Intelligent mining production management and control method based on ADP algorithm

Abstract

The invention provides an intelligent mining production management and control method based on an ADP algorithm, and relates to the technical field of intelligent mining, and the method comprises the steps: firstly obtaining vehicle data and wrong vehicle point data, carrying out the road division of a two-way single lane with a wrong vehicle point as the end points of the two ends of a road section, and carrying out the position judgment and display of a running vehicle in the two-way single lane; judging passing priorities of the vehicles, and checking whether vehicles running in opposite directions exist in adjacent road sections of the current vehicle or not; carrying out data preprocessing on the collected mine car scheduling data, dividing the data into a training data set and a test data set, carrying out classification processing on the preprocessed data, and establishing a classification model; identifying and dividing a naturally formed area of the underground mining area operation area by adopting a clustering algorithm; establishing a nonlinear system model, establishing a performance index, comprehensively considering transportation efficiency, cost saving and time benefit, measuring the effect of a scheduling strategy, and setting an optimization target; and establishing an ADP algorithm, and outputting an optimal scheduling scheme.

Description

Technical Field

[0001] The present invention relates to the field of intelligent mining technology, and in particular to an intelligent mining production management and control method based on an ADP algorithm. Background Art

[0002] Mineral resources, as an important strategic resource, have always been an important material basis for supporting the development of the national economy and play a very important role in resource development. However, traditional manual mining has defects such as high hidden dangers and low efficiency. The development and operation of underground locomotive unmanned driving systems can help reduce the probability of transportation accidents due to scheduling and operational errors and reduce labor costs, which is of great significance for improving the production scale and operation efficiency of mines. The optimization problem of road transportation scheduling system has always been one of the key issues that need to be focused on in mining system engineering, planning, and optimization modeling. Although the mine car scheduling system has gradually begun to be applied in various mining companies in China, the overall system is still in the initial promotion stage. For example, the traditional mine traffic planning is mainly based on the shortest path, without considering the road section's passing capacity and congestion, which may lead to the situation where the vehicle cannot actually complete the task according to the planning results.

[0003] Now there are a large number of scholars at home and abroad studying vehicle scheduling algorithms. Fatemah Al-Duoli and GhaithRabadi integrated data mining with meta-heuristics to improve the performance of Meta-RaPS, and formed a data mining hybrid method to solve vehicle scheduling problems. Xinguo Shui et al. proposed a vehicle scheduling method based on clonal selection algorithm to quickly generate satisfactory solutions to large-scale bus scheduling problems, and designed a fitness function to evaluate the quality of scheduling solutions. Pepin's team improved the three existing methods of heuristic mip solver, Lagrangian heuristic, and column generation heuristic for a given set of timetable tasks and multi-yard vehicle scheduling problems, and created two new methods of large neighborhood search heuristic and taboo search heuristic for neighborhood evaluation using column generation, summarizing the advantages and disadvantages of these methods. Ceder proposed a multi-vehicle scheduling heuristic algorithm based on deficit function theory. T.Xu et al. believed that the mine vehicle scheduling problem is a dynamic problem disturbed by dynamic information of the external environment, so the traditional solution method is not applicable, and then the ADP algorithm is used to solve the shortest completion time, fastest response time or shortest path problem. Yuan Mingcai organized the mine car scheduling problem into a mathematical model, and combined it with the particle swarm algorithm to mathematically solve it, formulated a scientific mine car scheduling strategy, and proposed the linkage between the mine car scheduling system and the ore distribution system. The two systems run on the same platform to achieve efficient and refined ore distribution in the production process.

[0004] Although there have been many studies and algorithms on vehicle scheduling, there is still a lack of research on the problem of ore truck scheduling in underground multi-layer, multi-mining areas, multi-vehicle types, and multi-vehicles, considering the passing capacity of road sections. In addition, there are few suitable operation platforms to achieve intelligent management and control. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, focusing on the scheduling problem of the underground crushing and loading operation cycle, improving the ADP optimization algorithm, and considering the actual situation of underground multi-layer mines, the present invention provides an intelligent mining production management and control method based on the ADP algorithm to achieve continuous, efficient, intelligent, safe, and visual production intelligent scheduling.

[0006] An intelligent mining production management and control method based on the ADP algorithm includes the following steps:

[0007] Step 1: Connect with the traffic management equipment at the passing point to obtain vehicle data and passing point data;

[0008] The specific content of Step 1 is as follows: Number all mining area operation areas, and use this number to represent the ore truck running route; Sequentially plan each ore truck according to the assigned task time sequence until the to-be-completed task pool is emptied, and then perform secondary optimization planning on the planning result; When performing the first sequential planning, only consider the task priority and resource allocation among ore trucks. When performing secondary optimization planning, comprehensively consider the total running distance, total waiting time, transportation task completion efficiency, and total cost index of all ore trucks for optimization; After the secondary optimization planning is completed, conduct a task priority review, adjust the situations that do not conform to the priority logic in the planned route, and generate the final scheduling plan in combination with the time axis data;

[0009] The specific mathematical expression method of the ore truck running route is as follows: Establish a ramp model. Assume that the model contains m nodes, and number them as M1, M2... Mm; For each trackless transport vehicle, assume its starting point is Mx0 and its ending point is Mx1, where 1 ≤ x0 < x1 ≤ m, x0, x1 ∈ Z, and Z is the set of integers. Create an array Cn of length m as the scheduling command, and assign 0 to c0 to cx0 and cx1 to cm; Cn = [c0, c1,..., cm-1]. For x ∈ [0, m-1], there are five forms of data settings in the array:

[0010]

[0011] Among them, the numbers of the path nodes are numbered using a three-dimensional space coordinate system, that is, numbered with (layer number z, x coordinate in the plane, y coordinate in the plane).

[0012] Step 2: Divide the two-way single lane into sections with the passing point as the two end points of the section, and determine and display the positions of the vehicles running in the two-way single lane;

[0013] In step 2, when determining the position of the running vehicle, an independent queue array with a data space of 2 is established for each trackless transport equipment. Every time the trackless transport equipment passes a passing point, it will be recorded by the video surveillance at the passing point and a passing signal will be fed back; the passing signal includes the passing point number and time, and the passing point number is input into the independent queue array. According to the queue principle, when the number of data in the independent queue array is greater than the data space, an overflow phenomenon occurs. According to the first-in-first-out principle, the data that first enters the array is deleted, and the data that enters subsequently is moved forward one position in sequence; assuming that when the vehicle travels from the M1-M2 road to the M2-M3 road, the original vehicle array is (1, 2). When entering the M2-M3 road, the digital number 3 enters from the back of the array, and the digital sequence number 1 will be squeezed out of the array. At this time, the vehicle array becomes (2, 3);

[0014] Step 3: Determine the vehicle's traffic priority and check whether there are oncoming vehicles on the adjacent road section of the current vehicle. If not, display a green light at the current passing point. If yes, display a waiting passing signal at the traffic light in the direction of the vehicle with lower priority and display a direct passing signal at the traffic light in the direction of the vehicle with higher priority according to the traffic priority Lv of the two oncoming vehicles.

