An arch dam cable machine group fine scheduling method for all task types

Abstract

The application provides an arch dam cable crane group fine scheduling method for all task types, relates to the technical field of water conservancy and hydropower construction and organization and management technology, and comprises the following steps: decomposing concrete pouring and auxiliary hoisting tasks into task microelements with single hoisting cycle as a basic unit, and constructing a whole task microelement set; generating all feasible initial allocation schemes that meet a safety distance by adopting a constraint baffle method after the microelements are sorted; modeling the scheduling problem as a Markov decision process, generating parallel search sub-trees with each scheme as a root node, and iteratively solving an optimal execution sequence by selection, expansion, simulation and backtracking; and selecting and outputting the shortest scheme according to the completion time of each scheme. The application can realize the collaborative optimization of task allocation and execution sequence, so as to improve the on-site applicability of the cable crane group scheduling scheme and the fine level of construction organization.

Description

Technical Field

[0001] This invention relates to the field of water conservancy and hydropower construction technology and organization and management technology, and in particular to a refined scheduling method for arch dam cable machine groups for all types of tasks. Background Technology

[0002] Cable cranes (referred to as "cable cranes") have become key equipment for hoisting building materials (especially concrete) and machinery into concrete arch dams due to their outstanding advantages such as large lifting capacity, high mobility, and wide span. In the complex terrain of high mountain and canyon areas, the operating efficiency of the cable crane group directly determines the concrete pouring strength and the dam's ascent speed. Therefore, developing a scientific and reasonable cable crane group scheduling plan is not only a core prerequisite for ensuring the safety of multiple cable cranes operating in parallel and improving equipment utilization efficiency, but also an important means to ensure that the arch dam pouring progress meets the construction schedule requirements and guarantees the quality of concrete pouring. However, existing research and engineering practices on cable crane group scheduling still have significant limitations in addressing the complexity of construction and the demands of all tasks. On the one hand, in terms of task type coverage, existing methods generally only consider concrete placement and hoisting tasks, neglecting a large number of auxiliary hoisting tasks such as mechanical equipment hoisting, formwork transfer, and surface cleaning. According to engineering measurements, auxiliary hoisting tasks account for approximately 25% of the total cable crane hoisting tasks. These tasks, due to their urgency and small workload per operation, are frequently inserted manually, which can disrupt the original pouring plan and hinder the pouring progress, or even cause serious consequences. This leads to excessive concrete layer coverage time, inducing serious quality risks. On the other hand, the scheduling granularity is relatively coarse. Existing models usually use 5-10m wide concrete strips as the basic scheduling unit for pouring tasks, and implicitly assume that the leveling operation is completed in a very short time after the cable crane completes unloading at the center point of the strip, ignoring the actual time spent by the leveling machine in laying the material. This results in large prediction errors when the strip is long or there are areas with dense reinforcement. Moreover, the strip-level scheduling results still rely on manual decision-making of specific unloading target points round by round when executed on site, which restricts the efficiency of cooperation between multiple cable cranes.

[0003] Chinese patent application CN120724609A discloses an intelligent scheduling method for cable crane groups that considers both quality and safety pre-control standards. This patent uses concrete pouring strips as the basic scheduling unit, dividing the pouring bay into continuous 5m wide strips along the dam axis. It constructs a multi-objective optimization model with the objectives of minimizing the maximum pouring time of the cable crane group and minimizing the root mean square error of the pouring time of each cable crane. A spatial squeeze method based on safety distance is used to determine the initial operating boundary of each cable crane. Then, a greedy-local search algorithm is used to solve for the optimal matching scheme between the cable crane and the strip. The final scheme is verified by dual constraints of the concrete layer coverage time and the safety distance between adjacent cable cranes, thus solving the coordination problem of task allocation when multiple cable cranes are operating in parallel. However, this patent still has the following limitations: First, in terms of task type coverage, the patent only considers the allocation and scheduling of concrete pouring strips, completely omitting auxiliary hoisting tasks such as mechanical equipment hoisting, formwork transfer, and surface cleaning. According to engineering measurements, auxiliary hoisting tasks are frequently inserted manually due to their urgency and small workload per operation. This can disrupt the original pouring plan or even cause the concrete layer coverage time to exceed the standard, leading to serious quality risks. Second, in terms of scheduling granularity, the patent uses the strip as the smallest scheduling unit, and the output scheduling scheme only provides the correspondence between the cable crane and the strip. On-site execution still relies on manual decision-making for the specific unloading target point for each hoisting round, which restricts the efficiency of collaboration between multiple cable cranes. Third, in terms of the completeness of scheduling decisions, the patent focuses on task allocation and does not provide a method for deciding the order of multiple tasks performed by each cable crane. The two-stage independent decision-making of task allocation and execution order may lead to local unreasonableness in the allocation scheme, thus restricting the optimization space of the execution order. Summary of the Invention

