Coal bulk cargo loading and unloading efficiency optimization method and system based on machine learning

Abstract

The invention provides a coal bulk loading and unloading efficiency optimization method and system based on machine learning, and relates to the technical field of coal bulk loading and unloading, and the method comprises the steps: recognizing key factors and quantitative association rules between the key factors and the coal bulk loading and unloading efficiency through a large language model; the storage yard decision topological graph is input into a reinforcement learning model, a coal bulk cargo loading and unloading scheme is determined by executing the process of state space coding, action selection and reward function calculation, and a reward function is obtained by taking a predicted value of the coal bulk cargo loading and unloading efficiency in the current period as a target and being constructed according to key factors and quantitative association rules; after the coal bulk cargo loading and unloading scheme is completed, the actual value of the coal bulk cargo loading and unloading efficiency in the current period is collected, the efficiency difference value is calculated, and the coal bulk cargo loading and unloading efficiency in the next period is optimized based on the efficiency difference value. According to the method, the efficiency can be effectively improved while various necessary conditions of loading and unloading operation are accurately met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal bulk cargo loading and unloading, and particularly relates to a coal bulk cargo loading and unloading efficiency optimization method and system based on machine learning. BACKGROUND

[0002] Coal bulk cargo loading and unloading is a key link in the coal logistics supply chain, wherein the loading and unloading operation as a core process of the connection between the port and the railway involves multi-dimensional coordination such as train scheduling, coal matching, stacker / reclaimer coordination, and stack yard stacking position management. In this scenario, necessary conditions such as coal and stacking position matching, stacker / reclaimer operation without conflict, stacking capacity upper limit, and empty stacking position disablement need to be met, while requirements such as efficient use of equipment (such as reducing waiting time), energy saving (such as shortening the moving distance of equipment), and safety (such as high temperature limitation and anti-collision) need to be considered to achieve the core goal of the shortest total operation time. With the increase of coal transportation volume and the expansion of the stack yard scale, the constraint conditions of the unloading operation are becoming more and more complex, and higher requirements are put forward for accurate and efficient operation scheme formulation.

[0003] In the prior art, the loading and unloading operation scheme depends on manual scheduling and experience decision. The scheduling personnel need to manually plan the unloading sequence, allocate stacking positions, and stack the reclaimer in combination with train forecast information (such as train number, coal type, tonnage), arrival time of the railway dispatching system, existing stacking position state of the stack yard (such as coal type, capacity, equipment association), and equipment operation state. In the decision-making process, the constraints such as coal matching and stacker / reclaimer conflict need to be manually checked, and the operation mode (such as preferentially processing small-capacity stacking positions or shortening the moving distance of equipment) is selected according to experience, and finally the operation plan is formed and issued to the field for execution.

[0004] However, the existing technology relies too much on manual experience, which has significant limitations: on the one hand, manual decision-making is difficult to handle multi-dimensional constraints (such as equipment failure, high temperature limitation, and stacking capacity upper limit) in real time, which easily leads to equipment occupation conflict, unreasonable stacking position allocation, and other problems, increasing the waiting time; on the other hand, manual decision-making lacks quantitative basis for efficiency target, and it is difficult to achieve global optimization. In complex scenarios such as concentrated arrival of trains and dynamic changes of equipment state, the decision-making efficiency is low, the error rate is high, and it is difficult to meet the efficient and refined needs of large-scale unloading operations. SUMMARY

[0005] The present application aims to provide a coal bulk cargo loading and unloading efficiency optimization method and system based on machine learning to solve the problem of low decision-making efficiency, high error rate, and inability to meet the efficient and refined needs of large-scale unloading operations in the prior art in complex scenarios such as concentrated arrival of trains and dynamic changes of equipment state.

[0006] To solve the above technical problems, in a first aspect, the application provides a coal bulk cargo loading and unloading efficiency optimization method based on machine learning, comprising: Collecting plan data, equipment state position data, stockyard resource occupation data and stockyard constraint data of coal bulk cargo loading and unloading operation to form a data set, wherein the stockyard constraint data includes coal type matching requirements, stacker-reclaimer operation conflict constraints, stack capacity upper limit, empty stack position usage restrictions, equipment failure correlation constraints and high-temperature operation restrictions; According to the data set, a constrained stockyard decision topology graph is generated; Using a large language model to identify key factors related to coal bulk cargo loading and unloading efficiency and quantitative association rules between key factors and coal bulk cargo loading and unloading efficiency; A deep double Q network is used as a framework of a reinforcement learning model, the stockyard decision topology graph is input into the reinforcement learning model, and through the processes of state space coding, action selection and reward function calculation, a coal bulk cargo loading and unloading scheme is determined, wherein the reward function aims to maximize the predicted value of the coal bulk cargo loading and unloading efficiency in the current period, and is constructed according to the key factors and the quantitative association rules; After the completion of the coal bulk cargo loading and unloading scheme, the actual value of the coal bulk cargo loading and unloading efficiency in the current period is collected, the efficiency difference value is calculated, and based on the efficiency difference value, the coal bulk cargo loading and unloading efficiency in the next period is optimized.

[0007] Optionally, the input of the stockyard decision topology graph into the reinforcement learning model, through the processes of state space coding, action selection and reward function calculation, to determine the coal bulk cargo loading and unloading scheme, comprises: The stockyard decision topology graph is input into the reinforcement learning model, and the node features and edge weights of the stockyard decision topology graph are aggregated by a multi-order neighborhood of a spatio-temporal graph convolution network of the reinforcement learning model to obtain a dynamic state vector that fuses real-time equipment positions, stack position coal type attributes, task queues and equipment failure states; Based on the deep double Q network, an action value vector corresponding to the dynamic state vector is calculated, and according to the action value with the maximum value in the action value vector, a corresponding action is selected, wherein the action includes a stack position coordinate allocation sequence of a stacker-reclaimer, a coal type identification configuration scheme of a conveyor belt and an equipment movement scheme; After the action is executed, observation data including actual loading and unloading amount, equipment movement distance and violation event count are determined; According to the observation data and the reward function, a reward value is calculated, wherein the reward function is constructed according to the key factors and the quantitative association rules, and includes an efficiency reward item, a compliance reward item, a movement cost penalty item and an energy saving reward item; Based on the reward value, the action value is updated using the time difference error, and when the fluctuation amplitude value of the action value of consecutive N decisions is less than a preset amplitude threshold, it is determined that the reinforcement learning model converges, and a coal bulk cargo handling scheme including a stack position coordinate allocation sequence of the stacker-reclaimer, a coal type identification configuration scheme of the conveyor belt, and a device movement scheme is output, and N is greater than or equal to 3.

[0008] Optionally, the deep double Q network is used to calculate the action value vector corresponding to the dynamic state vector, and the action corresponding to the maximum value in the action value vector is selected according to the maximum value in the action value vector, the action including: a stack position coordinate allocation sequence of the stacker-reclaimer, a coal type identification configuration scheme of the conveyor belt, and a device movement scheme, including: The dynamic state vector is input into the deep double Q network, and multiple types of action values are output, including: an action value of each stack position coordinate as a matching position of the stacker-reclaimer, an action value of each coal type identification activated and configured on the conveyor belt, and an action value of each path sequence as a device movement scheme; According to the train station arrival countdown or the ship berthing countdown in the plan data, the task emergency weight is calculated in combination with the user level, the task level, and the yard operation mode in the plan data; According to the task emergency weight and the priority corresponding to the yard operation mode, the action value corresponding to the stack position coordinate is adjusted; The stack position coordinate with the highest action value is selected as the matching position of the stacker-reclaimer, the coal type identification with the highest adjusted action value is selected as the configuration instruction of the conveyor belt, and the path sequence with the highest action value is selected as the device movement scheme, to obtain the corresponding action.

[0009] Optionally, the task emergency weight is calculated according to the train station arrival countdown or the ship berthing countdown in the plan data, in combination with the user level, the task level, and the yard operation mode in the plan data, including: A first difference value between the train station arrival estimated time and the current time or a second difference value between the ship berthing estimated time and the current time is calculated; The ratio of the first difference value to a preset time window is taken as the train station arrival countdown, or the ratio of the second difference value to the preset time window is taken as the ship berthing countdown; The user level, the task level, and the yard operation mode are quantified respectively to obtain a user level quantification value, a task level quantification value, and a yard operation mode quantification value; The superposition result of the train station arrival countdown, the user level quantification value, the task level quantification value, and the yard operation mode quantification value is taken as the emergency weight of the unloading task, and the superposition result of the ship berthing countdown, the user level quantification value, the task level quantification value, and the yard operation mode quantification value is taken as the emergency weight of the loading task.

[0010] Optionally, the key factors related to the coal bulk cargo loading and unloading efficiency and the quantitative correlation rules between the key factors and the coal bulk cargo loading and unloading efficiency are identified by using a large language model, including: calculating the Pearson correlation coefficients of all fields in the data set and the loading and unloading efficiency indicators, and screening structured fields and unstructured text fields with Pearson correlation coefficients greater than a preset coefficient threshold; mapping the structured fields and the unstructured text fields to knowledge graph nodes according to a space-time coordinate, calculating the edge weights between the nodes based on historical operation data, and forming a knowledge graph with a loading and unloading process chain as the skeleton and carrying the loading and unloading process dependency relationship and the edge weights; Based on the knowledge graph, combined with a domain knowledge base and a preset causal template, a multi-round chain reasoning oriented to the coal bulk cargo loading and unloading efficiency is performed, and a key factor list and a quantitative correlation rule are output, the key factor list including a plurality of key factors.

