Stock yard excavation progress prediction and control method based on interpretable integrated machine learning

Abstract

The invention discloses a stockyard excavation progress prediction and control method based on interpretable integrated machine learning, and the method mainly comprises the steps: firstly, building an excavation progress prediction simulation model through simulating an excavation block line production construction technology, and determining an excavation scheme through an orthogonal experiment, calculating and constructing an excavation progress multi-factor prediction data set based on the simulation model; secondly, based on the excavation progress multi-factor prediction data set, an excavation progress prediction machine learning model is trained, and rapid excavation progress prediction is achieved; and finally, the influence of different characteristic factors on the excavation progress is quantified in combination with an SHAP method, excavation progress cause analysis is achieved, and actual construction is guided. The explainable stock ground excavation construction progress prediction model established by the invention can comprehensively and accurately predict the stock ground excavation construction period; the influence of influence factors on the construction period can be quantified, and a stock yard excavation progress management and control suggestion is provided; the influence on the excavation construction progress of the stockyard under the multi-factor coupling effect is considered, and the result of predicting the influence factors is more reliable.

Description

Technical Field

[0001] The present invention relates to water conservancy project construction progress prediction and control, and specifically to a material yard excavation progress prediction and control method based on explainable integrated machine learning. Background Art

[0002] The areas where water conservancy and hydropower projects are carried out are often complex in natural environmental conditions. Water conservancy project construction is affected by factors such as high cold, high earthquake intensity and complex geological conditions. In addition, with the development of hydropower construction in my country and the improvement of construction technology, the scale of water conservancy construction has gradually increased, which has brought great challenges to construction progress management. Material yard excavation is a key component of water conservancy project construction. The excavation progress directly restricts the overall progress of the project and is the focus of water conservancy project construction progress management. [1] .

[0003] The excavation construction of the material yard is a complex system, and the construction progress is affected by many factors, mainly including construction technology, construction resources, construction environment, etc. From the perspective of construction technology, the excavation of the material yard is divided into zones, layers, strips, and blocks, and multiple processes are carried out in a flow-line operation on the excavation blocks, specifically involving site cleaning, drilling, charging and connection, blasting, slag transportation and other cyclic operations. The rules of zoning, layering, striping and block excavation of the material yard, the layout of the excavation surface, and the flow-line operation relationship of the process directly affect the excavation progress; from the perspective of construction resources, mechanical equipment such as excavators, transport vehicles, drilling rigs, and construction personnel are important factors affecting the excavation progress; from the perspective of the construction environment, weather, mechanical status, material source properties and other factors bring complex uncertainties to the excavation construction progress. Therefore, how to establish a prediction model for the impact of multiple factors on the excavation progress of the material yard, and quantitatively analyze the impact of different factors on the excavation progress, is an important challenge to achieve excavation construction progress control.

[0004] Some scholars have conducted relevant research on factors affecting construction progress.

[11] The orthogonal experimental design based on the global comprehensive sensitivity analysis method was used to conduct a multi-factor comprehensive sensitivity analysis on the factors affecting the uncertainty of the construction period.

[12] The combined weighting method based on the binomial coefficient method and the entropy method based on the distance function is used to analyze the factors affecting the excavation period of the pumping and storage project and evaluate the importance of the factors. However, the calculation of the construction period in these methods all uses the empirical formula method, which has the disadvantages of low accuracy and inability to consider complex random effects.

[0005] Construction progress simulation can comprehensively consider the influence of various complex factors to achieve accurate analysis of construction progress. Tong Dawei [4]A practical and feasible real-time control process for the progress and quality of concrete dam construction was established, and the construction progress monitoring and quality monitoring early warning model during the concrete dam construction process was studied. [5] Using data envelopment analysis, a multi-objective comprehensive evaluation model for underground cavern construction progress simulation that couples construction period and robustness was established, and the relative effectiveness of each optimization scheme was analyzed. [6] A construction progress risk analysis method for complex long shaft and long distance water diversion tunnel is proposed, which comprehensively considers the uncertainty of logical relationship and activity time. [7] Based on AutoCAD secondary development technology, a simulation-assisted modeling method for excavation construction was proposed, and a simulation model describing the excavation construction characteristics of a pumped storage power station was established. [8] A simulation model suitable for the construction technology of long-distance water diversion tunnel groups was established, and the entire construction process of long-distance water diversion tunnel groups was simulated using real-time simulation technology. [9] The system simulation technology was used to study the reservoir excavation process and realize the computer simulation of the excavation process.

[10] By using object-oriented methods and computer simulation technology, an object-oriented simulation model for spillway excavation construction was constructed, realizing the simulation calculation of the excavation process.

[11] Comprehensively analyze the factors that affect the excavation period of the inclined tunnel of the pumped storage project, establish an influencing index system for the construction period prediction, and use the principal component analysis method to establish a multivariate linear regression model for predicting the excavation period. Wang Renchao et al.

[12] A method for predicting the construction period of a diversion tunnel project is proposed by using engineering case analogy and fuzzy prediction and evaluation based on causal clustering.

