A method for predicting the late strength of a fill based on early monitoring data

Abstract

The present application relates to the technical field of mine filling mining, and more particularly to a filling body late strength prediction method based on early monitoring data. The content includes: collecting filling body sample data, introducing a time series residual driven strength evolution feature extraction mechanism, converting early strength observation values into structured features reflecting strength evolution rate and evolution stability, and obtaining an evolution factor vector; introducing a historical sample database, constructing an evolution factor vector set, based on the evolution factor vector set, introducing an evolution similarity weight mechanism based on an adaptive bandwidth control factor, and calculating the evolution similarity weight; based on the evolution similarity weight, constructing an evolution neighborhood subset of the filling body sample, and introducing a late strength reasoning mechanism based on evolution similarity and trend adjustment to predict the late strength of the filling body. The present application solves the problem of black box model prediction process and sensitivity to sample data quality in traditional filling body late strength prediction methods.

Description

Technical Field

[0001] This invention relates to the field of mine backfilling mining technology, and in particular to a method for predicting the later strength of backfill bodies based on early monitoring data. Background Technology

[0002] Backfill mining is one of the most widely used mining methods in underground mines today. It involves transporting cemented backfill materials into the goaf to form a backfill body, thereby achieving surrounding rock control, surface subsidence management, and tailings resource utilization. The strength performance of the backfill body is a key factor affecting mine safety and mining efficiency, and unconfined compressive strength is usually used as the main evaluation index in engineering.

[0003] In existing engineering practices, the strength of backfill is mainly obtained through indoor standard curing test blocks. However, the commonly used target age, including strength testing cycles of 28 days, 56 days or longer, is relatively long, which leads to a significant lag in the adjustment of backfill parameters and mining decisions. This is not conducive to the rapid feedback and dynamic adjustment of backfill ratio, and it is difficult to meet the needs of continuous and refined production in modern mines. In response to the above problems, existing technologies have proposed strength prediction methods based on empirical formulas or artificial intelligence models. Although some methods have improved the prediction accuracy, they still generally have the following shortcomings: (1) The internal structure of the model is complex and the feature contribution is unexplainable. It is a typical "black box" model, which cannot clearly trace the impact of each input feature on the prediction result, making it difficult to meet the needs of process transparency and timeliness in engineering practice; (2) It is sensitive to the quality of sample data, has poor robustness when the sample is sparse or the observation points are limited, the prediction result is unstable, and the model deployment threshold is high, making it difficult to widely promote and apply in actual mining environments.

[0004] Therefore, in order to address the above shortcomings, there is an urgent need to provide a method for predicting the later strength of filling materials based on early monitoring data. Summary of the Invention

[0005] This invention provides a method for predicting the later strength of infill bodies based on early monitoring data, in order to solve the technical problems of the black box nature of the model prediction process and the sensitivity to the quality of sample data in traditional methods for predicting the later strength of infill bodies.

[0006] The present invention provides a method for predicting the later strength of infill bodies based on early monitoring data, comprising the following steps:

[0007] S1. Collect infill sample data, construct the original ratio parameter vector and early intensity observation value set; introduce the intensity evolution feature extraction mechanism driven by time residual, combine the original ratio parameter vector, and transform the early intensity observation values ​​into structured features that reflect the intensity evolution rate and evolution stability to obtain the evolution factor vector;

[0008] S2. Introduce a historical sample database and construct an evolutionary factor vector set; based on the evolutionary factor vector set, introduce an evolutionary similarity weight mechanism based on adaptive bandwidth control factor to calculate the evolutionary similarity weight; based on the evolutionary similarity weight, construct an evolutionary neighborhood subset of the infill sample, and introduce a late-stage intensity inference mechanism based on evolutionary similarity and trend adjustment to predict the late-stage intensity of the infill sample, and obtain the predicted value of the late-stage intensity of the infill sample at the target age.

[0009] Preferably, the implementation process of the temporal residual-driven intensity evolution feature extraction mechanism specifically includes:

[0010] Based on early intensity observations at adjacent observation time points, the growth rate of early intensity observations is calculated, and a sequence of early intensity observation growth rates is constructed.