[0004] In view of this, the present invention provides a refined scheduling method for arch dam cable crane groups for all task types, which aims to integrate concrete pouring tasks and auxiliary hoisting tasks into the scheduling optimization framework, and use a single hoisting cycle as the basic decision unit to achieve coordinated optimization of task allocation and execution order, thereby improving the on-site applicability of the cable crane group scheduling scheme and the level of refinement of construction organization.

[0005] This invention provides a refined scheduling method for arch dam cable crane groups oriented towards all mission types, including: S1. Decompose the concrete pouring task and the auxiliary hoisting task into task micro-elements with a single hoisting cycle as the basic unit, and construct a set of task micro-elements. S2. Sort the task micro-element set according to the spatial coordinates of each task micro-element, generate several continuous segment division schemes using the constrained partition method and allocate them to each cable machine. After verifying the safe distance between adjacent cable machines, retain all feasible task allocation schemes. S3. Model the cable crane group scheduling problem as a Markov decision process. Based on each feasible task allocation scheme, construct the initial joint action, generate multiple parallel independent search tree root nodes, and use the Monte Carlo tree search algorithm to select, expand, simulate, and backtrack iteratively solve the optimal task execution order that satisfies the construction constraints, and obtain the completion time corresponding to each scheme. S4. Compare all feasible task allocation schemes with the completion time as the objective, and output the scheme with the shortest completion time as the final refined scheduling scheme for the cable crane group.

[0006] Preferably, step S1 includes: The concrete pouring task is divided into several unloading points according to the pouring chamber. Each unloading point corresponds to a single hoisting cycle, which constitutes a pouring task micro-element. All pouring task micro-elements form the pouring micro-element set. ; The auxiliary lifting task is decomposed into a single lifting cycle according to the implementation requirements. Each lifting operation constitutes an auxiliary task micro-element, and all auxiliary task micro-elements form an auxiliary micro-element set. ; Pick As a set of task micro-elements.

[0007] Preferably, the casting task micro-element is in triplets. Characterization, in which Number the pouring chamber. This is the serial number of the unloading point within the warehouse. Here are the spatial coordinates of the unloading point. To complete the required number of hoisting cycles at this unloading point, This refers to the duration of a single hoisting operation.

[0008] Preferably, auxiliary hoisting tasks are divided into three categories according to their implementation characteristics: The first category is tasks that can be intermittently performed within a specified time window. Internal interruption is allowed, superscript The start time of the time window, indicated by the superscript. This is the end time of the time window. Number the auxiliary tasks; The second category is interval-continuous tasks, within a time window. The process must be completed continuously to ensure continuous execution of the identifier. mark; The third type is intermittent single tasks, which have no time window constraints and can be executed at any time within the scheduling cycle; Auxiliary task micro-element in quintuple Characterization, in which The hoisting sequence for this task. The coordinates of the work location, This refers to the duration of a single hoisting operation. To assist in the execution window of hoisting tasks, This is a continuous execution flag.

[0009] Preferably, step S2 includes: According to the coordinates of the unloading point of each task micro-element Ascending order, same axis Press down on axis Axis ascending pair task infinitesimal set Sort to obtain an ordered task sequence ; Using the constrained diaphragm method to Divided into Each of the consecutive sub-segments is assigned to one of them. Taiwan cable machine, enumerate all partitioning schemes, among which This refers to the number of cable machines; For each partitioning scheme, the safety distance of the first task of adjacent cable cranes is checked, and schemes that do not meet the safety distance constraints are eliminated, while all feasible task allocation schemes are retained.

[0010] Preferably, adjacent cable cars and The method for verifying the safe distance between them is as follows: Take the tasks assigned to each cable machine The task element with the smallest coordinates is taken as the first task of the cable machine; if a cable machine is assigned to an empty task group, its safety distance check is skipped. Record the two adjacent cable cranes and The first task The coordinates are respectively and ,like If the partitioning scheme satisfies the safety distance constraint, it is discarded; otherwise, it is discarded. This is the minimum safe distance required for operation between adjacent cable machines.