[0006] However, existing progress simulation research lacks inverse analysis of factors affecting construction progress, making it difficult to determine factors affecting construction progress from the source, and unable to provide accurate material yard excavation decision-making management recommendations, limiting the guiding role of simulation models in the actual construction process. With the development of artificial intelligence, explainable machine learning provides a new approach for causal analysis. Explainable ensemble machine learning generally includes the following two steps: building a prediction model and calculating the contribution of features to the model. Among them, XGBoost (eXtreme gradient boosting) is a common method for building prediction models in explainable ensemble machine learning. XGBoost can quickly process problems with large-scale training data and high-dimensional features; it shows high accuracy and robustness on noisy data and missing data; it can output explanations of feature importance rankings to better understand the data; it supports various loss functions and custom objective functions, as well as the processing of missing values. Due to the above advantages of XGBoost, it has achieved good results in many prediction fields. [13-16]. SHAP (Shapley Additive Explanation) is a method for explaining integrated machine learning models. It can be used to explain the contribution of an individual prediction value relative to the benchmark value. SHAP is based on the concept of Shapley Values ​​and uses local interpretability methods to analyze and explain the input features corresponding to each sample. The prediction result can be regarded as the sum of the contributions of each feature, thereby quantifying the importance of each feature. The SHAP method can be used to analyze the marginal effect of one or two features on the prediction results of the machine learning model. It provides a relatively intuitive and accurate way to understand the important features of the prediction model and how they affect the output results of the model. [17-18] SHAP can better understand and explain the behavior of predictive models and reveal the complexity of the relationships within the model, leading to a better understanding of the data and better decision making.

[0007] [References]

[0008] [1] Ma Hongqi, Zhong Denghua, Zhang Zongliang, et al. Key technologies for real-time control of major water conservancy and hydropower projects and their engineering applications [J]. China Engineering Science, 2011, 13(12): 20-27+2.

[0009] [2] Wang Xiaoling, Zhao Mengqi, Hong Kun, et al. Global comprehensive sensitivity analysis of TBM construction period of water diversion tunnel[J]. Journal of Tianjin University (Natural Science and Engineering Technology Edition), 2015, 48(07): 569-577.

[0010] [3] Liu Fuqiang, Lu Zhihang, Lv Chengxin, et al. Analysis of factors affecting excavation period of pumped storage project based on combined weighting method [J]. Hydropower and New Energy, 2017(06):40-43+72. DOI:10.13622 / j.cnki.cn42-1800 / tv.1671-3354.2017.06.011.

[0011] [4] Tong Dawei. Research on real-time control of construction progress and quality of hydropower projects[D]. Tianjin University, 2009.

[0012] [5] Bi Lei. Research on simulation analysis and optimization of construction progress of underground caverns based on uncertainty analysis[D]. Tianjin University, 2015.

[0013] [6] Hong Kun. Research on construction progress risk analysis and simulation optimization of complex long shaft and long distance water diversion tunnel[D]. Tianjin University, 2016.

[0014] [7] Wang Renchao, Li Shipeng, Xu Yueming, et al. Simulation study on excavation construction of pumped storage power station[J]. Journal of Hydroelectric Engineering, 2018, 37(03): 18-27.

[0015] [8] Hu Lianxing. Research on key technologies of simulation optimization and progress control for the whole construction process of complex long-distance water diversion tunnel groups[D]. Tianjin University, 2013.

[0016] [9] Zhang Yan. Simulation study on the excavation construction process of the upper reservoir of a pumped storage power station[D]. Tianjin University, 2018.

[0017]

[10] Hu Yinghui. Modeling and simulation of spillway excavation construction system of Nuozhadu Hydropower Station[D]. Tianjin University, 2007.

[0018]

[11] Yang Jianzhou, Yang Jianglong, Lu Zhihang, et al. Prediction model of excavation period of water diversion tunnel for pumped storage power station based on principal component analysis [J]. Hydropower and New Energy, 2017(06): 3-5+51. DOI: 10.13622 / j.cnki.cn42-1800 / tv.1671-3354.2017.06.002.

[0019]

[12] Wang Renchao, Zhou Hongli. Research on fuzzy analogy prediction and evaluation method for diversion tunnel construction period[J]. Journal of Yangtze River Scientific Research Institute, 2013, 30(01): 80-85.

[0020]

[13] Liu Bo, Qin Chuan, Ju Ping et al. Short-term bus load forecasting based on fusion of XGBoost and Stacking model[J]. Electric Power Automation Equipment, 2020, 40(03): 147-153. DOI: 10.16081 / j.epae.202002024.

[0021]

[14] Zhang J, Ma

[0022]

[15] CHEN T,HE T.Higgs Boson Discovery with Boosted Trees[C] / / JMLR:Workshop and Conference Proceedings,2015,42:69-80.

[0023]

[16] CHEN T,GUESTRIN C.

[0024]

[17] MANGALATHU S,HWANG SH,JEON J S.Failure mode and effects analysis of RC members based on machine-learning-based SHapley Additive exPlanations(SHAP)approach[J].Engineering Structures,2020:1-10.

