CN-121999905-A - Rock hardness prediction method and device based on multi-modal pellet topology space learning

CN121999905ACN 121999905 ACN121999905 ACN 121999905ACN-121999905-A

Abstract

The invention discloses a rock hardness prediction method and device based on multi-mode pellet topology space learning, which solve the problem of poor inter-class separability caused by the fact that the traditional rock hardness prediction method does not introduce multi-mode data topology relevance, and effectively improve the accuracy and robustness of rock hardness prediction. The method comprises the steps of (1) constructing a pellet iterative splitting and optimizing module, adapting to non-uniform topological characteristics of rocks with different hardness, excavating internal distribution association of multi-modal data to realize pellet topological dynamic splitting optimization, (2) constructing a pellet quadratic optimization and sample topological graph, reinforcing topological relevance of multi-modal samples through topological dispersion quantification and sample consistency verification, fusing topological constraint sample topological relation, and (3) constructing an objective function of a multi-modal pellet topological space learning model, carrying out theoretical derivation on the objective function to obtain analytic solution of the multi-modal pellet topological space projection direction, thereby obtaining rock hardness pellet topological characteristics with good discrimination, and inputting the rock hardness pellet topological characteristics into a classifier to obtain a final rock hardness prediction result. The rock hardness prediction method fully utilizes the topological relation of the multi-mode data, and has higher accuracy in rock hardness prediction.

Inventors

ZHU YANMIN
Wu Kanghui
SU GUOYONG
SU SHUZHI

Assignees

安徽理工大学

Dates

Publication Date: 20260508
Application Date: 20260107

Claims (5)

1. Rock hardness prediction method and device based on multi-mode pellet topology space learning are characterized by comprising the following steps: (1) The sensor is used for collecting sound signals and vibration signals during rock drilling, 16 time domain features, 4 frequency domain features and 7 time-frequency domain features are obtained through feature extraction, and rock hardness prediction original sound samples are constructed Original vibration sample Wherein The dimensions of the sample are represented and, Representing the number of samples, the same dimension The number of samples below is expressed as ( ) Dividing an original sound sample and an original vibration sample into a training set and a testing set, and randomly extracting training set and testing set data of each experiment; (2) Constructing a pellet iterative splitting and optimizing module; (3) Constructing a pellet secondary optimization and sample topological graph; (4) Constructing an objective function of a multi-mode pellet topology space learning model, carrying out theoretical derivation on the objective function to obtain an analytic solution of the projection direction of the multi-mode pellet topology space, directly obtaining rock hardness pellet topology characteristics with good discrimination, and finally classifying by using a classifier to obtain a rock hardness prediction result.
2. The method and apparatus for predicting rock hardness based on multi-modal pellet topology space learning of claim 1, wherein the constructing pellet iterative splitting and optimizing module of step (2) is performed as follows: (2a) Is provided with Round iterative pellet aggregation Wherein For the number of the current split pellets, Represent the first Individual pellets, each pellet Wherein In order to be dimensional in number, Represent the first The number of samples in each pellet, Representing the first of the pellets A number of samples of the sample were taken, Represent the first The center of each particle ball is provided with a plurality of grooves, The ball set after single round splitting is : Wherein the method comprises the steps of And (3) with Is a single-wheel ball Two split pellets; Each pellet is given by the definition of a pellet , The following are provided: Wherein the method comprises the steps of Two sample points with the farthest intragranular distance: Wherein the method comprises the steps of Representing samples in pellets Distance from pellet center in euclidean space: (2b) If the number of ball samples is 1, Executing the following steps: wherein the pellet ball The number of samples included is Ball with seed The number of samples included is , , , , ; (2C) If the number of ball samples is Indicating that the pellet contains only a single sample or no sample, and cannot form an effective topological unit after splitting, skipping step (2 b) and retaining the original pellet The fragmentation of the topological structure is avoided, and each pellet is ensured to have statistically representative property.
3. The method and apparatus for predicting rock hardness based on multi-modal pellet topology space learning of claim 1, wherein the constructing the pellet quadratic optimization and sample topology of step (3) is performed as follows: (3a) Combining with a pellet iterative splitting and optimizing module, and constructing a sample topological graph: Wherein the method comprises the steps of As a parameter of the degree of similarity, , Representing the pellet inner radius function: Wherein the method comprises the steps of , Representing the topological center of the pellet, the effective pellet collection Some pellet The samples in (a) are , Indicating pellet Is used for the number of samples in the sample, Representing a sample in a valid pellet; (3b) After the pellet iterative splitting and optimizing module processes, partial pellets may still have topology dispersion problem due to overlarge radius, and may need further optimization through secondary splitting, and the definition splitting judgment is as follows: for each of The following judgment is made: Wherein the method comprises the steps of , Is the radius of the particle ball, Is the center of the pellet and is provided with a plurality of holes, For the split threshold, there is defined as follows: Wherein the method comprises the steps of Is that Is defined as the mean value of the radius, Is the median of the radius: Wherein the method comprises the steps of Representing the function of the radius within the pellet.
4. The method and apparatus for predicting rock hardness based on multi-modal pellet topology space learning according to claim 1, wherein the constructing the multi-modal pellet topology space learning model in step (4) is performed as follows: Constructing a target optimization function of a multi-mode pellet topology space learning model: Wherein the method comprises the steps of , Representation of samples Is provided with a projection direction matrix of (a), , , , Covariance matrices constructed for introducing the sample topology, Covariance matrix representing samples: Wherein the method comprises the steps of , Representing a sample mean; Wherein the method comprises the steps of , Representation of , The center matrix of the spheres of the sample, , , Representing a modality , Is used for the sample topology map of (a), , Representing a sample topology And (3) with A degree matrix of (2); i.e. the objective function is converted into: and then, carrying out optimization solution on the objective function to obtain an analytic solution of the multi-mode pellet topological space projection direction, thereby directly obtaining the rock hardness pellet topological characteristics with good discrimination, and finally, classifying by using a classifier to obtain a rock hardness prediction result.
5. Rock hardness prediction method and device based on multi-mode pellet topology space learning are characterized by comprising the following steps: The data acquisition module is used for acquiring original sound, vibration sample training data and sample test data when the rock is acquired, and generating a rock hardness prediction training sample set and a rock hardness prediction test sample set; the model construction module is used for constructing a target optimization function of the multi-mode pellet topology space learning model according to the rock hardness prediction training sample set; The projection matrix solving module is used for optimally solving the multi-mode pellet topological space projection matrix according to the target optimization function of the multi-mode pellet topological space learning model; The characteristic extraction module is used for respectively applying the multi-mode pellet topological space projection matrix to the rock hardness prediction training sample set and the rock hardness prediction test sample set to generate a training identification characteristic set of the rock hardness prediction training sample set and a test identification characteristic set of the rock hardness prediction test sample set; And the rock hardness prediction module is used for inputting the training identification feature set and the testing identification feature set into the support vector base classifier to generate a rock hardness prediction result.

