An intelligent scheduling method and system for underground mine ore extraction shovel-truck equipment

Abstract

This invention relates to the field of equipment scheduling technology and discloses an intelligent scheduling method and system for ore-carrying equipment in underground mines. The method includes: processing multi-source data during the equipment scheduling process to obtain initial feature representations of each scheduling element; processing the initial feature representations to obtain multi-dimensional constraint data for mine scheduling and constructing an objective function for an optimization algorithm; encoding the decision variables corresponding to the scheduling scheme into particle position vectors based on the objective function, and calculating the comprehensive performance of each particle using a fitness function; generating a population by employing different initialization strategies to improve the quality of the initial solution; introducing a deep deterministic strategy gradient algorithm in the particle swarm optimization iteration to achieve adaptive control of core parameters, and using tabu search to avoid local optima; generating a specific execution plan based on the optimal scheduling scheme obtained through training to achieve intelligent scheduling. This invention solves the problem of low solution efficiency in existing equipment scheduling methods.

Description

Technical Field

[0001] This invention relates to the field of equipment scheduling technology, and in particular to an intelligent scheduling method and system for ore extraction and transportation equipment in underground mines. Background Technology

[0002] In recent years, with the development of information technology in high-tech mining, the level of intelligent mine construction has been continuously improving, placing higher demands on the intelligent scheduling of underground ore flows. Current scheduling methods mainly rely on manual experience, resulting in inefficient operation and consequently decreased production efficiency and increased production costs. Existing research indicates that metaheuristic algorithms, such as genetic algorithms and particle swarm optimization algorithms, can effectively optimize the sequence of work processes and equipment resource allocation in typical manufacturing system scheduling scenarios, such as workshop scheduling, providing a feasible technical approach for intelligent scheduling of underground ore flows. However, traditional metaheuristic algorithms are mostly used for continuous space optimization, requiring additional discretization processing in discrete scheduling problems, which can easily lead to loss of solution accuracy or slower convergence speed. Secondly, traditional algorithm parameters, such as inertia weights and learning factors, are mostly fixed values, unable to be dynamically adjusted according to the complexity of the scheduling problem and the iteration process. In simple problems, this may lead to excessive iteration and wasted computing power, while in complex problems, it is prone to getting trapped in local optima, making it difficult to find the globally optimal scheduling solution.

[0003] Therefore, it is urgent to solve the technical problem that the key parameters in existing metaheuristic algorithms cannot be dynamically adjusted, which leads to the solution efficiency and speed not achieving the expected results. Summary of the Invention

[0004] This invention provides an intelligent scheduling method and system for ore extraction and transportation equipment in underground mines, in order to solve the problem of low solution efficiency in existing equipment scheduling methods.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] In a first aspect, the present invention provides an intelligent scheduling method for ore-carrying equipment in underground mines, comprising:

[0007] Acquire multi-source data during the scheduling process of mining equipment, and perform discretization encoding on the multi-source data to obtain the initial feature representation of each scheduling element;

[0008] The initial feature representation is processed to obtain multi-dimensional constraint data for mine equipment scheduling, and the objective function of the optimization algorithm is constructed based on the multi-dimensional constraint data.

[0009] Based on the objective function, the decision variables corresponding to the scheduling scheme are encoded into particle position vectors. Population initialization is performed on the particle position vectors to obtain an initial population, and the initial population is iteratively updated.

[0010] In the initial population iteration update, the path reconnection algorithm is used to select candidate solutions and construct a tabu search candidate set. The tabu search algorithm is then used to optimize the quality of the population in the iteration update based on the tabu search candidate set to obtain the optimal solution.

[0011] The optimal particle position vector is constructed using a deep deterministic gradient algorithm, and the comprehensive performance value of each particle position vector is calculated using a fitness function.

[0012] The particle with the highest overall performance value is used as the optimal scheduling scheme to generate a specific execution plan, thereby achieving intelligent scheduling.

[0013] Optionally, acquire multi-source data during the mine equipment scheduling process, including:

[0014] Collect production equipment status information, material storage information, and environmental safety parameters, and integrate the production equipment status information, material storage information, and environmental safety parameters through timestamp alignment to obtain multi-source data;

[0015] Discretizing and encoding the multi-source data yields initial feature representations for each scheduling element, including:

[0016] The mining ore flow transportation process is abstracted as a workpiece processing process of a flow workshop. A two-stage machine set is defined for the processing process, which includes: the machine set MC for the first stage of mining and loading and the machine set MD for the second stage of transportation.

[0017] The multi-source data is abstracted into workpieces. The two stages are traversed according to the processing order of all processes on each machine in the two stages to obtain the corresponding solution scheme. Each solution scheme includes m sub-sequences, which correspond to m machines in the stage respectively.

