Rock strength and crushability real-time prediction method and system

Abstract

The invention discloses a rock strength and crushability real-time prediction method and system, and the method comprises the steps: obtaining while-drilling data, while-digging data and while-mining data in an existing rock-soil construction process, and determining rock uniaxial compressive strength data according to an extracted sample; preprocessing the obtained data to construct a while-drilling data set, a while-digging data set and a while-mining data set; iterative parameter adjustment is performed on the lightweight gradient lifter based on a hiking optimization algorithm, and a rock strength and crushability prediction model is constructed and trained based on the hiking optimization algorithm and the lightweight gradient lifter; according to the method, a rock strength and crushability prediction model while drilling, a rock strength and crushability prediction model while digging and a rock strength and crushability prediction model while mining are obtained, and real-time rock strength and crushability prediction is performed on the process while drilling, the process while digging and the process while mining by adopting the obtained prediction models. The method is higher in reliability, better in accuracy and better in applicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of geotechnical construction, and particularly relates to a method and system for real-time prediction of rock strength and breakability. Background Art

[0002] With the development of economy and technology, there are more and more mining and tunnel boring projects. In the process of mining and tunnel boring, rock fragmentation is still the core task. In the deep hard rock environment, non-blasting mechanized mining has become the current mainstream rock-breaking technology due to its advantages such as continuous operation, excellent construction quality, and small environmental disturbance.

[0003] In the process of non-blasting mechanized mining, the prediction of rock strength and breakability has always been an important link in the mining process. Therefore, the real-time prediction of rock strength and breakability has always been one of the research focuses of researchers.

[0004] Current rock strength and breakability prediction schemes mainly include laboratory tests and empirical model predictions. The laboratory prediction scheme is to conduct corresponding experimental tests on the rock samples collected during the mining process in the laboratory, so as to determine the previous rock strength and breakability; although this kind of scheme has high accuracy, its process is cumbersome and the cost is high, and it is difficult to meet the real-time requirements. The empirical model prediction scheme is based on the empirical formula method and regression analysis method; however, this kind of scheme lacks universality and consistency, and cannot be applied to the current complex engineering environment and geological conditions, and its reliability and accuracy cannot be guaranteed either. Summary of the Invention

[0005] One of the purposes of the present invention is to provide a method for real-time prediction of rock strength and breakability with high reliability, good accuracy and wide application range.

[0006] Another purpose of the present invention is to provide a system for realizing the method for real-time prediction of rock strength and breakability.

[0007] The method for real-time prediction of rock strength and breakability provided by the present invention includes the following steps: S1. Obtain the data while drilling, while tunneling and while mining during the rock engineering construction process, and determine the uniaxial compressive strength data of the rock according to the extracted samples; S2. Preprocess the data obtained in step S1 to construct a data set while drilling, a data set while tunneling and a data set while mining; S3. Iteratively tune the parameters of the lightweight gradient booster based on the hiking optimization algorithm, and construct a rock strength and breakability prediction model based on the hiking optimization algorithm and the lightweight gradient booster; S4. Use the dataset constructed in step S2 to train the prediction models for rock strength and breakability constructed in step S3, so as to obtain the prediction models for in - drill rock strength and breakability, the prediction models for in - tunneling rock strength and breakability, and the prediction models for in - mining rock strength and breakability; S5. Use the prediction models obtained in step S4 to respectively predict the rock strength and breakability in real - time during the in - drill process, the in - tunneling process, and the in - mining process.

[0008] The said step S1 specifically includes the following steps: The in - drill data obtained includes the thrust force F, torque T1, rotational speed N, penetration rate V, and specific energy of rock breaking SE; The in - tunneling data obtained includes the cutterhead thrust force F, cutterhead torque T2, cutterhead rotational speed RPM, penetration rate PR, and specific energy of rock breaking SE; The in - mining data obtained includes the cutting angle , feed angle , rake angle , clearance angle , average cutting force MFC, peak chip force PFC, specific energy of rock breaking SE, and instantaneous cutting rate ICR; According to the extracted samples, conduct uniaxial compression tests to obtain the uniaxial compressive strength UCS of the rock.

[0009] The said step S2 specifically includes the following steps: Pre - process the data obtained in step S1; the said pre - processing includes data cleaning and data filling; Conduct the classification of breakability levels: If and , then the breakability level is easy to break; If and , then the breakability level is medium - difficult to break; If and , then the breakability level is relatively difficult to break; If and , then the breakability level is extremely difficult to break; Use the in - drill data as the input, and the uniaxial compressive strength UCS of the rock and the breakability level as the output data to construct the in - drill dataset; Use the in - tunneling data as the input, and the uniaxial compressive strength UCS of the rock and the breakability level as the output data to construct the in - tunneling dataset; Use the in - mining data as the input, and the uniaxial compressive strength UCS of the rock and the breakability level as the output data to construct the in - mining dataset.

[0010] The described step S3 includes the following steps: Build a prediction model for rock strength and breakability using a lightweight gradient booster; Use the hiking optimization algorithm to iteratively tune the parameters of the lightweight gradient booster to improve its performance; The described hiking optimization algorithm uses Tent chaotic mapping to generate the initial population to improve the model's early exploration ability; uses a stepped inertia weight strategy to balance the relationship between global exploration and local exploration; uses Lévy flight and Logistic chaotic perturbation to enhance the individual's exploration ability and avoid premature convergence to local optimal solutions; and finally uses a simulated annealing strategy to enhance the model's ability to escape local extreme value traps.