[0011] Preferably, the implementation process of the temporal residual-driven intensity evolution feature extraction mechanism further includes:

[0012] Based on the set of early intensity observations, an intensity evolution fitting curve is constructed using interpolation methods, and the deviation of the actual early intensity observations from the intensity evolution fitting curve is quantified to obtain the local evolution residual factor.

[0013] Preferably, the implementation process of the temporal residual-driven intensity evolution feature extraction mechanism further includes:

[0014] The original ratio parameter vector, the mean of the early intensity observation value growth rate sequence, and the maximum value of the early intensity observation value growth rate sequence are normalized to obtain the normalized original ratio parameter vector, the normalized mean of the early intensity observation value growth rate sequence, and the normalized maximum value of the early intensity observation value growth rate sequence. These are then horizontally concatenated with the local evolution residual factor according to the feature dimension to obtain the evolution factor vector.

[0015] Preferably, the implementation process of the evolutionary similarity weight mechanism based on adaptive bandwidth control factor specifically includes:

[0016] Based on the local evolution residual factor and combined with the exponential decay function, the distance metric of the evolution factor vectors between infill samples is mapped to the evolution similarity weight.

[0017] Preferably, the specific construction method of the evolutionary neighborhood subset of the filling body sample is as follows:

[0018] The evolutionary similarity weights are sorted by numerical value. Infill samples are selected based on the sorted evolutionary similarity weights, and an evolutionary neighborhood subset of the current infill sample is constructed.

[0019] Preferably, the implementation process of the late-stage intensity inference mechanism based on evolutionary similarity and trend adjustment specifically includes:

[0020] Based on evolutionary similarity weights, combined with the later intensity observations of historical samples in the evolutionary neighborhood subset, and by introducing a trend offset adjustment factor, the later intensity of the infill body is predicted, and the predicted later intensity of the infill body sample at the target age is obtained.

[0021] Preferably, the trend offset adjustment factor is constructed in the following way:

[0022] It is constructed by calculating the relative offset between the current infill sample and historical samples in the evolutionary neighborhood subset in terms of the early intensity growth rate trend, and is used to make directional corrections to the later intensity observations of historical samples.

[0023] The beneficial effects of the technical solution of the present invention are:

[0024] 1. This invention introduces a temporal residual-driven intensity evolution feature extraction mechanism, which transforms early intensity observations into structured features that reflect the intensity evolution rate and stability. This effectively avoids the problems of feature sparsity and missing evolutionary relationships caused by directly using early intensity observations, extracts temporal evolution structural features, and enhances the sensitivity and generalization ability of later intensity prediction to different ratio conditions and early response differences.

[0025] 2. Based on the traditional Gaussian kernel function structure, this invention introduces an evolutionary similarity weighting mechanism based on an adaptive bandwidth control factor. This mechanism can dynamically adjust the decay rate of evolutionary similarity according to the early intensity evolution stability of the current filling sample. It can automatically expand the similarity neighborhood on unstable samples and tighten the nearest neighbor range on stable samples, thereby effectively improving the rationality of similar sample selection.

[0026] 3. This invention proposes a late-stage intensity inference mechanism based on evolutionary similarity and trend adjustment. While maintaining evolutionary similarity weighting, it can further amplify or suppress the late-stage intensity of historical samples based on the relative differences in the intensity growth rate of infill bodies in the early age, significantly improving the prediction accuracy of the late-stage intensity of infill bodies. Moreover, the inference process does not rely on implicit weights or untraceable network structures, but constructs the prediction results item by item based on clearly calculable physical quantities. Attached Figure Description

[0027] Figure 1 This is a flowchart of a method for predicting the later strength of infill bodies based on early monitoring data, as described in this invention. Detailed Implementation

[0028] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] The following description, in conjunction with the accompanying drawings, details the specific scheme of a broadband oscillation suppression method for high-proportion wind power access systems based on impedance shaping provided by the present invention.

