CN-122022344-A - Mine mining process digital monitoring optimization method and system based on AI

CN122022344ACN 122022344 ACN122022344 ACN 122022344ACN-122022344-A

Abstract

The invention discloses a mine mining process digital monitoring optimization method and system based on AI, and relates to the technical field of digital monitoring optimization. The method breaks through the limitation of scattered traditional data acquisition and low association degree by synchronously acquiring multi-ring spectrum, working conditions and producing full-dimensional data, provides a complete data basis for digital monitoring, realizes the accurate judgment of material components and the optimal adaptation of process parameters by means of the accurate mining feature association of a fusion extraction model and constructing an AI working condition-process linkage regulation model, improves the adaptation capability of the model to different mining area working conditions by means of a closed-loop mechanism with deviation feedback, model iteration and continuously abundant historical data sets, solves the pain points with poor universality and difficulty in continuous optimization of the traditional model, and finally achieves the aims of monitoring digitization, process regulation and intellectualization, high efficiency of resource utilization and environment-friendly upgrading in the mining process, and remarkably reduces the yield of non-price minerals and improves the recovery rate of tailings and concentrate grade.

Inventors

ZHANG XIAHUI
YANG LONGFENG
ZHANG YU

Assignees

机数仪器(浙江)有限公司

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (10)