[0011] The step of building a prediction model for rock strength and breakability using a lightweight gradient booster specifically includes the following steps: The input D of the model is expressed as , where is the ii-th feature data of the input, is the true label; N is the total number of input data; The model is initialized using the following formula: In the formula is the predicted value during model initialization; c is the initial prediction constant; is the set loss function; The model is iteratively updated; each round of update includes the following steps: During the m-th iteration, the negative gradient of each sample is calculated using the following formula: In the formula is the negative gradient of the i-th sample during the m-th iteration; is the partial derivative of the loss function with respect to the predicted value ; is the set loss function used to calculate the difference between the predicted value and the true label; is the predicted value of the i-th sample during the m-th iteration; is the model predicted value during the m-1-th iteration; Use the data to fit a new decision tree and divide the input features into J leaf node regions ; For each leaf node region , calculate the best predicted value that minimizes the loss function, expressed as: In the formula is the set loss function; For the current leaf node region The best prediction adjustment value calculated; by optimizing the loss function, the prediction value most suitable for this leaf node region is calculated; Adopt a learning rate Control the step size and update the model, expressed as: In the formula Is an indicator function used to indicate whether the sample x belongs to the leaf node , if the sample x belongs to the leaf node Then , if the sample x does not belong to the leaf node Then ; Iteratively update the model until the set maximum number of iterations is reached, or the iteration error decrease amplitude is less than the set threshold. At this time, the final model is obtained, expressed as: In the formula Is the final prediction model; Is the decision tree learned in the m-th iteration; Is the learning rate; Is the prediction model at initialization.

[0012] The above-mentioned iterative parameter tuning of the lightweight gradient booster using the hiking optimization algorithm specifically includes the following steps: (1) Population initialization based on Tent chaotic mapping: Generate the initial population using Tent chaotic mapping, expressed as: In the formula Is the position of the current population; After Bernoulli shift transformation, get , where mod is the modulo operation; The population position is initialized as: In the formula Is the position of the i-th individual at initialization; Is the initial value of the j-th dimension; Is the end value of the j-th dimension; Initialize the speed using the Tobler hiking function, expressed as: In the formula Is the speed of the traveler at iteration i or time t; Is the slope, and , Is the angle of the slope; (2) Fitness calculation: Calculate the fitness value using the following formula: In the formula Is the fitness value of the i-th individual; is the set loss function; Determine the current optimal position according to the fitness value: In the formula is the current optimal position; (3) Based on the stepped inertia weight strategy, Levy flight, and Logistic chaotic perturbation, update the velocity of the traveler: The stepped inertia weight is expressed as: In the formula is the inertia weight at the current time step t; is the maximum inertia weight; is the minimum inertia weight; t is the current iteration number; is the maximum iteration number; is the parameter for controlling the transition; Use the following formula to update the velocity of the traveler: In the formula is the dynamic adjustment factor of the i-th traveler at the t-th iteration; is the adjustment factor; is the optimal position; is the control factor; is the Levy flight variable, and , is the exponential parameter of the Levy distribution; is the perturbation coefficient; is the Logistic chaotic perturbation variable, and , is the chaotic control parameter, is the chaotic value of the previous round; (4) Update the position of the traveler: The formula for updating the position of the traveler is expressed as ; (5) Introduce the simulated annealing strategy: Whether the new solution is accepted is judged according to the following probability function: In the formula, P is the acceptance probability; is the objective function value of the new solution; is the objective function value of the current best solution; Use the following formula to update the annealing temperature: In the formula, T is the current temperature; is the initial temperature; t is the iteration number; (6) Output the optimal solution: If the set maximum iteration number is reached, output the optimal solution; the optimal solution is the current and .

[0013] The training described in step S4 specifically includes the following steps: For each dataset, divide the dataset into several subsets. Each time, select one subset as the test set and the remaining subsets as the training set, and perform five-fold cross-validation; For each trained model, conduct performance comparison and analysis: Select NN evaluation metrics and set the scoring rule for the model as: In a certain evaluation metric, the model with the best performance gets NN points, and the remaining models get scores from NN - 1 to 1 in descending order of performance; Finally, select the model with the highest total score in the NN evaluation metrics as the final model to complete the training process.

[0014] The described step S5 specifically includes the following steps: Calculate the weighted average of rock strength using the following formula : In the formula is the rock strength predicted while drilling; is the weight coefficient of the rock strength predicted while drilling; is the rock strength predicted while driving; is the weight coefficient of the rock strength predicted while driving; is the rock strength predicted while mining; is the weight coefficient of the rock strength predicted while mining; Calculate the weighted average of rock breakability using the following formula : In the formula is the rock breakability value predicted while drilling; is the weight coefficient of the rock breakability value predicted while drilling; is the rock breakability value predicted while driving; is the weight coefficient of the rock breakability value predicted while driving; is the rock breakability value predicted while mining; is the weight coefficient of the rock breakability value predicted while mining; During final prediction, the larger the value of the weighted average of rock strength , the higher the rock strength; the larger the value of the weighted average of rock breakability , the better the rock breakability.