[0018] The solution scheme is discretized to obtain the initial feature representation. The solution space corresponding to the discretization encoding method satisfies the following relationship:

[0019] ;

[0020] In the formula, Indicates the number of stages. For the number of workpieces, Let be the number of machines in the i-th stage. It represents factorial.

[0021] Optionally, the objective function of the optimization algorithm is constructed based on the multi-dimensional constraint data, including:

[0022] Based on multi-dimensional constraint data, with minimizing transportation time as the optimization objective, an objective function F of the optimization algorithm is constructed, which satisfies the following relationship:

[0023] ;

[0024] In the formula, This represents the end time of the i-th workpiece on the machine during the transportation phase. This indicates the start time of the i-th workpiece on the machine during the loading and unloading phase.

[0025] Optionally, the particle position vector is initialized with a population to obtain an initial population, and the initial population is iteratively updated, including:

[0026] A heuristic algorithm is used to encode half of the particle position vectors to obtain permuted coded individuals, and a random strategy is used to encode the other half of the particle position vectors to obtain randomly coded individuals. The permuted coded individuals and the randomly coded individuals are combined to obtain the initial population.

[0027] A particle update mechanism is established to update the position vector of each particle, resulting in the updated position vector of each particle.

[0028] The update mechanisms include: a speed update mechanism and a position update mechanism. The speed update mechanism satisfies the following relationship:

[0029] ;

[0030] The location update mechanism satisfies the following relationship:

[0031] ;

[0032] In the formula, This represents the velocity vector of particle i at the t-th iteration. This represents the position of particle i at the t-th iteration. Let represent the optimal position of particle i in the t-th iteration. This indicates the global optimum of the particle swarm at iteration t. Indicates inertia weight, For individual learning factors, As a social learning factor, and It is a random number in [0,1].

[0033] Optionally, during the initial population iterative update, a path reconnection algorithm is used to select candidate solutions to construct a tabu search candidate set, including:

[0034] Using the individual's current solution, the individual's historical best solution, and the population's historical best solution as core nodes, we construct search paths from the current solution to the individual's historical best solution and from the current solution to the population's historical best solution.

[0035] For each candidate solution on the search path, calculate its sequence similarity S(P,Q) with the individual's current solution, the individual's historical best solution, and the population's historical best solution.

[0036] A preset similarity threshold range is set, and candidate solutions with similarity within this threshold range are selected and added to the tabu search candidate set.

[0037] The Kendall-Tau distance is introduced during the process of adding sequences to the tabu search candidate set. Given two sequences P and Q of length n, the Kendall-Tau distance is defined as follows:

[0038] ;

[0039] In the formula, Let represent the i-th element of sequence P. This represents the i-th element of sequence Q. The logical AND operator, The operator represents the logical OR. Indicates the Kendall-Tau distance;

[0040] The similarity calculation formula is defined as follows:

[0041] ;

[0042] In the formula, n represents the number of elements in the sequence, and S(P,Q) is the similarity between the two sequences.

[0043] Optionally, the optimal solution can be obtained by using the tabu search algorithm to optimize the quality of the population during iterative updates based on the tabu search candidate set, including:

[0044] The candidate solutions in the tabu search candidate set are initialized, and the candidate solutions are used as the starting search points for the tabu search algorithm.

[0045] A neighborhood search is performed based on the starting search point to obtain an updated solution. The neighborhood search includes:

[0046] Obtain the workpiece processing process corresponding to the candidate solution in the tabu search candidate set, identify the longest process in the workpiece processing as the key block, take turns using the steps in the process within the key block as the first or last step of the key block to obtain the neighborhood search process, determine the duration of the neighborhood search process, compare the duration of the neighborhood search process with the duration of the key block, and when the duration of the neighborhood search process is less than the duration of the key block, take the corresponding neighborhood search process as the updated new solution;

[0047] We construct taboo and amnesty criteria to judge the updated solution. When the updated solution satisfies the amnesty criterion, it is taken as the optimal solution. When the updated solution does not satisfy the amnesty criterion, the optimal solution is selected from the solutions that are not taboo.

[0048] Among them, the taboo and amnesty criteria include taboo criteria and amnesty criteria. The taboo criteria include: establishing a process taboo table through manual construction, comparing the neighborhood search process with the process taboo table, and when the neighborhood search process is in the process taboo table, the updated solution corresponding to the neighborhood search process is the taboo solution, and when the neighborhood search process is not in the process taboo table, the updated solution corresponding to the neighborhood search process is the untabbed solution.