[0011] Preferably, step S3 includes: The cable crane group scheduling problem is modeled as a Markov decision process, with the task completion status, the current operation status of each cable crane and the current system time constituting the system state, and the task micro-element combination of each cable crane to be executed next constituting the joint action. The initial state is defined as the condition where all task elements are incomplete and all idle cable cranes are awaiting task assignment. Based on each feasible allocation scheme, the first task of each cable machine is combined into an initial joint action. After executing this action, the system... Transition to the corresponding new state To generate multiple parallel, independent search tree root nodes; For each independent root node of the search tree, the Monte Carlo tree search algorithm is used to perform four stages of iteration: selection, expansion, simulation, and backtracking, until the preset number of iterations is reached and then terminated. The task execution order corresponding to the leaf node path with the highest cumulative reward during the iteration process is taken as the optimal execution order of the scheme, and the corresponding completion time is recorded.

[0012] Preferably, the selection phase employs a confidence upper bound tree algorithm to select child nodes layer by layer, starting from the root node. Select child nodes At that time, calculate the values ​​of each candidate child node. value: ; in child node Cumulative rewards child node Number of visits, For its parent node Number of visits, To explore coefficients; Each time you choose The child node with the largest value expands downwards until it reaches a node that has not been fully expanded.

[0013] Preferably, in the expansion stage, at the current node, based on the continuity constraints of the unloading point, the time constraints of the billet layer coverage, the safety distance constraints of adjacent cable cranes, and the constraints of the continuous execution and window period of auxiliary tasks, the set of legal task micro-elements of each cable crane in the current state is screened, the Cartesian product of the legal actions of each cable crane is taken to construct the joint action space, and a child node corresponding to a joint action is randomly selected from it and added to the search tree. The simulation phase starts from the newly expanded node and rapidly deduces the final state according to the predetermined order of the ordered task sequence I, for continuous execution markers. The auxiliary task micro-element is forced to execute continuously. For time intervals caused by conflicts in the safety distance between adjacent cable machines, waiting actions are inserted. The simulated completion time is obtained after the simulation is completed. And use a nonlinear piecewise mapping function to Converted into reward value: ; in, As the base time, This is the operating condition factor, with a value ranging from 1.2 to 1.3.

[0014] Preferably, the backtracking phase will simulate the obtained reward value. Propagate along the simulation path from leaf node to root node, node by node, for any node passed along the path. Update its access count and cumulative rewards After completing the update, return to the selection phase to begin the next iteration; After reaching the preset number of iterations, proceed along each layer. The largest child node greedily backtracks to the terminal node, and the sequence of joint actions along the path is the optimal task execution order for this feasible task allocation scheme.

[0015] The present invention has the following advantages over the prior art: (1) This invention integrates auxiliary hoisting tasks such as mechanical equipment hoisting, template transfer, and slag removal from the warehouse with concrete pouring tasks into a unified scheduling optimization framework. It further classifies and models these auxiliary hoisting tasks based on their implementation characteristics, thereby fundamentally eliminating the problem of frequent plan interruptions caused by traditional methods that only schedule pouring tasks and rely on manual "squeezing in" for auxiliary tasks. This helps ensure that the concrete billet covering time is within the specified process range and reduces the quality risk caused by temporary interference with the scheduling plan.

[0016] (2) This invention takes a single "loading-entry-unloading-return" hoisting cycle as the basic unit of scheduling, models the operation process corresponding to each unloading point independently, and considers the actual collaborative timing of the cable crane entry and leveling operations in the constraint design, as well as the complex execution characteristics such as continuous execution of auxiliary tasks and window periods. Compared with the traditional strip-level scheduling model, it is more accurate in the task timing description, effectively reducing the deviation between the scheduling simulation results and the actual engineering operation process. The output scheduling scheme can be accurate to the target position and execution time of each cable crane in each hoisting cycle, reducing the need for secondary decision-making by on-site personnel.