[0011] Optionally, the multi-round chain reasoning oriented to the coal bulk cargo loading and unloading efficiency based on the knowledge graph, combined with the domain knowledge base and the preset causal template, includes: Similarity matching between the knowledge graph nodes and the preset causal template is performed to screen out a candidate node set affecting the coal bulk cargo loading and unloading efficiency; Efficiency influence weights of the candidate nodes are calculated; For candidate nodes with efficiency influence weights greater than a preset weight threshold, gradient significance test values corresponding to the candidate nodes are differentially calculated based on historical operation data; The candidate nodes with gradient significance test values greater than a preset test threshold are screened as key factors to output a key factor list; The correlation rule between the key factors and the coal bulk cargo loading and unloading efficiency is quantified by a preset fitting method to obtain a quantitative correlation rule.

[0012] Optionally, the efficiency difference value is used to optimize the coal bulk cargo loading and unloading efficiency in the next period, including: When the efficiency difference value is less than or equal to a preset difference value, the weights corresponding to the efficiency reward item, the compliance reward item, the movement cost penalty item and the energy saving reward item in the reward function are adjusted according to the offset direction and size of the efficiency difference value, so that the coal bulk cargo loading and unloading efficiency in the next period is optimized according to the adjusted reward function; Or, when the efficiency difference value is greater than the preset difference value, the large language model is adjusted according to the prediction bias sample set, so that the coal bulk cargo loading and unloading efficiency in the next period is optimized according to the adjusted large language model.

[0013] In a second aspect, the present application provides a coal bulk cargo loading and unloading efficiency optimization system based on machine learning, including: The collection module is configured to collect plan data, equipment state position data, stockyard resource occupation data and stockyard constraint data of the coal bulk cargo loading and unloading operation to form a data set; The generation module is configured to generate a stockyard decision topology graph with constraints according to the data set; The identification module is configured to identify, by using a large language model, key factors related to the coal bulk cargo loading and unloading efficiency and quantitative association rules between the key factors and the coal bulk cargo loading and unloading efficiency; The reinforcement learning module is configured to use a deep double Q network as a framework of a reinforcement learning model, input the stockyard decision topology graph into the reinforcement learning model, and determine a coal bulk cargo loading and unloading scheme by performing a process of state space coding, action selection and reward function calculation, wherein the reward function is constructed according to the key factors and the quantitative association rules and aims to maximize a predicted value of the coal bulk cargo loading and unloading efficiency in a current period. The calculation optimization module is configured to collect an actual value of the coal bulk cargo loading and unloading efficiency in the current period after the coal bulk cargo loading and unloading scheme is completed, calculate an efficiency difference value, and optimize the coal bulk cargo loading and unloading efficiency in a next period based on the efficiency difference value.

[0014] In a third aspect, the present application provides an electronic device, comprising: A memory configured to store a computer program; A processor configured to execute the computer program to implement the steps of the coal bulk cargo loading and unloading efficiency optimization method based on machine learning according to the first aspect.

[0015] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executable by a processor to implement the steps of the coal bulk cargo loading and unloading efficiency optimization method based on machine learning according to the first aspect.

[0016] The coal bulk cargo loading and unloading efficiency optimization method based on machine learning provided in the application comprises the following steps: collecting plan data, equipment state position data, stockyard resource occupation data and stockyard constraint data of coal bulk cargo loading and unloading operation, forming a data set, the stockyard constraint data comprising coal type matching requirements, stacker-reclaimer operation conflict constraints, stack capacity upper limit, empty stack position use restriction, equipment fault correlation constraints and high-temperature operation restriction; generating a stockyard decision topology graph with constraints according to the data set; identifying key factors related to the coal bulk cargo loading and unloading efficiency and the quantitative association rules between the key factors and the coal bulk cargo loading and unloading efficiency by using a large language model; taking a deep double Q network as a framework of a reinforcement learning model, inputting the stockyard decision topology graph into the reinforcement learning model, and determining a coal bulk cargo loading and unloading scheme through the processes of state space coding, action selection and reward function calculation, the reward function being constructed according to the key factors and the quantitative association rules and aiming to maximize the predicted value of the coal bulk cargo loading and unloading efficiency in the current period; after the coal bulk cargo loading and unloading scheme is completed, collecting the actual value of the coal bulk cargo loading and unloading efficiency in the current period, calculating an efficiency difference value, and optimizing the coal bulk cargo loading and unloading efficiency in the next period based on the efficiency difference value.

[0017] The application has the following advantages by adopting the above technical solution: (1) The application generates a stockyard decision topology graph with constraints by collecting stockyard constraint data including coal type matching requirements, stacker-reclaimer operation conflict constraints, stack capacity upper limit (27000 tons), empty stack position use restriction, equipment fault correlation constraints and high-temperature operation restriction (not unloading at a temperature above 40 degrees Celsius), and makes a scheme decision based on the graph, which can ensure that all unloading operations strictly follow the necessary conditions and avoid illegal operations (such as loading coal into empty stack positions and simultaneous operation of stackers and reclaimers at the same stack position) caused by human judgment omissions, thereby fundamentally guaranteeing the compliance and stability of the operation.

[0018] (2) Based on the key factors (such as stacker movement distance, stack capacity, equipment occupation conflict, etc.) and the quantitative association rules identified by the large language model, and combined with the reward function (covering efficiency, energy saving, movement cost, etc.) aiming to maximize the loading and unloading efficiency, the reinforcement learning model can automatically generate an optimal unloading scheme: preferentially realizing "all trains operating at the same time" to reduce waiting time and shorten total operation time, which meets the core index of "the shortest total operation time being the best"; minimizing stacker movement distance and preferentially selecting the smallest stack to directly respond to energy saving needs and reduce equipment energy consumption and loss; balancing equipment occupation conflicts to avoid efficiency loss caused by resource competition and improve equipment utilization.

[0019] (3) The reinforcement learning model of the present application can be flexibly adapted to different operation modes according to the actual situation of the yard: in the preparation mode, the pile position with small stacking capacity is preferentially selected, in the balance mode, the cargo volume of the BQ line pile position is evenly realized, and in the high-yield mode, the unloading time is shortened, without manual switching of decision logic, which significantly improves the adaptability to complex yard scenes and ensures that efficient solutions can be output under different operation objectives.

[0020] (4) By collecting the actual efficiency value after the operation is completed, calculating the efficiency difference value and feeding back to the model optimization process (adjusting the reward function weight or large language model parameter), the present application can realize dynamic iteration of the decision scheme. In view of sudden situations such as equipment failure and coal type change, the model can continuously optimize the decision logic through continuous learning, avoid the limitations caused by fixed experience, and long-term guarantee the efficiency and adaptability of unloading operation.

[0021] (5) The scheme includes anti-collision consideration (such as one pile position interval between stacking and taking material) and high-temperature operation restriction, etc. safety conditions, and automatically checks through the model to ensure execution, reduces the safety hazards that may be ignored by manual decision, and provides stable safety guarantee for unloading operation.

[0022] (6) Further, the method adjusts the action value according to the task emergency weight and operation mode by fusing dynamic information such as equipment, pile position, task, etc., selects the optimal pile position distribution, conveyor belt configuration and equipment movement scheme, optimizes the decision by combining the reward function including efficiency, compliance, movement cost and energy saving, can meet the necessary conditions of unloading operation, reduce the waiting and movement cost, adapt to multi-mode demand, and improve the efficiency and energy saving effect. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical scheme of the embodiments of the present application or the prior art, the drawings needed to be used in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creating any creative labor.

[0024] Figure 1 A flowchart of a coal bulk cargo loading and unloading efficiency optimization method based on machine learning provided by an embodiment of the present application; Figure 2 An application scenario diagram of a coal bulk cargo loading and unloading efficiency optimization method based on machine learning provided by an embodiment of the present application; Figure 3 A structure diagram of a coal bulk cargo loading and unloading efficiency optimization system based on machine learning provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to solve the problems of low decision efficiency, high error rate and inability to meet the efficient and refined needs of large-scale unloading operation in the prior art under complex scenarios such as train concentration in port and dynamic change of equipment state, the embodiment of the present application provides a coal bulk cargo loading and unloading efficiency optimization method based on machine learning. The method adopts the following concept: through system collection of operation plans, equipment states, stockyard resources and various constraint conditions (such as coal type matching, equipment conflict, high-temperature operation restriction, etc.), overall decision basis is formed; then, intelligent model is used to analyze key influencing factors and rules, and finally the optimal loading and unloading scheme is calculated, and can be continuously optimized according to the actual effect. This way, on the one hand, through automatic processing of complex constraints, it avoids the omissions of manual decision and reduces equipment conflict and waiting time; on the other hand, with the help of data-driven quantitative analysis, global optimal decision is realized, which can maintain efficiency and accuracy even in complex scenarios, thereby effectively making up for the shortcomings of traditional manual scheduling.

[0026] In order for those skilled in the art to better understand the present application, the present application will be further described in detail below in conjunction with the drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] The core of the present application is to provide a coal bulk cargo loading and unloading efficiency optimization method based on machine learning. The flowchart of one specific embodiment is shown in Figure 1 The method comprises: S11, collecting plan data, equipment state position data, stockyard resource occupation data and stockyard constraint data of coal bulk cargo loading and unloading operation, forming a data set, and the stockyard constraint data includes coal type matching requirement, stacker-reclaimer operation conflict constraint, stack capacity upper limit, empty stack position use restriction, equipment fault correlation constraint and high-temperature operation restriction.