[0049] The amnesty criteria include: when the updated solution is better than the global optimal solution, it can be directly selected as the optimal solution without being affected by the taboo criteria;

[0050] The global optimal solution is the candidate solution with the shortest process time in the tabu search candidate set. The condition for the updated solution to be better than the global optimal solution is that the process time of the updated solution is shorter than the process time of the global optimal solution.

[0051] Optionally, the particle position vector of the optimal solution is constructed using a depth deterministic gradient algorithm, including:

[0052] The optimal solution is divided into a state set, which contains three types of core indicators.

[0053] One type of core indicator is population diversity. The standard deviation of the particle position vector is used to quantify the dispersion of the particle distribution in the solution space. The calculation formula is as follows:

[0054] ;

[0055] In the formula, Indicates the iteration number t. This represents the value of the d-th dimension for the i-th individual, where n is the population size and D is the dimension.

[0056] The second type of core indicator is the improvement rate of the optimal solution. The change in the global optimal fitness over k consecutive generations reflects the optimization efficiency. The calculation formula is:

[0057] ;

[0058] In the formula, This represents the optimal solution for the population at the t-th iteration, based on the optimal solution k generations ago;

[0059] The three core metrics are convergence progress P: the ratio of the current iteration number to the total number of iterations, expressed by the formula:

[0060] ;

[0061] In the formula, This indicates the maximum number of iterations preset by the algorithm;

[0062] Based on three core indicators, a set of states is constructed, and the parameters are normalized to the [-1, 1] interval using a mapping formula:

[0063] ;

[0064] In the formula, param represents the parameters included in the aforementioned three types of indicators, and the value of i ranges from 0 to 5. This represents the normalized mapping parameters;

[0065] Based on the normalized mapping parameters, a 15-dimensional state input vector is finally constructed.

[0066] Inertia weights, a key parameter for population iterative updates Cognitive acceleration coefficient Social acceleration coefficient To regulate the inertial weights, a neural network and reward function are constructed. Cognitive acceleration coefficient Social acceleration coefficient renew;

[0067] Regarding the design of the reward function: used to calculate the reward value after performing a certain action, the reward function is designed as follows:

[0068] ;

[0069] In the formula, Indicates the first The optimal solution in the next iteration. This represents the reward value in the t-th iteration;

[0070] For neural network design: Neural networks include actor networks. Target Actor Network Action Value Network and target action value network ;

[0071] To train the action value network, minimize the loss function:

[0072] ;

[0073] Subsequently, the actor network is trained using the action value network via policy gradient, and the corresponding policy gradient formula is as follows:

[0074] ;

[0075] In the formula, As a discount factor, Let the loss function be the action value network. For the strategy gradient of the actor network, Let be the reward value in the t-th iteration. For the target action value network in the (t+1)th iteration, For gradient operators, For the parameter vector of the actor network, The objective function of the actor network, For action vectors, Let t be the action vector for the t-th iteration. For state vectors, Let be the state vector for the t-th iteration. The output value of the action value network. For the gradient of the action value network, For the gradient of the actor network, The parameter adjustment action of the actor network based on the state vector output at the t-th iteration;

[0076] Inertial weight Cognitive acceleration coefficient Social acceleration coefficient The updated parameters and inertial weights are obtained by inputting the 15-dimensional state input vector into the trained actor network. Cognitive acceleration coefficient Social acceleration coefficient ;

[0077] Updated parameter inertia weights Cognitive acceleration coefficient Social acceleration coefficient Substitute the particle update mechanism to update and iterate the optimal solution to obtain the final solution;

[0078] The final solution is input into the objective function to obtain the particle position vector of the optimal solution.

[0079] Optionally, the overall performance value of each particle's position vector can be calculated using the fitness function, including:

[0080] Input each particle position vector into the objective function, calculate the objective function F, and then substitute the objective function F into the fitness function to calculate the comprehensive performance value of each particle position vector;

[0081] The fitness function satisfies the following relationship:

[0082] ;

[0083] In the formula, The objective function value, This represents the overall performance value for each particle's position vector.

[0084] Optionally, the particle with the highest overall performance value is selected as the optimal scheduling scheme to generate a specific execution plan, including:

[0085] Output the optimal code corresponding to the particle with the highest overall performance value, and generate a Gantt chart based on the optimal code;

[0086] The Gantt chart of the process is decoded and transformed into a specific execution plan.

[0087] Secondly, embodiments of this application provide an intelligent scheduling system for underground mine ore extraction and transportation equipment, including a processor and a memory;

[0088] Memory, used to store computer programs;

[0089] When a processor executes a program stored in memory, it implements any of the steps of the method described in the first aspect.