[0015] The present invention also provides a system for implementing the real-time prediction method of rock strength and breakability, including a data acquisition module, a data processing module, a model construction module, a model training module, and a prediction module; the data acquisition module, the data processing module, the model construction module, the model training module, and the prediction module are connected in series in sequence; the data acquisition module is used to acquire the data while drilling, the data while tunneling, and the data while mining during the construction of the existing rock project, determine the uniaxial compressive strength data of the rock according to the extracted samples, and upload the data information to the data processing module; the data processing module is used to preprocess the acquired data according to the received data information to construct a dataset while drilling, a dataset while tunneling, and a dataset while mining, and upload the data information to the model construction module; the model construction module is used to iteratively tune the parameters of the lightweight gradient booster based on the hiking optimization algorithm according to the received data information, and construct a prediction model for rock strength and breakability based on the hiking optimization algorithm and the lightweight gradient booster, and upload the data information to the model training module; the model training module is used to train the constructed prediction model for rock strength and breakability by using the constructed dataset according to the received data information to obtain a prediction model for rock strength and breakability while drilling, a prediction model for rock strength and breakability while tunneling, and a prediction model for rock strength and breakability while mining, and upload the data information to the prediction module; the prediction module is used to respectively perform real-time prediction of rock strength and breakability during the process of drilling while, tunneling while, and mining while according to the received data information by using the obtained prediction model.

[0016] The real-time prediction method and system for rock strength and breakability provided by the present invention construct a training dataset by processing the data while drilling, the data while tunneling, and the data while mining during the construction process, and train and implement a prediction model combined with the hiking optimization algorithm and the lightweight gradient booster. Therefore, the present invention can not only realize the real-time prediction of rock strength and breakability, but also has higher reliability, better accuracy, and better applicability. Brief Description of the Drawings

[0017] Figure 1 It is a schematic flow chart of the method of the present invention.

[0018] Figure 2 It is a schematic diagram of the functional modules of the system of the present invention. Detailed Embodiments

[0019] As Figure 1 shown is a schematic flow chart of the method of the present invention: The real-time prediction method for rock strength and breakability disclosed by the present invention includes the following steps: S1. Obtain the data while drilling, while tunneling, and while mining during the construction of rock engineering, and determine the uniaxial compressive strength data of the rock according to the extracted samples; specifically, it includes the following steps: The data obtained while drilling includes the thrust force F, torque T1, rotational speed N, penetration rate V, and specific energy of rock breaking SE; The data obtained while tunneling includes the cutterhead thrust force F, cutterhead torque T2, cutterhead rotational speed RPM, penetration rate PR, and specific energy of rock breaking SE; The data obtained while mining includes the cutting angle , feed angle , rake angle , clearance angle , average cutting force MFC, peak chip force PFC, specific energy of rock breaking SE, and instantaneous cutting rate ICR; According to the extracted samples, conduct a uniaxial compression test to obtain the uniaxial compressive strength UCS of the rock.

[0020] S2. Preprocess the data obtained in step S1 to construct a dataset while drilling, a dataset while tunneling, and a dataset while mining; specifically, it includes the following steps: Preprocess the data obtained in step S1; the preprocessing includes data cleaning and data filling; Completely obtain the full-hole core during the operation as the core sample, and then analyze the core sample in combination with geological professional knowledge to obtain information on rock type, integrity, basic structure and texture, weathering degree, and core recovery rate. Catalog the drilled core and record information such as rock type, integrity, basic structure and texture, weathering degree, and core recovery rate to obtain the final catalog data; According to the specific energy of rock breaking and the uniaxial compressive strength, divide the breakability grade: If and , the breakability grade is easy to break; If and , the breakability grade is medium difficulty to break; If and , the breakability grade is difficult to break; If and , the breakability grade is extremely difficult to break; Use the data while drilling as the input, and the uniaxial compressive strength UCS of the rock and the breakability grade as the output data to construct a dataset while drilling; The statistical data of the dataset while drilling constructed by the present invention is shown in Table 1: Table 1 Statistical Table of the Database While Drilling

[0021] Using the data obtained during excavation as input and the uniaxial compressive strength UCS of the rock and the breakability grade as output data, a dataset during excavation is constructed; For the dataset during excavation constructed in the present invention, its statistical data is shown in Table 2: Table 2 Statistical table of the database during excavation

[0022] Using the data obtained during mining as input and the uniaxial compressive strength UCS of the rock and the breakability grade as output data, a dataset during mining is constructed; For the dataset during mining constructed in the present invention, its statistical data is shown in Table 3: Table 3 Statistical table of the database during mining

[0023] Finally, three datasets are constructed for subsequent model training.

[0024] S3. Iteratively tune the parameters of the lightweight gradient booster based on the hiking optimization algorithm, and construct a rock strength and breakability prediction model based on the hiking optimization algorithm and the lightweight gradient booster; the method includes the following steps: Use the lightweight gradient booster to construct a rock strength and breakability prediction model; specifically, it includes the following steps: The lightweight gradient booster is an ensemble learning method that combines weak classifiers (decision trees) and Boosting technology; its core idea is to gradually reduce the training error so that multiple weak classifiers gradually improve the prediction ability of the overall model; The input D of the model is expressed as , where is the ii-th feature data of the input, is 's true label; N is the total number of input data; the goal is to find a function to minimize the loss function; The following formula is used for model initialization: In the formula is the predicted value during model initialization; c is the initial prediction constant; is the set loss function, usually the gap between the predicted value and the true label; Iteratively update the model; each round of update includes the following steps: During the m-th iteration, calculate the negative gradient of each sample using the following formula: In the formula is the negative gradient of the i-th sample during the m-th iteration; is the loss function with respect to the predicted value Partial derivative; is a set loss function for calculating the difference between the predicted value and the true label; is the predicted value of the i-th sample in the m-th iteration; is the model predicted value in the (m - 1)-th iteration; Adopt data to fit a new decision tree and divide the input features into J leaf node regions ; For each leaf node region calculate the best predicted value that minimizes the loss function , expressed as: In the formula is the set loss function; is the best predicted adjustment value calculated in the current leaf node region ; By optimizing the loss function, the predicted value most suitable for this leaf node region is calculated; Adopt the learning rate to control the step size and update the model, expressed as: In the formula is an indicator function used to represent whether the sample x belongs to the leaf node , if the sample x belongs to the leaf node then , if the sample x does not belong to the leaf node then ; Iteratively update the model until the set maximum number of iterations is reached, or the iteration error decrease amplitude is less than the set threshold. At this time, the final model is obtained, expressed as: In the formula is the final prediction model, the model output obtained after M rounds of iteration; is the decision tree (base learner) learned in the m-th iteration; is the learning rate used to control the influence of each round of iteration on the final model; is the prediction model at initialization.