[0015] Step 4: Preprocess the collected mine car dispatching data, including data cleaning, normalization, feature extraction and dimensionality reduction;

[0016] Step 5: Divide the preprocessed data into a training data set and a test data set, and classify the preprocessed data to establish a classification model;

[0017] Step 5.1: Decision point prediction: Use classification algorithm to predict the best decision point;

[0018] Step 5.2: Classification of minecart attributes: Use classification algorithms to automatically classify minecarts;

[0019] Step 5.3: Classification model establishment: According to the characteristics of the mine car scheduling problem and the data characteristics, the classification algorithm random forest model is selected, and the parameters of the selected classification algorithm are configured, including setting regularization parameters, adjusting the depth and number of trees, and training it using the training data set. During the training process, the random forest model learns the patterns and relationships in the data to achieve data classification.

[0020] Step 5.4: Classification model tuning: Tune the model parameters through cross-validation and grid search;

[0021] Step 5.5: Classification model verification: Use the test data set to verify the trained classification model, evaluate the prediction accuracy and stability of the classification model, and calculate the classification accuracy, recall rate, and precision performance indicators. If the corresponding thresholds are met, the classification model is qualified, otherwise continue to tune;

[0022] Step 6: Based on the classification results, a clustering algorithm is used to effectively identify and divide the multiple parts or groups naturally formed in the underground mining area;

[0023] Step 6.1: Area division: Using clustering algorithm, the mining area is divided into different operating areas according to the operating characteristics or functional requirements of each mining area;

[0024] Step 6.2: Similarity identification: Use clustering algorithm to classify the mine carts into different categories according to their driving paths or tasks;

[0025] Step 6.3: Multi-dimensional clustering: Considering the different attributes of the mine car and the multi-dimensional characteristics of the operating area, a comprehensive cluster analysis is performed;

[0026] Step 6.4: Clustering effect evaluation: clustering effect is evaluated by clustering accuracy index;

[0027] Step 6.5: Algorithm integration: Combine the clustering algorithm with the ADP algorithm.

[0028] Step 7: Establish a nonlinear system model. The state update equation defines the evolution of the system between time steps k and k+1:

[0029] The nonlinear system model includes:

[0030] State vector: The state vector x(k) represents the state of the system at a specific time step k;

[0031] Control input vector: The control input vector u(k) represents the control action applied to the system at time step k;

[0032] State transition rate: The state transition rate f(x(k),u(k),k) is a nonlinear function that describes how the system state transitions to the next state given the current state and control input;

[0033] State Update Equation: The state update equation defines the evolution of the system state over time and has the following form:

[0034] x(k+1)=f(x(k),u(k),k)+w(k)

[0035] w(k) represents the random disturbance or noise of the model;

[0036] System dynamics: The state update equation reflects the dynamic behavior of the system, including the movement of the mine carts in the mine, the influence of the scheduling strategy, and the role of environmental factors.

[0037] Time step: The time step Δk represents the time interval between state updates;

[0038] Initial state: At the initial moment of the state update equation, the state of the system is determined by the initial state x(0);

[0039] Boundary conditions and constraints: In practical applications, the state update equation usually needs to satisfy certain boundary conditions and constraints, including the maximum speed of the mine car, the maximum load, and the geographical restrictions of the mining area.

[0040] Model parameters: The parameters in the state update equation include the dynamic parameters of the mine car, road conditions, and traffic rules;

[0041] Step 8: Establish performance indicators, comprehensively consider transportation efficiency, cost savings, and time benefits, measure the effectiveness of scheduling strategies, and set optimization goals.

[0042] The performance index J is as follows:

[0043]

[0044] Where: x T (t) is the transpose of the state vector x(t) at time t, Q(t) is the state penalty weight matrix, u T (t) is the transpose of the control input vector u(t) at time t, and R(t) is the control input penalty weight matrix;

[0045] The optimization goal is specifically set to be a scheduling strategy that minimizes transportation time, maximizes transportation efficiency, and minimizes transportation costs.

[0046] Step 9: Establish the ADP algorithm and output the optimal scheduling solution;

[0047] Step 9.1: Initialization

[0048] Initialize the network: set the initial parameters of the evaluation network, model network, and execution network;

[0049] The Critic Network is used to evaluate the performance of the current strategy, that is, to evaluate the long-term return of taking a specific action in a specific state; the Critic Network is implemented by learning a value function V(x), which predicts the expected return that can be obtained starting from the current state x and following the current strategy;

[0050] The model network Model Network is used to learn and predict the dynamics of the environment, that is, how to transfer from the current state to the next state. The model network is implemented by learning the state transfer function f(x(k),u(k),k), which describes how the system state transfers to the next state given the current state and control input.

[0051] The Actor Network is used to generate control inputs, i.e., to select the best action in a given state. The Actor Network is implemented by learning a policy function π(x), which represents the optimal control input taken in a given state x.

[0052] Step 9.2: Design value function and policy function;

[0053] Value function V(x): represents the expected return from starting from state x and following strategy π;

[0054] Strategy function π(x): represents the optimal control input taken under a given state x;

[0055] Minimize the total distance traveled by all shifts, the formula is as follows:

[0056]

[0057] In the formula is rounded up; the decision variable is x ij and n ij , where n ij Calculated by formula (1)

[0058] Constraints include

[0059]

[0060] Constraint (3) stipulates that the total transport volume of the transport truck cannot be less than the target ore volume transported during the shift, constraint (4) stipulates that the loading capacity of the transport truck at the loading point during the shift working time shall not exceed the ore volume stored at the loading point, constraint (5) stipulates that the ore volume transported to each unloading point during the shift working time is within the target range, and constraint (6) controls the grade of the ore transported to each unloading point within the required range;

[0061] The optimal transport task sequence of transport trucks is determined by solving the transport truck scheduling solution; the objective function minimizes the total waiting time of the transport trucks using the following formula:

[0062]

[0063] The decision variable mk represents the number of encounters and waits of truck k on the downhill road section of the current transport task sequence, and its value is affected by the truck transport task sequence. After determining the number of trucks, the empty driving time of trucks, the loaded driving time, and the interval time between trucks, the following calculation is performed:

[0064]

[0065] Constraint (8) ensures that the working time of each transport truck cannot exceed their working time. In the formula, the cycle time T of the transport truck between the loading point i and the unloading point ij Calculated by the following formula:

[0066] T ij =T1+T2+T3+T4+T5+T6 (10)

[0067] T1 is the driving time of the loaded vehicle in the drifting section, T2 is the driving time of the loaded vehicle in the ramp section, T3 is the driving time of the loaded vehicle in the ground road section, T4 is the driving time of the ground highway section, T5 is the driving time of the empty vehicle in the ramp section, and T6 is the driving time of the empty vehicle in the drifting section;

[0068] Constraint (9) ensures that the number of transportation times between the unloading and loading points of the transport truck matches the result obtained in step 1;

[0069] The above formulas (1), (2), and (7) are objective functions, i.e., value functions, and formulas (3), (4), (5), (6), (8), and (9) are constraints, i.e., strategy functions.