[0090] Beneficial effects:

[0091] This invention provides an intelligent scheduling method for underground mine ore handling equipment. It utilizes reinforcement learning to intelligently update w, c1, and c2 in the discrete particle swarm optimization algorithm, enhancing solution efficiency and speed. By dividing the state set into chromosomes and using the average fitness and diversity of the population as a basis for reflecting the overall population state, it facilitates improved population quality and makes it easier to obtain superior individuals. Furthermore, it introduces a tabu search algorithm to avoid excessively rapid local convergence. The proposed scheduling optimization algorithm significantly reduces vehicle waiting time during pickup, increases throughput, and indirectly shortens the maximum operating time for underground ore flow transportation, meeting the efficiency requirements of underground ore flow transportation. The proposed optimization algorithm solves the problems of slow response, low accuracy, and poor adaptability of traditional scheduling methods in complex scenarios, providing a standardized and scalable technical solution for intelligent scheduling in various industries. It belongs to the cutting-edge technology field of interdisciplinary integration between artificial intelligence and industrial engineering, transportation engineering, and logistics engineering. Attached Figure Description

[0092] Figure 1 A flowchart of a preferred embodiment of the intelligent scheduling method for underground mine ore extraction and transportation equipment;

[0093] Figure 2 This is a flowchart illustrating the algorithm of the intelligent scheduling method for underground mine ore extraction and transportation equipment according to a preferred embodiment of the present invention.

[0094] Figure 3 A flowchart of the tabu search algorithm provided in a preferred embodiment of the present invention;

[0095] Figure 4 This is a schematic diagram of the DDPG algorithm neural network training provided in a preferred embodiment of the present invention. Detailed Implementation

[0096] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0097] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0098] Please see Figure 1-2 This application provides an intelligent scheduling method for underground mine ore extraction and transportation equipment, including:

[0099] S1. Collect multi-source data involved in the mine scheduling process, perform discretization coding processing, and obtain the initial feature representation of each scheduling element;

[0100] In a preferred embodiment of the present invention, data acquisition is achieved through real-time sensors, intelligently collecting multi-source data across three dimensions: production equipment status, material storage information, and environmental safety parameters. A three-tiered architecture of "real-time sensor acquisition + edge computing preprocessing + cloud aggregation" is employed to achieve data acquisition: the front end collects raw data via sensors, which is then transmitted to the mining system via industrial Ethernet through an edge gateway. Asynchronous data from multiple devices are aligned using timestamps, ultimately forming a standardized raw dataset.

[0101] S2. Based on the initial feature embedding obtained in S1, the data noise is removed by filtering, and the multi-dimensional constraints of mine scheduling are quantified to construct the objective function of the particle swarm optimization algorithm.

[0102] In a preferred embodiment of the present invention, considering machine loading efficiency and the order of stage operations, and constrained by process sequence and equipment capacity, an optimization objective function F that minimizes transportation time is constructed. The constructed objective function is as follows:

[0103]

[0104] In the formula, This represents the end time of the i-th workpiece on the machine during the transportation phase. This indicates the start time of the i-th workpiece on the machine during the loading and unloading phase.

[0105] S3, a multi-objective function built on S2, is used to train the model through particle swarm optimization algorithm to achieve global optimization of the scheduling scheme;

[0106] In a preferred embodiment of the present invention, the encoding method specifically takes the form of the processing sequence of all operations on each machine at each stage. Each solution includes m subsequences, each corresponding to one of the m machines, and the order in the subsequences represents the processing order of the operations on the machines. Because each workpiece needs to flow through each stage, all machine sequences at each stage should contain all operations. The solution space corresponding to the encoding method is:

[0107]

[0108] In the formula, Indicates the number of stages. For the number of workpieces, Let be the number of machines in the i-th stage. It represents factorial.

[0109] In a preferred embodiment of the present invention, after encoding individuals in the population using a specific encoding method, the algorithm iteration and model training include:

[0110] Population initialization involves using four heuristic algorithms (SPTB, NEHLPT(λ), bLPTB, and NHEbSPT(λ)) to generate high-quality permutation codes with half of the population having different characteristics. To enhance population diversity, a random strategy is adopted to initialize the remaining individuals, which helps increase the differences among individuals within the population and ensures population diversity. The generated permutation codes are then combined with the first available machine criteria for full-stage coding generation.

[0111] The fitness function of an individual is calculated to determine the individual's current solution, the individual's historical best solution, and the population's historical best solution. Here, the fitness is calculated in the following form:

[0112]

[0113] In the formula, The objective function value, This represents the overall performance value for each particle's position vector.