[0025] In the lightweight gradient boosting model, there are four core parameters that jointly determine its performance and generalization ability: max_depth controls the maximum depth of each decision tree, restricting the model's complexity to prevent overfitting; n_estimators represents the number of iterations of the model, that is, the number of decision trees in the ensemble model, affecting the learning ability and computational cost; learning_rate (learning rate) is used to adjust the contribution of each tree to the final prediction. A smaller learning rate usually requires more n_estimators to achieve the desired effect, while a larger learning rate may lead to unstable training; min_samples_split (minimum number of samples for splitting) determines the minimum number of samples required for internal nodes to be divided. A larger value helps prevent overfitting but may reduce the flexibility of the model, while a smaller value will make the tree deeper, increasing complexity and computational cost. These four parameters cooperate with each other and are crucial for the performance of the lightweight gradient boosting model. Therefore, in practical applications, reasonable tuning is required according to specific data to achieve the best balance between fitting ability and generalization ability.

[0026] The hiking optimization algorithm is used to iteratively tune the parameters of the lightweight gradient booster to improve its performance. For the described hiking optimization algorithm, Tent chaotic mapping is used to generate the initial population to improve the model's early exploration ability; a stepped inertia weight strategy is used to balance the relationship between global exploration and local exploration; Levy flight and Logistic chaotic perturbation are used to enhance the exploration ability of individuals and avoid premature convergence to local optimal solutions; finally, a simulated annealing strategy is adopted to enhance the model's ability to escape from local extreme value traps.

[0027] The improved hiking optimization algorithm provided by the present invention has significant advantages in terms of convergence speed, stability, and global optimization ability, and can be more effectively applied to complex optimization problems.

[0028] Specifically, the implementation includes the following steps: (1) Initialize the population based on Tent chaotic mapping: Use Tent chaotic mapping to generate the initial population, expressed as: In the formula is the position of the current population; After Bernoulli shift transformation, we get , where mod is the modulo operation, used to ensure that the generated population positions are within the range; The population positions are initialized as: In the formula is the position of the i-th individual during initialization; is the initial value of the j-th dimension; is the end value of the j-th dimension; The Tobler hiking function is used for speed initialization, expressed as: In the formula is the speed of the traveler at iteration i or time t; is the slope, and , is the angle of the slope, ranging from to; (2) Perform fitness calculation: The fitness value is calculated using the following formula: In the formula is the fitness value of the i-th individual; is the set loss function, used to measure the difference between the model prediction value and the true label; Based on the fitness value, determine the current optimal position: In the formula is the current optimal position; (3) Based on the stepped inertia weight strategy, Lévy flight, and Logistic chaotic perturbation, update the traveler's speed: The stepped inertia weight is expressed as: In the formula is the inertia weight at the current time step t; is the maximum inertia weight; is the minimum inertia weight; t is the current iteration number; is the maximum iteration number; is a parameter for controlling the transition, used to determine the critical point of the change in the inertia weight; The following formula is used to update the traveler's speed: In the formula is the dynamic adjustment factor of the i-th traveler at the t-th iteration. As the iteration progresses, it may adjust the speed update step of the traveler to enhance the flexibility of the search; is the adjustment factor, used to control the amplitude of the speed update of the i-th traveler at the t-th iteration, and it can be used to guide the traveler to converge to the optimal position; is the optimal position, representing the currently found optimal solution or position; is the control factor, usually used to adjust the weight of certain influencing factors and control the intensity of the speed update; is the Lévy flight variable, and , is the exponential parameter of the Lévy distribution (generally taking values in to); is the perturbation coefficient; is the Logistic chaotic perturbation variable, and , is the chaos control parameter, is the chaos value of the previous round; (4)Update the traveler's position: The arithmetic expression for updating the traveler's position is expressed as ; (5)Introduce the simulated annealing strategy: Whether the new solution is accepted is judged according to the following probability function: where P is the acceptance probability; is the objective function value of the new solution; is the objective function value of the current best solution; The core of simulated annealing lies in the gradual decrease of temperature, simulating the cooling process in the metal annealing process; the temperature controls the search range of the algorithm, widely exploring at high temperatures and focusing on local search near the current solution at low temperatures; as the temperature decreases, the system gradually reduces the probability of accepting poor solutions and finally converges to the global optimal solution; The following arithmetic expression is used to update the annealing temperature: where T is the current temperature; is the initial temperature; t is the number of iterations; (6)Output the optimal solution: If the set maximum number of iterations is reached, the optimal solution is output; the optimal solution is the current and .