1. The mine mining process digital monitoring optimization method based on AI is characterized by comprising the following steps: Acquiring laser-induced breakdown spectrum data of minerals of each link, and synchronously acquiring time sequence data of the picking working condition and resource output data to form an original multi-source heterogeneous data set; Carrying out space-time alignment and pretreatment on an original multi-source heterogeneous data set, extracting element spectrum feature vectors, working condition dynamic feature vectors and cross-correlation feature vectors of the element spectrum feature vectors and the working condition dynamic feature vectors based on a fusion extraction model, and outputting a high-dimensional fusion feature set; Based on a historical mining multisource heterogeneous data set, adopting an algorithm of fusion of transfer learning and deep reinforcement learning to construct an AI working condition-process linkage regulation model; inputting the high-dimensional fusion feature set into an AI working condition-process linkage regulation model, and outputting the accurate content of all elements and the judgment result of the compound components of the materials at each stage of current picking through feature matching and reinforcement learning reasoning, and adapting to the optimal process parameter regulation scheme of the current working condition; according to the optimal technological parameter regulation scheme, parameter adjustment is carried out on the picking equipment, and a deviation coefficient is determined based on a preset technological regulation effect; And simultaneously supplementing the newly acquired multi-source heterogeneous data set to a historical data set, and updating a working condition adaptation knowledge base of the AI working condition-process linkage regulation model through transfer learning.
2. The AI-based mine mining process digital monitoring optimization method of claim 1, wherein the fusion extraction model comprises a spectral feature extraction sub-model, a working condition feature extraction sub-model and an associated feature fusion sub-model; the spectral feature extraction sub-model is formed by connecting a wave band attention module and a 1D convolutional neural network in series; The working condition characteristic extraction sub-model is formed by connecting a time sequence attention module and a time sequence convolution neural network in series; And the associated feature fusion sub-model is a weight guide type feature interaction network.
3. The AI-based mine mining process digital monitoring optimization method of claim 2, wherein the process of outputting the high-dimensional fusion feature set based on the element spectrum feature, the working condition dynamic feature and the cross-correlation feature extracted by the fusion extraction model is as follows: Inputting the preprocessed laser-induced breakdown spectroscopy data into a spectral feature extraction sub-model, completing the weight distribution of a key wave band through a wave band attention module, and outputting element spectral feature vectors with uniform dimensions through convolution operation, pooling operation and full-connection layer mapping of a 1D convolution neural network; Inputting the preprocessed time sequence data of the selected working condition into a working condition characteristic extraction sub-model, completing the distribution of sensitive parameters and abrupt time sequence node weights through a time sequence attention module, and outputting working condition dynamic characteristic vectors with uniform dimensions through time sequence convolution, cavity convolution operation and characteristic compression mapping of a time sequence convolution neural network; inputting the element spectrum feature vector and the working condition dynamic feature vector into a correlation feature fusion sub-model, mining the coupling correlation mode of the element spectrum feature vector and the working condition dynamic feature vector through matrix point multiplication operation of a feature interaction network, optimizing correlation feature response by combining an attention weight feedback adjustment mechanism, and outputting a cross correlation feature vector with the dimension consistent with the feature vector; And after carrying out dimension consistency verification on the element spectrum feature vector, the working condition dynamic feature vector and the cross-correlation feature vector, adopting a feature stitching algorithm to integrate to form a high-dimensional fusion feature set.
4. The method for digitally monitoring and optimizing the mining process based on the AI according to claim 1, wherein the process for constructing the AI working condition-process linkage regulation model by adopting an algorithm integrating transfer learning and deep reinforcement learning based on the historical mining multisource heterogeneous data set is as follows: acquiring a historical mining multi-source heterogeneous data set, wherein the historical mining multi-source heterogeneous data set comprises mineral type data, working condition data, process parameter data and corresponding resource output effect data of different mining areas; preprocessing a multi-source heterogeneous dataset of a historical mining industry, dividing the multi-source heterogeneous dataset into a source domain dataset and a target domain dataset according to the production maturity of a mining area, and dividing the source domain dataset and the target domain dataset into a training subset and a verification subset according to a set proportion respectively; Constructing a transfer learning-depth reinforcement learning fusion algorithm framework comprising a quantization transfer learning module and a depth reinforcement learning decision module, wherein the quantization transfer learning module is configured with a feature transfer unit and a model parameter transfer unit, the feature transfer unit is used for mining a common feature space of a source domain and a target domain data set, and the model parameter transfer unit is used for weighting and transferring model parameters obtained by source domain training to a target domain according to domain similarity; Inputting the source domain training subset into a quantization transfer learning module, learning the cross-mining-area commonality characteristics through a characteristic transfer unit, and training by a model parameter transfer unit to obtain a source domain basic model; Inputting the target domain training subset into the source domain basic model, and adopting a domain self-adaptive loss function to fine tune model parameters to obtain a target domain initialization model; Parameters of a target domain initialization model are imported into a deep reinforcement learning decision module through a parameter sharing layer, state input, action output, rewarding feedback and parameter updating are used as loops, process control interaction in a picking process is simulated based on a target domain training subset, and strategy network and value network parameters of the model are iteratively updated through a near-end strategy optimization algorithm; inputting the source domain verification subset and the target domain verification subset into an optimized transfer learning-deep reinforcement learning fusion algorithm, and testing the deviation between a predicted value and an actual value of a resource output effect corresponding to a technological parameter regulation scheme output by the transfer learning-deep reinforcement learning fusion algorithm; And if the deviation exceeds a preset threshold, readjusting the weight coefficient of the transfer learning and the weight of the reward function of the reinforcement learning until the deviation value meets the threshold requirement, and obtaining the AI working condition-process linkage regulation and control model.