[0114] The solution is updated using a particle swarm optimization algorithm, resulting in multiple particle position matrices and multiple particle velocity matrices. These matrices are used to update particle positions and define the fitness function. The particle position transformations and velocity changes are shown below:

[0115]

[0116] In the formula, This represents the velocity vector of particle i at the t-th iteration. This represents the position of particle i at the t-th iteration. Let represent the optimal position of particle i in the t-th iteration. This indicates the global optimum of the particle swarm at iteration t. Indicates inertia weight, For individual learning factors, As a social learning factor, and It is a random number in [0,1], used to increase randomness.

[0117] Using a path reconnection algorithm, search paths are constructed from the current solution to the individual's historical best solution and from the current solution to the population's historical best solution, with the individual's current solution, the individual's historical best solution, and the population's historical best solution as core nodes. Through specific selection, solutions on some search paths are added to the tabu search candidate set, and the similarity between two solutions is obtained by calculating the Kendall-Tau distance (KTD).

[0118] In this preferred embodiment, for two given sequences P and Q of length n, the KTD distance is defined as follows:

[0119]

[0120]

[0121] In the formula, Let represent the i-th element of sequence P. This represents the i-th element of sequence Q. The logical AND operator, The operator represents the logical OR. This represents the Kendall-Tau distance. Furthermore, the similarity calculation formula is defined as follows:

[0122]

[0123] In the formula, n represents the number of elements in the sequence, and S(P,Q) is the similarity between the two sequences.

[0124] After formula transformation, the similarity is within the range of [0,1], which facilitates the control of similarity analysis for problems of different scales. When two sequences are completely identical, the similarity is 1; when two sequences are completely opposite, the similarity is 0.

[0125] By selectively choosing some solutions, preferred solutions are added to the candidate set, and quality optimization is performed through tabu search, such as... Figure 3 As shown, a tabu search algorithm is introduced to improve local optimization capabilities and avoid premature convergence. Specifically, the main search steps include initialization, neighborhood search, tabu and pardon criterion determination, and solution updating. The initialization process starts with the aforementioned candidate solutions, providing the algorithm with a starting search point. The neighborhood search process generates new solutions in a specific way to expand the search space. The tabu and pardon criterion determination process helps ensure the rationality of the search direction; when a new solution is better than the current optimal solution, the pardon criterion is satisfied; otherwise, the optimal solution is selected from the solutions not tabulated. The solution updating process involves adding the new solution to the current solution set, updating the tabu list, and continuously optimizing the quality of the solutions.

[0126] Specific neighborhood searches include: attempting to select all critical blocks and move the operation to before its head or after its tail, or inserting the critical block head operation or tail operation into the block; or removing a critical operation from the machine in the current stage and inserting it into another available machine in the same stage.

[0127] In this preferred embodiment, reinforcement learning is used to achieve adaptive control of key parameters, such as... Figure 4 As shown, the main steps of reinforcement learning include:

[0128] A deep deterministic gradient algorithm is introduced to achieve adaptive parameter control:

[0129] State set partitioning: The state set contains three core metrics:

[0130] (1) Population diversity The dispersion of particle distribution in the solution space is quantified based on the standard deviation of the particle position vector. The calculation formula is as follows:

[0131]

[0132] in, Indicates the iteration number t. This represents the value of the d-th dimension of the i-th individual, where n is the population size and D is the dimension.

[0133] (2) Improvement rate of the optimal solution The change in the global optimal fitness over k consecutive generations reflects the optimization efficiency.

[0134]

[0135] in, Let represent the population optimal solution at the t-th iteration. Based on the optimal solution k generations ago, calculate the relative improvement ratio of the current optimal solution. This indicates a performance improvement. This indicates that improvement has stalled.

[0136] (3) Convergence progress P: The ratio of the current iteration number to the total number of iterations. It is used to indicate the iteration stage of the algorithm, reflecting whether the algorithm is in the exploration or development stage. The formula is shown below, where This indicates the maximum number of iterations preset by the algorithm.

[0137]

[0138] In the formula, This indicates the maximum number of iterations preset by the algorithm.

[0139] Based on the aforementioned three types of indicators, a set of states is constructed. Given that some parameters have extremely small values ​​and their dynamic changes are even more subtle, to enable the neural network to effectively perceive and learn these subtle changes, their values ​​are normalized to the interval [-1, 1] using the following mapping formula:

[0140]

[0141] Here, param represents the parameters included in the aforementioned three types of indicators, and the value of i ranges from 0 to 5; based on this, a 15-dimensional state input vector is finally constructed.