[0028] In step S11, the coal bulk cargo loading and unloading operation includes coal bulk cargo unloading operation and coal bulk cargo loading operation. The coal bulk cargo unloading operation refers to temporarily storing coal on trains into specific stacks in the stockyard by various intelligent devices, and the coal bulk cargo loading operation refers to loading coal in specific stacks in the stockyard onto ships by various intelligent devices. The intelligent devices include stacker, reclaimer, conveyor belt, car dumper, ship loader, intelligent monitoring camera, Internet of Things sensor, central control system, etc. Figure 2As shown, the dumper is provided with 4, respectively CD1, CD2, CD3 and CD4, after the train arrives, start working. The belt conveyor (referred to as the belt) below the dumper funnel is also provided with 4, respectively BF1, BF2, BF3 and BF4, for conveying coal; each belt is provided with a corresponding belt scale of the dumper, the code is BSBF1, BSBF2 and BSBF3 and BSBF4, for obtaining the relevant electronic scale information of the corresponding dumper, the relevant electronic scale information includes instantaneous value, cumulative value and dumper number. In order to transport safely, the application embodiment is provided with a belt conveyor switching tower at different belt positions: T4, T5, T6, T7, T8, T9, T10, T11 and the like; BH1-3, BH1-4, BH2-3, BH2-4, BH3-3, BH3-4, BH4-3, BH4-4 represent the code of the middle belt in the corresponding longitudinal different stacking line, and the middle belt of the taking line is provided with 3: BJ1-1, BJ2-1 and BJ3-1; MSBQ1, MSBQ2 and MSBQ3 represent the corresponding electromagnetic separator respectively, the stacker belt has 4, respectively BD1, BD2, BD3 and BD4. The reclaimer belt has 4, respectively BQ1, BQ2, BQ3 and BQ4. The ship loader has three: SL1, SL2 and SL3; the connecting conveyor belt of the ship loader has three: BM1, BM2 and BM3. The application is provided with 4 stackers in the stockyard, respectively S1, S2, S3 and S4. The application is provided with a plurality of reclaimer in the stockyard: R1-1, R1-2, R2-1, R2-2, R3-1, R3-2. Exemplarily, the number of rows of the stockyard can be 6, and the number of columns can be 4, so the number of the stockyard can be 24, such as the stockyard 101, the stockyard 102, the stockyard 103,..., the stockyard 604, and the type of coal has many types, for example: coal A, coal B, coal C, coal D, coal E. Among them, the stacker is used to stack the coal after unloading to the designated stockyard; the reclaimer is responsible for taking coal from the stockyard for ship loading; the conveyor belt undertakes the transportation of coal between different equipment and stockyard, and between ship; the dumper is used to automatically unload the coal in the train compartment to the conveyor belt; the ship loader accurately loads the conveyed coal into the ship cabin; the intelligent monitoring camera monitors the working site state in real time; the internet of things sensor collects equipment operating parameters, coal flow, stockyard capacity and other data; the central control system coordinates and remotely controls all kinds of equipment to ensure the automatic and intelligent operation of unloading and loading.

[0029] And, the plan data includes forecast information of the train / ship, such as train number, ship name, coal type, tonnage, estimated arrival time, etc.; the equipment state position data can indicate the real-time running state (such as normal / fault) and the position (such as pile position coordinates, track position, etc.) of the equipment such as the stacker-reclaimer and the conveyor belt; the stockyard resource occupation data is the information of the current coal type, the occupied amount, the remaining capacity, etc. of each pile position; the stockyard constraint data is the rule that must be followed in the operation, including the coal type matching requirement (including: specific coal type can only be stacked in the corresponding pile position, the same pile position can only stack the same coal type, etc.), the stacker-reclaimer operation conflict constraint (such as the same pile position cannot simultaneously carry out stacking and reclaiming), the stacking capacity upper limit (that is, the maximum coal loading amount of a single pile position, such as 27000 tons), the empty pile position use limitation (such as the pile position without associated coal type cannot load coal), the equipment fault association constraint (such as the pile position corresponding to the fault equipment cannot be used), and the high-temperature operation limitation (above a specific temperature, such as 40 degrees Celsius, no unloading is carried out).

[0030] In the embodiment of the present application, first, the plan data is collected from the train dispatching system and the ship forecasting system, for example, the train number K, the coal type bituminous coal, the tonnage 5000 tons, and the estimated arrival time 10:00 are obtained; second, the equipment state position data is collected through the equipment sensor, such as the state (normal operation) and the position (currently in pile position 2) of the stacker-reclaimer B, and the state (normal) of the conveyor belt C; third, the pile position resource data is called from the stockyard management system, such as the pile position 1 (coal type bituminous coal, occupied 10000 tons, upper limit 27000 tons), the pile position 2 (coal type non-bituminous coal, occupied 8000 tons), and the pile position 3 (empty pile position, no associated coal type) and the like stockyard resource occupation data; finally, the preset stockyard constraint data is arranged, such as “bituminous coal can only be stacked in pile positions 1, 4, and 5”, “the stacker-reclaimer cannot simultaneously operate in the same pile position”, “no unloading above 40℃” and the like, and all the data is summarized to form a data set.

[0031] S12, generating a stockyard decision topology graph with constraints according to the data set.

[0032] In step S12, the stockyard decision topology graph with constraints is a structure that displays the related elements and constraint relationships of the stockyard operation in a graphical way, wherein the “node” can represent the train / ship task, the equipment (stacker-reclaimer, conveyor belt), the pile position, etc., and the “edge” indicates the association or constraint between the nodes (such as a certain stacker-reclaimer can reach a certain pile position, a certain coal type matches a certain pile position, a certain equipment has a conflict with a certain pile position, etc.).

[0033] In the embodiments of the present application, based on the data set formed by S11, various elements are converted into nodes of the topology graph, for example, train K (task node), stacker-reclaimer B (device node), and stack position 1 (stack position node) are taken as nodes; then, edges between nodes are drawn according to the constraint relationship in the data, for example, the coal type of train K (bituminous coal) matches the coal type of stack position 1, and a solid line edge is connected and marked as "coal type matching"; stacker-reclaimer B is currently at stack position 2, and the moving path to stack position 1 has no conflict, a dashed line edge is connected and marked as "movable"; stack position 3 is an empty stack position without associated coal type, and is connected with train K (bituminous coal) by an edge with "X" and marked as "empty stack position restriction"; finally, a topology graph containing all tasks, devices, stack positions and constraint relationships is formed.

[0034] S13, using a large language model to identify key factors related to coal bulk cargo loading and unloading efficiency and quantitative association rules between key factors and coal bulk cargo loading and unloading efficiency.

[0035] In step S13, the large language model is an artificial intelligence model capable of processing structured data and text information and identifying rules; the key factor refers to a factor that has a significant impact on loading and unloading efficiency (such as device moving distance, stack position remaining capacity, task urgency, etc.); the quantitative association rule refers to the mathematical relationship between the key factor and the loading and unloading efficiency (such as "every 50 meters increase in device moving distance, 3 minutes increase in loading and unloading time"); the Pearson correlation coefficient is used to measure the degree of linear correlation between two variables (the closer the value is to 1 or -1, the stronger the correlation).

[0036] In the embodiments of the present application, the specific description of this step S13 can refer to steps 131-133 described below, which will not be repeated here.

[0037] S14, taking a deep double Q network as the framework of a reinforcement learning model, inputting the stockyard decision topology graph into the reinforcement learning model, determining a coal bulk cargo loading and unloading scheme through the processes of state space coding, action selection and reward function calculation, and constructing the reward function according to the key factors and the quantitative association rules, with the goal of maximizing the predicted value of the coal bulk cargo loading and unloading efficiency in the current period.

[0038] In S14, the reinforcement learning model is an artificial intelligence model that optimizes decisions by interacting with the environment and learning through trial and error, and the deep double Q network as its framework is composed of two neural networks to reduce decision bias. The reward function is used to evaluate the goodness of actions, including efficiency reward items (add points for shortening time), compliance reward items (add points for complying with constraints), moving cost penalty items (deduct points for long moving distance), and energy saving reward items (add points for energy saving actions).

[0039] In the embodiments of the present application, the specific description of this step S14 can refer to steps 141-145 described below, which will not be repeated here.

[0040] S15, after the coal bulk cargo handling scheme is completed, the actual value of the coal bulk cargo handling efficiency in the current period is collected, the efficiency difference value is calculated, and the coal bulk cargo handling efficiency in the next period is optimized based on the efficiency difference value.

[0041] The size of the preset difference value can be set comprehensively according to the accuracy requirement of the actual operation scene, the fluctuation range of the historical data and the business target, and is not a fixed value. For example, in the coal bulk cargo handling operation, if the "operation time" is taken as the efficiency measurement index, and combined with the common operation accuracy requirement in the industry, the preset difference value can be set to 5-10 minutes (which can be adjusted according to the daily operation volume of the port, the running stability of the equipment, etc.): when the difference between the actual operation time and the predicted time is within 5 minutes, it is considered that the model prediction deviation is small, and only the reward function weight needs to be fine-tuned; if the difference exceeds 10 minutes, it is determined that the deviation is large, and the large language model needs to be adjusted to optimize the key factor identification logic. The core is to make the preset difference value not only reflect the acceptable error range of the model prediction, but also trigger an effective correction mechanism in time to ensure that the model continues to adapt to the actual operation changes.

[0042] Specifically, when the efficiency difference value is less than or equal to the preset difference value, the weights corresponding to the efficiency reward item, the compliance reward item, the moving cost penalty item and the energy saving reward item in the reward function are adjusted according to the offset direction and size of the efficiency difference value, so as to optimize the coal bulk cargo handling efficiency in the next period according to the adjusted reward function. Or, when the efficiency difference value is greater than the preset difference value, the large language model is adjusted according to the prediction deviation sample set, so as to optimize the coal bulk cargo handling efficiency in the next period according to the adjusted large language model. The preset deviation sample set contains samples corresponding to mode selection deviation, constraint satisfaction deviation, etc. It should be noted that in the reward function, the formula or expression of each item is not specifically limited in the present application. Moreover, based on the efficiency difference value in the current period, the coal bulk cargo handling efficiency in the next period can be optimized. Similarly, the reward function, the large language model, etc. used in the current period are obtained by modifying the efficiency difference value in the last period.