5. The method for digitally monitoring and optimizing mining and selecting processes based on AI according to claim 4, wherein the deep reinforcement learning decision module defines a state space as a fusion vector of current working conditions, mineral type characteristics and real-time process parameters, an action space as a process parameter adjustment quantity set, and a reward function as a comprehensive evaluation index of resource output effect.
6. The digital monitoring and optimizing method for mine mining process based on AI according to claim 4, wherein the process of outputting the determination result of the total element accurate content and the compound component of the materials in each current mining stage and adapting the optimal technological parameter regulation scheme of the current working condition is as follows: invoking a cross-mining-area common feature space obtained through the training of the quantitative transfer learning module, and calculating the distribution distance between a high-dimensional fusion feature set of the current mining area and source domain mining area features in the cross-mining-area common feature space by adopting a feature distribution alignment algorithm based on the maximum mean value difference; generating a characteristic adaptive weight matrix based on the distribution distance, and fine-tuning characteristic mapping layer parameters of an AI working condition-process linkage regulation model through the characteristic adaptive weight matrix; the AI working condition-process linkage regulation and control model after the parameter adjustment of the feature mapping layer obtains a Top-N historical feature sample set which is most similar to the current high-dimensional fusion feature set based on the high-dimensional fusion feature set; taking the adapted high-dimensional fusion feature set and Top-N historical feature sample set as input, carrying out probability reasoning on the material components through a value network of a deep reinforcement learning decision module, simultaneously calling a full-band intelligent peak searching algorithm and a spectral peak area algorithm to analyze element spectral feature vectors, and outputting the full-element accurate content and compound component judgment result of the materials in each current picking stage; and traversing a preset technological parameter adjustment action space through a strategy network of the deep reinforcement learning decision module by taking the dynamic characteristic vector of the current working condition and the judgment result of the material components as state input, calculating expected benefits of all actions based on a preset reward function, screening out an action combination with optimal expected benefits, and outputting an optimal technological parameter regulation scheme adapting to the current working condition.
7. The method for digitally monitoring and optimizing the mining process based on the AI according to claim 6, wherein the process of obtaining the Top-N historical feature sample set most similar to the current high-dimensional fusion feature set by the AI working condition-process linkage regulation and control model after the feature mapping layer parameter adjustment based on the high-dimensional fusion feature set is as follows: And searching and matching the high-dimensional fusion feature set with the preprocessed mining multi-source heterogeneous data set, and screening a Top-N historical feature sample set which is most similar to the current high-dimensional fusion feature set by adopting a k-nearest neighbor algorithm and taking Euclidean distance of feature vectors as a matching basis.
8. The method for digitally monitoring and optimizing mining and selecting processes of mines based on AI as set forth in claim 6, wherein the process of performing parameter adjustment on mining and selecting equipment according to an optimal process parameter adjustment scheme and determining deviation coefficients based on a preset process adjustment effect is as follows: Multi-source data extraction is carried out on the whole process of extraction and selection after regulation and control, the concentrate grade lifting amplitude, the tailing component change trend and the medicament consumption reduction are recorded, and a process regulation and control effect real-time monitoring ledger is formed; calculating individual deviation rates of data in the process control effect real-time monitoring ledgers and data in the preset process control effect respectively; and carrying out linear weighted summation on the individual deviation rate to obtain a deviation coefficient.
9. The method for digitally monitoring and optimizing the mining process based on AI according to claim 4, wherein the process of feeding back deviation data to the AI working condition-process linkage regulation model and performing model parameter iteration is as follows: Carrying out structural arrangement on various data related to deviation to form a complete deviation feedback data set, and inputting the complete deviation feedback data set to a deep reinforcement learning decision module of an AI working condition-process linkage regulation model; The deep reinforcement learning decision module analyzes the reasons for the deviation based on the deviation feedback data set and adjusts the reward function; the deep reinforcement learning decision module adopts a near-end strategy optimization algorithm, takes a deviation feedback data set as a training sample, and updates the strategy network and the value network weight of the model.
10. AI-based mine mining process digital monitoring optimization system, which is characterized by comprising: The data set acquisition module is used for acquiring laser-induced breakdown spectrum data of minerals in each link, synchronously acquiring time sequence data of the picking working condition and resource output data, and forming an original multi-source heterogeneous data set; the fusion characteristic acquisition module performs space-time alignment and pretreatment on the original multi-source heterogeneous data set, extracts element spectrum characteristic vectors, working condition dynamic characteristic vectors and cross correlation characteristic vectors of the element spectrum characteristic vectors and the working condition dynamic characteristic vectors based on the fusion extraction model, and outputs a high-dimensional fusion characteristic set; the model construction module is used for constructing an AI working condition-process linkage regulation and control model by adopting an algorithm of fusion of transfer learning and deep reinforcement learning based on a historical mining multi-source heterogeneous data set; the prediction module is used for inputting the high-dimensional fusion feature set into an AI working condition-process linkage regulation model, outputting the judgment result of the total element accurate content and the compound component of the materials in each current picking stage through feature matching and reinforcement learning reasoning, and adapting to the optimal process parameter regulation scheme of the current working condition; The deviation determining module is used for carrying out parameter adjustment on the mining and selecting equipment according to an optimal process parameter regulation and control scheme and determining a deviation coefficient based on a preset process regulation and control effect; The model updating module is used for feeding deviation data back to the AI working condition-process linkage regulation and control model to iterate model parameters when the deviation coefficient exceeds a preset allowable range, and meanwhile supplementing a newly acquired multi-source heterogeneous data set to the historical data set, and updating a working condition adaptation knowledge base of the AI working condition-process linkage regulation and control model through transfer learning.

