenance and reinforcement plans.

[0151] Step S210: training the impact energy analysis model.

[0152] Training the impact energy analysis model is a prerequisite for ensuring that it can accurately analyze the energy state. The training process includes multiple steps such as data preparation, model construction, and parameter optimization.

[0153] Step S211: collecting a set of impact characteristic signals of historical impact events and corresponding initial impact energy characteristics, buffer dissipation energy characteristics, and residual energy characteristics as a training data set.

[0154] The impact characteristic signal set of historical impact events comes from the pressure sensor signals and vibration sensor signals recorded during the past shaft construction process. These signals undergo the same processing process as steps S110 to S113 to form a standardized impact characteristic signal set. The corresponding initial impact energy characteristics, buffer dissipation energy characteristics, and residual energy characteristics are obtained through energy analysis and calculation of historical events. The initial impact energy characteristics are calculated and determined based on parameters such as the mass and speed of the material. The buffer dissipation energy characteristics are obtained by analyzing the deformation and energy consumption data of the buffer structure. The residual energy characteristics are determined based on the energy monitoring data transmitted to the well wall. These data are divided into training sets and validation sets according to a certain ratio. The training set is used to learn the model parameters, and the validation set is used to evaluate the training effect of the model.

[0155] Step S212: Constructing a network structure of an impact energy analysis model, wherein the network structure includes an input layer, an embedding layer, a bidirectional long short-term memory network layer, an attention mechanism module, and a fully connected layer.

[0156] The number of neurons in the input layer matches the dimensionality of the energy dynamic feature sequence and is used to receive the input feature sequence. The embedding layer contains several neurons, whose function is to convert the input feature sequence into a high-dimensional feature vector. The number of neurons in the embedding layer is determined by the complexity of the feature sequence. The bidirectional long short-term memory network layer consists of multiple memory cells, each of which includes an input gate, a forget gate, and an output gate. This allows it to process the temporal dependencies of sequence data. The number of layers and the number of memory cells in each layer are set based on the length and complexity of the temporal data. The attention mechanism module is connected to the output of the bidirectional long short-term memory network layer to perform weighted processing on the temporal correlation feature matrix. The fully connected layer consists of multiple neurons, and the number of neurons in its output layer matches the total dimensionality of the initial impact energy feature, the buffer dissipation energy feature, and the residual energy feature. It outputs the resolved energy features. Each layer is connected by weight parameters to form a complete network structure.

[0157] Step S213: Convert the impact characteristic signal set in the training set into an energy dynamic characteristic sequence, input it into the constructed impact energy analysis model, and use the corresponding initial impact energy characteristics, buffer dissipation energy characteristics and residual energy characteristics as target values ​​to train the model using the back propagation algorithm.

[0158] First, the set of impact signature signals in the training set is converted into a sequence of dynamic energy signatures according to the method in steps S120 to S124. These dynamic energy signature sequences are then input into the impact energy analysis model, which processes them and outputs the predicted initial impact energy signature, buffer dissipation energy signature, and residual energy signature. The loss function between the predicted value and the target value is calculated using the mean square error function, which is the average of the sum of the squares of the differences between the predicted value and the target value.

[0159] Using a backpropagation algorithm, the weight parameters of each model layer are adjusted based on the gradient of the loss function. Through multiple iterative training, the loss function value is gradually reduced until the preset convergence condition is reached. During the training process, the model is evaluated using a validation set after a certain number of iterations. If the loss function value of the validation set no longer decreases, training is stopped to avoid overfitting of the model.

[0160] Step S214: The performance of the trained impact energy analysis model is evaluated using a validation set. When the prediction error of the model is within a preset range, the model training is determined to be complete.

[0161] Performance evaluation metrics include the mean, maximum, and standard deviation of the prediction error, which is the difference between the energy signature output by the model and the target value. When these metrics are within the preset range, the model accurately interprets the impact energy signature and training is complete.

[0162] If the prediction error exceeds the preset range, it is necessary to adjust the network structure parameters of the model, such as increasing the number of memory units in the bidirectional long short-term memory network layer, adjusting the number of neurons in the embedding layer, etc., and retrain until the model performance meets the requirements.