[0031] See attached document Figure 1 The diagram illustrates a flowchart of a method for predicting the later strength of infill bodies based on early monitoring data, according to an embodiment of the present invention. The method includes the following steps:

[0032] S1. Collect infill sample data, construct the original ratio parameter vector and early intensity observation value set; introduce the intensity evolution feature extraction mechanism driven by time residual, combine the original ratio parameter vector, and transform the early intensity observation values ​​into structured features that reflect the intensity evolution rate and evolution stability to obtain the evolution factor vector;

[0033] Assume there is a total One filled body sample, sample number is , collect the first Data from several infill body samples were collected, and an original ratio parameter vector was constructed. and early intensity observation set The original proportioning parameter vector ,in Indicates the first The first filling sample The mix proportion parameters include the water-cement ratio, the cement-sand ratio, and the cement admixture ratio. For the proportioning parameter dimension, For the ratio parameter index; the set of early intensity observations ,in Indicates the first Early intensity observation time points, unit: days, such as the third day, the seventh day, etc. Indicates the first The first filling sample The unconfined compressive strength values ​​at each early strength observation time point, i.e., early strength observation values, in MPa. Index of early intensity observation time points, This represents the number of early intensity observation time points;

[0034] Due to the limited number of early intensity observation time points, and the potential for noise in early intensity observations due to experimental errors and environmental fluctuations, the set of early intensity observations is used directly. Since a complete description of later intensity changes is not possible, a temporal residual-driven intensity evolution feature extraction mechanism is introduced. The specific implementation process is as follows: The early intensity observation values ​​at adjacent early intensity observation time points are temporally differencing to calculate the growth rate of the early intensity observation values, and an early intensity observation value growth rate sequence is constructed. Based on the set of early intensity observation values, an existing local Lagrange interpolation method is used to construct an intensity evolution fitting curve, and the deviation of the actual early intensity observation values ​​from the intensity evolution fitting curve is calculated to obtain the local evolution residual factor. Finally, based on the original ratio parameter vector, the mean and maximum values ​​of the early intensity observation value growth rate sequence, and the local evolution residual factor, an evolution factor vector that can characterize the later intensity evolution features is constructed.

[0035] Specifically, the growth rate of early intensity observations per unit time is calculated by comparing the increment of early intensity observation values ​​at adjacent early intensity observation time points with the time increment, as shown in the following formula:

[0036]

[0037] in, Indicates the first One filling sample in to The rate of increase of early intensity observations within the time period, in MPa / day; Indicates the first Early intensity observation time points, Indicates the first Early intensity observation time points; Indicates the first Early intensity observation values ​​at each early intensity observation time point; Indicates the first Early intensity observations at the first early intensity observation time point; based on the growth rate of the early intensity observations, construct the first... Early intensity observation growth rate sequence of individual infill samples , Represents the growth rate sequence of early intensity observations The index;

[0038] Based on early intensity observation sets Using existing local Lagrange interpolation methods for An intensity evolution fitting curve is constructed using early and mid-stage intensity observation time points and observed values ​​to approximately represent the ideal evolution trend of actual early-stage intensity observations over time. Furthermore, to characterize the overall deviation between actual early-stage intensity observations and the intensity evolution fitting curve, a local evolution residual factor is defined, as follows:

[0039]

[0040] in, Indicates the first The local evolution residual factor of each infill sample is used to reflect the overall deviation between the early intensity observation changes and the intensity evolution fitting curve, and can reflect the stability level of early intensity evolution. The smaller the value, the more stable the early strength evolution of the infill material, which is more beneficial for later strength prediction. The larger the value, the more unstable the early intensity evolution is, which will lead to greater uncertainty in the later intensity evolution; Index of early intensity observation time points, This represents the number of early intensity observation time points; Indicates the first The deviation of early intensity observation values ​​from fitted early intensity observation values ​​at each early intensity observation time point; Indicates the first Early intensity observation values ​​at each early intensity observation time point; Indicates the first The fitted values ​​of early intensity observations at each early intensity observation time point are obtained from the intensity evolution fitting curve; To prevent extremely small constants with a denominator of 0, such as 0.01 MPa;

[0041] The original mix proportion parameter vector, the mean of the early intensity observation growth rate sequence, and the maximum of the early intensity observation growth rate sequence are normalized using the minimum-maximum normalization method to obtain the normalized original mix proportion parameter vector. Mean of the growth rate sequence of normalized early intensity observations Maximum value of the growth rate sequence of early intensity observations after normalization Then, it is horizontally concatenated with the local evolution residual factor along the feature dimension to obtain the evolution factor vector. ,in ;

[0042] The above process introduces a temporal residual-driven intensity evolution feature extraction mechanism, which transforms early intensity observations into structured features that reflect the intensity evolution rate and stability. This effectively avoids the problems of feature sparsity and missing evolutionary relationships caused by directly using early intensity observations, extracts temporal evolution structural features, and enhances the sensitivity and generalization ability of later intensity predictions to different ratio conditions and early response differences.