[0017] (3) This invention uses a constrained partition method to systematically enumerate all feasible task allocation schemes that satisfy the safety distance constraint in the initial stage, and constructs Monte Carlo tree search in parallel with each scheme as an independent root node. Under the premise of considering the continuity constraint of the unloading point, the billet layer coverage time constraint, the safety distance constraint, and the auxiliary task window period constraint, the task execution order under each allocation scheme is optimized, and finally the completion time is used as the indicator to select the best output. This mechanism transforms the two-stage independent decision of "allocation first and sorting later" in traditional scheduling into a collaborative optimization process, which helps to avoid the problem of limited space for optimization of execution order due to local unreasonableness of allocation scheme.

[0018] (4) The final output scheduling scheme of this invention includes the allocation of pouring and unloading points and auxiliary hoisting micro-elements for each cable crane, as well as the specific task micro-elements, hoisting cycle sequence and waiting period executed by each cable crane at each moment. The scheduling decision accuracy reaches the hoisting cycle level and can be directly used as on-site operation instructions. Compared with the traditional strip-level scheduling scheme, it has improved in terms of executability and can provide clearer operation basis for construction management personnel. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the micro-elemental decomposition of the concrete pouring and auxiliary hoisting tasks in this invention; Figure 3 This is a schematic diagram of the task micro-element sorting method in this invention; Figure 4 This is a schematic diagram of the grouping process generated by the constrained partition method in this invention; Figure 5 This is a schematic diagram illustrating the parallel execution of the Monte Carlo tree search process on each independent subtree in this invention; Figure 6 This is a schematic diagram illustrating the three types of motion for each cable machine as defined in this invention; Figure 7 This is a schematic diagram of the fast simulation stage in the Monte Carlo tree search of the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, this invention provides a refined scheduling method for arch dam cable crane groups oriented towards all mission types, including: S1. Decompose the concrete pouring task and the auxiliary hoisting task into task micro-elements with a single hoisting cycle as the basic unit, and construct a set of task micro-elements. S2. Sort the task micro-element set according to the spatial coordinates of each task micro-element, generate several continuous segment division schemes using the constrained partition method and allocate them to each cable machine. After verifying the safe distance between adjacent cable machines, retain all feasible task allocation schemes. S3. Model the cable crane group scheduling problem as a Markov decision process. Based on each feasible task allocation scheme, construct the initial joint action, generate multiple parallel independent search tree root nodes, and use the Monte Carlo tree search algorithm to select, expand, simulate, and backtrack iteratively solve the optimal task execution order that satisfies the construction constraints, and obtain the completion time corresponding to each scheme. S4. Compare all feasible task allocation schemes with the completion time as the objective, and output the scheme with the shortest completion time as the final refined scheduling scheme for the cable crane group.

[0023] In one embodiment of the present invention, step S1 includes: First, obtain the construction status information at the current scheduling moment, including all slabs to be poured, auxiliary hoisting tasks to be executed, and currently available cable crane resources.

[0024] Let the current set of slabs to be poured be . ,in The total number of slabs to be poured; the set of auxiliary hoisting tasks to be executed is... ,in To assist in the total number of hoisting tasks; the current set of available idle cable cranes is: ,in This represents the total number of available cable cars.

[0025] Using a single "loading-warehouse-unloading-return" hoisting cycle as the basic scheduling unit, the two types of tasks are decomposed into micro-elemental units.

[0026] like Figure 2 As shown, for a concrete pouring task, the concrete pouring task is divided into several unloading points according to the pouring silo. Each unloading point corresponds to a single hoisting cycle, which constitutes a pouring task micro-element. All pouring task micro-elements form a pouring micro-element set. .

[0027] Specifically, for any sump to be poured The process involves stacking several 0.5m thick layers of billet using a layered casting technique. Each layer has the same planar projection area. The projection plane of the current billet layer to be cast is divided into several unloading points based on the average coverage area of ​​a single unloading operation by the cable crane's hoisting tank. (Based on 9...) Taking the hanging tank as an example, its unloading coverage area is approximately a square area with sides of 4.2m, meaning the planar control area of ​​a single unloading point is approximately... .

[0028] Record No. The first warehouse The micro-element of the pouring task corresponding to each unloading point is: With triples The representation, and the definitions of each attribute are as follows: Let be the spatial coordinates of the unloading point, where The axis is in the direction of the main rail of the cable crane (upstream and downstream). The axis is the direction of the main cable of the cable machine (left and right bank direction). The axis represents the elevation direction, and the coordinate values ​​are provided by the construction survey control network.