[0043] For example, if the actual efficiency is better than the predicted efficiency and mainly benefits from the short moving distance of the equipment, the weight of the moving cost penalty item can be appropriately increased, so that the model pays more attention to shortening the moving distance of the equipment in the next period; if the efficiency difference value is greater than the preset difference value, it indicates that the model prediction deviation is large, at this time, samples with large prediction deviation are collected to form a prediction deviation sample set, and the large language model used to identify key factors and quantify association rules in S13 is adjusted based on the sample set, such as retraining the model to correct the judgment of the key factors or the quantization relationship of the association rules, so that the adjusted large language model can more accurately mine the factors and rules affecting the efficiency, and then make the handling scheme in the next period more suitable for the actual operation situation, and finally realize the optimization of the coal bulk cargo handling efficiency in the next period.

[0044] Based on the above process, the following specific examples are provided: Port A receives the unloading task of train K, the train number of train K is A01, the transported coal type is bituminous coal, the tonnage is 5000 tons, and it is expected to arrive at the port at 10:00. Subsequently, port A starts the coal bulk cargo handling efficiency optimization process based on machine learning.

[0045] First, collect data comprehensively at S11. Obtain the planned data of train K from the railway transportation system, collect the status and position of stacker-reclaimers B and D through Internet of Things devices, where stacker-reclaimer B is in normal status and located at bay 2, and stacker-reclaimer D is malfunctioning and disabled for bay 6. Retrieve the bay information from the stockyard system, bay 1 stores bituminous coal, has stored 10000 tons, and the upper limit of capacity is 27000 tons; bay 4 stores bituminous coal, has stored 15000 tons; bay 3 is an empty bay and not associated with a coal type. At the same time, sort out the constraint rules, bituminous coal can only be stored in bays 1, 4, and 5, stacker-reclaimers cannot work in the same bay at the same time, and work cannot be performed when the environmental temperature reaches 40°C or above. Finally, a complete data set is formed.

[0046] Next, generate a constrained stockyard decision topology graph based on the data at S12. Take train K as the task node, stacker-reclaimer B as the equipment node, and bays 1, 4, and 3 as the bay nodes, which are collectively the core nodes. Connect train K with bays 1 and 4 with "coal type matching" edges, connect train K with bay 3 with "empty bay restriction" edges, and label the paths of stacker-reclaimer B to bays 1 and 4 with "movable" edges, while labeling the remaining capacities of bays 1 and 4, which all meet the 5000-ton storage demand.

[0047] Subsequently, at S13, key factors are mined through a large language model. Based on the historical bituminous coal unloading data of port A, it is calculated that the Pearson correlation coefficients of "equipment moving distance", "bay remaining capacity" and total operation time are 0.7 and 0.65 respectively, both of which are higher than the threshold value of 0.6. After mapping them as knowledge graph nodes, it is determined through multiple rounds of reasoning that both are key factors, and the quantitative association rule is total operation time = 20 minutes + 0.1 minute / meter x moving distance - 0.001 minute / ton x remaining capacity, i.e. the shorter the moving distance and the larger the remaining capacity, the shorter the operation time.

[0048] In S14, the reinforcement learning model generates an optimal solution. First, the topological graph features are aggregated by the spatio-temporal graph convolution network to obtain a dynamic state vector, which indicates that the stacker B is at bay 2, there are 17,000 tons of bituminous coal left in bay 1, and the train K has 30 minutes left. The initial output of the deep double Q network is that the value of “stacker B to bay 1” is 80 points, and the value of “to bay 4” is 70 points. After adjusting the emergency weight of train K to 0.8, the action value of bay 1 is increased to 85 points. After executing this action, it is observed that the moving distance is 50 meters, there is no violation, and the loading and unloading amount is 5,000 tons, and the reward value is calculated to be 30 points, including 15 points for efficiency, 10 points for compliance, and 5 points for moving cost. After 3 iterations, the model decision fluctuation is less than 3%, and the converged output solution is that the stacker B moves from bay 2 to bay 1 along the straight path, and the conveyor belt is configured with the bituminous coal identifier.

[0049] Finally, in S15, dynamic optimization is performed. The actual unloading time of train K is 28 minutes, which is -2 minutes different from the predicted value of 30 minutes. Since the difference is less than the preset difference of 5 minutes, the weight of the efficiency reward item in the reward function is increased. In the next cycle of processing the same type of task, the actual time of the model output solution is further shortened to 26 minutes, realizing the continuous optimization of efficiency.

[0050] The above method provided by the present application can have the following advantages by executing all the above steps: (1) Intelligent decision-making throughout the whole process is realized, breaking through the limitations of artificial experience. Specifically, by S11, comprehensive data (plans, equipment, yards, and constraints) are collected to provide a complete information base for decision-making; S12 converts the data into a visual topological graph to intuitively present the association and constraints of tasks, equipment, and bays; S13 uses a large language model to mine key factors (such as equipment moving distance and bay remaining capacity) and quantitative rules (such as “every 50 meters of moving distance is shortened, 5 minutes of operation time is reduced”), replacing artificial experience as a basis for decision-making; S14 automatically generates an optimal solution by a reinforcement learning model, which selects actions (such as the optimal path of the stacker and the configuration of the conveyor belt) to achieve efficiency, compliance, energy saving, etc. The whole process does not require human intervention, completely eliminating the dependence on experience, making the decision more accurate and scientific.

[0051] (2) Improve the efficiency of loading and unloading and resource utilization, and reduce operating costs. Specifically, through the reinforcement learning model of S14, the scheme can preferentially select efficient actions such as shortening the movement distance of the equipment, matching the remaining capacity of the stack, etc. (such as the example where the stacker-reclaimer selects the stack 1 with a shorter movement distance, reducing invalid movement), directly shortening the total operation time. At the same time, the model adjusts the action value through the task emergency weight, ensuring that high-priority tasks (such as the train arriving soon) are handled first, reducing waiting time. Combined with the dynamic optimization of S15 (such as increasing the weight of the efficiency reward term), the scheme is continuously iterated, further shortening the operation time of subsequent similar tasks (from 28 minutes to 26 minutes in the example), significantly improving the overall loading and unloading efficiency. In addition, the allocation of resources is more reasonable (such as avoiding equipment conflicts and making full use of effective stacks), reducing equipment idle and resource waste, and indirectly reducing operating costs.

[0052] (3) Strictly follow the operation constraints to ensure safety and standardization of the operation. Specifically, the scheme explicitly includes constraint data such as coal type matching, stacker-reclaimer conflicts, and high-temperature restrictions in S11, and the topological graph of S12 visually marks the constraint relationship (such as empty stack restrictions and equipment unreachable paths). The reward function of S14 uses compliance rewards (points for complying with constraints) and penalties for violations to ensure that the scheme strictly complies with the rules (such as no violations in the example). This effectively avoids the risk of violations caused by human errors (such as loading coal into empty stacks or high-temperature operations), ensuring the safety of the operation. At the same time, the unified constraint execution standard makes the operation process more standardized, reducing errors caused by inconsistent operations.

[0053] (4) Consider energy-saving goals and promote green operation modes. Specifically, the method sets an energy-saving reward term in the reward function of S14 (such as giving points for short equipment movement distances), and S13 identifies key factors related to energy saving (such as equipment movement distance), guiding the model to preferentially select energy-saving actions (such as the example where the stacker-reclaimer selects a short path). By reducing equipment idle movement and optimizing resource allocation, energy consumption and equipment wear and tear are reduced, promoting the transformation of port loading and unloading operations towards a green and low-carbon mode.

[0054] (5) Dynamically adapt to complex scenarios and have continuous optimization capabilities. Specifically, in the face of complex scenarios such as the concentration of trains arriving at the port, equipment failures, and changes in environmental temperature, the dynamic optimization mechanism of S15 can adjust the model according to the actual efficiency and the predicted difference: when the deviation is small, fine-tune the reward function weight (such as focusing more on efficiency), and when the deviation is large, modify the key factor identification logic of the large language model. This closed-loop optimization capability enables the scheme to continuously adapt to changes in scenarios, maintaining an efficient decision-making state in the long term, solving the problem of delayed reaction and poor adaptability of traditional manual scheduling in dynamic scenarios.

[0055] (6) Provide technical support for the construction of smart ports, with industry promotion value. Specifically, the scheme realizes the unmanned and fine operation of coal bulk cargo handling through the whole-link intelligentization of "data collection-intelligent analysis-automatic decision-dynamic optimization", which meets the demand for improving the intelligent level in the construction of smart ports. The technical ideas (such as constraint visualization and machine learning driven decision) can be popularized to other bulk cargo ports or industrial scenes (such as production organizations with "plan-execution" architecture), providing reusable solutions for the intelligent upgrading of the industry.

[0056] In summary, the method improves the efficiency, safety, standardization and energy saving of coal bulk cargo handling through an intelligent, data-driven and dynamic decision-making mechanism, while having strong adaptability and promotion value, providing strong support for the efficient operation and intelligent transformation of bulk cargo ports.

[0057] In some possible embodiments, S13, using a large language model to identify key factors related to coal bulk cargo handling efficiency and the quantitative correlation between the key factors and the coal bulk cargo handling efficiency, includes the following processes: Step 131, calculate the Pearson correlation coefficient of all fields in the data set and the handling efficiency index, and screen the structured fields and unstructured text fields with a Pearson correlation coefficient greater than a preset coefficient threshold.