[0025]

[18] PARSA AB, MOVAHEDI A, TAGHIPOUR H, et al. Toward safer highways, application of XGBoost and SHAP for real-time accident detection and feature analysis [J]. Accident Analysis and Prevention, 2020, 136 (2020): 1-8. Summary of the invention

[0026] In view of the lack of an explainable excavation progress prediction model in the above-mentioned existing technologies, it is difficult to quantify the impact of different factors on the construction progress, which leads to challenges in the control of excavation construction progress. A method for predicting and controlling the excavation progress of a material yard based on explainable integrated machine learning is proposed.

[0027] In order to solve the above technical problems, the present invention proposes a method for predicting and controlling the excavation progress of a material field based on interpretable integrated machine learning, which mainly includes:

[0028] S1. Construction of multi-factor prediction data set for material field excavation progress based on excavation simulation;

[0029] S2. Construction of excavation progress prediction model based on integrated machine learning;

[0030] S3. Based on the results of the explainable analysis of the factors affecting the material yard excavation progress, a decision on the control of the material yard excavation progress is proposed.

[0031] The content of step S1 includes:

[0032] 1-1) Extract simulation parameters based on parametric modeling, use the three-dimensional model of the material yard and the excavation block data output by parametric cutting of the material yard to calculate and obtain the information of the excavation blocks of each construction unit;

[0033] 1-2) According to meteorological data statistics or engineering experience or construction organization design, the simulation parameters of monthly effective construction days, daily effective construction hours, driving speed, road width and road slope are determined; in the simulation calculation, the driving speed of the dump truck and the road technical standards are taken from "DLT5397-2007 Hydropower Engineering Construction Organization Design Specification";

[0034] 1-3) Based on the CYCLONE model, an excavation discrete event simulation model is established to describe the transition of the excavation state of the excavation face from non-excavation, excavation to excavation completion, and the material yard excavation face is used as the smallest unit to promote the simulation process;

[0035] The material field is divided into layers, strips, and blocks according to the size of the excavation block. A single excavation block is used as the minimum unit to advance the simulation process. The state transfer equation of the simulation process is shown in formula (1):

[0036] T(i,a+1)=T(i,a)+ΔT,a=1,2,...,n (1)

[0037] In formula (1), i is the number of the excavation block in the material field, T(i,a) is the time when the excavation state of the i-th excavation block a is reached, and △T is the time taken from the construction state a to a+1;

[0038] The construction state variables of the excavation block include the current excavation state and the current excavation process. The current excavation process reflects the process of the excavation block at the current simulation moment. The current excavation state includes unexcavated, excavating, and excavation completed. For each working face, the process conversion is judged at each simulation moment. The working face is the core simulation object for completing the excavation construction process. Each working face is combined with a separate excavation block, construction personnel, and mechanical team to realize the advancement and operation of the excavation process. According to the similarity of different process conversions, its construction state variables are divided into: state variables of the drilling and charging connection construction process, and blasting process state variables.

[0039] The state variables of the drilling and charging line construction process include the process plan duration T pl 、The duration of the process is T al , process waiting time T wl When T pl >Tal , it indicates that the process has not been completed.

[0040] T al =T al +Δt (2)

[0041] When T pl <T al When , it indicates that the process is completed and satisfies both equations (3) and (4).

[0042] T al =T al +Δt (3)

[0043] T wl =T wl +Δt (4)

[0044] The blasting time is fixed within a certain time period. The blasting process state variables include the current simulation time T, the start time t1 and the end time t2 of each blasting process, and the blasting process is controlled by formula (5);

[0045] T∈[t1,t2](5)

[0046] 1-4) Clarify the test objectives and factors that affect the test results, construct an orthogonal table, and build a multi-factor prediction data set X for the excavation progress of the material field through orthogonal experiments:

[0047] First, determine the factors that affect the progress of material yard excavation, as follows:

[0048]

[0049]

[0050] Then, select the orthogonal table L M (Q N ) to arrange the orthogonal experiment, L M (Q N ) represents an orthogonal array with N factors and Q levels, M is the number of level combinations, which is also equal to the number of rows in the orthogonal array; the orthogonal array established satisfies the following conditions: the same number of experiments are conducted for each level of any factor, and the same number of experiments are conducted for the level combination of any two factors;

[0051] The content of step S2 includes:

[0052] 2-1) The multi-factor prediction data set X of the material field excavation progress constructed based on the orthogonal test in step 1-4 is divided into a training set and a test set in a ratio of 7:3, and the data is cleaned and preprocessed;

[0053] 2-2) The input of the material field excavation construction progress prediction model based on the XGBOOST model is defined as the material field excavation progress multi-factor prediction data set X, and the output of the excavation construction progress prediction model based on the XGBOOST model is defined as the objective function O bj and excavation progress forecast And set the regression model parameters, including the number of trees, depth, learning rate, and regularization parameter. The larger the number of trees, the easier it is for the material yard construction excavation progress prediction model to overfit. The depth is set to 4-6, the learning rate is set to 0.01-0.2, and the regularization parameter is 1.

[0054] 2-3) Use the training set to build an initial tree for model training and obtain the excavation progress prediction value The real value y obtained by monitoring i Then, in each iteration, a tree is added to fit the residual of the model's last prediction until the model learning process is terminated, thereby forming an iterative residual tree set composed of many tree models.