Description

Mine mining process digital monitoring optimization method and system based on AI Technical Field The invention relates to the technical field of digital monitoring and optimizing, in particular to a digital monitoring and optimizing method and system for a mine mining and selecting process based on AI. Background The digital monitoring and intelligent process control of the mining and selecting process are core links of improving the resource utilization rate and reducing the energy consumption, however, in the actual mining and selecting production scene of the mine, the factors such as data acquisition mode, analysis technical limitation, model construction logic and the like are mutually restricted, and the following defects are exposed in the prior art: Firstly, the existing data lacks a uniform synchronous acquisition mechanism, a single multi-focusing link is not realized, and the whole flow coverage of mining, mineral dressing and tailing treatment is not realized, so that the problems of space-time dislocation and dimension incomplete of the data exist, and a fragmented data island is formed. And secondly, the monitoring analysis depends on the traditional offline detection means, the online real-time detection technology support is lacked, the data processing is mostly manual arrangement and summarization, and the automatic time-space alignment, outlier rejection and feature extraction processes are not realized, so that key information feedback lag such as mineral components, working condition changes and the like is caused, and a timely basis cannot be provided for process regulation. Finally, the prior art only completes a one-way flow of parameter adjustment-effect monitoring, does not construct a deviation feedback mechanism, cannot reversely input deviation data of an actual regulation effect and a preset target into a model for parameter iteration, does not continuously supplement new acquired data to update training samples, and causes that the model cannot adapt to the change of conditions, and lacks full-flow closed-loop optimization logic. Therefore, a digital monitoring optimization scheme combining multisource synchronous acquisition, AI feature fusion extraction, cross-mining-area adaptation model and closed-loop iteration mechanism is needed to solve the technical bottleneck and realize real-time monitoring, accurate regulation and continuous optimization of the mining and selecting process. Disclosure of Invention Aiming at the defects of the prior art, the invention provides the mine mining process digital monitoring and optimizing method and system based on AI, which solve the problems of fragmentation of data acquisition, lag in monitoring and analysis, lack of accurate basis for process regulation and control, poor adaptability across mining areas and no closed-loop optimizing mechanism in the traditional mine mining process. The invention aims to realize the purpose by adopting the following technical scheme that the mine mining and selecting process digital monitoring and optimizing method based on AI comprises the following steps: Acquiring laser-induced breakdown spectrum data of minerals of each link, and synchronously acquiring time sequence data of the picking working condition and resource output data to form an original multi-source heterogeneous data set; Carrying out space-time alignment and pretreatment on an original multi-source heterogeneous data set, extracting element spectrum feature vectors, working condition dynamic feature vectors and cross-correlation feature vectors of the element spectrum feature vectors and the working condition dynamic feature vectors based on a fusion extraction model, and outputting a high-dimensional fusion feature set; Based on a historical mining multisource heterogeneous data set, adopting an algorithm of fusion of transfer learning and deep reinforcement learning to construct an AI working condition-process linkage regulation model; inputting the high-dimensional fusion feature set into an AI working condition-process linkage regulation model, and outputting the accurate content of all elements and the judgment result of the compound components of the materials at each stage of current picking through feature matching and reinforcement learning reasoning, and adapting to the optimal process parameter regulation scheme of the current working condition; according to the optimal technological parameter regulation scheme, parameter adjustment is carried out on the picking equipment, and a deviation coefficient is determined based on a preset technological regulation effect; And simultaneously supplementing the newly acquired multi-source heterogeneous data set to a historical data set, and updating a working condition adaptation knowledge base of the AI working condition-process linkage regulation model through transfer learning. Mine mining process digital monitoring optimizing system based on AI includes: The data set acquisition