[0058] In step 131, the handling efficiency index includes total operation time, equipment energy consumption, etc. The Pearson correlation coefficient is an index for measuring the linear correlation between two variables (such as a field and handling efficiency), and the closer the value is to 1 or -1, the stronger the correlation. Structured fields refer to quantifiable numerical data (such as device movement distance, remaining capacity of stacking position, real-time operation rate of device, switching time of conveyor belt, turnaround time stamp of yard, temperature data, etc.); unstructured text fields refer to descriptive information (such as handling operation instructions, device failure records, operation logs, scheduling delay logs, ship stowage feedback sheets, etc.); the preset coefficient threshold is a standard for judging whether the correlation is significant (such as 0.6, which is considered to be related if it exceeds this value); the handling efficiency index is a quantitative index for measuring operation efficiency (such as total operation time, unit time handling capacity).

[0059] In the embodiments of the present application, first, all fields (including structured fields such as "stacker movement distance" and "remaining capacity of stacking position", and unstructured text fields such as "number of device failures" and "operation environment temperature records") are extracted from the data set formed by S11; second, the Pearson correlation coefficient of each field and the handling efficiency index (such as total operation time) is calculated, and the calculation formula of the Pearson correlation coefficient in the present embodiment is not specifically limited; finally, the fields with a correlation coefficient greater than a preset coefficient threshold (such as 0.6) are selected as candidate factors for subsequent analysis.

[0060] Step 132, mapping the structured fields and unstructured text fields to knowledge graph nodes according to the spatiotemporal coordinates, calculating the edge weights between the nodes based on the historical operation data, and forming a knowledge graph taking the loading and unloading process chain as the framework and carrying the loading and unloading process dependency relationship and edge weights.

[0061] In step 132, the spatiotemporal coordinates are identifiers combining time and spatial position (such as "9:30, stacker reclaimer at pile position 2"); the knowledge graph is a graph representing entities and relationships in the form of nodes and edges, where the nodes are the filtered fields, and the edges represent the dependency relationship and weight between the fields. The knowledge graph nodes are the mapped structured / unstructured fields (such as "stacker reclaimer moving distance" and "pile position remaining capacity"); the edge weight is a numerical value that measures the strength of the dependency relationship between nodes (the higher the weight, the stronger the dependency relationship); and the loading and unloading process chain refers to the complete process of coal loading and unloading (such as unloading truck → stacking → taking material → loading ship). The loading and unloading process dependency relationship can include coal type matching dependency relationship, equipment conflict dependency relationship, etc.

[0062] In the embodiments of the present application, first, the structured fields and unstructured text fields filtered in step 131 are mapped according to the spatiotemporal coordinates, for example, "stacker reclaimer moving distance 50 meters" is mapped to the node "10:00, stacker reclaimer B, moving distance 50 meters", and "pile position 1 remaining capacity 17000 tons" is mapped to the node "10:00, pile position 1, remaining capacity 17000 tons"; second, the edge weights between the nodes are calculated based on the historical operation data, for example, by analyzing the co-occurrence frequency and influence degree of "stacker reclaimer moving distance" and "total operation time" in historical data, the edge weight between the two nodes is determined to be 0.8; and finally, taking the loading and unloading process chain (such as "train unloading → stacker reclaimer stacking → pile position storage") as the framework, the nodes are connected according to the process flow to form a knowledge graph carrying the process dependency relationship and edge weight.

[0063] Step 133, based on the knowledge graph, combining the domain knowledge base and the preset causal templates, performing multi-round chain reasoning oriented to the efficiency of coal bulk cargo loading and unloading, outputting a key factor list and a quantitative association rule, and the key factor list including a plurality of key factors.

[0064] In step 133, the domain knowledge base is a database for storing professional knowledge of the coal loading and unloading industry, such as "the shorter the moving distance of the stacker-reclaimer, the higher the work efficiency", and can also include equipment technical parameters, loading and unloading process rules, historical optimization cases, unloading energy saving specifications, safety operation standards, etc.; the preset causal template is a standardized sentence for reasoning the causal relationship, such as "if factor X increases, efficiency indicator Y changes", for example: [weather type] affects [equipment efficiency indicator, and further affects [total loading and unloading time]]; the multi-round chain reasoning refers to a process of gradually deducing deep correlations based on initial relationships; the gradient significance test value is a statistical quantity for measuring the significance of the influence of a factor on efficiency; and the quantitative correlation rule is a mathematical expression between the key factor and the efficiency (such as "total work time = a + b x factor 1 + c x factor 2"), wherein a, b and c are coefficients, and the numerical values are constants, which are not limited in the embodiments of the present application. The plurality of key factors in the key factor list are, for example, the stacker-reclaimer idling rate, the conveyor belt switching frequency, the difference between adjacent stockyard coal types, the equipment moving distance, the waiting time for changing the coal type, the coal type matching degree, the equipment moving distance, the stacking capacity utilization rate, the operation mode parameter, the environmental temperature, etc.

[0065] Specifically, step 133 can include the following processes: step a1, similarity matching the knowledge graph nodes with the preset causal templates to filter out a candidate node set affecting the coal bulk cargo loading and unloading efficiency; step a2, calculating the efficiency influence weight of the candidate nodes; step a3, for the candidate nodes with the efficiency influence weight greater than a preset weight threshold, based on the historical operation data, differentially calculating the corresponding gradient significance test value; step a4, filtering out the candidate nodes with the gradient significance test value greater than a preset test threshold as key factors to output a key factor list; and step a5, quantifying the correlation rule between the key factors and the coal bulk cargo loading and unloading efficiency by a preset fitting method to obtain a quantitative correlation rule.

[0066] In the embodiments of the present application, firstly, the knowledge graph nodes formed in step 132 are similarity matched with the preset causal templates (such as "if the moving distance of the stacker-reclaimer increases, the total work time is prolonged") to filter out a candidate node set that can affect the loading and unloading efficiency; secondly, the efficiency influence weight of the candidate nodes is calculated (based on the edge weight of the knowledge graph and the domain knowledge base), for example, the weight of the "stacker-reclaimer moving distance" is 0.7, and the weight of the "remaining capacity of the stacking position" is 0.6; thirdly, for the candidate nodes with the weight higher than a preset weight threshold (such as 0.5), the gradient significance test value is differentially calculated based on the historical operation data, and the nodes with the test value greater than a preset test threshold (such as 1.96) are filtered out as key factors; and finally, the correlation rule between the key factors and the loading and unloading efficiency is quantified by a preset fitting method such as linear regression to obtain an expression such as "total work time = 20 + 0.1 x stacker-reclaimer moving distance - 0.001 x remaining capacity of the stacking position", and the calculation formula of the gradient significance test value is not limited in the embodiments.

[0067] In steps 131-133, the Pearson correlation coefficients of all fields (such as device movement distance, stack position capacity, coal matching degree, etc.) in the data set and the loading and unloading efficiency index (such as total operation time) are calculated, and the fields with correlation coefficients greater than 0.6 (preset threshold) are screened out (such as device movement distance, stack position remaining capacity); secondly, these fields are mapped into knowledge graph nodes according to the space-time coordinates (such as "10:00 stacker-reclaimer B in stack position 2"), the edge weights between nodes are calculated based on historical operation data (such as the edge weight between "device movement distance" and "total operation time" is 0.8), and a knowledge graph with loading and unloading process chain as the skeleton is formed; finally, based on the knowledge graph, combined with the port operation field knowledge base (such as "stacker-reclaimer movement distance affects efficiency") and the preset causal template (such as "if X increases, Y changes"), multiple rounds of reasoning are carried out: first, match and screen out the candidate node set (such as device movement distance, stack position capacity), calculate the efficiency influence weight, and perform gradient significance test value on the nodes with high weight, and screen out the nodes with test value up to standard as key factors, and then obtain the quantitative association rules (such as "total operation time = basic time + 0.06 x device movement distance") through linear fitting and other methods.

[0068] In one specific example, the A port data set contains the fields of "stacker moving distance", "remaining capacity of pile position", "operation environment temperature", "number of equipment failures", etc. with total operation time as the efficiency indicator. The correlation coefficient of "stacker moving distance" and total operation time is 0.7, the correlation coefficient of "remaining capacity of pile position" is 0.65, the correlation coefficient of "operation environment temperature" is 0.3, and the correlation coefficient of "number of equipment failures" is 0.55. The preset coefficient threshold is 0.6, so the fields of "stacker moving distance" and "remaining capacity of pile position" are selected. The A port maps the selected "stacker moving distance" and "remaining capacity of pile position" to nodes according to the space-time coordinates, calculates the edge weight of "stacker moving distance" and "total operation time" as 0.75, the edge weight of "remaining capacity of pile position" and "total operation time" as 0.6, and the edge weight of "stacker moving distance" and "remaining capacity of pile position" as 0.2 (because both of them affect the operation time but have weak direct correlation), takes the process chain of "train unloading → stacker moving → stacking to pile position" as the skeleton, and forms a knowledge graph. The candidate node set of the A port is "stacker moving distance" and "remaining capacity of pile position". After matching the cause-effect template, it is calculated that the efficiency influence weights of the two are 0.7 and 0.6 (both > 0.5) respectively; the difference calculation gradient significance test values are 3.2 and 2.8 (both > 1.96) respectively, which are determined as key factors; the quantified association rule is obtained by linear regression fitting: total operation time (minutes) = 20 + 0.1 x stacker moving distance (meters) - 0.001 x remaining capacity of pile position (tons), that is, the operation time increases by 1 minute for every 10 meters increase in moving distance; the operation time decreases by 1 minute for every 1000 tons increase in remaining capacity.