[0043] S2. Introduce a historical sample database and construct an evolutionary factor vector set; based on the evolutionary factor vector set, introduce an evolutionary similarity weight mechanism based on adaptive bandwidth control factor to calculate the evolutionary similarity weight; based on the evolutionary similarity weight, construct an evolutionary neighborhood subset of the infill sample, and introduce a late-stage intensity inference mechanism based on evolutionary similarity and trend adjustment to predict the late-stage intensity of the infill sample, and obtain the predicted value of the late-stage intensity of the infill sample at the target age.

[0044] Construct an evolutionary factor vector set based on a historical sample database. , ,in Indicates the historical sample index. The number of historical samples is used as a reference. Based on the traditional Gaussian kernel function structure, and considering the instability of the early intensity evolution path of the infill body, an evolutionary similarity weight mechanism based on an adaptive bandwidth control factor is introduced. This mechanism maps the distance metric to the evolutionary similarity weight using an exponential decay function. The formula is:

[0045]

[0046] in, Indicates the first The first filling sample and the first The evolutionary similarity weights between historical samples are mapped to the distance metric of the evolutionary factor vectors between samples using an exponential decay function. Evolutionary similarity weights within the interval; Indicates the first The evolution factor vector of the first infill sample and the first The squared Euclidean distance of the evolution factor vectors of historical samples is used to measure the degree of difference between two infill samples in the early intensity evolution path. Indicates the first The evolution factor vector of each infill sample; Indicates the first The evolution factor vector of each historical sample; This is an adaptive bandwidth control factor used to dynamically adjust the bandwidth in the evolutionary similarity weight calculation based on the stability of the early intensity evolution of the current filling sample, thereby affecting the range of similar samples. The smaller the value, the more stable the early intensity evolution process of the sample. The smaller the bandwidth, the more focused the evolutionary similarity weight distribution is on highly similar samples, thus tightening the nearest neighbor range on stable samples. The basic smoothing parameter is a positive real constant used to adjust the decay rate of the evolutionary similarity weights, and is obtained through existing cross-validation methods. For the first The local evolution residual factor of each infill sample is used to reflect the stability level of early intensity evolution;

[0047] Based on the traditional Gaussian kernel function structure, the above formula introduces an evolutionary similarity weight mechanism based on an adaptive bandwidth control factor. This mechanism can dynamically adjust the decay rate of evolutionary similarity according to the early intensity evolution stability of the current filling sample. It can automatically expand the similarity neighborhood on unstable samples and tighten the nearest neighbor range on stable samples, thereby effectively improving the rationality of similar sample selection.

[0048] Evolutionary similarity weights Sort by value from largest to smallest, and select the top... The infill sample corresponding to the evolutionary similarity weight is used to construct the first... Evolutionary neighborhood subset of each filling sample And generate an evolutionary neighborhood subset. The corresponding sample index set is The size of the evolutionary neighborhood subset is This was obtained through cross-validation.

[0049] To predict the late-stage intensity of infill bodies at the target age, a late-stage intensity inference mechanism based on evolutionary similarity and trend adjustment is proposed. This mechanism multiplies the evolutionary similarity weights by the observed late-stage intensity values ​​of historical samples, corrects this by introducing a trend offset adjustment factor, and then sums the results. The sum is then normalized based on the sum of the evolutionary similarity weights to obtain the predicted late-stage intensity value of the infill body sample at the target age. The formula is as follows:

[0050]