[0029] The number of hoisting cycles required to complete the unloading at this point is calculated by dividing the volume of concrete at the unloading point by the rated capacity of the hoisting tank and rounding up. The formula is as follows: ; in, For the first Cangdi The volume of concrete corresponding to the unloading point ( ), Rated capacity of the hanging tank ( ), This is the floor operator.

[0030] The estimated execution time for a single hoisting cycle consists of the sum of the times for loading, transporting, unloading, and returning. ; in, The loading operation time includes the stages of hopper positioning, loading, and hook lifting, and is taken as the average of on-site statistics; The unloading operation time includes the stages of placement in the warehouse, unloading, and waiting for the warehouse to be leveled, and the average value is taken from the on-site statistics. These are the coordinates of the cable car's current location; For the cable car from the current position Run to unloading point Walking time, For the cable crane from the unloading point Return to starting point The travel time is estimated by taking the larger value between the displacement of the trolley in the main rail direction (calculated as the trolley's travel speed) and the displacement of the trolley in the main cable direction (calculated as the trolley's travel speed), and then adding the hook lifting time. ; in, For the speed of the large vehicle, The speed of the car For takeoff and landing elevation difference, For lifting speed, The coordinates of the starting point along the main cable direction are: Here are the coordinates of the starting point along the main track. The coordinates of the target point along the main cable direction. Let the coordinates of the target point be along the main track direction. This indicates the displacement of the trolley along the main rail. This indicates the displacement of the trolley in the direction of the main cable. All parameters mentioned above were obtained from the cable crane equipment nameplate parameters and on-site measured data. Specifically, in the calculation... At that time, the starting point was The target point is In calculation At the same time, it is also adopted The formula, where the starting point is The target point is .

[0031] The auxiliary lifting task is decomposed into a single lifting cycle according to the implementation requirements. Each lifting operation constitutes an auxiliary task micro-element, and all auxiliary task micro-elements form an auxiliary micro-element set. .

[0032] Based on long-term statistical analysis of on-site cable crane lifting data, auxiliary lifting tasks are categorized into three types according to their implementation characteristics: The first category is intermittent tasks within a specified time window, such as the hoisting of materials like formwork and reinforcing bars. Internal interruption is allowed, superscript The start time of the time window, indicated by the superscript. This is the end time of the time window. Number the auxiliary tasks; The second category is tasks that occur continuously within a given time window, such as urgent tasks like cleaning slag from storage areas. The process must be completed continuously without interruption, ensuring continuous execution of the markings. mark; The third category is intermittent single tasks, such as hoisting equipment like leveling vibratory machines and electrical workshops. These tasks are small and can be completed in a single hoisting operation. There are no time window constraints, and they can be executed at any time within the scheduling cycle.

[0033] For any auxiliary hoisting task Based on its total number of hoisting cycles Decompose it into Each auxiliary task micro-element, recorded as the first The first auxiliary task The micro-element corresponding to the next hoisting is , with quintuple Characterization, in which The hoisting sequence for this task; These are the coordinates of the work location, such as the coordinates of the template installation point, the coordinates of the slag hopper placement point, etc. The estimated execution time for a single hoisting cycle can be obtained by sampling the statistical distribution of historical operation data for this type of task, and is considered a known parameter in the model; To assist in the execution window specified for hoisting tasks, the superscript... The start time of the time window, indicated by the superscript. This is the end time of the time window. Number the auxiliary tasks; To ensure continuous execution of the flag, This indicates that the infinitesimal element belongs to an interval-continuous task. This indicates that the micro-element belongs to an interval-intermittent task. For intermittent single tasks, the micro-element has no time window attribute, and the scheduling model does not impose a window constraint on its execution time.

[0034] After completing the above decomposition, take As a unified set of all-task micro-elements Total number of micro-element tasks This set serves as the fundamental decision-making object for subsequent scheduling, defining the boundary conditions for solving the model.

[0035] In one embodiment of the present invention, step S2 includes: like Figure 3 As shown, firstly, the task's infinitesimal set... Sort them. Figure 3 Green represents the micro-element for pouring tasks, and yellow represents the micro-element for auxiliary hoisting. The coordinates of the unloading points for each micro-element are used for... Ascending order, same axis Press down on axis The principle of ascending order of axes, for All Sort the individual task elements to obtain an ordered task sequence. .