[0070] Parameter Description:

[0071] x ij is the amount of ore transported by the truck between loading point i and unloading point j; n ij represents the number of transport tasks performed by the truck between loading point i and unloading point j, is the number of transport tasks between loading point i and unloading point j in the transport task sequence of the first k transport trucks, m k is the number of times truck k waits on the ramp under the current transport task sequence, NL is the number of loading points, NS is the number of unloading points, K is the number of transport trucks running simultaneously during the shift, C is the load capacity of the transport truck, D is the target ore volume transported during the shift, and L ij is the length of the transport path from loading point i to unloading point j, p i is the ore grade at loading point i, is the highest grade ore required for unloading point j, is the minimum ore grade allowed at unloading point j, WL i is the amount of ore at loading point i, WSj is the minimum amount of ore required for unloading point j, T w is the waiting time when empty cars meet on the slope section, T In T is the interval between each truck when the convoy starts running. ij is the cycle time between loading point i and unloading point j, T D Working shifts for transport trucks

[0072] Step 9.3: Design the Bellman equation to define the optimal strategy, linking immediate returns to future returns:

[0073] V(x k )=min{f(x k ,u k )+γV(x k+1 )}

[0074] Where x k is the state vector, which represents the state of the system at a specific time step k, including the position, speed, load status, and task allocation information of the mine car. k is the control input vector, which represents the control action applied to the system at time step k, including acceleration, deceleration, turning and going straight, waiting operation, or task switching and path selection of the mine cart; V(x) is the value function, which represents the expected return that can be obtained by starting from state x and following strategy π; R(x,u) is the immediate return, which is a function of the current state and the control input; γ is the discount factor, which is used to discount the future return to the current state; f(x,u,k) is the subsequent state, which is discounted to the current state by the discount factor γ; r k is the immediate reward, taking action u at time step k k The immediate reward, r k =R(x k ,u k ), where R is the reward function, and the future reward, i.e., the discounted reward or cumulative reward, is the sum of all immediate rewards from time step k to the terminal state. The formula for future reward is:

[0075]

[0076] Among them, G k is the total future reward starting from time step k, T is the total number of time steps or terminal state of the plan, and r k+i+1 is the immediate reward obtained at time step k+i+1.

[0077] Step 9.4: Perform strategy iteration and value iteration;

[0078] The policy iteration starts with the initial policy π and iteratively improves until convergence; in each iteration, the current policy π is used to calculate the value function V, and then this value function is used to update the policy π, expressed as: π new = argmax π [R(x,π(x))+γΣ x' P(x'|x,π(x))V(x')]

[0079] where π new is the updated policy, R(x,π(x)) is the immediate return, γ is the discount factor, P(x'|x,π(x)) is the probability of transitioning from state x to state x', and V(x') is the value function of the next state;

[0080] The value iteration starts with the initial value function V and is updated iteratively until convergence; in each iteration, the Bellman equation is used to update the value function: V new (x)=max u [R(x,u)+γΣ x' P(x'|x,u)V(x')] where V new (x) is the updated value function, R(x,u) is the immediate reward, γ is the discount factor, and P(x'|x,u) is the probability of transitioning from state x to state x';

[0081] Step 9.5: Function Approximation

[0082] Use function approximation to approximate V(x) or π(x):

[0083] V(x)≈φ(x) T θ

[0084] π(x)≈φ(x) T ω

[0085] Where φ(x) is the basis function vector, θ and ω are the parameter vectors to be learned;

[0086] Step 9.6: Adjust the learning rate and set the convergence conditions;

[0087] The convergence condition is that the change in parameter update is less than a certain threshold or reaches a maximum number of iterations.

[0088] Step 9.7: Evaluate the performance of the calculated solution;

[0089] Performance indicator calculation: Calculate performance indicators, total transportation time, cost savings.

[0090] Strategy effect evaluation: Evaluate the impact of the current strategy function on performance indicators.

[0091] Step 9.8: Output the optimal scheduling solution;

[0092] Output the optimal mine car scheduling plan and display the scheduling plan in a visual way;

[0093] The beneficial effects of adopting the above technical solution are:

[0094] The present invention provides an intelligent mining production control method based on the ADP algorithm, which has the following beneficial effects:

[0095] Improving the safety of mine operations: The present invention significantly reduces the probability of transportation accidents caused by human scheduling and operational errors by introducing an underground locomotive unmanned driving system and an intelligent scheduling algorithm, thereby greatly improving the safety of mine operations.

[0096] Improve operating efficiency and output: By optimizing the mine car running route and task allocation through the intelligent scheduling system, the present invention reduces the total running distance and waiting time of the mine car, improves the completion efficiency of the transportation task, and thus improves the output scale and operating efficiency of the mine.

[0097] Reduce labor costs and operating costs: By reducing the reliance on manual operations, the present invention reduces labor costs. At the same time, the optimized scheduling scheme reduces energy consumption and maintenance costs, thus saving operating costs.

[0098] Realize intelligent decision-making and dynamic adaptation: Combining the ADP algorithm and advanced data processing technology, the present invention can realize accurate mine car scheduling decisions and adapt to the mine vehicle scheduling problem, which is a dynamic problem disturbed by the dynamic information of the external environment. It provides an effective dynamic scheduling solution and enhances the intelligent decision-making ability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 It is an overall flow chart of the intelligent mining production control method in a specific embodiment of the present invention; DETAILED DESCRIPTION

[0100] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0101] An intelligent mining production control method based on the ADP algorithm is implemented through an intelligent scheduling algorithm. Figure 1 As shown, the following steps are included:

[0102] Step 1: Connect with the traffic management equipment at the passing point to obtain vehicle data and passing point data;

[0103] The specific steps of Step 1 are as follows: Number all mining operation areas and use this number to represent the running route of the ore trucks; Sequentially plan each ore truck according to the assigned task time sequence until the task pool to be completed is emptied, and then perform secondary optimization planning on the planning results; When performing the first sequential planning, only consider the task priorities and resource allocations among the ore trucks. When performing secondary optimization planning, comprehensively consider the total running distance, total waiting time, transportation task completion efficiency, and total cost indicators of all ore trucks for optimization; After the secondary optimization planning is completed, conduct a review of task priorities, adjust the situations that do not conform to the priority logic in the planned route, and generate the final scheduling plan in combination with the time-axis data;

[0104] The specific mathematical expression method for the running route of the ore trucks is as follows: Establish a ramp model. Assume that the model contains m nodes, and number them as M1, M2... Mm; For each trackless transportation vehicle, assume its starting point is Mx0 and its ending point is Mx1, where 1 ≤ x0 < x1 ≤ m, and x0, x1 ∈ Z, and Z is the set of integers. Create an array Cn of length m as the scheduling command, and assign 0 to c0 to cx0 and cx1 to cm; Cn = [c0, c1,..., cm-1]. For x ∈ [0, m-1], there are five forms of data settings in the array:

[0105]

[0106] Among them, the numbers of the path nodes are numbered using a three-dimensional space coordinate system, that is, numbered with (layer number z, x coordinate in the plane, y coordinate in the plane); In this embodiment, the layer number z is 1, 2, 3; The x coordinate in the plane is 1, 2, 3, 4, 5, 6; The y coordinate in the plane is 1, 2, 3, 4, 5, 6, 7, 8.