[0029] S4. Use the dataset constructed in step S2 to train the rock strength and breakability prediction models constructed in step S3 to obtain the rock strength and breakability prediction models while drilling, the rock strength and breakability prediction models while tunneling, and the rock strength and breakability prediction models while mining; The training process specifically includes the following steps: During training, use the hiking optimization algorithm to optimize the hyperparameter configuration of the lightweight gradient booster, including but not limited to the depth of the tree, the learning rate, the subsample ratio, etc.; by continuously adjusting the hyperparameters, ensure that the lightweight gradient booster model has better robustness and generalization ability in prediction performance; For each dataset, divide the dataset into several subsets, select one subset as the test set each time, and select the remaining subsets as the training set for five-fold cross-validation; For each trained model, conduct performance comparison and analysis: Select NN evaluation indicators (which can include the coefficient of determination ( ), mean absolute error (MAE), root mean square error RMSE, variance explanation rate (VAF), etc.), and set the scoring rule of the model as: In a certain evaluation index, the best-performing model gets NN points, and the remaining models get scores from NN - 1 to 1 in descending order of performance; Finally, select the model with the highest total score in the NN evaluation indexes as the final model to complete the training process.

[0030] S5. Use the prediction model obtained in step S4 to predict the rock strength and breakability in real time during the drilling, tunneling, and mining processes respectively; specifically, it includes the following steps: Calculate the weighted average of rock strength using the following formula : In the formula is the rock strength predicted during drilling; is the weight coefficient of the rock strength predicted during drilling; is the rock strength predicted during tunneling; is the weight coefficient of the rock strength predicted during tunneling; is the rock strength predicted during mining; is the weight coefficient of the rock strength predicted during mining; Calculate the weighted average of rock breakability using the following formula : In the formula is the rock breakability value predicted during drilling; is the weight coefficient of the rock breakability value predicted during drilling; is the rock breakability value predicted during tunneling; is the weight coefficient of the rock breakability value predicted during tunneling; is the rock breakability value predicted during mining; is the weight coefficient of the rock breakability value predicted during mining; During the final prediction, the larger the value of the weighted average of rock strength , the higher the rock strength; the larger the value of the weighted average of rock breakability , the better the breakability of the rock; In specific implementation, a preferred solution is , , , and its value-taking rule is: In the case of a pure drilling scenario, the weight coefficients of the prediction results of the drilling data, tunneling data, and mining data are set to 1, 0, and 0 respectively; For a pure tunneling scenario, the weight coefficients of the prediction results of the drilling data, tunneling data, and mining data are set to 0, 1, and 0 respectively; For the pure mining scenario, the weight coefficients of the prediction results of the data while drilling, the data while tunneling, and the data while mining are set to 0, 0, and 1 respectively; For the scenario where drilling and tunneling coexist, the weight coefficients of the prediction results of the data while drilling, the data while tunneling, and the data while mining are set to 0.5, 0.5, and 0 respectively; For the scenario where drilling and mining coexist, the weight coefficients of the prediction results of the data while drilling, the data while tunneling, and the data while mining are set to 0.5, 0, and 0.5 respectively; For the scenario where mining and tunneling coexist, the weight coefficients of the prediction results of the data while drilling, the data while tunneling, and the data while mining are set to 0, 0.5, and 0.5 respectively; For the scenario where the three operation modes of drilling, tunneling, and mining coexist, the weight coefficients of the prediction results of the data while drilling, the data while tunneling, and the data while mining are set to 1 / 3, 1 / 3, and 1 / 3 respectively; During specific implementation, in the non-blasting mechanized mining process, there are respectively a drilling process, a tunneling process, and a mining process, and these three processes are independent of each other; therefore, during specific application, for the drilling process, a prediction model for the in-situ rock strength and breakability while drilling is used for prediction, for the tunneling process, a prediction model for the in-situ rock strength and breakability while tunneling is used for prediction, and for the mining process, a prediction model for the in-situ rock strength and breakability while mining is used for prediction; the three prediction models are independent of each other; for different application scenarios, a weighted average algorithm is applied for comprehensive evaluation to obtain the in-situ rock strength and rock breakability values; Workers can adjust the construction parameters and construction plans of the corresponding process according to the prediction results of the in-situ rock strength and breakability obtained from each prediction model and the comprehensive evaluation. Specifically, for the drilling process, the bit type, drilling speed, and drilling angle are adjusted according to the prediction results; during the tunneling process, the working parameters of the roadheader, such as speed, cutterhead rotation speed, and thrust magnitude, are adjusted to improve the tunneling efficiency and prevent equipment overload; during the mining process, the operation mode of the mining equipment is adjusted according to the strength and breakability of the rock, the mining sequence is optimized, and special auxiliary equipment is used. In terms of equipment control, the operating parameters of the equipment are automatically adjusted using real-time data to ensure equipment load balance and extend the service life. The construction plan is flexibly adjusted according to different rock stratum characteristics to ensure efficient and safe operation, and the construction strategy is optimized through real-time monitoring and feedback throughout the process to ensure the smooth progress of the non-blasting mechanized mining process and the maximization of resource utilization.