[0142] Action set construction: using key parameters of the particle swarm optimization algorithm (inertia weight) Cognitive acceleration coefficient Social acceleration coefficient The algorithm dynamically optimizes parameter configurations based on the algorithm's iteration process, targeting parameters such as inertia weight and social acceleration coefficient. In the early iteration stage, it tends to increase inertia weight and decrease social acceleration coefficient to enhance the population's global search capability and avoid premature entrapment in local optima. In the middle iteration stage, it dynamically balances cognitive acceleration coefficient and social acceleration coefficient to achieve coordinated advancement of global exploration and local development, balancing solution diversity and optimization efficiency. In the later iteration stage, it decreases inertia weight and increases cognitive acceleration coefficient to guide the population to focus on the current optimal region for refined local optimization, improving solution accuracy. Based on these parameter control requirements, a four-dimensional action set is output through the Deep Deterministic Policy Gradient (DDPG) algorithm. This action set corresponds to the quantified adjustment signals of key parameters, providing precise control basis for the adaptive optimization of particle swarm optimization parameters.

[0143] Reward function design: Used to calculate the reward value after performing a certain action, its goal is to encourage the particle swarm optimization algorithm to obtain a better global optimum. The reward function is designed as follows:

[0144]

[0145] in Indicates the first The optimal solution in the next iteration. This represents the reward value for the t-th iteration.

[0146] In the deep deterministic policy gradient algorithm, four neural networks are designed to obtain the optimal policy, namely the Actor Network. Target Actor Network Action Value Network (Critic Network) And the Target Critic Network. Among them, the Actor network With Target Actor Network Used for state-based action selection; Critic network With TargetCritic network This is used to evaluate the value of actions selected by the actor's network. To train the action value network, the following loss function needs to be minimized:

[0147]

[0148] Subsequently, the Critic network is used to train the Actor network via policy gradient, and the corresponding policy gradient formula is as follows:

[0149]

[0150] In the formula, As a discount factor, Let the loss function be the action value network. For the strategy gradient of the actor network, Let be the reward value in the t-th iteration. For the target action value network in the (t+1)th iteration, For gradient operators, For the parameter vector of the actor network, The objective function of the actor network, For action vectors, Let t be the action vector for the t-th iteration. For state vectors, Let be the state vector for the t-th iteration. The output value of the action value network. For the gradient of the action value network, For the gradient of the actor network, For the parameter adjustment action of the actor network based on the state vector output at the t-th iteration.

[0151] Repeat the above algorithm iteration steps until the maximum number of iterations is reached. Output the historical best solution to generate a Gantt chart of the optimal scheduling scheme, and generate a specific execution plan to achieve intelligent scheduling.

[0152] This application also provides an intelligent scheduling system for underground mine ore extraction and transportation equipment, including a processor and a memory;

[0153] Memory, used to store computer programs;

[0154] When a processor executes a program stored in memory, it implements any of the steps described in the intelligent scheduling method for ore extraction and transportation equipment in underground mines.

[0155] The aforementioned intelligent scheduling system for underground mine ore extraction and transportation equipment can realize various embodiments of the aforementioned intelligent scheduling method for underground mine ore extraction and transportation equipment, and can achieve the same beneficial effects. Here, it will not be elaborated further.