[0107] Step 2: Divide the two-way single-lane road with passing points as the two end points of the road section, and determine and display the positions of the vehicles running in the two-way single-lane road;

[0108] In Step 2, when determining the position of the running vehicle, establish an independent queue array with a data space of 2 for each trackless transportation equipment. Every time a passing point is passed, the trackless transportation equipment will be recorded by the video monitoring at the passing point and a passing signal will be fed back; The passing signal includes the passing point number and time. Input the passing point number into the independent queue array. According to the queue principle, when the number of data in the independent queue array is greater than the data space, an overflow phenomenon occurs. According to the first-in, first-out principle, the data that enters the array first is deleted, and the subsequent entered data moves forward one position in sequence; Assume that when the vehicle travels from the M1-M2 road to the M2-M3 road, the original array of the vehicle is (1, 2). When entering the M2-M3 road, the digital number 3 enters from the rear of the array, and the digital serial number 1 will be squeezed out of the array. At this time, the vehicle array becomes (2, 3);

[0109] Step 3: Determine the vehicle's traffic priority and check whether there are oncoming vehicles on the adjacent road section of the current vehicle. If not, display a green light at the current passing point. If yes, display a waiting passing signal at the traffic light in the direction of the vehicle with lower priority and display a direct passing signal at the traffic light in the direction of the vehicle with higher priority according to the traffic priority Lv of the two oncoming vehicles.

[0110] Step 4: Preprocess the collected mine car dispatching data (target mine site, mine car data), including data cleaning, normalization, feature extraction and dimensionality reduction, to enhance the robustness and efficiency of the algorithm. Although the ADP algorithm itself does not directly involve data preprocessing, in practical applications, the following data preprocessing steps are crucial to solving the optimization problem and can help improve the performance and accuracy of the ADP algorithm, including:

[0111] a. Data cleaning: Remove outliers and duplicate records in the data set to ensure the consistency and reliability of the data set.

[0112] b. Normalization: Standardize or normalize the data so that features of different dimensions and magnitudes can contribute information equally in the algorithm.

[0113] c. Feature extraction: Extract useful information from the raw data and construct a feature set that can represent the state and behavior of the mine cart, such as the capacity of the mine cart, current load, speed, etc.

[0114] d. Dimensionality reduction: Use methods such as principal component analysis (PCA), kernel principal component analysis (KPCA) and linear discriminant analysis (LDA) to reduce the dimension of the data, remove redundant information, and retain the main change trend of the data.

[0115] e. Feature selection: Select the most informative features through statistical tests or model-based methods to reduce the complexity of the model and improve the generalization ability of the algorithm.

[0116] f. Dataset division: The preprocessed dataset is divided into a training dataset and a test dataset for learning and evaluating the algorithm to ensure the effectiveness of the algorithm on unknown data.

[0117] Through the above data preprocessing steps, a more accurate and efficient input can be provided to the ADP algorithm, thereby achieving better mine car scheduling decisions on the intelligent mining production management and control platform.

[0118] Step 5: Divide the preprocessed data into a training data set and a test data set, and classify the preprocessed data to establish a classification model;

[0119] The classification process includes:

[0120] (1) Decision point prediction: In the case of scheduling problems that require determining when and where to allocate specific resources or tasks to mining vehicles, this method uses classification algorithms, such as support vector machines (SVMs) and random forests, to predict the best decision points; these algorithms can analyze historical scheduling data, identify key factors that affect scheduling efficiency, and predict the best time and location for future scheduling.

[0121] (2) Classification of mine cart attributes: This method uses a classification algorithm to automatically classify mine carts based on their performance, load capacity, or other key attributes. This includes, but is not limited to, the capacity, speed, current load, technical status, and driver experience of the mine cart. Through this classification, the most suitable task type for each mine cart can be quickly identified, thereby improving the efficiency and accuracy of resource allocation.

[0122] (3) Classification model establishment: According to the characteristics of the mine car scheduling problem and the data characteristics, the classification algorithm random forest model is selected, and the parameters of the selected classification algorithm are configured to adapt to the specific data set, including setting regularization parameters to prevent overfitting and adjusting the depth and number of trees to optimize the random forest model. It is trained using the training data set. During the training process, the random forest model learns the patterns and relationships in the data so that it can accurately classify new and unseen data.

[0123] (4) Model tuning: Model parameters are tuned through cross-validation and grid search to improve the model's predictive performance and generalization ability.

[0124] (5) Model validation: Use the test data set to validate the trained model, evaluate the model’s prediction accuracy and stability, and calculate the classification accuracy, recall rate, and precision performance indicators.

[0125] Step 6: Based on the classification results, clustering algorithms are used to effectively identify and divide the multiple parts or groups naturally formed in the underground mining operation area;

[0126] Step 6.1: Region division: Through clustering algorithms, such as fuzzy C-means (FCM), the mining area is divided into different operating areas according to the operating characteristics or functional requirements of each mining area. Each area has similar operating characteristics or functional requirements to facilitate centralized management and optimize resource allocation.

[0127] Step 6.2: Similarity identification: Through the clustering algorithm, the mine carts are divided into different categories according to their travel paths or tasks, so as to simplify and optimize the state space in the ADP algorithm.

[0128] Step 6.3: Multi-dimensional clustering: Considering the different attributes of the mine cars and the multi-dimensional characteristics of the operating area, a comprehensive clustering analysis is conducted to achieve more comprehensive area management and task optimization.

[0129] Step 6.4: Clustering effect evaluation: Evaluate the clustering effect through the clustering accuracy index to ensure that the clustering results can effectively support the scheduling decision of the ADP algorithm.

[0130] Step 6.5: Algorithm integration: Combine the clustering algorithm with the ADP algorithm.

[0131] Step 7: Establish a nonlinear system model. The state update equation defines the evolution of the system between time steps k and k+1:

[0132] The nonlinear system model includes:

[0133] State vector: In a nonlinear system model, the state vector x(k) represents the state of the system at a specific time step k. For the intelligent mining production control platform, the state vector includes information such as the position, speed, load status, and task allocation of the mining car.

[0134] Control input vector: The control input vector u(k) represents the control action applied to the system at time step k, including acceleration, deceleration, steering and other operations of the mine car.