[0031] The following further illustrates the method of the present invention in combination with a comparative example: Real-time prediction of in-situ rock strength and drillability based on the monitoring parameters while drilling: Before performing optimization iteration, set the built-in parameters of the model; preferably, the built-in parameters include the number of loop iterations, population size, value range of parameters to be optimized, etc.; the specific values are shown in Table 4: Table 4 Schematic Table of Model Parameters

[0032] The performance of the model under different population sizes was demonstrated, and according to the proposed scoring method, the optimal population size parameter setting was determined, as shown in Table 5: Table 5 Data Table of Model Performance with Different Population Size Settings (Based on the While-Drilling Database)

[0033] Through iterative calculation, the optimal combination of HOA-LGBM hyperparameter settings is obtained as shown in Table 6 below: Table 6 Schematic Table of Model Parameters Based on the While-Drilling Database

[0034] The performance of the optimal model of the present invention in predicting rock strength was compared with the classical model to highlight the superiority of the present invention; the comparison data are shown in Table 7: Table 7 Schematic Table for Evaluating the Monitoring Effect of Rock Strength Based on the While-Drilling Database

[0035] Among them, the LGBM scheme was proposed by Guoliun Ke in the paper "A Highly Efficient Gradient Boosting Decision Tree" in 2017; the KNN scheme was proposed by Cover and Hart in the paper "Nearest neighbor pattern classification" in 1967; The performance of the optimal model of the present invention in predicting rock breakability was compared with the classical model to highlight the superiority of the present invention; the comparison data are shown in Table 8: Table 8 Schematic Table for Evaluating the Monitoring Effect of Rock Breakability Based on the While-Drilling Database

[0036] It can be seen from Table 7 and Table 8 that the solution of the present invention can effectively predict the breakability of rocks in real time based on the while-drilling data, which is significantly higher than the unoptimized LBGM model and the classical classification KNN model.

[0037] Real-time Prediction of Rock Strength and Excavability Based on While-Excavation Monitoring Parameters: Before performing optimization iteration, set the built-in parameters of the model of the present invention; preferably, the built-in parameters include the number of cyclic iterations, population size, value range of parameters to be optimized, etc., as specifically shown in Table 9: Table 9 Schematic Table of Model Built-in Parameter Settings

[0038] The performance of the model of the present invention under different population numbers is shown, and according to the proposed scoring method, the optimal population number parameter setting is determined, as shown in Table 10: Table 10 Data Table of the Performance of the Model of the Present Invention under Different Population Number Settings (Based on the Database of Excavation Along with Mining)

[0039] Through iterative calculation, the optimal combination of HOA-LGBM hyperparameter settings is obtained, as shown in Table 11 below: Table 11 Schematic Table of Model Parameter Settings of the Present Invention Based on the Database of Excavation Along with Mining

[0040] The performance of the optimal model of the present invention in predicting rock strength is compared with that of the classical model to highlight the superiority of the present invention; the specific comparison data are shown in Table 12: Table 12 Schematic Table for Evaluating the Monitoring Effect of Rock Strength Based on the Database of Excavation Along with Mining

[0041] The performance of the optimal model of the present invention in predicting excavability is compared with that of the classical model to highlight the superiority of the present invention; the specific comparison data are shown in Table 13: Table 13 Schematic Table for Evaluating the Monitoring Effect of Rock Fragmentability Based on the Database of Excavation Along with Mining

[0042] It can be seen from Table 12 and Table 13 that the solution of the present invention can effectively predict the rock fragmentability in real time based on the data of excavation along with mining, which is significantly higher than the unoptimized LBGM model and the classical classification KNN model.

[0043] Real-time Prediction of Rock Strength and Mining Feasibility Based on Parameters of Mining Along with Monitoring: Before performing optimization iteration, set the built-in parameters of the model of the present invention; preferably, the built-in parameters include the number of cyclic iterations, population size, value range of parameters to be optimized, etc.; specifically shown in Table 14: Table 14 Schematic Table of Model Built-in Parameter Settings of the Present Invention

[0044] The performance of the model of the present invention under different population sizes is shown, and according to the proposed scoring method, the optimal parameter settings of the population size are determined, as shown in Table 15 specifically: Table 15 Schematic table of the performance of the model of the present invention under different population size settings (based on the random sampling database)

[0045] Through the iterative calculation of HOA, the optimal combination of hyperparameter settings of the model of the present invention is obtained as shown in Table 16 below: Table 16 Schematic table of the parameter settings of the model of the present invention based on the random sampling database

[0046] The performance of the optimal model of the present invention in predicting rock strength is compared with that of the classical model to highlight the superiority of the present invention; the comparison data is shown in Table 17: Table 17 Schematic table of the evaluation data of the monitoring effect of rock strength based on the random sampling database

[0047] The performance of the optimal model of the present invention in predicting rock strength is compared with that of the classical model to highlight the superiority of the present invention; the specific comparison data is shown in Table 18: Table 18 Schematic table of the evaluation data of the monitoring effect of rock breakability based on the random sampling database

[0048] It can be seen from Table 17 and Table 18 that the solution of the present invention can effectively predict the breakability of rocks in real time based on the random sampling data, which is significantly higher than that of the unoptimized LBGM model and the classical classification KNN model.