[0163] Step S310: training the wellbore damage prediction model.

[0164] The training of the wellbore damage prediction model aims to establish a mapping relationship between the dynamic changes of impact energy and the wellbore damage risk. The training process can include steps such as data preparation, model construction and parameter optimization.

[0165] Step S311: collecting the dynamic change curve of impact energy of historical impact events, the characteristic parameters of the wellbore material and the corresponding wellbore damage status data as a training data set, wherein the wellbore damage status data is quantified to obtain a damage risk index.

[0166] The dynamic change curve of impact energy during historical impact events is generated by processing and analyzing historical impact characteristic signals. Shaft wall material characteristic parameters are obtained from material inspection reports during shaft wall construction. Shaft wall damage data is obtained through inspection and assessment of the shaft wall after historical events. This data includes information such as the location, extent, and severity of the damage. This information is quantified to generate a damage risk index. This quantification process assigns different scores based on the severity of the damage, with higher scores indicating greater damage risk. This data is divided into training and validation sets for model training and evaluation.

[0167] Step S312: constructing a network structure of a wellbore damage prediction model, wherein the network structure includes a feature interaction layer, a convolutional attention module, a long short-term memory network layer, and an output layer.

[0168] The feature interaction layer, composed of multiple fully connected sublayers, is used to model the nonlinear associations of multidimensional input feature vectors. Its number of neurons is determined by the dimensionality of the input features. The convolutional attention module comprises a convolutional layer and an attention mechanism layer. The convolutional kernel size and number of the convolutional layers are set according to the dimensionality of the interaction feature matrix to extract local features. The attention mechanism layer is used to assign spatial attention weights. The long-short-term memory network layer, composed of multiple memory units, is used to capture temporal damage characteristics. Its structural parameters are determined by the temporal characteristics of the data. The output layer is a fully connected layer containing one neuron, which outputs the damage risk index. The activation function, for example, can be a sigmoid function, ensuring that the output value is between 0 and 1.

[0169] Step S313: Convert the dynamic change curve of impact energy and the characteristic parameters of wellbore material in the training set into a multi-dimensional input feature vector, input it into the constructed wellbore damage prediction model, and use the back propagation algorithm to train the model with the corresponding damage risk index as the target value.

[0170] Following the methods in steps S141 to S143, the dynamic change curves of impact energy and the characteristic parameters of the wellbore material in the training set are converted into a multidimensional input feature vector. This multidimensional input feature vector is then fed into the wellbore damage prediction model, which outputs a predicted damage risk index. A loss function is calculated between the predicted and target values. The cross-entropy loss function, suitable for probabilistic prediction problems, is used. A backpropagation algorithm is used to adjust the weight parameters of each model layer, and the loss function is minimized through iterative training. During training, model performance is regularly evaluated using a validation set to prevent overfitting.

[0171] Step S314: The performance of the trained wellbore damage prediction model is evaluated using a validation set. When the prediction accuracy of the model is within a preset range, the model training is determined to be complete.

[0172] Performance evaluation metrics include prediction accuracy, precision, and recall. Prediction accuracy measures the degree of agreement between the damage risk index predicted by the model and the target value. Model training is complete when these metrics meet preset requirements. If not, the model's network structure is adjusted, such as by changing the size of the convolution kernel or increasing the number of long-short-term memory (LSTM) layers, and training is repeated until the model can accurately predict the wellbore damage risk index.