[0135] State transition rate: The state transition rate f(x(k),u(k),k) is a nonlinear function that describes how the system state transitions to the next state given the current state and control input. This function reflects the dynamic characteristics and control strategy of the system.

[0136] State Update Equation: The state update equation defines the evolution of the system state over time and has the following form:

[0137] x(k+1)=f(x(k),u(k),k)+w(k)

[0138] w(k) represents the random disturbance or noise of the model, reflecting the inevitable uncertainty in the actual system.

[0139] System dynamics: The state update equation reflects the dynamic behavior of the system, including the movement of the mine carts in the mine, the influence of the scheduling strategy, and the role of environmental factors.

[0140] Time step: The time step Δk represents the time interval between state updates and can be adjusted according to the needs of the actual application. Smaller time steps can provide finer control but may also increase the computational burden.

[0141] Initial state: At the initial moment of the state update equation, the state of the system is determined by the initial state x(0). The choice of initial state has an important impact on the long-term behavior of the system and the scheduling results.

[0142] Boundary conditions and constraints: In practical applications, the state update equation usually needs to satisfy certain boundary conditions and constraints, including the maximum speed of the mine car, the maximum load, and the geographical restrictions of the mining area.

[0143] Model parameters: The parameters in the state update equation include the dynamic parameters of the mine car, road conditions, and traffic rules, which have a direct impact on the system dynamics.

[0144] Step 8: Establish performance indicators, comprehensively consider transportation efficiency, cost savings, and time benefits, measure the effectiveness of scheduling strategies, and set optimization goals.

[0145] The performance index J is usually used to measure system performance. It is an integral of the state and control input, and is usually expected to be minimized, as follows:

[0146]

[0147] Where: x T (t) The transpose of the state vector x(t) at time t: records the position, speed, load status, task allocation, etc. of the minecart. Q(t) is the state penalty weight matrix, usually diagonalized, representing the penalty for state variables, such as the minecart's load, speed, position, etc. T (t) is the transpose of the control input vector u(t) at time t: records the control signals or decision variables that affect the behavior of the mine car, such as the acceleration and deceleration instructions, steering operations, task switching, path selection, etc. R(t) is the control input penalty weight matrix, usually diagonalized, representing the penalty for the control input, which is related to the acceleration, deceleration, steering and other operations of the mine car.

[0148] The optimization goal is specifically set to be a scheduling strategy that minimizes transportation time, maximizes transportation efficiency, and minimizes transportation costs.

[0149] Step 9: Establish the ADP algorithm and output the optimal scheduling solution;

[0150] Step 9.1: Initialization

[0151] Initialize the network: set the initial parameters of the evaluation network, model network, and execution network;

[0152] The critic network is used to evaluate the performance of the current strategy, that is, to evaluate the long-term rewards of taking a specific action in a specific state; the critic network is implemented by learning a value function V(x), which predicts the expected rewards that can be obtained starting from the current state x and following the current strategy;

[0153] The evaluation network provides a feedback mechanism to guide the optimization direction of the policy network, that is, how to improve the strategy to obtain better long-term returns.

[0154] The model network is used to learn and predict the dynamics of the environment, that is, how to transfer from the current state to the next state. The model network is implemented by learning the state transfer function f(x(k),u(k),k), which describes how the system state transfers to the next state given the current state and control input.

[0155] The model network enables the ADP algorithm to predict future state changes, which is crucial for developing long-term effective strategies.

[0156] The Actor Network is used to generate control inputs, i.e., to select the best action in a given state; the Actor Network is implemented by learning a policy function π(x), which represents the optimal control input taken in a given state x;

[0157] Role: The executive network is directly involved in the decision-making process, with the goal of finding strategies that maximize long-term returns.

[0158] Step 9.2: Design value function and policy function;

[0159] Value function V(x): represents the expected return from starting from state x and following strategy π;

[0160] Strategy function π(x): represents the optimal control input taken under a given state x;

[0161] Minimize the total distance traveled by all shifts, the formula is as follows:

[0162]

[0163] In the formula is rounded up; the decision variable is x ij and n ij , where n ij Calculated by formula (1)

[0164] Constraints include

[0165]

[0166] Constraint (3) stipulates that the total transport volume of the transport truck cannot be less than the target ore volume transported during the shift, ensuring that the transportation task specified by the mine can be completed within the working hours of the shift. Constraint (4) stipulates that the loading capacity of the transport truck at the loading point during the shift working hours shall not exceed the amount of ore stored at the loading point. Constraint (5) stipulates that the amount of ore transported to each unloading point during the shift working hours is within the target range. Constraint (6) controls the grade of the ore transported to each unloading point within the required range and meets the grade requirements of the ore processing plant.

[0167] The optimal transport task sequence of transport trucks is determined by solving the transport truck scheduling solution; the objective function minimizes the total waiting time of the transport trucks using the following formula:

[0168]

[0169] The decision variable mk represents the number of encounters and waits of truck k on the downhill road section of the current transport task sequence, and its value is affected by the truck transport task sequence. After determining the number of trucks, the empty driving time of trucks, the loaded driving time, and the interval time between trucks, the following calculation is performed:

[0170]

[0171] Constraint (8) ensures that the working time of each transport truck cannot exceed their working time. In the formula, the cycle time T of the transport truck between the loading point i and the unloading point ij Calculated by the following formula:

[0172] T ij =T1+T2+T3+T4+T5+T6 (10)

[0173] T1 is the driving time of the loaded vehicle in the drifting section, T2 is the driving time of the loaded vehicle in the ramp section, T3 is the driving time of the loaded vehicle in the ground road section, T4 is the driving time of the ground highway section, T5 is the driving time of the empty vehicle in the ramp section, and T6 is the driving time of the empty vehicle in the drifting section;

[0174] Constraint (9) ensures that the number of transportation times between the unloading and loading points of the transport truck matches the result obtained in step 1;

[0175] The above formulas (1), (2), and (7) are objective functions, i.e., value functions, and formulas (3), (4), (5), (6), (8), and (9) are constraints, i.e., strategy functions.