[0049] Such as Figure 2The following is a schematic diagram of the functional modules of the system of the present invention: The system for implementing the method for real-time prediction of rock strength and breakability disclosed in the present invention includes a data acquisition module, a data processing module, a model construction module, a model training module, and a prediction module; the data acquisition module, the data processing module, the model construction module, the model training module, and the prediction module are connected in series in sequence; the data acquisition module is used to acquire the data while drilling, the data while tunneling, and the data while mining during the construction of the existing rock engineering, and determine the uniaxial compressive strength data of the rock according to the extracted samples, and upload the data information to the data processing module; the data processing module is used to preprocess the acquired data according to the received data information to construct a dataset while drilling, a dataset while tunneling, and a dataset while mining, and upload the data information to the model construction module; the model construction module is used to iteratively tune the parameters of the lightweight gradient booster based on the hiking optimization algorithm according to the received data information, and construct a rock strength and breakability prediction model based on the hiking optimization algorithm and the lightweight gradient booster, and upload the data information to the model training module; the model training module is used to train the constructed rock strength and breakability prediction model by using the constructed dataset according to the received data information to obtain a rock strength and breakability prediction model while drilling, a rock strength and breakability prediction model while tunneling, and a rock strength and breakability prediction model while mining, and upload the data information to the prediction module; the prediction module is used to respectively perform real-time prediction of the rock strength and breakability during the process of drilling while, the process of tunneling while, and the process of mining while according to the received data information by using the obtained prediction model.

Claims

1. A real-time prediction method for rock strength and breakability, characterized in that It includes the following steps: S1. Obtain the data while drilling, while tunneling, and while mining during the rock engineering construction process, and determine the uniaxial compressive strength data of the rock according to the extracted samples; S2. Preprocess the data obtained in step S1 to construct a dataset while drilling, a dataset while tunneling, and a dataset while mining; S3. Iteratively tune the parameters of the lightweight gradient booster based on the hiking optimization algorithm, and construct a prediction model for rock strength and breakability based on the hiking optimization algorithm and the lightweight gradient booster; S4. Use the datasets constructed in step S2 to train the prediction model for rock strength and breakability constructed in step S3 to obtain a prediction model for rock strength and breakability while drilling, a prediction model for rock strength and breakability while tunneling, and a prediction model for rock strength and breakability while mining; S5. Use the prediction models obtained in step S4 to respectively predict the rock strength and breakability in real time during the drilling process, the tunneling process, and the mining process.

2. The real-time prediction method of rock strength and breakability according to claim 1, characterized in that The step S1 specifically includes the following steps: The data obtained while drilling includes the thrust force F, torque T1, rotational speed N, penetration rate V, and specific energy of rock breaking SE; The data obtained while tunneling includes the cutterhead thrust force F, cutterhead torque T2, cutterhead rotational speed RPM, penetration rate PR, and specific energy of rock breaking SE; The acquired data during cutting include cutting angle , feed angle , rake angle , clearance angle , average cutting force MFC, peak chip force PFC, specific energy SE for rock breaking, and instantaneous cutting rate ICR; According to the extracted samples, conduct a uniaxial compression test to obtain the uniaxial compressive strength UCS of the rock.

3. The real-time prediction method of rock strength and breakability according to claim 2, characterized in that The step S2 specifically includes the following steps: Preprocess the data obtained in step S1; the preprocessing includes data cleaning and data filling; Conduct the division of breakability levels: If and , the breakability level is easy to break; If and , the breakability level is medium difficulty to break; If and , the breakability level is difficult to break; If and , the breakability level is extremely difficult to break; Use the data while drilling as the input, and use the uniaxial compressive strength UCS of the rock and the breakability level as the output data to construct a dataset while drilling; Use the data while tunneling as the input, and use the uniaxial compressive strength UCS of the rock and the breakability level as the output data to construct a dataset while tunneling; Use the data while mining as the input, and use the uniaxial compressive strength UCS of the rock and the breakability level as the output data to construct a dataset while mining.

4. The real-time prediction method of rock strength and breakability according to claim 3, characterized in that The step S3 includes the following steps: Use the lightweight gradient booster to construct a prediction model for rock strength and breakability; Use the hiking optimization algorithm to iteratively tune the parameters of the lightweight gradient booster to improve the performance of the lightweight gradient booster; For the hiking optimization algorithm, use the Tent chaotic map to generate the initial population to improve the early exploration ability of the model; use the stepped inertia weight strategy to balance the relationship between global exploration and local exploration; use Levy flight and Logistic chaotic perturbation to enhance the exploration ability of individuals and avoid premature convergence to local optimal solutions; finally, use the simulated annealing strategy to enhance the ability of the model to escape from local extreme value traps.

5. The real-time prediction method of rock strength and breakability according to claim 4, characterized in that The use of the lightweight gradient booster to construct a prediction model for rock strength and breakability specifically includes the following steps: The input D of the model is represented as , where is the i-th feature data of the input, is 's true label; N is the total number of input data; The model is initialized using the following formula: In the formula is the predicted value during model initialization; c is the initial prediction constant; is the set loss function; Iteratively update the model; each round of update includes the following steps: At the m-th iteration, the negative gradient of each sample is calculated using the following formula: where is the negative gradient of the i-th sample at the m-th iteration; is the partial derivative of the loss function with respect to the predicted value ; is the set loss function, which is used to calculate the difference between the predicted value and the true label; is the predicted value of the i-th sample at the m-th iteration; is the model predicted value at the (m - 1)-th iteration; Adopt data to fit a new decision tree , and divide the input features into J leaf node regions ; For each leaf node region , calculate the best prediction value that minimizes the loss function , expressed as: In the formula is the set loss function; is the best prediction adjustment value calculated in the current leaf node region ; by optimizing the loss function, the prediction value most suitable for this leaf node region is calculated. Adopt a learning rate to control the step size and update the model, expressed as: In the formula is an indicator function used to represent whether the sample x belongs to the leaf node , if the sample x belongs to the leaf node then , if the sample x does not belong to the leaf node then ; Iteratively update the model until the set maximum number of iterations is reached, or the decrease in the iteration error is less than the set threshold. At this point, the final model is obtained and expressed as: In the formula is the final prediction model; is the decision tree learned in the m-th iteration; is the learning rate; is the prediction model at initialization.