[0055] The step S3 includes: ranking the single factor overall SHAP values ​​f(xj) by importance, and guiding the actual construction according to the positive and negative influence of the influencing factors with higher importance ranking on the model prediction results.

[0056] Furthermore, the method for predicting and controlling the excavation progress of a material field according to the present invention comprises:

[0057] In step 1-1), the three-dimensional model of the material yard is imported into Rhino7.0, and the three-dimensional model is layered from top to bottom according to the height of the excavation block using the material yard parametric cutting program. The reference points that satisfy the same grid division direction and excavation rules are selected, and a grid division whose length and width match the size of the excavation block is established. The three-dimensional model is automatically divided according to the size of the excavation block. After the model is automatically divided, the information of each excavation block is saved, including the excavation block layer number, excavation block row number, excavation block column number and the volume of the excavation block.

[0058] In step 2-3), the process of each iteration is as follows:

[0059] 2-3-1) Use the training set to train the material field excavation progress prediction model based on the XGBOOST model. The objective function calculation formula of XGBoost is:

[0060]

[0061] In formula (6), m is the total amount of sample data imported into the kth tree; the first term is the loss function, which is used to measure the true value y i and excavation progress prediction The error between them; the second term is the regularization term, which is used for model complexity to prevent overfitting; the complexity of each tree is defined as:

[0062]

[0063] In formula (7), γ is the difficulty of node segmentation, T is the number of leaf nodes, λ is the L2 regularization coefficient to prevent overfitting, and ω is the modulus of the leaf node vector;

[0064] 2-3-2) Use the trained model to predict the excavation progress of the test set and output the predicted value of the excavation progress

[0065]

[0066] In formula (8), K represents all CART trees established; xi represents the characteristics of the i-th sample; f k (x i ) represents the predicted value of the kth tree.

[0067] In step S3, the single factor population SHAP value f(xj) is obtained according to the following steps:

[0068] 3-1) For each set of excavation construction parameters and the corresponding excavation construction progress prediction value Calculate the single factor individual SHAP value f(x ij );

[0069]

[0070] In formula (9), F is the sample x i The complete set of all influencing factors, S is the sample x i The subset formed by any number of influencing factors in the subset S, v(S) is the contribution of the influencing factors included in the subset S, and v(S∪{j})-v(S) is the contribution of influencing factor j to the joint effect;

[0071] 3-2) The influencing factor j of all samples is accumulated and averaged to obtain the single factor overall SHAP value f(x j ),

[0072]

[0073] In step S3, the predicted value of excavation construction progress for each excavation construction influencing factor is calculated. The impact of the impact can be analyzed and displayed visually to assist yard managers in making decisions.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] A material yard excavation construction progress prediction model was established to predict the material yard excavation construction period more comprehensively and accurately. Based on the excavation progress decision-making control of explainable analysis of progress influencing factors, the impact of the influencing factors of the material yard excavation construction progress on the construction period was quantified, and suggestions for material yard excavation progress control were put forward. The impact of multiple factors coupling on the material yard excavation construction progress was considered. In the process of material yard excavation construction progress prediction and causal analysis decision-making, the feature importance ranking of XGBoost and the influencing factors under the SHAP method were simultaneously considered, and the results of predicting the influencing factors were more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow chart of the prediction and control of the excavation progress of the material field based on interpretable integrated machine learning of the present invention;

[0077] Figure 2 It is a visualized schematic diagram of the prediction result (construction period) of the material field excavation progress of the present invention;

[0078] Figure 3 The SHAP value of the entire model is obtained by adding the SHAP values ​​of the excavation construction parameters. It is a visual diagram used to explain the results of the excavation construction progress prediction model to fully understand the process of excavation construction progress prediction and the importance of excavation construction parameters. DETAILED DESCRIPTION

[0079] The design concept of the method for predicting and controlling the progress of material yard excavation based on interpretable integrated machine learning proposed in the present invention is: since the material yard excavation construction is an intricate system, it not only involves the cyclic operation of multiple processes such as site cleaning, drilling, charging and connection, blasting, slag transportation, etc., but also involves multiple aspects such as excavation surface layout, process, mechanical equipment and transportation of excavated materials. Traditional excavation construction simulation is limited to sensitivity analysis, and the positive and negative effects of single characteristic factors on simulation results are unclear. At the same time, it also lacks comprehensive consideration of the effects of multiple factors coupling on simulation results. Therefore, applying interpretable integrated machine learning to conduct causal analysis and decision-making on the progress of material yard excavation construction is an effective means to achieve reasonable control of construction progress. It mainly includes: first, by simulating the construction process of excavation in layers, strips and blocks, a simulation model for excavation progress prediction is established, and multiple groups of excavation schemes are determined through orthogonal experiments. A multi-factor prediction data set for excavation progress is constructed based on the simulation model calculation; second, based on the multi-factor prediction data set for excavation progress constructed by excavation simulation, a machine learning model for excavation progress prediction under the influence of multiple factors is trained to achieve rapid prediction of excavation progress; finally, the SHAP method is used to quantify the impact of different characteristic factors on excavation progress, to achieve causal analysis of excavation progress, and to scientifically put forward management suggestions to guide actual construction.