[0176] Parameter Description:

[0177] x ij is the amount of ore transported by the truck between loading point i and unloading point j; nij represents the number of transport tasks performed by the truck between loading point i and unloading point j, is the number of transport tasks between loading point i and unloading point j in the transport task sequence of the first k transport trucks, m k is the number of times truck k waits on the ramp under the current transport task sequence. Its value is affected by the order of transport tasks. After determining the number of trucks, empty travel time, heavy travel time, and truck interval time. This formula is used to determine the total waiting time for truck operation. NL is the number of loading points, NS is the number of unloading points, K is the number of transport trucks running simultaneously during the shift work time, C is the load capacity of the transport truck, D is the target ore volume transported during the shift work time, and L ij is the length of the transport path from loading point i to unloading point j, p i is the ore grade at loading point i, is the highest grade ore required for unloading point j, is the minimum ore grade allowed at unloading point j, WL i is the amount of ore at loading point i, WS j is the minimum amount of ore required for unloading point j, T w is the waiting time when empty cars meet on the slope section, T In T is the interval between each truck when the convoy starts running. ij is the cycle time between loading point i and unloading point j, T D Working shifts for transport trucks

[0178] Step 9.3: Design the Bellman equation to define the optimal strategy and link immediate returns to future returns. The Bellman equation is the basis of dynamic programming and its approximate form is also found in ADP:

[0179] V(x k )=min{f(x k ,u k )+γV(x k+1 )}

[0180] Where x k is the state vector, which represents the state of the system at a specific time step k, including the position, speed, load status, and task allocation information of the mine car. These information together describe the specific situation of the system at time step k. kis the control input vector, which represents the control action applied to the system at time step k, including acceleration, deceleration, turning and going straight, waiting operation, or task switching and path selection of the mine cart; V(x) is the value function, which represents the expected return that can be obtained by starting from state x and following strategy π; R(x,u) is the immediate return, which is a function of the current state and the control input; γ is the discount factor, which is used to discount the future return to the current state; f(x,u,k) is the subsequent state, which is discounted to the current state by the discount factor γ; r k is the immediate reward, taking action u at time step k k The immediate reward, r k =R(x k ,u k ); where R is the reward function, which calculates the immediate reward based on the current state and the action taken. The future reward, also known as the discounted reward or cumulative reward, is the sum of all immediate rewards from time step k to the terminal state, discounted by a discount factor γ (0≤γ≤1) to reflect the uncertainty and time value of future rewards. The formula for future rewards is:

[0181]

[0182] Among them, G k is the total future reward starting from time step k, T is the total number of time steps or terminal state of the plan, and r k+i+1 is the immediate reward obtained at time step k+i+1.

[0183] Step 9.4: Perform strategy iteration and value iteration;

[0184] The policy iteration starts with the initial policy π and iteratively improves until convergence; in each iteration, the current policy π is used to calculate the value function V, and then this value function is used to update the policy π, expressed as: π new = argmax π [R(x,π(x))+γ∑ x' P(x'|x,π(x))V(x')]

[0185] where π new is the updated policy, R(x,π(x)) is the immediate return, γ is the discount factor, P(x'|x,π(x)) is the probability of transitioning from state x to state x', and V(x') is the value function of the next state;

[0186] The value iteration starts with the initial value function V and is updated iteratively until convergence; in each iteration, the Bellman equation is used to update the value function: V new (x)=max u [R(x,u)+γ∑x' P(x'|x,u)V(x')] where V new (x) is the updated value function, R(x,u) is the immediate reward, γ is the discount factor, and P(x'|x,u) is the probability of transitioning from state x to state x';

[0187] Step 9.5: Function Approximation

[0188] Since the state space may be continuous, the ADP algorithm usually needs to use function approximation methods to approximate V(x) or π(x):

[0189] V(x)≈φ(x) T θ

[0190] π(x)≈φ(x) T ω

[0191] Where φ(x) is the basis function vector, which is used to approximate the value function or policy function, and θ and ω are the parameter vectors to be learned; the parameters of the basis function are continuously updated through the learning process to better approximate the true value function and policy function.

[0192] Step 9.6: Adjust the learning rate and set the convergence conditions;

[0193] Learning rate adjustment: Adjust the learning rate to control the speed of parameter updates and ensure the stability and convergence of the algorithm.

[0194] The convergence condition is that the change in parameter update is less than a certain threshold or reaches a maximum number of iterations.

[0195] Step 9.7: Evaluate the performance of the calculated solution;

[0196] Performance indicator calculation: Calculate performance indicators, total transportation time, cost savings.

[0197] Strategy effect evaluation: Evaluate the impact of the current strategy function on performance indicators.

[0198] Step 9.8: Output the optimal scheduling solution;

[0199] Output the optimal mine car scheduling plan and display the scheduling plan in a visual way to facilitate understanding and operation.

[0200] According to the acquired vehicle data, this scheme uses the improved approximate dynamic programming (ADP) algorithm to make mining car scheduling decisions. The ADP algorithm optimizes large-scale state space problems through the collaborative work of the evaluation network, model network and execution network.

[0201] The first step of this embodiment is to establish a communication connection with the mine traffic management equipment to obtain vehicle data and missed vehicle point data in real time. These data include the location, speed, load, status information and task type of the mine car. Through the integrated interface and communication protocol, the system can seamlessly connect with the existing monitoring and management system of the mine to ensure the accuracy and real-time nature of the data.

[0202] Preprocess the collected raw data. This step includes data cleaning to remove invalid or erroneous data records; data normalization to unify the format of data from different sources and types; and feature extraction to identify key information that has a significant impact on mine car scheduling decisions. In addition, data preprocessing also includes dimensionality reduction, using techniques such as principal component analysis (PCA), kernel principal component analysis (KPCA), and linear discriminant analysis (LDA) to reduce the dimension of the data while retaining the most important information.

[0203] The state space is defined using the preprocessed data and the tasks are assigned. Each state in the state space consists of a set of features that describe the mine cart and the environment. Based on these states, the system uses an improved approximate dynamic programming (ADP) algorithm to calculate the optimal mine cart scheduling strategy. The ADP algorithm evaluates each possible decision point and selects the optimal control input through the collaborative work of the evaluation network, model network, and execution network.

[0204] The clustering algorithm analyzes the operating areas of the mine and groups areas with similar characteristics. This clustering helps simplify the state space of the ADP algorithm and more effectively manages the dispatch of mine cars in different areas by identifying natural groupings of operating areas. The clustering results also support the identification of similar paths or tasks, thereby reducing the number of potential paths that the algorithm needs to consider and improving dispatch efficiency.

[0205] The algorithm initializes the evaluation network, model network, and execution network, which are responsible for evaluating the performance indicators of the state, approximating the current system state, and selecting the current control input. Then, the algorithm enters the iterative process. In each iteration step, the evaluation network calculates the performance indicators based on the current state and the predicted next state, the model network is used to predict the transition of the system state, and the execution network selects the optimal control strategy based on the current state. These networks are continuously updated through the learning process to better approximate the optimal strategy and performance indicators. In each iteration, the algorithm calculates the weighted sum of the immediate return and the future return, and uses the discount factor to discount the future return to the current state to form an approximation of the value function. Through the method of policy iteration or value iteration, the algorithm gradually optimizes the policy function to minimize or maximize the performance indicator. At the same time, function approximation techniques, such as neural networks, are used to deal with continuous state space problems, and the real value function and policy function are approximated by adjusting the network parameters.

[0206] In addition, the algorithm also includes a feedback mechanism to monitor the scheduling effect in real time and adjust the control strategy based on the feedback. This process is repeated until the algorithm converges to the optimal solution or reaches a predetermined number of iterations. Ultimately, the ADP algorithm provides an optimized mine car scheduling solution to ensure high efficiency and low cost of mine operations.

[0207] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) to form a technical solution.