[0173] The method provided in the embodiments of the present application continuously collects pressure and vibration sensor signals from different impact surfaces of the shaft buffer structure contact interface to construct a multi-source heterogeneous impact characteristic signal set. Dynamic feature extraction technology is then used to analyze and generate an energy dynamic feature sequence containing impact duration, signal amplitude change, and frequency distribution characteristics, thereby comprehensively characterizing the complete evolution of impact energy from generation to attenuation. A pre-trained impact energy analysis model is used to deeply analyze the energy dynamic feature sequence, accurately extract the initial impact energy, buffer dissipation energy, and residual energy characteristics, and generate a dynamic change curve, thereby achieving quantitative characterization and dynamic tracking of the impact energy state. Furthermore, a damage risk index is generated using a pre-trained wellbore damage prediction model in combination with the characteristic parameters of the wellbore material, thereby establishing a closed-loop prediction system from energy monitoring to structural damage assessment. Finally, a graded risk response is achieved through a graded early warning and control mechanism. By combining multi-dimensional data fusion with intelligent model collaboration, this method overcomes the limitations of single-parameter analysis in traditional monitoring methods and significantly improves the accuracy, real-time performance, and reliability of shaft material impact energy monitoring, thereby effectively reducing the risk of damage to the wellbore structure caused by impact energy accumulation and ensuring the safe and stable operation of the shaft project.

[0174] Figure 2 This is a structural diagram of a shaft material impact energy monitoring system 100 based on machine learning provided in an embodiment of the present application. Figure 2 As shown, the processor 120 can be used in the shaft material impact energy monitoring system 100 based on machine learning and used to perform the functions of the present invention.

[0175] The machine learning-based shaft material impact energy monitoring system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the machine learning-based shaft material impact energy monitoring method of the present invention. Although only one server is shown in the present invention, for convenience, the functions described in the present invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0176] For example, the shaft material impact energy monitoring system 100 based on machine learning may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the shaft material impact energy monitoring system 100 based on machine learning may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The shaft material impact energy monitoring system 100 based on machine learning also includes an input / output (I / O) interface 150 between the computer and other input and output devices.

[0177] For ease of explanation, only one processor is described in the shaft material impact energy monitoring system 100 based on machine learning. However, it should be noted that the shaft material impact energy monitoring system 100 based on machine learning in the present invention may also include multiple processors, so the steps performed by one processor described in the present invention may also be performed jointly or individually by multiple processors. For example, if the processor of the shaft material impact energy monitoring system 100 based on machine learning executes step A and step B, it should be understood that step A and step B may also be executed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0178] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the shaft material impact energy monitoring method based on machine learning of the aforementioned embodiment.

[0179] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute the steps in the vertical shaft material impact energy monitoring method based on machine learning of the aforementioned embodiment.

[0180] The device embodiments described above are merely illustrative. Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network modules. Some or all of these modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0181] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the above technical solution, in essence, or the portion that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of storing or storing data.

[0182] Finally, it should be noted that what is disclosed above is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for monitoring the impact energy of shaft materials based on machine learning, characterized in that: include: Continuously collecting a set of impact characteristic signals of the contact interface of the shaft buffer structure, wherein the set of impact characteristic signals includes pressure sensing signals and vibration sensing signals of different impact action surfaces; Performing dynamic feature extraction processing on the impact characteristic signal set to generate an energy dynamic feature sequence reflecting the impact energy change process, wherein the energy dynamic feature sequence includes an impact duration feature, a signal amplitude change feature, and a frequency distribution feature; Calling a pre-trained impact energy analysis model to perform energy state analysis on the energy dynamic feature sequence, and generating an impact energy dynamic change curve related to the initial impact energy feature, buffer dissipation energy feature, and residual energy feature obtained by analysis; Calling a pre-trained well wall damage prediction model, inputting the impact energy dynamic change curve and preset well wall material characteristic parameters into the well wall damage prediction model, and generating a damage risk index for the well wall structure; A graded warning and control operation is performed according to the damage risk index.

2. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 1 is characterized in that: The continuous collection of the impact characteristic signal set of the contact interface of the shaft buffer structure includes: Generating a raw sensor signal stream with a time stamp based on the pressure sensor signal and the vibration sensor signal collected from the contact interface of the shaft buffer structure, wherein the sampling frequency of the raw sensor signal stream matches the characteristic frequency of the impact event; Adopting an adaptive filtering algorithm to perform noise elimination processing on the original sensor signal stream to obtain an effective signal component containing impact energy characteristics; After the valid signal components pass the signal integrity check, the checked valid signal components are grouped based on the spatial position of the impact surface to generate the impact characteristic signal set containing multi-channel signals, and the number of channels of the multi-channel signals is consistent with the number of partitions of the impact surface.

3. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 1 is characterized in that: The step of performing dynamic feature extraction processing on the impact feature signal set to generate an energy dynamic feature sequence reflecting the impact energy change process includes: performing time domain waveform analysis on the pressure sensing signal in the shock characteristic signal set, extracting signal rising edge slope and peak interval parameters of the pressure sensing signal, and generating a shock duration feature, wherein the shock duration feature is used to describe the time span of a single shock event; performing amplitude spectrum analysis on the vibration sensor signals in the impact characteristic signal set, calculating a ratio of a signal peak amplitude to a root mean square value of the vibration sensor signals, and generating a signal amplitude variation feature, wherein the signal amplitude variation feature is used to characterize intensity fluctuations of impact energy; Performing frequency domain decomposition on the vibration sensor signal by fast Fourier transform, extracting the power spectrum density distribution of the main frequency components, and generating frequency distribution features, wherein the frequency distribution features are used to reflect the distribution of the impact energy in different frequency bands; The impact duration feature, the signal amplitude change feature, and the frequency distribution feature are spliced ​​in chronological order to form the energy dynamic feature sequence with time sequence correlation.

4. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 1 is characterized in that: The calling of the pre-trained impact energy analysis model performs energy state analysis on the energy dynamic feature sequence to generate an impact energy dynamic change curve related to the initial impact energy feature, buffer dissipation energy feature, and residual energy feature obtained by the analysis, including: Inputting the energy dynamic feature sequence into the input layer of the impact energy analysis model, and converting the energy dynamic feature sequence into a high-dimensional feature vector through the embedding layer of the impact energy analysis model, wherein the dimension of the high-dimensional feature vector matches the number of neurons in the hidden layer of the impact energy analysis model; Performing temporal dependency modeling on the high-dimensional feature vectors through the bidirectional long short-term memory network layer of the impact energy analysis model, extracting feature correlation relationships at different time steps, and generating a temporal correlation feature matrix; Using the attention mechanism module of the impact energy analysis model to dynamically weight the time series correlation feature matrix, strengthen the feature contribution of key time nodes, and generate an attention weighted feature vector; Performing nonlinear mapping processing on the attention weighted feature vector through the fully connected layer of the impact energy analysis model to output the initial impact energy feature, the buffer dissipation energy feature, and the residual energy feature; The impact energy dynamic change curve is generated based on the peak amplitude of the initial impact energy feature, the attenuation coefficient of the buffer dissipation energy feature, and the frequency spectrum distribution parameter of the residual energy feature.

5. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 1 is characterized in that: The method of calling a pre-trained well wall damage prediction model, inputting the impact energy dynamic change curve and preset well wall material characteristic parameters into the well wall damage prediction model, and generating a well wall structure damage risk index includes: Obtaining, by analyzing the impact energy dynamic change curve, a peak energy feature, an attenuation rate feature, and a spectrum distribution feature of the impact energy dynamic change curve, wherein the peak energy feature corresponds to the peak amplitude of the initial impact energy feature, the attenuation rate feature corresponds to the attenuation coefficient of the buffer dissipation energy feature, and the spectrum distribution feature corresponds to the spectrum distribution density of the residual energy feature; Acquiring preset characteristic parameters of the well wall material, wherein the characteristic parameters of the well wall material include an elastic modulus parameter, a Poisson's ratio parameter, a compressive strength parameter, and a tensile strength parameter; Performing feature fusion processing on the peak energy feature, the attenuation rate feature, the spectrum distribution feature and the wellbore material characteristic parameters to generate a multi-dimensional input feature vector; Inputting the multidimensional input feature vector into the feature interaction layer of the wellbore damage prediction model, performing nonlinear correlation modeling on the multidimensional input feature vector through a feature cross network, and generating an interaction feature matrix, wherein the interaction feature matrix represents the interaction characteristics between wellbore material and energy; Performing spatial attention weight allocation on the interaction feature matrix through the convolutional attention module of the wellbore damage prediction model, strengthening the feature components corresponding to the weak areas of the wellbore material, and generating a weighted interaction feature map; Performing long-short term memory modeling on the weighted interactive feature map to capture the cumulative damage effect of the dynamic change curve of impact energy on the wellbore structure and generate a time series damage feature vector; The damage risk index is generated by performing probability conversion on the time series damage feature vector through the output layer of the wellbore damage prediction model.

6. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 4 is characterized in that: The generating of the impact energy dynamic change curve based on the peak amplitude of the initial impact energy characteristic, the attenuation coefficient of the buffer dissipation energy characteristic, and the spectrum distribution parameter of the residual energy characteristic includes: Extracting the peak amplitude of the initial impact energy characteristic to determine the initial value of the energy intensity at the initial stage of the impact energy action; Obtaining an attenuation coefficient of the buffer dissipation energy characteristic, analyzing an attenuation law of the attenuation coefficient over time, and generating energy attenuation trend data; Analyzing the spectrum distribution parameters of the residual energy feature, identifying the energy proportions corresponding to different frequency intervals, and generating a spectrum energy distribution feature; The energy intensity initial value is used as the starting point of the curve, and combined with the energy attenuation trend data, a main impact energy attenuation curve is generated; Based on the spectrum energy distribution characteristics, the frequency energy fluctuation component superposition processing is performed on the impact energy attenuation main curve to generate the impact energy dynamic change curve containing multi-frequency energy fluctuation characteristics.

7. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 6 is characterized in that: The method of performing frequency energy fluctuation component superposition processing on the impact energy attenuation main curve based on the spectrum energy distribution characteristics to generate an impact energy dynamic change curve containing multi-frequency energy fluctuation characteristics includes: Extracting energy fluctuation amplitudes and fluctuation periods corresponding to different frequency intervals in the spectrum energy distribution characteristics; generating frequency energy fluctuation components corresponding to each frequency interval according to the energy fluctuation amplitude and the fluctuation period; Analyze the energy proportion of each frequency interval and assign corresponding fluctuation weight coefficients to the frequency energy fluctuation components of the corresponding frequency interval; Performing weighted processing on the frequency energy fluctuation component of each frequency interval and the corresponding fluctuation weight coefficient to generate a weighted frequency energy fluctuation component; The weighted frequency energy fluctuation components of all the frequency intervals are synchronously superimposed on the impact energy attenuation main curve to generate the impact energy dynamic change curve containing multi-frequency energy fluctuation characteristics.

8. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 1, characterized in that: The step of performing a graded early warning and control operation according to the injury risk index includes: Comparing the injury risk index with a preset warning threshold range to determine the warning level of the current injury risk index, wherein the warning level includes a first warning level, a second warning level, and a third warning level; According to the warning level, executing a warning control operation corresponding to the warning level, the warning control operation including at least one of outputting a warning prompt message, triggering an audible and visual alarm, restricting construction equipment construction behavior, and initiating emergency support; According to the damage risk index, at least one of the pre-tightening force of the composite plate included in the buffer structure and the angle of the guide surface of the buffer structure is adjusted to dynamically optimize the buffering efficiency of the buffer structure.

9. The method for monitoring the impact energy of shaft materials based on machine learning according to claim 5, characterized in that: After generating the damage risk index of the wellbore structure, the method further includes: Extracting damage sensitive features of each area of ​​the wellbore output by the middle layer of the wellbore damage prediction model; Correlating and mapping the damage-sensitive features of each area of ​​the well wall with the spatial position information of the well wall structure to establish a corresponding relationship between the damage features and the spatial coordinates; Based on the corresponding relationship between the damage characteristics and the spatial coordinates, constructing a grid model of the wellbore structure in a three-dimensional spatial coordinate system; color-coding grid cells of different damage risk levels according to the distribution of the damage risk index in each grid cell in the grid model; The color-coded grid model is rendered to generate a three-dimensional damage distribution map of the wellbore structure, and potential high-risk areas with higher damage risk indexes are marked in the three-dimensional damage distribution map.

10. A shaft material impact energy monitoring system based on machine learning, characterized in that: It includes a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a computer, the vertical shaft material impact energy monitoring method based on machine learning as described in any one of claims 1 to 9 is implemented.

Images