[0080] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but the following embodiments are by no means intended to limit the present invention in any way.

[0081] The present invention proposes a method for predicting and controlling the excavation progress of a material field based on interpretable integrated machine learning, the steps are as follows: Figure 1 As shown, the specific implementation is as follows:

[0082] Step 1: Construction of a multi-factor prediction dataset for the excavation progress of the material yard based on excavation simulation. First, the simulation parameters are extracted based on parametric modeling, and the excavation block data output by the material yard three-dimensional model and the parametric cutting of the material yard are used to calculate the excavation block volume and excavation block number of each construction unit; secondly, other simulation parameters are determined based on engineering experience or construction organization design, such as the effective construction days per month and the effective construction hours per day; thirdly, an excavation discrete event simulation model is established based on the CYCLONE (Cycle Operation Network) model to describe the transfer of the excavation state of the excavation working face from unexcavated, excavated to completed excavation, and the material yard excavation working face is used as the smallest unit to promote the simulation process; finally, the test objectives that need to be investigated and the factors that affect the test results are clarified, an orthogonal table is constructed, and a data set is constructed through orthogonal experiments.

[0083] 1.1 Extract simulation parameters based on parametric modeling. Import the 3D model of the material yard into Rhino7.0, use the material yard parametric cutting program to layer the 3D model from top to bottom according to the height of the excavation block, select reference points that meet the same grid division direction as the excavation rule, establish a grid division whose length and width match the size of the excavation block, and automatically divide the 3D model according to the size of the excavation block; after the model is automatically divided, save the information of each excavation block, including the excavation block layer number, excavation block row number, excavation block column number and excavation block volume.

[0084] 1.2 Determine other simulation parameters based on engineering experience or construction organization design. The mining of the material field is greatly affected and restricted by climate factors. It is necessary to formulate corresponding shutdown standards based on relevant factors. The simulation parameters of effective construction days per month, effective construction hours per day, driving speed, road width, and road slope should be determined based on meteorological data statistics or engineering experience or construction organization design; in the simulation calculation, the driving speed of the dump truck and the road technical standards are taken from "DLT5397-2007 Hydropower Engineering Construction Organization Design Specifications".

[0085] 1.3 Establish an excavation discrete event simulation model. When studying the simulation of earthwork excavation process, the object of concern is the state changes of each construction object at each time point, such as the cumulative excavation volume, excavation working surface, etc., and these variables are non-continuous changes over time, so they can be studied according to discrete system simulation. An excavation discrete event simulation model is established based on the CYCLONE (Cycle Operation Network) model to describe the transfer of the excavation state of the excavation working surface from non-excavation, excavation to excavation completion. According to the size of the excavation block, the material yard is divided into layers, strips, and blocks, and a single excavation block is used as the minimum unit to promote the simulation process. The state transfer equation of the simulation process is shown in formula (1),

[0086] T(i,a+1)=T(i,a)+ΔT,a=1,2,...,n (1)

[0087] In formula (1), i is the number of the excavation block in the material yard, T(i,a) is the time when the i-th excavation block a is in excavation state, and △T is the time taken from construction state a to a+1.

[0088] The construction state variables of the excavation block include the current excavation state and the current excavation process; the current excavation process reflects the process of the excavation block at the current simulation moment; the current excavation state includes unexcavated, excavating, and excavation completed; for each working face, the process conversion is judged at each simulation moment; the working face is the core simulation object for completing the excavation construction process, and each working face is combined with a separate excavation block, construction personnel and mechanical team to realize the advancement and operation of the excavation process; according to the similarities of different process conversions, its construction state variables are divided into: state variables of drilling and charging connection construction process, and blasting process state variables.

[0089] The state variables of the drilling and charging line construction process include the process plan duration T pl 、The duration of the process is T al , process waiting time T wl .

[0090] When T pl >T al , it indicates that the process is not yet completed.

[0091] T al =T al +Δt (2)

[0092] When T pl <T al When , it indicates that the process has been completed and the process can be switched. However, due to constraints, the process cannot be switched and needs to wait. In this case, there are:

[0093] Tal =T al +Δt (3)

[0094] T wl =T wl +Δt (4)

[0095] Since this simulation fixes the blasting time within a certain time period, the blasting process state variables include the current simulation time T, the specified start time t1 and end time t2 of each blasting process, and the blasting process is controlled by the following formula.

[0096] T∈[t1,t2](5)

[0097] 1.4 Constructing a data set through orthogonal experiments. First, the test objectives and factors affecting the test results are clearly defined. The factors affecting the test results are determined based on experience and theoretical analysis, and these factors are numbered from 1 to 18, as shown in Table 1. Secondly, an orthogonal table is constructed and a multi-factor prediction data set X for the excavation progress of the material field is constructed through orthogonal experiments, that is,

[0098] Choose a suitable orthogonal array L M (Q N ) to arrange orthogonal experiments and infer the best level combination through factor analysis. M (Q N ) represents an orthogonal array with N factors and Q levels, M is the number of level combinations, which is also equal to the number of rows in the orthogonal array. The constructed orthogonal array must meet two requirements: 1. The same number of experiments are performed for each level of any factor; 2. The same number of experiments are performed for any combination of the levels of any two factors.