Description

Rock hardness prediction method and device based on multi-modal pellet topology space learning Technical Field The invention relates to a rock hardness prediction method and device based on multi-mode pellet topology space learning, belonging to the fields of pattern recognition and rock hardness prediction. Background Rock hardness is a key mechanical parameter in the fields of geotechnical engineering, mineral resource development and the like, and directly influences evaluation of underground drilling efficiency, blasting design, tunneling safety, rock drillability and the like. Development of underground tunneling work is affected by factors such as rock surface weathering degree, uneven mineral particle distribution and the like in different areas, and rock hardness prediction becomes particularly important for improving tunneling efficiency. Among the numerous rock hardness prediction methods, the spatial learning method has been attracting attention by means of a solid theoretical basis and good engineering practicality. The typical correlation analysis is used as a core multivariate statistical analysis method, and has the core value of mining potential correlation among bimodal data such as vibration and acoustics and the like and providing an effective path for multimodal feature fusion. However, a core difficulty faced by typical correlation analysis in rock hardness prediction scenarios is that mining of topological relevance of sound, vibration bimodal data is not introduced. Typical correlation analysis only builds a covariance matrix based on original sensing data, inherent topological correlation characteristics between two types of modal data cannot be captured, characterization of effective hardness information cannot be enhanced through topological cooperativity between the modalities, core characteristics which are strongly correlated with rock hardness are difficult to refine, and finally a sample cannot be effectively projected into a characteristic space to realize efficient classification. Therefore, the invention combines the space learning theory, designs the conversion strategy of the original data space and the multi-mode pellet topological space, and introduces the distribution association among the multi-mode data. The invention constructs a pellet iterative splitting and optimizing module and digs the internal distribution association of multi-mode data. Based on the module, a sample topological relation of topological constraint is fused, and a pellet secondary optimization and sample topological graph is constructed. And finally, constructing an objective function of the multi-mode pellet topological space learning model by minimizing pellet weighting dispersion and maximizing modal pairwise correlation, carrying out theoretical derivation on the objective function to obtain an analytical solution of the multi-mode pellet topological space projection direction, directly obtaining rock hardness pellet topological characteristics with good discrimination, and finally, classifying by using a classifier to obtain a rock hardness prediction result. Disclosure of Invention In order to introduce distribution association among multi-modal data, firstly, a pellet iterative splitting and optimizing module is constructed, on the basis, a sample topological relation of pellet secondary optimization and a sample topological graph to fuse topological constraint is constructed, and finally, an objective function of a multi-modal pellet topological space learning model is constructed by minimizing pellet weighting distribution and maximizing modal pairwise correlation. And further carrying out theoretical deduction on the objective function to obtain an analytic solution of the multi-mode pellet topological space projection direction, thereby directly obtaining the rock hardness pellet topological characteristic with good discrimination. The specific implementation steps of the invention are as follows: 1. The sensor is used for collecting sound signals and vibration signals during rock collection, 16 time domain features, 4 frequency domain features and 7 time-frequency domain features are obtained through feature extraction, and rock hardness prediction original sound samples are constructed Original vibration sampleWhereinThe dimensions of the sample are represented and,The number of samples is represented and the number of samples,() Representing the same dimensionAnd dividing the original sound sample and the original vibration sample into a training set and a testing set according to the number of the samples, and randomly extracting training set and testing set data of each experiment. 2. And constructing a pellet iterative splitting and optimizing module. The specific construction steps of the pellet iterative splitting and optimizing module are as follows: In the same mode, set Round iterative pellet aggregationWhereinFor the number of the current split pellets,Represent the firstI