[0036] Then, as Figure 4 As shown, a task allocation scheme is generated using a constrained partition method that allows empty groups. An ordered task micro-element is assigned to The problem of idle cable cars is equivalent to the problem of the sequence of Insert into each gap (including the beginning and end). Block partitions divide the sequence into Each of the consecutive sub-segments is assigned to one of them. Taiwan cable machine, enumerate all partitioning schemes, among which This represents the number of cable cranes. Two partitions are allowed to be in the same gap; in this case, the corresponding cable crane is assigned to an empty task group, indicating that the cable crane will not participate in operations during this scheduling. The total number of partitioning schemes is [number missing]. kind.

[0037] For each allocation scheme, the safe distance between adjacent cable cranes is checked. The assigned task for each cable crane is then taken. The task element with the smallest coordinates is designated as the first task of that cableway; if a cableway is assigned to an empty task group, its safety distance check is skipped. Record the values ​​of two adjacent cableways. and The first task The coordinates are respectively and If the following conditions are met: ; If the partitioning scheme satisfies the safety distance constraint, it is discarded; where... This is the minimum safe distance required for operation between adjacent cable machines. The structural parameters of the cable crane equipment and the engineering safety regulations are jointly determined: in the direction of the main track, The spacing must be no less than the minimum allowable distance between adjacent cable crane towers. Its calculation requires comprehensive consideration of the lateral spacing of the cable crane main cables, the lateral sway limit displacement of the trolley, and safety redundancy. The typical value ranges from 20m to 30m, with specific values ​​specified in the engineering design documents. Iterate through all possible diaphragm insertion combinations, retaining those that pass the above verification. There are 1 feasible allocation schemes, denoted as _____. .

[0038] In one embodiment of the present invention, step S3 includes: The cable crane group scheduling problem is modeled as a Markov decision process. The system state is composed of the task completion state, the current operation state of each cable crane, and the current system time. The joint action is composed of the combination of the task elements to be executed by each cable crane next.

[0039] Specifically, the system state is defined as the completion state of all current task micro-elements. Operating status of each cable machine and system time combination ,in This includes the set of completed infinitesimal elements, the infinitesimal elements currently being executed and their remaining loop counts, and the set of infinitesimal elements to be executed. This includes the working or idle status of each cable crane, its current location coordinates, and the remaining time for the current task.

[0040] Each cable machine selects its next action based on its current status. The combined action of the cable machine group The cable crane group performed coordinated operations. Then, the system time advances to the moment when the earliest cable machine completes its current action, and the state of each micro-element and each cable machine is updated synchronously to complete the system state transition. This continues until all task micro-elements have completed their execution, reaching the system final state.

[0041] The initial state is defined as the condition where all task elements are incomplete and all idle cable cranes are awaiting task assignment. Each feasible allocation scheme generated by S2 Based on this, the initial tasks of each cable machine are combined to form the initial joint action. After executing this action, the system will... Transition to the corresponding new state Since each allocation scheme corresponds to a different new state, multiple parallel child nodes can be generated in the initial state, and each child node subsequently serves as the root node of an independent search tree. For example... Figure 5 As shown, this step ultimately generates a number of feasible allocation schemes. Each subtree has a parallel independent subtree, and subsequent Monte Carlo tree searches are performed independently of each other.

[0042] For each independent root node of the search tree, the Monte Carlo tree search algorithm is used to perform four stages of iteration: selection, expansion, simulation, and backtracking, until the preset number of iterations is reached and then terminated. The task execution order corresponding to the leaf node path with the highest cumulative reward during the iteration process is taken as the optimal execution order of the scheme, and the corresponding completion time is recorded.

[0043] (1) Selection stage Starting from the root node of the current search tree, recursively apply the confidence upper bound tree strategy to select the most promising child node and traverse downwards.

[0044] For any node in the search tree Record the number of times it was accessed. The cumulative reward value is From node Select child nodes At that time, calculate the values ​​of each candidate child node. value: ; in, child node Cumulative rewards child node Number of visits, For its parent node Number of visits, This is the exploration coefficient, used to balance the utilization and exploration. The first term in the formula is the utilization term, reflecting the child nodes. The average reward level from historical simulations guides the search towards known, high-performing action paths; the second term is the exploration term, providing opportunities to select child nodes with fewer visits, thus avoiding getting trapped in local optima. Each selection... The child node with the largest value expands downwards until it reaches a node that has not been fully expanded.

[0045] (2) Expansion phase like Figure 6 As shown, in the expansion phase, at the current node, dynamic actions are generated for each cable machine based on the current system state.