[0099] Table 1 Factors affecting the progress of material yard excavation

[0100] serial number factor unit serial number factor unit 1 Excavation block length m 10 Drilling machine quantity tower 2 Excavation block width m 11 Comprehensive utilization rate of drilling machinery ％ 3 Excavation block height m 12 Drilling uncertainty buffer time h 4 Site cleaning efficiency <![CDATA[h / m 2 ]]> 13 Deep hole charging efficiency h / pc 5 Hole distance m 14 Shallow hole charging efficiency h / pc 6 Row spacing m 15 Connection efficiency h / pc 7 Measuring pay-off efficiency h / pc 16 Daily fixed blasting time h 8 Drilling length m 17 Explosion duration h 9 Drilling machine efficiency m / h 18 Blasting inspection efficiency h / pc

[0101] Step 2: Construction of excavation progress prediction model based on integrated machine learning. First, the excavation construction data set constructed based on orthogonal experiment is divided into training set and test set, and the data is cleaned and preprocessed. Secondly, the material yard excavation construction progress prediction model is defined, and the regression model parameters, such as the number of trees, depth, learning rate, regularization parameter, etc., are set. Thirdly, the XGBoost model is trained with the training set, and the performance of the material yard excavation construction progress prediction model on the cross-validation data is monitored to adjust the parameters and improve the model performance; then, during the training process, the best parameter combination is selected according to the performance of the model on the validation data set to improve the generalization ability of the model. Finally, the trained model is used to predict the excavation construction progress of the test set, and the prediction results are output.

[0102] 2.1 The excavation construction data set X constructed based on the orthogonal test in the above steps is divided into a training set and a test set, generally in a ratio of 7:3, and the data is cleaned and preprocessed.

[0103] 2.2 Define the input of the material field excavation construction progress prediction model based on the XGBOOST model as the material field excavation progress multi-factor prediction data set X, and define the output of the excavation construction progress prediction model based on the XGBOOST model as the objective function O bj and excavation progress forecast And set the regression model parameters, such as the number of trees, depth, learning rate, regularization parameter, etc. Generally, the larger the number of trees, the easier it is for the material yard construction excavation progress prediction model to overfit. In the present invention, the depth is usually set to 4-6, the learning rate is usually set to 0.01-0.2, and the regularization parameter defaults to 1.

[0104] 2.3 Use the training set to train the XGBoost model and monitor the performance of the material yard excavation construction progress prediction model on the cross-validation data in order to adjust parameters and improve model performance; then during the training process, select the best parameter combination based on the performance of the model on the validation data set to improve the generalization ability of the model and construct an excavation progress prediction agent model based on the XGboost model.

[0105] Use the training set to build an initial tree for model training and obtain the excavation progress prediction value The real value y obtained by monitoring i The residual of the last prediction is then added in each iteration until the model learning process is terminated, thereby forming an iterative residual tree set composed of many tree models.

[0106] The process of each iteration is as follows:

[0107] The training set is used to train the material field excavation progress prediction model based on the XGBOOST model. The objective function calculation formula of XGBoost is:

[0108]

[0109] In formula (6), m is the total amount of sample data imported into the kth tree; the first term is the loss function, which is used to measure the true value y i and excavation progress prediction The error between them; the second term is the regularization term, which is used for model complexity to prevent overfitting; the complexity of each tree is defined as:

[0110]

[0111] In formula (7), γ is the difficulty of node segmentation, T represents the number of leaf nodes, λ is the L2 regularization coefficient to prevent overfitting, and ω is the modulus of the leaf node vector.

[0112] Use the trained model to predict the excavation progress of the test set, build an excavation progress prediction agent model based on the XGboost model, and output the excavation progress prediction value

[0113]

[0114] In formula (8), K represents all CART trees established; xi represents the characteristics of the i-th sample; f k (x i ) represents the predicted value of the kth tree.

[0115] 2.4 Input the test schemes in the orthogonal table into the excavation progress prediction model based on machine learning one by one for simulation calculation, output and count the construction simulation duration, and its visualization is as follows: Figure 2 shown.

[0116] Step 3: Excavation progress decision-making and control based on interpretable analysis of progress influencing factors. First, for each set of excavation construction parameters and the corresponding excavation construction progress prediction results, the SHAP value of each excavation construction influencing factor is calculated; secondly, the degree of influence of each excavation construction influencing factor on the excavation construction progress prediction results is calculated and visualized to understand the process of excavation construction progress prediction and the importance of excavation construction influencing factors; finally, the SHAP values ​​of the excavation construction influencing factors are added to obtain the SHAP value of the entire model, which is used to explain the results of the excavation construction progress prediction model. According to the SHAP interpretation results, scientific decisions on the excavation progress control of the material yard are proposed.