[0156] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for intelligent scheduling of ore-carrying equipment in underground mines, characterized in that, include: Acquire multi-source data during the scheduling process of mining equipment, and perform discretization encoding on the multi-source data to obtain the initial feature representation of each scheduling element; The initial feature representation is processed to obtain multi-dimensional constraint data for mine equipment scheduling, and the objective function of the optimization algorithm is constructed based on the multi-dimensional constraint data. Based on the objective function, the decision variables corresponding to the scheduling scheme are encoded into particle position vectors. Population initialization is performed on the particle position vectors to obtain an initial population, and the initial population is iteratively updated. In the initial population iterative update, the path reconnection algorithm is used to select candidate solutions and construct a tabu search candidate set. The tabu search algorithm is then used to optimize the quality of the population in the iterative update based on the tabu search candidate set to obtain the optimal solution. The optimal particle position vector is constructed using a deep deterministic gradient algorithm, and the comprehensive performance value of each particle position vector is calculated using a fitness function. The particle with the highest overall performance value is used as the optimal scheduling scheme to generate a specific execution plan, thereby achieving intelligent scheduling. Among them, the optimal solution is obtained by using the tabu search algorithm to optimize the quality of the population during iterative updates based on the tabu search candidate set, including: The candidate solutions in the tabu search candidate set are initialized, and the candidate solutions are used as the starting search points for the tabu search algorithm. A neighborhood search is performed based on the starting search point to obtain an updated solution. The neighborhood search includes: Obtain the workpiece processing process corresponding to the candidate solution in the tabu search candidate set, identify the longest process in the workpiece processing as the key block, take turns using the steps in the process within the key block as the first or last step of the key block to obtain the neighborhood search process, determine the duration of the neighborhood search process, compare the duration of the neighborhood search process with the duration of the key block, and when the duration of the neighborhood search process is less than the duration of the key block, take the corresponding neighborhood search process as the updated new solution; We construct taboo and amnesty criteria to judge the updated solution. When the updated solution satisfies the amnesty criterion, it is taken as the optimal solution. When the updated solution does not satisfy the amnesty criterion, the optimal solution is selected from the solutions that are not taboo. Among them, the taboo and amnesty criteria include taboo criteria and amnesty criteria. The taboo criteria include: establishing a process taboo table through manual construction, comparing the neighborhood search process with the process taboo table, and when the neighborhood search process is in the process taboo table, the updated solution corresponding to the neighborhood search process is the taboo solution, and when the neighborhood search process is not in the process taboo table, the updated solution corresponding to the neighborhood search process is the untabbed solution. The amnesty criteria include: when the updated solution is better than the global optimal solution, it can be directly selected as the optimal solution without being affected by the taboo criteria; The global optimal solution is the candidate solution with the shortest process time in the tabu search candidate set. The condition for the updated solution to be better than the global optimal solution is that the process time of the updated solution is shorter than the process time of the global optimal solution. The particle position vector for constructing the optimal solution is generated using a deep deterministic gradient algorithm, including: The optimal solution is divided into a state set, which contains three types of core indicators. One type of core indicator is population diversity. The standard deviation of the particle position vector is used to quantify the dispersion of the particle distribution in the solution space. The calculation formula is as follows: ； In the formula, Indicates the iteration number t. This represents the value of the d-th dimension for the i-th individual, where n is the population size and D is the dimension. The second type of core indicator is the improvement rate of the optimal solution. The change in the global optimal fitness over k consecutive generations reflects the optimization efficiency. The calculation formula is: ； In the formula, This represents the population-optimal solution at the t-th iteration. This represents the optimal solution for the population at the tk-th iteration, with the optimal solution from k generations ago as the benchmark. The three core metrics are convergence progress P: the ratio of the current iteration number to the total number of iterations, expressed by the formula: ； In the formula, This indicates the maximum number of iterations preset by the algorithm; Based on three core indicators, a set of states is constructed, and the parameters are normalized to the [-1, 1] interval using a mapping formula: ； In the formula, param represents the parameters included in the aforementioned three types of indicators, and the value of i ranges from 0 to 5. This represents the normalized mapping parameters; Based on the normalized mapping parameters, a 15-dimensional state input vector is finally constructed. Inertia weights, a key parameter for population iterative updates Cognitive acceleration coefficient Social acceleration coefficient To regulate the inertial weights, a neural network and reward function are constructed. Cognitive acceleration coefficient Social acceleration coefficient renew; Regarding the design of the reward function: used to calculate the reward value after performing a certain action, the reward function is designed as follows: ； In the formula, Indicates the first The optimal solution in the next iteration. Indicates the first The optimal solution in +1 iterations This represents the reward value in the t-th iteration; For neural network design: Neural networks include actor networks. Target Actor Network Action Value Network and target action value network ; To train the action value network, minimize the loss function: ； Subsequently, the actor network is trained using the action value network via policy gradient, and the corresponding policy gradient formula is as follows: ； In the formula, As a discount factor, Let the loss function be the action value network. For the strategy gradient of the actor network, Let be the reward value in the t-th iteration. For the target action value network in the (t+1)th iteration, For gradient operators, For the parameter vector of the actor network, The objective function of the actor network, For action vectors, Let t be the action vector for the t-th iteration. For state vectors, Let be the state vector for the t-th iteration. The output value of the action value network. For the gradient of the action value network, For the gradient of the actor network, The parameter adjustment action of the actor network based on the state vector output at the t-th iteration; Inertial weight Cognitive acceleration coefficient Social acceleration coefficient The updated parameters and inertial weights are obtained by inputting the 15-dimensional state input vector into the trained actor network. Cognitive acceleration coefficient Social acceleration coefficient ; Updated parameter inertia weights Cognitive acceleration coefficient Social acceleration coefficient Substitute the particle update mechanism to update and iterate the optimal solution to obtain the final solution; The final solution is input into the objective function to obtain the particle position vector of the optimal solution.