[0046] Figure 6 This illustrates the action allocation of three cable cranes at a certain scheduling moment: For the cable cranes If a cable crane is currently executing a task element that requires multiple loops and still has remaining hoisting attempts, its only action is to "continue the original task." The cable crane does not need to move and continues to complete the subsequent hoisting loops of the current element. At this time, the real-time distance between this cable crane and adjacent cable cranes must be maintained. ; If the cable car If it is in an idle state, first determine whether it needs to be moved to the target task by a large vehicle. The coordinate range is then used to filter the set of legal task micro-elements in the current state based on constraints such as the continuity of the unloading point, the time constraint of the billet layer coverage, the safety distance constraint between adjacent cable cranes, and the constraints of continuous execution and window period for auxiliary tasks. The continuity constraint of the unloading point requires that tasks be assigned to the cable cranes. The casting micro-elements must remain spatially continuous; the time constraint for covering the lower layer of the concrete must not exceed the initial setting time of the concrete. The safety distance constraint between adjacent cable cranes requires that the real-time distance between the execution position of the candidate micro-element and the current position of the adjacent cable crane be no less than [amount missing]. ; Consecutive execution constraints of auxiliary tasks The continuous auxiliary task micro-element must be executed continuously by the same cable machine; The auxiliary task window constraint requires that the infinitesimal element must be within the time window of its assigned task. Completed within the specified time.

[0047] Each candidate micro-element that passes the screening corresponds to a "execute new task" action, with the target position being the coordinates of that micro-element; if no candidate micro-element passes the screening, the only action the cableway can take is "wait in place," which is usually due to the real-time distance between the cableway and adjacent cableways. The next task can only be executed after the safety distance is met after the adjacent cable cranes have been moved.

[0048] After completing the above determination, construct the joint motion space by taking the Cartesian product of the legal actions of each cable machine: ; in For cableway In the current state The set of legal individual actions.

[0049] from Randomly select a joint action The new state is obtained according to the state transition rules. : Each cable machine operates according to its mechanism. Update its own status, including location, remaining execution time, and work status flags; for cable cars assigned new micro-elements, the ownership and status of their corresponding micro-elements are updated; the system time advances to the earliest moment when a cable car completed its current action. This creates child nodes. storage Then add it to the search tree as a child node of the current node to complete the expansion.

[0050] (3) Simulation phase like Figure 7 As shown, the newly expanded child nodes during the simulation phase Starting from this point, the simulation proceeds in one go according to the established rapid simulation strategy, until the final node where all tasks are completed. The rapid simulation strategy includes the following four rules: Rule a, according to the ordered task sequence Selecting the currently pending task elements in a predetermined order ; Rule b, if the currently pending microelement Continuous execution identifier Once the microelement is scheduled, the simulation strategy forces the continuous execution of all remaining microelements of the task until the task is fully completed, and then returns to the sequence. Continue to the next item; Rule c: For time intervals caused by conflicts in the safety distance between adjacent cable machines, insert necessary waiting actions until the safety conditions are met before executing the current task element. If there are no safety conflicts, execute directly. Rule d: Repeat the above process, updating the task micro-element state until all task micro-elements are in a completed state. Record the completion time of the last task micro-element as the total completion time of this simulation. Output the scheduling scheme.

[0051] Will Transformed into reward value using a non-linear piecewise mapping function : ; in, As the base time, This is the operating condition factor, with a value ranging from 1.2 to 1.3.

[0052] The formula for calculation is: ; In the formula, For task micro-element The estimated execution time for a single hoisting cycle, The number of cable cranes participating in this dispatch. This represents the theoretical shortest completion time under ideal working conditions, ignoring all non-operational time. The properties of this mapping function are as follows: when the completion time is better than... A full score is awarded when the completion time is twice the baseline time; when the completion time is longer than twice the baseline time, the reward is zero; the reward decreases in the middle range according to a quadratic function. In the early stages of the search, most simulation results... When the reward is relatively large and close to zero, then... The exploration term of the formula dominates node selection, promoting the algorithm to explore in breadth. As the iteration progresses, a high-quality scheduling scheme is discovered, the utilization term increases significantly, the algorithm automatically shifts to deep utilization, and quickly converges to the global optimal solution.