[0117] 3.1 For each set of excavation construction parameters and the corresponding excavation construction progress prediction results Calculate the SHAP value of each excavation simulation parameter. Let the i-th sample be x i , the jth feature of the i-th sample is x ij , the model's predicted value for this sample is y i , the baseline of the entire model (the mean of the prediction results when the model does not have any input features) is y base ,f(x ij ) is the SHAP value of the jth feature of the i-th sample, then the SHAP value obeys the following formula:

[0118] y i =y base +f(x i1 )+f(x i2 )+…+f(xij )

[0119] For each set of excavation construction parameters and the corresponding excavation construction progress prediction value Calculate the single factor individual SHAP value f(x ij );

[0120]

[0121] In formula (9), F is the sample x i The complete set of all influencing factors, S is the sample x i The subset formed by any number of influencing factors in the subset S, v(S) is the contribution of the influencing factors included in the subset S, and v(S∪{j})-v(S) is the contribution of influencing factor j to the joint effect;

[0122] 3.2 Calculate the predicted value of excavation construction progress for each excavation construction influencing factor The cumulative average of the influencing factors j of all samples is the single factor overall SHAP value f(x j ). The formula is as follows:

[0123]

[0124] 3.3 Add the SHAP values ​​of the excavation construction parameters to get the SHAP value of the entire model, and rank the importance of the single factor overall SHAP value f(xj) to explain the results of the excavation construction progress prediction model and visualize it, as shown in Figure 3. Figure 3 As shown, it is used to understand the process of excavation construction progress prediction and the importance of excavation construction parameters, so as to assist the material yard management personnel to make relevant decisions on the material yard excavation progress control according to the positive and negative impact of the influencing factors ranked at the top of importance on the model prediction results, and guide the actual construction.

[0125] by Figure 3Taking the SHAP diagram shown in the figure as an example, the number of drill rods, temperature, borehole depth and elevation have a great influence on the model prediction results. Among them, the number of drill rods has the greatest influence on the model prediction results, and when the value of the number of drill rods increases gradually, the corresponding SHAP value is greater than 0 and will increase with the increase of the characteristic value. The positive effect of the number of drill rods on the model prediction results is more significant; when the value of the number of drill rods decreases gradually, the corresponding SHAP value is less than 0 and will decrease with the decrease of the characteristic value, and the negative effect of the number of drill rods on the model prediction results is more significant. For temperature, when its characteristic value increases, its corresponding SHAP value is less than 0, but the decrease in SHAP value is small, indicating that the negative effect of temperature on the model prediction results is not obvious; when the characteristic value of temperature decreases, its corresponding SHAP value is greater than 0 and the corresponding SHAP value is also increasing, and the positive effect of temperature on the model prediction results is greater than its negative effect on the model prediction results. For the drilling depth, when its characteristic value increases, the corresponding SHAP value is greater than 0, but the increase in the SHAP value is small, indicating that the positive effect of the drilling depth on the model prediction results is not obvious. When the characteristic value of the elevation decreases, the corresponding SHAP value is less than 0. For the elevation, when its characteristic value increases, the corresponding SHAP value is greater than 0, and when the characteristic value of the elevation decreases, the corresponding SHAP value is less than 0, but the decrease in the SHAP value is small, indicating that the negative effect of the elevation on the model prediction results is not obvious, and the positive effect of the elevation on the model prediction results is more obvious than the negative effect. Therefore, in this case, appropriately increasing the number of drill rods, lowering the temperature, increasing the drilling depth and increasing the elevation will improve the drilling efficiency.

[0126] Although the present invention has been described above in conjunction with the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can make many improvements and changes without departing from the purpose of the present invention, which are all within the protection of the present invention.