[0069] By performing steps a1~a5, the embodiment of the present application can accurately locate the key factors that have a significant impact on the efficiency of coal bulk cargo handling from the knowledge graph, and establish the quantified association rule between the key factors and the efficiency. Specifically, the candidate factors are preliminarily locked by matching the cause-effect template, and are filtered layer by layer by combining weight calculation and gradient significance test, so as to ensure that the selected key factors have actual influence significance rather than accidental correlation; then the relationship between the key factors and the efficiency is converted into a calculable quantified rule by a preset fitting method, which not only provides a scientific basis for constructing a reward function for subsequent reinforcement learning model, but also changes the core logic affecting the efficiency from an empirical judgment to a data-driven precise rule, thereby improving the pertinence and effectiveness of the entire handling efficiency optimization scheme.

[0070] By performing steps 131-133, the embodiment of the application can preliminarily screen out the fields significantly related to the loading and unloading efficiency through quantitative analysis, exclude irrelevant or weakly related factors, reduce redundant information in subsequent analysis, and lay a foundation for accurately identifying key factors. The screened fields are converted into a structured knowledge graph, which intuitively presents the dependency between factors and efficiency, and provides clear structural support for subsequent reasoning of key factors. Through multiple rounds of reasoning and quantitative analysis, the key factors and correlation rules that have a significant impact on the loading and unloading efficiency are accurately identified, providing quantitative basis for the decision of the subsequent reinforcement learning model, replacing experience judgment, and improving the scientific nature of the decision.

[0071] In some possible embodiments, in step S14, the yard decision topology graph is input into the reinforcement learning model, and a coal bulk cargo loading and unloading scheme is determined by performing the processes of state space coding, action selection, and reward function calculation, including: Step 141: inputting the yard decision topology graph into the reinforcement learning model, performing multi-order neighborhood aggregation on the node features and edge weights of the yard decision topology graph by using a spatio-temporal graph convolution network of the reinforcement learning model, and obtaining a dynamic state vector that fuses real-time positions of equipment, coal type attributes of stacks, task queues, and equipment fault states.

[0072] In step 141, the spatio-temporal graph convolution network is used to process the spatio-temporal features of the topology graph, and the multi-order neighborhood aggregation refers to the features of the node and its surrounding multi-layer nodes. The dynamic state vector is a vector that fuses real-time positions of equipment (such as coordinates of the stacker-reclaimer), coal type attributes of stacks (such as coal type and capacity), task queues (such as information of a train to be unloaded), and equipment fault states (such as normal / fault), and is used to reflect the dynamic state of the current work scene.

[0073] In the embodiment of the application, the yard decision topology graph with constraints is input into the reinforcement learning model; secondly, by using the spatio-temporal graph convolution network in the model, the features of each node (such as the coal type of stack 1 being bituminous coal and the remaining capacity being 17000 tons) and the edge weights (such as the weight of the “coal type matching” edge being 0.9) in the topology graph are aggregated in multi-order neighborhoods, that is, not only the information of the node itself is considered, but also the features of the adjacent nodes (such as the association between the stacker-reclaimer B and the stack 1) and the adjacent nodes of the adjacent nodes (such as the indirect association between the train K and the stacker-reclaimer B) are fused; and finally, a dynamic state vector is obtained, which comprehensively reflects real-time information such as current equipment positions, stack attributes, task queues, and equipment faults.

[0074] Step 142: calculating an action value vector corresponding to the dynamic state vector based on a deep double Q network, and selecting a corresponding action according to the action value with the maximum value in the action value vector, the action including: a stack coordinate allocation sequence of the stacker-reclaimer, a coal type identification configuration scheme of the conveyor belt, and an equipment movement scheme.

[0075] In step 142, the deep double Q network is a reinforcement learning network composed of two independent Q networks (Q1, Q2) for reducing the action value estimation bias; the action value vector is a set of scores evaluating the pros and cons of each possible action (such as the stacker-reclaimer going to a certain stack position). The stack position coordinate allocation sequence of the stacker-reclaimer refers to the order of stack positions to which the stacker-reclaimer should go, meeting the requirements of the stacker-reclaimer not working at the same stack position at the same time and the anti-collision interval; the coal type identification configuration scheme of the conveyor belt refers to the coal type matching identification (such as “bituminous coal”) set for the conveyor belt; and the equipment movement scheme refers to the movement path planning of the stacker-reclaimer.

[0076] Specifically, step 142 can include the following processes: step b1, inputting the dynamic state vector into the deep double Q network to output multiple types of action values, including the action value of each stack position coordinate as the matching position of the stacker-reclaimer, the action value of each coal type identification activated and configured on the conveyor belt, and the action value of each path sequence as the equipment movement scheme; step b2, calculating the task urgency weight according to the train station arrival countdown or the ship berthing countdown in the plan data, in combination with the user level, task level and yard operation mode in the plan data; step b3, adjusting the action value corresponding to the stack position coordinate according to the task urgency weight and the priority corresponding to the yard operation mode; step b4, selecting the stack position coordinate with the highest action value as the matching position of the stacker-reclaimer, selecting the coal type identification with the highest adjusted action value as the configuration instruction of the conveyor belt, and selecting the path sequence with the highest action value as the equipment movement scheme, to obtain the corresponding action.

[0077] In the above process, during the unloading operation, the yard operation mode includes the preparation mode, the balance mode and the high yield mode. Among them, the high yield mode preferentially shortens the unloading time, and the preparation mode preferentially selects a small capacity stacking position. And the stacking position coordinates satisfy the constraints of empty stacking position disable and stacking position unavailable corresponding to faulty equipment, and satisfy the anti-collision interval requirement. The action values of multiple types include three categories: stacking position coordinate matching value (such as the value of stacking position 1 as the target of the stacker-reclaimer), coal type identification configuration value (such as the value of activating the "bituminous coal" identification of the conveyor), and path sequence value (such as the straight line path value of the stacker-reclaimer from stacking position 2 to stacking position 1). The task emergency weight is a quantitative value for measuring the degree of task urgency (range 0~1, the higher the value, the more urgent); the train arrival countdown is the difference between the expected arrival time of the train and the current time, and the proportion of the preset window; the ship berthing countdown is similar; the user level (such as VIP / ordinary), the task level (such as urgent / regular), and the yard operation mode (such as preparation / balance / high yield) need to be quantified into numerical values (such as 1~5 points). The priority of the yard operation mode is the decision tendency in different modes (such as high yield mode preferentially shortening the time, and balance mode preferentially evenly distributing); the adjusted action value is the action value after being corrected in combination with the emergency weight and the mode priority (such as the stacking position value of the urgent task is increased). The matching position is the stacking position where the stacker-reclaimer should go to work; the configuration instruction is the coal type identification that the conveyor needs to activate; and the movement scheme is the optimal movement path of the stacker-reclaimer.

[0078] In step b2, the following process can be included: step b21, calculating a first difference between the expected arrival time of the train and the current time, or a second difference between the expected arrival time of the ship and the current time; step b22, taking the ratio of the first difference to the preset time window as the train arrival countdown, or taking the ratio of the second difference to the preset time window as the ship berthing countdown; step b23, quantifying the user level, the task level and the yard operation mode respectively to obtain the user level quantification value, the task level quantification value and the yard operation mode quantification value; step b24, taking the superposition result of the train arrival countdown, the user level quantification value, the task level quantification value and the yard operation mode quantification value as the emergency weight of the unloading task, and taking the superposition result of the ship berthing countdown, the user level quantification value, the task level quantification value and the yard operation mode quantification value as the emergency weight of the loading task.

[0079] In the above process, the first difference value is the difference between the train's expected arrival time and the current time (e.g., 30 minutes); the second difference value is the difference between the ship's expected arrival time and the current time. The preset time window is a standard time range set according to the work scene (e.g., 60 minutes for train unloading, 120 minutes for ship loading); the train arrival countdown is the ratio of the first difference value to the preset window (range 0~1, the smaller the value, the more urgent); the ship arrival countdown is similar. The user level quantitative value is the numerical value converted from the user level (e.g., VIP, ordinary) (e.g., VIP=5, ordinary=3); the task level quantitative value is the numerical value converted from the task level (e.g., urgent, regular) (e.g., urgent=5, regular=2); the yard operation mode quantitative value is the numerical value converted from the operation mode (e.g., high yield, balanced) (e.g., high yield=4, balanced=3). The unloading task emergency weight is the quantitative value of the emergency degree of the train unloading task (obtained by superimposing the train arrival countdown, user level quantitative value, etc.); the loading task emergency weight is the quantitative value of the emergency degree of the ship loading task (obtained by superimposing the ship arrival countdown, etc.).

[0080] By performing steps b21~b24, the embodiment of the present application can provide basic data for subsequent measurement of task urgency by calculating time difference. The absolute time difference is converted into a relative proportion to unify the measurement scale of the urgency, which facilitates subsequent weight calculation. The non-numerical level and mode are converted into calculable numerical values to support the quantitative calculation of the emergency weight. Through multi-dimensional superposition calculation, the task urgency is comprehensively measured to ensure that high-urgency tasks are given priority.

[0081] By performing steps b1~b4, the embodiment of the present application can provide clear basis for subsequent action selection by outputting the quantitative value of each action through the deep double Q network, reduce the estimation deviation of a single network, and then adjust the priority of the action value; by adjusting the action value, the decision is more in line with the task urgency and the operation mode target, improving the scene adaptability of the scheme; by selecting the highest value action, the decision is ensured to be optimal in the current scene, taking into account efficiency, urgency and operation mode.

[0082] Step 143, after performing the action, determining observation data including actual loading and unloading amount, equipment moving distance, and violation event count.

[0083] In step 143, the observation data is the actual result data generated after the action is performed, including actual loading and unloading amount (e.g., actual unloading of 5000 tons of coal), equipment moving distance (e.g., 50 meters of stacker-reclaimer movement), and violation event count (e.g., 0 times of violation). Violation events can include high-temperature work, empty stack position coal loading, coal type mismatch, and other violation behaviors.