Claims

1. An intelligent mining production control method based on ADP algorithm, characterized in that: It includes the following steps: Step 1: Connect to the passing point traffic management device to obtain vehicle data and passing point data; Step 2: Divide the two-way single lane with the passing point as the two end points of the road section, and determine and display the positions of the vehicles running in the two-way single lane; Step 3: Judge the passing priority of the vehicle, and at the same time check whether there are oncoming vehicles in the adjacent road sections of the current vehicle. If not, display a green light at the current passing point. If so, according to the passing priority Lv of the two oncoming vehicles, display a waiting for passing signal at the signal light in the direction of the vehicle with the lower priority, and display a direct passing signal at the signal light in the direction of the vehicle with the higher priority; Step 4: First perform data preprocessing on the collected mine car scheduling data, including data cleaning, normalization, feature extraction, and dimensionality reduction processing; Step 5: Divide the preprocessed data into a training data set and a test data set, and at the same time perform classification processing on the preprocessed data to establish a classification model; Step 6: According to the classified results, for the multiple parts or groups naturally formed in the underground mining area operation area, use a clustering algorithm to effectively identify and divide these areas; Step 7: Establish a non-linear system model, and the state update equation defines the evolution of the system between time steps k and k + 1: Step 8: Establish a performance index, comprehensively consider transportation efficiency, cost savings, and time benefits, measure the effect of the scheduling strategy, and set an optimization goal; Step 9: Establish an ADP algorithm and output an optimal scheduling plan.

2. According to claim 1, a smart mining production control method based on ADP algorithm is characterized in that: The specific content of Step 1 is as follows: Number all the mining area operation areas, and use this number to represent the mine car running route; Sequentially plan each mine car according to the assigned task time sequence until the to-be-completed task pool is emptied, and then perform a secondary optimization plan on the planning result; When performing the first sequential plan, only consider the task priority and resource allocation among the mine cars. When performing the secondary optimization plan, comprehensively consider the total running distance, total waiting time, transportation task completion efficiency, and total cost index of all mine cars for optimization; After the secondary optimization plan is completed, perform a task priority review, adjust the situations that do not conform to the priority logic in the planned route, and generate a final scheduling plan in combination with the time axis data; The specific mathematical expression method of the mine car running route is as follows: Establish a ramp model. Suppose the model contains m nodes, and number them as M1, M2... Mm; For each trackless transport vehicle, suppose its starting point is Mx0 and its ending point is Mx1, where 1 ≤ x0 < x1 ≤ m, and x0, x1 ∈ Z, and Z is the set of integers. Create an array Cn of length m as the scheduling command, and assign 0 to c0 to cx0 and cx1 to cm; Cn = [c0, c1,..., cm - 1]. There exists x ∈ [0, m - 1], and the data in the array is set in five forms: Among them, the numbers of the path nodes are numbered using a three-dimensional space coordinate system, that is, numbered with (layer number z, x coordinate in the plane, y coordinate in the plane).

3. The intelligent mining production control method based on ADP algorithm according to claim 1 is characterized in that: In step 2, when determining the position of the running vehicle, an independent queue array with a data space of 2 is established for each trackless transport equipment. Every time the trackless transport equipment passes a passing point, it will be recorded by the video surveillance at the passing point and a passing signal will be fed back; the passing signal includes the passing point number and time, and the passing point number is input into the independent queue array. According to the queue principle, when the number of data in the independent queue array is greater than the data space, an overflow phenomenon occurs. According to the first-in-first-out principle, the data that first enters the array is deleted, and the data that enters subsequently is moved forward one position in order; Assume that when a vehicle moves from the M1-M2 road to the M2-M3 road, the original vehicle array is (1, 2). When entering the M2-M3 road, number 3 enters from the back of the array and number 1 is squeezed out of the array. At this time, the vehicle array becomes (2, 3).

4. The intelligent mining production control method based on ADP algorithm according to claim 1 is characterized in that: The step 5 specifically comprises the following steps: Step 5.1: Decision point prediction: Use classification algorithm to predict the best decision point; Step 5.2: Classification of minecart attributes: Use classification algorithms to automatically classify minecarts; Step 5.3: Classification model establishment: According to the characteristics of the mine car scheduling problem and the data characteristics, the classification algorithm random forest model is selected, and the parameters of the selected classification algorithm are configured, including setting regularization parameters, adjusting the depth and number of trees, and training it using the training data set. During the training process, the random forest model learns the patterns and relationships in the data to achieve data classification; Step 5.4: Classification model tuning: Tune the model parameters through cross-validation and grid search; Step 5.5: Classification model verification: Use the test data set to verify the trained classification model, evaluate the prediction accuracy and stability of the classification model, and calculate the classification accuracy, recall rate, and precision performance indicators. If the corresponding threshold is met, the classification model is qualified, otherwise continue to tune.

5. The intelligent mining production control method based on ADP algorithm according to claim 1 is characterized in that: The step 6 specifically comprises the following steps: Step 6.1: Area division: Using clustering algorithm, the mining area is divided into different operating areas according to the operating characteristics or functional requirements of each mining area; Step 6.2: Similarity identification: Use clustering algorithm to classify the mine carts into different categories according to their driving paths or tasks; Step 6.3: Multi-dimensional clustering: Considering the different attributes of the mine car and the multi-dimensional characteristics of the operating area, a comprehensive cluster analysis is performed; Step 6.4: Clustering effect evaluation: clustering effect is evaluated by clustering accuracy index; Step 6.5: Algorithm integration: Combine the clustering algorithm with the ADP algorithm.

6. The intelligent mining production control method based on ADP algorithm according to claim 1 is characterized in that: The nonlinear system model in step 7 includes: State vector: The state vector x(k) represents the state of the system at a specific time step k; Control input vector: The control input vector u(k) represents the control action applied to the system at time step k; State transition rate: The state transition rate f(x(k),u(k),k) is a nonlinear function that describes how the system state transitions to the next state given the current state and control input; State Update Equation: The state update equation defines the evolution of the system state over time and has the following form: x(k+1)=f(x(k),u(k),k)+w(k) w(k) represents the random disturbance or noise of the model; System dynamics: The state update equation reflects the dynamic behavior of the system, including the movement of the mine car in the mine, the influence of the scheduling strategy, and the role of environmental factors; Time step: The time step Δk represents the time interval between state updates; Initial state: At the initial moment of the state update equation, the state of the system is determined by the initial state x(0); Boundary conditions and constraints: In practical applications, the state update equation usually needs to satisfy certain boundary conditions and constraints, including the maximum speed of the mine car, the maximum load, and the geographical restrictions of the mining area; Model parameters: The parameters in the state update equation include the dynamic parameters of the mine car, road conditions, and traffic rules.

7. The intelligent mining production control method based on ADP algorithm according to claim 1 is characterized in that: The performance index J described in step 8 is as follows: Where: x T (t) is the transpose of the state vector x(t) at time t, Q(t) is the state penalty weight matrix, u T (t) is the transpose of the control input vector u(t) at time t, and R(t) is the control input penalty weight matrix; The optimization goal is specifically set to be a scheduling strategy that minimizes transportation time, maximizes transportation efficiency, and minimizes transportation costs.