6. The real-time prediction method of rock strength and breakability according to claim 5, characterized in that The use of the hiking optimization algorithm to iteratively tune the parameters of the lightweight gradient booster specifically includes the following steps: (1) Initialize the population based on the Tent chaotic map: Generate the initial population using the Tent chaotic map, expressed as: In the formula is the position of the current population; After a Bernoulli shift transformation, we obtain , where mod represents the remainder operation; The initial population position is set as follows: In the formula is the position of the i-th individual at initialization; is the initial value of the j-th dimension; is the end value of the j-th dimension; The Tobler hiking function is used for speed initialization, expressed as: In the formula is the traveler's speed at iteration i or time t; is the slope, and , is the angle of the slope; (2) Calculate the fitness: The fitness value is calculated using the following formula: In the formula is the fitness value of the i-th individual; is the set loss function; Determine the current optimal position according to the fitness value: In the formula is the current optimal position; (3) Update the velocity of the traveler based on the stepped inertia weight strategy, Lévy flight, and Logistic chaotic perturbation: The stepped inertia weight is expressed as: In the formula is the inertia weight at the current time step t; is the maximum inertia weight; is the minimum inertia weight; t is the current iteration number; is the maximum iteration number; is the parameter for controlling the transition; The speed of the traveler is updated using the following equation: where is the dynamic adjustment factor of the i-th traveler at the t-th iteration; is the adjustment factor; is the optimal position; is the control factor; is the Lévy flight variable, and , is the exponential parameter of the Lévy distribution; is the perturbation coefficient; is the Logistic chaotic perturbation variable, and , is the chaotic control parameter, is the chaotic value of the previous round; (4) Update the position of the traveler: The arithmetic expression for updating the traveler's position is expressed as ; (5) Introduce the simulated annealing strategy: Whether the new solution is accepted is determined according to the following probability function: where P is the acceptance probability; is the objective function value of the new solution; is the objective function value of the current best solution; The annealing temperature is updated using the following formula: where T is the current temperature; is the initial temperature; t is the number of iterations; (6) Output the optimal solution: If the set maximum number of iterations is reached, the optimal solution is output; the optimal solution is the current and .

7. The real-time prediction method of rock strength and breakability according to claim 6, characterized in that The training described in step S4 specifically includes the following steps: For each dataset, divide the dataset into several subsets, select one subset as the test set each time, and select the remaining subsets as the training set for five-fold cross-validation; For each trained model, conduct performance comparison and analysis: Select NN evaluation indicators and set the scoring rules for the model as: In a certain evaluation indicator, the model with the best performance gets NN points, and the remaining models get scores from NN - 1 points to 1 point in order of performance from good to bad; Finally, select the model with the highest total score in the NN evaluation indicators as the final model to complete the training process.

8. The real-time prediction method of rock strength and breakability according to claim 7, characterized in that The described step S5 specifically includes the following steps: The weighted average of the rock strength is calculated using the following formula : where is the rock strength predicted while drilling; is the weight coefficient of the rock strength predicted while drilling; is the rock strength predicted while driving; is the weight coefficient of the rock strength predicted while driving; is the rock strength predicted while mining; is the weight coefficient of the rock strength predicted while mining; The weighted average of rock breakability is calculated using the following formula : In the formula is the predicted value of rock breakability while drilling; is the weight coefficient of the predicted value of rock breakability while drilling; is the predicted value of rock breakability while tunneling; is the weight coefficient of the predicted value of rock breakability while tunneling; is the predicted value of rock breakability while mining; is the weight coefficient of the predicted value of rock breakability while mining; At the final prediction, the weighted average value of rock strength The larger the value is, the higher the rock strength is; the weighted average value of rock breakability The larger the value is, the better the rock breakability is.

9. A system for implementing the real-time prediction method of rock strength and breakability according to any one of claims 1 to 8, characterized in that It includes a data acquisition module, a data processing module, a model construction module, a model training module, and a prediction module; the data acquisition module, the data processing module, the model construction module, the model training module, and the prediction module are connected in series in sequence; the data acquisition module is used to acquire the data while drilling, while tunneling, and while mining during the construction process of the existing rock engineering, determine the uniaxial compressive strength data of the rock according to the extracted samples, and upload the data information to the data processing module; The data processing module is used to preprocess the acquired data according to the received data information to construct a dataset while drilling, a dataset while tunneling, and a dataset while mining, and upload the data information to the model construction module; The model construction module is used to iteratively tune the parameters of the lightweight gradient booster based on the hiking optimization algorithm according to the received data information, and construct a rock strength and breakability prediction model based on the hiking optimization algorithm and the lightweight gradient booster, and upload the data information to the model training module; The model training module is used to train the constructed rock strength and breakability prediction model using the constructed dataset according to the received data information to obtain a rock strength and breakability prediction model while drilling, a rock strength and breakability prediction model while tunneling, and a rock strength and breakability prediction model while mining, and upload the data information to the prediction module; The prediction module is used to use the obtained prediction model to respectively predict the rock strength and breakability in real time during the process of drilling, tunneling, and mining.

Images