[0084] In the embodiment of the present application, after the action determined in step 142 (such as the stacker-reclaimer B moving to the pile position 1 and the conveyor belt being configured with the bituminous coal identifier) is performed, the actual results are collected by the Internet of Things sensor and monitoring system: the actual loading and unloading tonnage of coal is recorded, the actual distance moved by the equipment is counted, whether there is a violation (such as coal type mismatching and equipment conflict) is checked and counted, and the observation data is formed by summarizing.

[0085] In step 144, the reward value is calculated according to the observation data and the reward function, the reward function is constructed according to the key factors and the quantitative association rules, and includes the efficiency reward item, the compliance reward item, the moving cost penalty item and the energy saving reward item.

[0086] In step 144, the reward value is a quantitative score for measuring the pros and cons of the action; the efficiency reward item is a score given according to the actual loading and unloading efficiency (such as time saving), which focuses on the optimization of the total operation time; the compliance reward item is a score given for complying with the constraints (such as no violation), such as 0 violation plus 10 points; the moving cost penalty item is a score deducted for long distance moving (such as 5 points for short distance), and vice versa, with minimum stacking priority and short distance moving reward; the energy saving reward item is a score given for energy saving action (such as short path), such as 5 points for short distance. All scores can be directly summed or weighted summed to obtain the reward value.

[0087] In the embodiment of the present application, the reward value is calculated according to the observation data of step 143 and the key factors and the quantitative association rules determined in S13: the efficiency reward item is calculated according to the difference between the actual operation time and the standard time (such as 2 minutes in advance plus 15 points); the compliance reward item is calculated according to the violation count (such as 0 violation plus 10 points); the moving cost penalty item is calculated according to the moving distance (such as 5 points for 50 meters and 5 points deducted for 100 meters); the energy saving reward item is calculated in combination with the moving distance and the energy consumption association rule (such as 5 points for short distance), and the sum of the four items is the total reward value.

[0088] In step 145, the action value is updated using the time series difference error based on the reward value, and when the fluctuation amplitude value of the action value of continuous N times of decision making is less than a preset amplitude threshold, it is determined that the reinforcement learning model converges, and the coal bulk cargo handling scheme including the pile position coordinate allocation sequence of the stacker-reclaimer, the coal type identifier configuration scheme of the conveyor belt and the equipment moving scheme is output, and N is greater than or equal to 3.

[0089] In step 145, the time series difference error is an error value (reflecting the difference between the predicted value and the actual reward + the next state value) for updating the action value; the fluctuation amplitude value is the change amount of the action value in continuous decision making (such as the absolute value of the difference between the values of two times of decision making); the preset amplitude threshold is a standard for judging the convergence of the model (such as 3%); N is the number of continuous convergence (≥3).

[0090] In the embodiment of the present application, first, the action value is updated using the timing difference error formula; second, the fluctuation range of the action value of the continuous N times of decision-making (such as 80 minutes for the first time, 82 minutes for the second time, and fluctuation 2.5%) is calculated; finally, when the fluctuation range is less than a preset threshold (such as 3%) and the continuous N≥3 times, it is determined that the model converges, and the unloading scheme including the stacker reclaimer distribution, the conveyor belt configuration, and the equipment movement is output.

[0091] For example, for the unloading task of train K in port A, the yard decision topology graph contains nodes such as "train K (bituminous coal 5000 tons, 9:30 countdown 30 minutes)", "stacker B (normal, stacking position 2)", "stacking position 1 (bituminous coal, remaining 17000 tons)", etc., and edge weights such as "train K-stacking position 1 (coal type matching, 0.9)", "stacker B-stacking position 1 (movable, 0.8)". After the spatio-temporal graph convolution network aggregates this information, the dynamic state vector is output: "stacker B is located at stacking position 2 (normal), stacking position 1 stores bituminous coal with a remaining amount of 17000 tons, and train K (bituminous coal 5000 tons) is waiting to be unloaded in 30 minutes", and there is no equipment failure. After the dynamic state vector of port A is input into the deep double Q network, the output is: the matching value of stacking position 1 is 80 points, the matching value of stacking position 4 is 70 points; the configuration value of the conveyor belt "bituminous coal" label is 90 points, and the "non-bituminous coal" label is 0 points; the straight-line path value of the stacker from stacking position 2 to stacking position 1 is 85 points, and the detour path value is 60 points. Train K in port A is expected to arrive at 10:00, and the current time is 9:30, so the first difference = 10:00-9:30=30 minutes; if ship B is handled for loading, it is expected to arrive at 14:00, and the current time is 13:00, so the second difference = 60 minutes. The first difference of train K in port A is 30 minutes, the preset window is 60 minutes, the train arrival countdown = 30 / 60=0.5; the second difference of ship B is 60 minutes, the preset window is 120 minutes, and the ship arrival countdown = 60 / 120=0.5. The user of train K in port A is VIP (quantitative value 4), the task level is relatively urgent (quantitative value 3), and the current yard operation mode is high yield (quantitative value 4), so the user level quantitative value = 4, the task level quantitative value = 3, and the yard operation mode quantitative value = 4. The train arrival countdown of train K in port A = 0.5 (take 1-0.5=0.5 to participate in calculation), the user level quantitative value = 4, the task level quantitative value = 3, the mode quantitative value = 4, and the emergency weight = 0.25×0.5+0.25×4+0.25×3+0.25×4=0.125+1+0.75+1=2.875. The emergency weight of train K in port A is 2.875, the high-yield mode is preferred to shorten the time, the original value of stacking position 1 is 80 points (short distance, short time), and the adjusted value = 80×(1+2.875×0.1)=80×1.2875=103 points; the original value of stacking position 4 is 70 points (long distance), and the adjusted value = 70×(1+2.875×0.1)=89.125 points. After step b3 in port A is adjusted, the value of stacking position 1 is 103 points (the highest), so the stacker B matches stacking position 1; the value of the conveyor belt "bituminous coal" label is 90 points (the highest), and the configuration instruction is "bituminous coal"; the straight-line path value of the stacker B from stacking position 2 to stacking position 1 is 85 points (the highest), and the movement scheme is the straight-line path.After the port performs the action, 5000 tons of coal are actually unloaded (consistent with the plan), the stacker-reclaimer B moves a distance of 50 meters, no coal type mismatch, equipment conflict, or other violations occur, the violation event count is 0, and the observation data is (5000 tons, 50 meters, 0 times). The observation data of the A port is (5000 tons, 50 meters, 0 times), the actual operation time is 28 minutes (the standard is 30 minutes), the efficiency reward item is +15 points; 0 violations, the compliance reward item is +10 points; the moving distance is 50 meters (shorter), the moving cost penalty item is +5 points; short distance energy saving, the energy saving reward item is +5 points, and the total reward value is 35 points. The A port updates the action value by 35 points of reward value, after 3 iterations, the action value is from 80 points to 82 points to 83 points, and the fluctuation amplitudes are 2.5% and 1.2% (both <3%), the model converges, and the output scheme is: the stacker-reclaimer B moves from the pile position 2 to the pile position 1 along the straight line path, and the conveyor belt is configured with the bituminous coal identifier.

[0092] By performing steps 141-145, the embodiment of the application can convert complex topological graph information into a concise dynamic state vector through multi-order aggregation of the spatio-temporal graph convolution network, retain key scene information, facilitate subsequent model processing, provide a basis for accurate decision-making, form a specific scheme of stacker-reclaimer allocation, conveyor belt configuration and equipment movement, reflect the actual effect of action execution through observation data, provide real feedback for subsequent reward calculation and model optimization, quantify the action effect through the reward value, guide the model to preferentially select efficient, compliant and energy-saving actions, and improve decision-making quality, update the action value and judge convergence, continuously optimize the decision-making of the model, and finally output a stable and optimal loading and unloading scheme to ensure efficient operation.

[0093] Figure 3 A structural schematic diagram of a coal bulk cargo loading and unloading efficiency optimization system based on machine learning provided by the embodiment of the application is shown in FIG. 1. Figure 3 The system can include: The acquisition module 31 is configured to acquire plan data, equipment state position data, stockyard resource occupation data and stockyard constraint data of the coal bulk cargo loading and unloading operation, and form a data set.

[0094] The generation module 32 is configured to generate a constrained stockyard decision topological graph according to the data set.

[0095] The identification module 33 is configured to identify key factors related to the coal bulk cargo loading and unloading efficiency and the quantitative association rules between the key factors and the coal bulk cargo loading and unloading efficiency by using a large language model.

[0096] The reinforcement learning module 34 is configured to input the yard decision topology into a reinforcement learning model as a framework of a deep double Q network, to determine the coal bulk cargo loading and unloading scheme by performing a process of state space coding, action selection and reward function calculation, and to maximize a predicted value of the coal bulk cargo loading and unloading efficiency in the current period, and to construct the reward function according to key factors and quantitative association rules.

[0097] The calculation optimization module 35 is configured to collect an actual value of the coal bulk cargo loading and unloading efficiency in the current period after the coal bulk cargo loading and unloading scheme is completed, to calculate an efficiency difference value, and to optimize the coal bulk cargo loading and unloading efficiency in the next period based on the efficiency difference value.

[0098] The coal bulk cargo loading and unloading efficiency optimization system based on machine learning according to the embodiments of the present application is used to implement the coal bulk cargo loading and unloading efficiency optimization method based on machine learning as described above, and therefore the specific embodiments of the coal bulk cargo loading and unloading efficiency optimization system based on machine learning can be seen from the foregoing embodiments of the coal bulk cargo loading and unloading efficiency optimization method based on machine learning, and the specific embodiments can be described with reference to the descriptions of the respective embodiments, which will not be repeated here.