[0051] in, Indicates the first Predicted late-stage strength values ​​of individual infill samples at the target age, in MPa; Indicates the first The set of sample indices corresponding to the evolutionary neighborhood subset of each infill sample; Indicates the first The sample and the first Evolutionary similarity weights between historical samples; Indicates the first The late intensity observation values ​​of historical samples at the target age, in MPa, were obtained directly from the historical sample database. Indicates the first Mean of the growth rate sequence of early intensity observations in a historical sample; This is a trend offset adjustment factor, representing the relative offset between the current infill sample and historical samples in the early intensity growth rate trend. It is used to directionally correct the later intensity observations of historical samples. hour, If the value is greater than 1, it is an upward correction. , If less than 1, it is a downward correction. , Approaching 1, no correction is needed; This means multiplying the evolutionary similarity weights by the later intensity observations of historical samples and correcting them using a trend offset adjustment factor; This is a trend offset adjustment coefficient, used to control the adjustment strength of the relative offset of the early intensity growth rate trend. The larger the value, the more sensitive it is to differences in the early intensity growth rate trend, which was obtained through cross-validation. For the first The mean of the early intensity observation growth rate sequence of the first infill sample and the first A measure of the relative difference between the mean values ​​of the early intensity growth rate sequences of historical samples, representing the relative offset of the early intensity growth rate trend; To prevent extremely small constants with a denominator of 0, such as 0.01 MPa / day;

[0052] The above formula proposes a late-stage intensity inference mechanism based on evolutionary similarity and trend adjustment. While maintaining evolutionary similarity weighting, it can further amplify or suppress the late-stage intensity observations of historical samples based on the relative differences in the intensity growth rate of infill bodies in the early age, significantly improving the prediction accuracy of the late-stage intensity of infill bodies. Moreover, the inference process does not rely on implicit weights or untraceable network structures, but constructs the prediction results item by item based on clearly calculable physical quantities.

[0053] In summary, a method for predicting the later strength of infill bodies based on early monitoring data has been developed.

[0054] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0055] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0056] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting the later strength of infill bodies based on early monitoring data, characterized in that, Includes the following steps: S1. Collect infill sample data to construct the original ratio parameter vector and early intensity observation set; introduce a time-series residual-driven intensity evolution feature extraction mechanism, calculate the growth rate of early intensity observations based on early intensity observations at adjacent observation time points, and construct an early intensity observation growth rate sequence; based on the early intensity observation set, construct an intensity evolution fitting curve through interpolation methods, and quantify the deviation of the actual early intensity observations from the intensity evolution fitting curve to obtain the local evolution residual factor; The original ratio parameter vector, the mean of the growth rate sequence of early intensity observations, and the maximum value of the growth rate sequence of early intensity observations are normalized to obtain the normalized original ratio parameter vector, the normalized mean of the growth rate sequence of early intensity observations, and the normalized maximum value of the growth rate sequence of early intensity observations. These are then horizontally concatenated with the local evolution residual factor according to the feature dimension to transform the early intensity observations into structured features that reflect the intensity evolution rate and evolution stability, thus obtaining the evolution factor vector. S2. A historical sample database is introduced to construct an evolutionary factor vector set. Based on the evolutionary factor vector set, an evolutionary similarity weight mechanism based on an adaptive bandwidth control factor is introduced. Using the local evolutionary residual factor and combined with the exponential decay function, the distance metric of the evolutionary factor vectors between infill samples is mapped to the evolutionary similarity weight. Based on the evolutionary similarity weight, an evolutionary neighborhood subset of infill samples is constructed. A late-stage intensity inference mechanism based on evolutionary similarity and trend adjustment is introduced. Combining the late-stage intensity observations of historical samples in the evolutionary neighborhood subset and the trend offset adjustment factor, the late-stage intensity of infill samples is predicted to obtain the predicted late-stage intensity value of infill samples at the target age. The trend offset adjustment factor is constructed by calculating the relative offset between the current infill sample and historical samples in the evolutionary neighborhood subset in the early intensity growth rate trend, and is used to directionally correct the late-stage intensity observations of historical samples.

2. The method for predicting the later strength of infill bodies based on early monitoring data according to claim 1, characterized in that, The specific method for constructing the evolutionary neighborhood subset of the filling body sample is as follows: The evolutionary similarity weights are sorted by numerical value. Infill samples are selected based on the sorted evolutionary similarity weights, and an evolutionary neighborhood subset of the current infill sample is constructed.

Images