8. The intelligent mining production control method based on ADP algorithm according to claim 1 is characterized in that: The step 9 specifically comprises the following steps: Step 9.1: Initialization; Initialize the network: set the initial parameters of the evaluation network, model network, and execution network; The Critic Network is used to evaluate the performance of the current strategy, that is, to evaluate the long-term return of taking a specific action in a specific state; the Critic Network is implemented by learning a value function V(x), which predicts the expected return that can be obtained starting from the current state x and following the current strategy; The model network is used to learn and predict the dynamics of the environment, that is, how to transfer from the current state to the next state. The model network is implemented by learning the state transfer function f(x(k),u(k),k), which describes how the system state transfers to the next state given the current state and control input; The Actor Network is used to generate control inputs, i.e., to select the best action in a given state. The Actor Network is implemented by learning a policy function π(x), which represents the optimal control input taken in a given state x. Step 9.2: Design value function and policy function; Value function V(x): represents the expected return from starting from state x and following strategy π; Strategy function π(x): represents the optimal control input taken under a given state x; Minimize the total distance traveled by all shifts, the formula is as follows: In the formula is rounded up; the decision variable is x ij and n ij , where n ij Calculated by formula (1) Constraints include Constraint (3) stipulates that the total transport volume of the transport truck cannot be less than the target ore volume transported during the shift, constraint (4) stipulates that the loading capacity of the transport truck at the loading point during the shift working time shall not exceed the ore volume stored at the loading point, constraint (5) stipulates that the ore volume transported to each unloading point during the shift working time is within the target range, and constraint (6) controls the grade of the ore transported to each unloading point within the required range; The optimal transport task sequence of transport trucks is determined by solving the transport truck scheduling solution; the objective function minimizes the total waiting time of the transport trucks using the following formula: The decision variable mk represents the number of encounters and waits of truck k on the downhill road section of the current transport task sequence, and its value is affected by the truck transport task sequence. After determining the number of trucks, the empty driving time of trucks, the loaded driving time, and the interval time between trucks, the following calculation is performed: Constraint (8) ensures that the working time of each transport truck cannot exceed their working time; where the cycle time T of the transport truck between the loading point i and the unloading point is ij Calculated by the following formula: <h2 style=";text-align:left;direction:ltr">T<h2 style=";text-align:left;direction:ltr"> ij <h2 style=";text-align:left;direction:ltr"> (10) T1+T2+T3+T4+T5+T6 T1 is the driving time of the loaded vehicle in the drifting section, T2 is the driving time of the loaded vehicle in the ramp section, T3 is the driving time of the loaded vehicle in the ground road section, T4 is the driving time of the ground highway section, T5 is the driving time of the empty vehicle in the ramp section, and T6 is the driving time of the empty vehicle in the drifting section; Constraint (9) ensures that the number of transportation times between the unloading and loading points of the transport truck matches the result obtained in step 1; The above formulas (1), (2), and (7) are objective functions, i.e., value functions, and formulas (3), (4), (5), (6), (8), and (9) are constraints, i.e., strategy functions. Parameter Description: x ij is the amount of ore transported by the truck between loading point i and unloading point j; n ij represents the number of transport tasks performed by the truck between loading point i and unloading point j, is the number of transport tasks between loading point i and unloading point j in the transport task sequence of the first k transport trucks, m k is the number of times truck k waits on the ramp under the current transport task sequence, NL is the number of loading points, NS is the number of unloading points, K is the number of transport trucks running simultaneously during the shift, C is the load capacity of the transport truck, D is the target ore volume transported during the shift, and L ij is the length of the transport path from loading point i to unloading point j, p i is the ore grade at loading point i, is the highest grade ore required for unloading point j, is the minimum ore grade allowed at unloading point j, WL i is the amount of ore at loading point i, WS j is the minimum amount of ore required for unloading point j, T w is the waiting time when empty cars meet on the slope section, T In T is the interval between each truck when the convoy starts running. ij is the cycle time between loading point i and unloading point j, T D Working hours for transport truck shifts Step 9.3: Design the Bellman equation to define the optimal strategy, linking immediate returns to future returns: V(x k )=min{f(x k ,u k )+γV(x k+1 )} Where x k is the state vector, which represents the state of the system at a specific time step k, including the position, speed, load status, and task allocation information of the mine car. k is the control input vector, which represents the control action applied to the system at time step k, including acceleration, deceleration, turning and going straight, waiting operation, or task switching and path selection of the mine cart; V(x) is the value function, which represents the expected return that can be obtained by starting from state x and following strategy π; R(x,u) is the immediate return, which is a function of the current state and the control input; γ is the discount factor, which is used to discount the future return to the current state; f(x,u,k) is the subsequent state, which is discounted to the current state by the discount factor γ; r k is the immediate reward, taking action u at time step k k The immediate reward, r k =R(x k ,u k ), where R is the reward function, and the future reward, i.e., the discounted reward or cumulative reward, is the sum of all immediate rewards from time step k to the terminal state. The formula for future reward is: Among them, G k is the total future reward starting from time step k, T is the total number of time steps or terminal state of the plan, and r k+i+1 is the immediate reward obtained at time step k+i+1; Step 9.4: Perform strategy iteration and value iteration; The policy iteration starts with the initial policy π and iteratively improves until convergence; in each iteration, the current policy π is used to calculate the value function V, and then this value function is used to update the policy π, expressed as: π new = argmax π [R(x,π(x))+γ∑ x' P(x'|x,π(x))V(x')] where π new is the updated policy, R(x,π(x)) is the immediate return, γ is the discount factor, P(x'|x,π(x)) is the probability of transitioning from state x to state x', and V(x') is the value function of the next state; The value iteration starts with the initial value function V and is updated iteratively until convergence; in each iteration, the Bellman equation is used to update the value function: V new (x)=max u [R(x,u)+γΣ x' P(x'|x,u)V(x')] where V new (x) is the updated value function, R(x,u) is the immediate reward, γ is the discount factor, and P(x'|x,u) is the probability of transitioning from state x to state x'; Step 9.5: Function Approximation Use function approximation to approximate V(x) or π(x): V(x)≈φ(x) T i π(x)≈φ(x) T oh Where φ(x) is the basis function vector, θ and ω are the parameter vectors to be learned; Step 9.6: Adjust the learning rate and set the convergence conditions; The convergence condition is that the change in parameter update is less than a certain threshold or reaches the maximum number of iterations; Step 9.7: Evaluate the performance of the calculated solution; Performance index calculation: calculate performance index, total transportation time, cost savings; Strategy effect evaluation: evaluate the impact of the current strategy function on performance indicators; Step 9.8: Output the optimal scheduling solution; Output the optimal mine car scheduling plan and display the scheduling plan in a visual way.

Images