[0099] The present application also provides an electronic device, comprising a memory for storing a computer program, and a processor for executing the computer program to implement the steps of any of the coal bulk cargo loading and unloading efficiency optimization methods based on machine learning as described above.

[0100] The present application also provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the steps of any of the coal bulk cargo loading and unloading efficiency optimization methods based on machine learning as described above.

[0101] In an exemplary embodiment, the computer readable storage medium as described above can include, but is not limited to, a U disk, a read-only memory, a random access memory, a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0102] The embodiments of the present application also provide a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of any of the coal bulk cargo loading and unloading efficiency optimization methods based on machine learning as described above.

[0103] Those skilled in the art will further realize that the mere concepts, teachings, and embodiments described herein are merely meant to provide an enabling description of the claimed application. Accordingly, modifications and / or additions, other than those explicitly described herein, can be obvious to those skilled in the art in the light of this disclosure. The claimed application is intended to embrace all such modifications and / or additions.

[0104] The above provides a coal bulk cargo loading and unloading efficiency optimization method and system based on machine learning. The principles and implementation modes of the present application are described by applying specific examples in this paper. The above description of the embodiments is only used to help understand the method and its core idea of the present application. It should be pointed out that for those skilled in the art, without departing from the principles of the present application, some improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the present application.

Claims

1. A method for optimizing coal bulk loading and unloading efficiency based on machine learning, characterized in that: include: Collecting planned data, equipment status and location data, yard resource occupancy data, and yard constraint data for bulk coal loading and unloading operations to form a data set. The yard constraint data includes: coal type matching requirements, stacker-reclaimer operation conflict constraints, stack capacity limits, empty stack usage restrictions, equipment failure-related constraints, and high-temperature operation restrictions; generating a storage yard decision topology diagram with constraints based on the data set; Using large language models to identify key factors related to bulk coal loading and unloading efficiency, and to quantify association rules between key factors and bulk coal loading and unloading efficiency; A deep double-Q network is used as the framework of a reinforcement learning model. The yard decision topology map is input into the reinforcement learning model. The bulk coal loading and unloading plan is determined by executing a process of state space encoding, action selection, and reward function calculation. The reward function is constructed based on the key factors and the quantitative association rules, with the goal of maximizing the predicted value of the bulk coal loading and unloading efficiency in the current cycle. After the bulk coal loading and unloading plan is completed, the actual value of the bulk coal loading and unloading efficiency in the current cycle is collected, the efficiency difference is calculated, and based on the efficiency difference, the bulk coal loading and unloading efficiency in the next cycle is optimized.

2. The method according to claim 1, characterized in that The method includes inputting the yard decision topology map into the reinforcement learning model, and determining the coal bulk loading and unloading plan by executing the process of state space encoding, action selection, and reward function calculation, including: The stockpile decision topology graph is input into the reinforcement learning model, and the node features and edge weights of the stockpile decision topology graph are subjected to multi-order neighborhood aggregation by the spatiotemporal graph convolutional network of the reinforcement learning model to obtain a dynamic state vector that integrates the real-time location of the equipment, the coal type attributes of the stack, the task queue, and the equipment fault status; Based on the deep double-Q network, the action value vector corresponding to the dynamic state vector is calculated. According to the action value with the largest value in the action value vector, the corresponding action is selected. The actions include: the stacking coordinate allocation sequence of the stacker-reclaimer, the coal type identification configuration scheme of the conveyor belt, and the equipment movement scheme; After performing the actions, determining observation data including actual loading and unloading volume, equipment movement distance, and violation event counts; Calculating a reward value based on the observation data and a reward function, wherein the reward function is constructed based on the key factors and the quantitative association rules and includes an efficiency reward item, a compliance reward item, a mobility cost penalty item, and an energy-saving reward item; Based on the reward value, the action value is updated using the temporal difference error. When the fluctuation amplitude of the action value of N consecutive decisions is less than a preset amplitude threshold, the reinforcement learning model is determined to have converged, and a coal bulk loading and unloading plan including a stacker coordinate allocation sequence for the stacker and reclaimer, a coal type identification configuration plan for the conveyor belt, and an equipment movement plan is output, where N is greater than or equal to 3.

3. The method according to claim 2, characterized in that The deep double-Q network is based on calculating the action value vector corresponding to the dynamic state vector, and selecting the corresponding action according to the action value with the largest value in the action value vector. The actions include: the stacking coordinate allocation sequence of the stacker and reclaimer, the coal type identification configuration scheme of the conveyor belt, and the equipment movement scheme, including: The dynamic state vector is input into a deep double-Q network to output multiple types of action values, including the action value of each stack coordinate as the matching position of the stacker-reclaimer, the action value of each coal type identifier activated on the conveyor belt, and the action value of each path sequence as the equipment movement plan; Calculate the task urgency weight based on the train arrival countdown or ship berthing countdown in the plan data, combined with the user level, task level and yard operation mode in the plan data; Adjust the action value corresponding to the stack coordinate according to the task urgency weight and the priority corresponding to the yard operation mode; The stack coordinates with the highest action value are selected as the matching position of the stacker-reclaimer, the coal type identifier with the highest adjusted action value is selected as the configuration instruction of the conveyor belt, and the path sequence with the highest action value is selected as the equipment movement plan to obtain the corresponding action.

4. The method according to claim 3, characterized in that The calculation of the task urgency weight according to the train arrival countdown or ship berthing countdown in the plan data, combined with the user level, task level and yard operation mode in the plan data, includes: Calculating a first difference between the estimated time of arrival of a train and the current time or a second difference between the estimated time of arrival of a ship and the current time; Using the ratio of the first difference to the preset time window as the train arrival countdown, or using the ratio of the second difference to the preset time window as the ship berthing countdown; quantifying the user level, the task level, and the yard operation mode respectively to obtain a user level quantization value, a task level quantization value, and a yard operation mode quantization value; The superposition result of the train docking countdown, user level quantization value, task level quantization value and yard operation mode quantization value is used as the emergency weight of the unloading task, and the superposition result of the ship berthing countdown, user level quantization value, task level quantization value and yard operation mode quantization value is used as the emergency weight of the loading task.

5. The method according to claim 1, wherein The large language model is used to identify key factors related to bulk coal loading and unloading efficiency, as well as quantitative association rules between key factors and bulk coal loading and unloading efficiency, including: Calculating the Pearson correlation coefficients between all fields in the data set and the loading and unloading efficiency index, and screening structured fields and unstructured text fields whose Pearson correlation coefficients are greater than a preset coefficient threshold; Mapping the structured fields and the unstructured text fields into knowledge graph nodes according to spatiotemporal coordinates, and calculating edge weights between the nodes based on historical operation data to form a knowledge graph with a loading and unloading process chain as the skeleton and carrying loading and unloading process dependencies and edge weights; Based on the knowledge graph, combined with the domain knowledge base and the preset causal template, multi-round chain reasoning oriented towards the efficiency of bulk coal loading and unloading is performed to output a list of key factors and quantitative association rules, wherein the list of key factors includes multiple key factors.

6. The method according to claim 5, characterized in that Based on the knowledge graph, combined with the domain knowledge base and the preset causal template, multi-round chain reasoning oriented towards the efficiency of bulk coal loading and unloading is performed to output a list of key factors and quantitative association rules, including: Performing similarity matching between the knowledge graph nodes and the preset causal template to screen out a set of candidate nodes that affect the efficiency of bulk coal loading and unloading; Calculate the efficiency impact weight of candidate nodes; For candidate nodes whose efficiency impact weight is greater than the preset weight threshold, the corresponding gradient significance test value is differentially calculated based on historical operation data; Filter candidate nodes whose gradient significance test values ​​are greater than the preset test threshold as key factors to output a list of key factors; The association rules between key factors and coal bulk loading and unloading efficiency were quantified by a preset fitting method, and quantitative association rules were obtained.

7. The method according to claim 1, characterized in that Optimizing the bulk coal loading and unloading efficiency in the next cycle based on the efficiency difference includes: When the efficiency difference is less than or equal to the preset difference, adjusting the weights corresponding to the efficiency reward item, the compliance reward item, the movement cost penalty item, and the energy-saving reward item in the reward function according to the offset direction and magnitude of the efficiency difference, so as to optimize the coal bulk loading and unloading efficiency in the next cycle according to the adjusted reward function; Alternatively, when the efficiency difference is greater than a preset difference, the large language model is adjusted according to the prediction deviation sample set to optimize the coal bulk loading and unloading efficiency in the next cycle according to the adjusted large language model.

8. A coal bulk loading and unloading efficiency optimization system based on machine learning, characterized in that: include: The collection module is used to collect the planning data of bulk coal loading and unloading operations, equipment status and location data, yard resource occupancy data, and yard constraint data to form a data set; A generating module, configured to generate a storage yard decision topology diagram with constraints based on the data set; An identification module for using a large language model to identify key factors related to bulk coal loading and unloading efficiency, and quantitative association rules between key factors and bulk coal loading and unloading efficiency; a reinforcement learning module, configured to use a deep double-Q network as a framework for a reinforcement learning model, input the yard decision topology map into the reinforcement learning model, and determine a bulk coal loading and unloading plan by executing a process of state space encoding, action selection, and reward function calculation, wherein the reward function is constructed based on the key factors and the quantitative association rules, with the goal of maximizing the predicted value of bulk coal loading and unloading efficiency in the current cycle; The calculation optimization module is used to collect the actual value of the bulk coal loading and unloading efficiency in the current cycle after the bulk coal loading and unloading plan is completed, calculate the efficiency difference, and optimize the bulk coal loading and unloading efficiency in the next cycle based on the efficiency difference.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of a method for optimizing bulk coal loading and unloading efficiency based on machine learning as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, can implement a method for optimizing coal bulk loading and unloading efficiency based on machine learning according to any one of claims 1 to 7.

Images