2. The intelligent scheduling method for underground mine ore extraction and transportation equipment according to claim 1, characterized in that, Acquire multi-source data during the mine equipment scheduling process, including: Collect production equipment status information, material storage information, and environmental safety parameters, and integrate the production equipment status information, material storage information, and environmental safety parameters through timestamp alignment to obtain multi-source data; Discretizing and encoding the multi-source data yields initial feature representations for each scheduling element, including: The mining ore flow transportation process is abstracted as a workpiece processing process of a flow workshop. A two-stage machine set is defined for the processing process, which includes: the machine set MC for the first stage of mining and loading and the machine set MD for the second stage of transportation. The multi-source data is abstracted into workpieces. The two stages are traversed according to the processing order of all processes on each machine in the two stages to obtain the corresponding solution scheme. Each solution scheme includes m sub-sequences, which correspond to m machines in the stage respectively. The solution scheme is discretized to obtain the initial feature representation. The solution space corresponding to the discretization encoding method satisfies the following relationship: ； In the formula, Indicates the number of stages. For the number of workpieces, Let be the number of machines in the i-th stage. It represents factorial.

3. The intelligent scheduling method for underground mine ore extraction and transportation equipment according to claim 1, characterized in that, The objective function of the optimization algorithm is constructed based on the multi-dimensional constraint data, including: Based on multi-dimensional constraint data, with minimizing transportation time as the optimization objective, an objective function F of the optimization algorithm is constructed, which satisfies the following relationship: ； In the formula, This represents the end time of the i-th workpiece on the machine during the transportation phase. This indicates the start time of the i-th workpiece on the machine during the loading and unloading phase.

4. The intelligent scheduling method for underground mine ore extraction and transportation equipment according to claim 1, characterized in that, The particle position vectors are initialized to obtain an initial population, and the initial population is iteratively updated, including: A heuristic algorithm is used to encode half of the particle position vectors to obtain permuted coded individuals, and a random strategy is used to encode the other half of the particle position vectors to obtain randomly coded individuals. The permuted coded individuals and the randomly coded individuals are combined to obtain the initial population. A particle update mechanism is established to update the position vector of each particle, resulting in the updated position vector of each particle. The update mechanisms include: a speed update mechanism and a position update mechanism. The speed update mechanism satisfies the following relationship: ； The location update mechanism satisfies the following relationship: ； In the formula, This represents the velocity vector of particle i at the (t+1)th iteration. This represents the velocity vector of particle i at the t-th iteration. This represents the position of particle i in the (t+1)th iteration. This represents the position of particle i at the t-th iteration. Let represent the optimal position of particle i in the t-th iteration. This indicates the global optimum of the particle swarm at iteration t. Indicates inertia weight, For individual learning factors, As a social learning factor, and It is a random number in [0,1].

5. The intelligent scheduling method for underground mine ore extraction and transportation equipment according to claim 1, characterized in that, In the initial population iterative update, the path reconnection algorithm is used to select candidate solutions and construct a tabu search candidate set, including: Using the individual's current solution, the individual's historical best solution, and the population's historical best solution as core nodes, we construct search paths from the current solution to the individual's historical best solution and from the current solution to the population's historical best solution. For each candidate solution on the search path, calculate its sequence similarity S(P,Q) with the individual's current solution, the individual's historical best solution, and the population's historical best solution. A preset similarity threshold range is set, and candidate solutions with similarity within this threshold range are selected and added to the tabu search candidate set. The Kendall-Tau distance is introduced during the process of adding sequences to the tabu search candidate set. Given two sequences P and Q of length n, the Kendall-Tau distance is defined as follows: ； In the formula, This represents the i-th element of sequence P. This represents the i-th element of sequence Q. The logical AND operator, The operator represents the logical OR. Indicates the Kendall-Tau distance; The similarity calculation formula is defined as follows: ； In the formula, n represents the number of elements in the sequence, and S(P,Q) is the similarity between the two sequences.

6. The intelligent scheduling method for underground mine ore extraction and transportation equipment according to claim 1, characterized in that, The overall performance value of each particle's position vector is calculated using the fitness function, including: Input each particle position vector into the objective function, calculate the objective function F, and then substitute the objective function F into the fitness function to calculate the comprehensive performance value of each particle position vector; The fitness function satisfies the following relationship: ； In the formula, The objective function value, This represents the overall performance value for each particle's position vector.

7. The intelligent scheduling method for underground mine ore extraction and transportation equipment according to claim 1, characterized in that, The particle with the highest overall performance value is used as the optimal scheduling scheme to generate a specific execution plan, including: Output the optimal code corresponding to the particle with the highest overall performance value, and generate a Gantt chart based on the optimal code; The Gantt chart of the process is decoded and transformed into a specific execution plan.

8. An intelligent scheduling system for ore-carrying equipment in underground mines, characterized in that, Including processor and memory; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method described in any one of claims 1-7.

Images