Claims

1. A method for predicting and controlling the excavation progress of a material field based on interpretable integrated machine learning, characterized in that: The following steps are involved: S1. Construction of multi-factor prediction data set for material field excavation progress based on excavation simulation: including: 1-1) Extract simulation parameters based on parametric modeling, use the three-dimensional model of the material yard and the excavation block data output by parametric cutting of the material yard to calculate and obtain the information of the excavation blocks of each construction unit; 1-2) According to meteorological data statistics or engineering experience or construction organization design, the simulation parameters of monthly effective construction days, daily effective construction hours, driving speed, road width and road slope are determined; in the simulation calculation, the driving speed of the dump truck and the road technical standards are taken from "DLT5397-2007 Hydropower Engineering Construction Organization Design Specification"; 1-3) Based on the CYCLONE model, an excavation discrete event simulation model is established to describe the transition of the excavation state of the excavation face from non-excavation, excavation to excavation completion, and the material yard excavation face is used as the smallest unit to promote the simulation process; The material field is divided into layers, strips, and blocks according to the size of the excavation block. A single excavation block is used as the minimum unit to advance the simulation process. The state transfer equation of the simulation process is shown in formula (1): T(i,a+1)=T(i,a)+ΔT,a=1,2,...,n (1) In formula (1), i is the number of the excavation block in the material yard, T(i,a) is the time when the excavation state of the i excavation block a is reached, and △T is the time taken from the construction state a to a+1; The construction state variables of the excavation block include the current excavation state and the current excavation process. The current excavation process reflects the process of the excavation block at the current simulation moment. The current excavation state includes unexcavated, excavating, and excavation completed. For each working face, the process conversion is judged at each simulation moment. The working face is the core simulation object for completing the excavation construction process. Each working face is combined with a separate excavation block, construction personnel, and mechanical team to realize the advancement and operation of the excavation process. According to the similarity of different process conversions, its construction state variables are divided into: state variables of the drilling and charging connection construction process, and blasting process state variables. The state variables of the drilling and charging line construction process include the process plan duration T pl 、The duration of the process is T al , process waiting time T wl When T pl >T al , it indicates that the process has not been completed. T al =T al +Δt (2) When T pl <T al When , it indicates that the process is completed and satisfies equations (3) and (4) at the same time. T al =T al +Δt (3) T wl =T wl +Δt (4) The blasting time is fixed within a certain time period. The blasting process state variables include the current simulation time T, the start time t1 and the end time t2 of each blasting process, and the blasting process is controlled by formula (5); T∈[t1,t2] (5) 1-4) Clarify the test objectives and factors that affect the test results, construct an orthogonal table, and build a multi-factor prediction data set X for the excavation progress of the material field through orthogonal experiments: First, determine the factors that affect the progress of material yard excavation, as follows: Then, select the orthogonal table L M (Q N ) to arrange the orthogonal experiment, L M (Q N ) represents an orthogonal array with N factors and Q levels, M is the number of level combinations, which is also equal to the number of rows in the orthogonal array; the orthogonal array established satisfies the following conditions: the same number of experiments are conducted for each level of any factor, and the same number of experiments are conducted for the level combination of any two factors; S2. Construction of excavation progress prediction model based on integrated machine learning, including: 2-1) The multi-factor prediction data set X of the material field excavation progress constructed based on the orthogonal test in step 1-4 is divided into a training set and a test set in a ratio of 7:3, and the data is cleaned and preprocessed; 2-2) The input of the material field excavation construction progress prediction model based on the XGBOOST model is defined as the material field excavation progress multi-factor prediction data set X, and the output of the excavation construction progress prediction model based on the XGBOOST model is defined as the objective function O bj and excavation progress forecast And set the regression model parameters, including the number of trees, depth, learning rate, and regularization parameter. The larger the number of trees, the easier it is for the material yard construction excavation progress prediction model to overfit. The depth is set to 4-6, the learning rate is set to 0.01-0.2, and the regularization parameter is 1. 2-3) Use the training set to build an initial tree for model training and obtain the excavation progress prediction value The real value y obtained by monitoring i Then, in each iteration, a tree is added to fit the residual of the model's last prediction until the model learning process is terminated, thereby forming an iterative residual tree set composed of many tree models. S3. Based on the results of the explainable analysis of the factors affecting the progress of material yard excavation, a decision on the control of the progress of material yard excavation is proposed, including: ranking the overall SHAP values ​​f(xj) of the single factors by importance, and guiding the actual construction according to the positive and negative impact of the factors with the highest importance ranking on the model prediction results.

2. The method for predicting and controlling the excavation progress of a material field according to claim 1, characterized in that: In step 1-1), the three-dimensional model of the material yard is imported into Rhino7.0, and the three-dimensional model is layered from top to bottom according to the height of the excavation block using the material yard parametric cutting program. The reference points that satisfy the same grid division direction and excavation rules are selected, and a grid division whose length and width match the size of the excavation block is established. The three-dimensional model is automatically divided according to the size of the excavation block. After the model is automatically divided, the information of each excavation block is saved, including the excavation block layer number, excavation block row number, excavation block column number and the volume of the excavation block.

3. The method for predicting and controlling the excavation progress of a material field according to claim 1, characterized in that: In step 2-3), the process of each iteration is as follows: 2-3-1) Use the training set to train the material field excavation progress prediction model based on the XGBOOST model. The objective function calculation formula of XGBoost is: In formula (6), m is the total amount of sample data imported into the kth tree; the first term is the loss function, which is used to measure the true value y i and excavation progress prediction The error between them; the second term is the regularization term, which is used for model complexity to prevent overfitting; the complexity of each tree is defined as: In formula (7), γ is the difficulty of node segmentation, T is the number of leaf nodes, λ is the L2 regularization coefficient to prevent overfitting, and ω is the modulus of the leaf node vector; 2-3-2) Use the trained model to predict the excavation progress of the test set and output the predicted value of the excavation progress In formula (8), K represents all CART trees established; xi represents the characteristics of the i-th sample; f k (x i ) represents the predicted value of the kth tree.

4. The method for predicting and controlling the excavation progress of a material field according to claim 1, characterized in that: In step S3, the single factor population SHAP value f(xj) is obtained according to the following steps: 3-1) For each set of excavation construction parameters and the corresponding excavation construction progress prediction value Calculate the single factor individual SHAP value f(x ij ); In formula (9), F is the sample x i The complete set of all influencing factors, S is the sample x i The subset formed by any number of influencing factors in the subset S, v(S) is the contribution of the influencing factors included in the subset S, and v(S∪{j})-v(S) is the contribution of influencing factor j to the joint effect; 3-2) The influencing factor j of all samples is accumulated and averaged to obtain the single factor overall SHAP value f(x j ), 5. The method for predicting and controlling the excavation progress of a material field according to claim 4, characterized in that: In step S3, the predicted value of excavation construction progress for each excavation construction influencing factor is calculated. The impact of the impact can be analyzed and displayed visually to assist yard managers in making decisions.

Images