[0053] (4) Retrospective phase The reward value obtained from the simulation Propagate along the simulation path from leaf node to root node, node by node, for any node passed along the path. Update its visit count and cumulative rewards: ; After the update is complete, return to the selection phase to begin the next iteration.

[0054] After reaching the preset number of iterations, proceed along each layer. The largest child node greedily backtracks to the terminal node. The sequence of joint actions along the path represents the optimal task execution order for this feasible task allocation scheme, and the system time of the corresponding terminal node is the completion time of the scheme. .

[0055] In the Monte Carlo tree modeling and solution process described above, the auxiliary hoisting task is incorporated into a unified constraint system and its exploration and utilization tendencies are controlled through an adaptive nonlinear reward mapping algorithm. Compared to a fixed linear reward mapping method, this piecewise mapping function can effectively expand the exploration range in the early stages of the search and guide the algorithm to focus on utilizing the discovered high-quality scheduling paths in the later stages of the search, thus improving both the solution quality and the computational convergence speed. The search trees corresponding to multiple allocation schemes are solved independently in parallel, avoiding the serial coupling between allocation decisions and sequential decisions, and effectively reducing the overall computational overhead.

[0056] In one embodiment of the present invention, step S4 includes: For S2 generated One feasible allocation scheme After solving the Monte Carlo tree in S3, the optimal task execution order and completion time corresponding to each solution are obtained. All schemes are ranked based on completion time as the evaluation index. The task allocation scheme with the shortest completion time and its corresponding optimal task execution order are output as the final refined scheduling scheme for the cable crane group.

[0057] The solution comprises two parts: a task allocation scheme that clarifies the pouring and unloading points and auxiliary hoisting micro-elements that each cable crane is responsible for; and a task execution sequence that is precise down to the task micro-elements executed by each cable crane at each moment, the number of hoisting cycles in that cycle, and the waiting periods triggered by safety distances or process constraints. This achieves a refined leap in scheduling decision granularity from the strip level to the hoisting cycle level, providing directly executable and refined operation plans for on-site construction organization.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A refined scheduling method for arch dam cable crane groups oriented towards all task types, characterized in that, include: S1. Decompose the concrete pouring task and the auxiliary hoisting task into task micro-elements with a single hoisting cycle as the basic unit, and construct a set of task micro-elements. S2. Sort the task micro-element set according to the spatial coordinates of each task micro-element, generate several continuous segment division schemes using the constrained partition method and allocate them to each cable machine. After verifying the safe distance between adjacent cable machines, retain all feasible task allocation schemes. S3. Model the cable crane group scheduling problem as a Markov decision process. Based on each feasible task allocation scheme, construct the initial joint action, generate multiple parallel independent search tree root nodes, and use the Monte Carlo tree search algorithm to select, expand, simulate, and backtrack iteratively solve the optimal task execution order that satisfies the construction constraints, and obtain the completion time corresponding to each scheme. S4. Compare all feasible task allocation schemes with the completion time as the objective, and output the scheme with the shortest completion time as the final refined scheduling scheme for the cable crane group.

2. The method according to claim 1, characterized in that, Step S1 includes: The concrete pouring task is divided into several unloading points according to the pouring chamber. Each unloading point corresponds to a single hoisting cycle, which constitutes a pouring task micro-element. All pouring task micro-elements form the pouring micro-element set. ; The auxiliary lifting task is decomposed into a single lifting cycle according to the implementation requirements. Each lifting operation constitutes an auxiliary task micro-element, and all auxiliary task micro-elements form an auxiliary micro-element set. ; Pick As a set of task micro-elements.

3. The method according to claim 2, characterized in that, The casting task micro-element is a triplet Characterization, in which Number the pouring chamber. This is the serial number of the unloading point within the warehouse. Here are the spatial coordinates of the unloading point. To complete the required number of hoisting cycles at this unloading point, This refers to the duration of a single hoisting operation.

4. The method according to claim 2, characterized in that, Auxiliary hoisting tasks are classified into three categories based on their implementation characteristics: The first category is tasks that can be intermittently performed within a specified time window. Internal interruption is allowed, superscript The start time of the time window, indicated by the superscript. This is the end time of the time window. Number the auxiliary tasks; The second category is interval-continuous tasks, within a time window. The process must be completed continuously to ensure continuous execution of the identifier. mark; The third type is intermittent single tasks, which have no time window constraints and can be executed at any time within the scheduling cycle; Auxiliary task micro-element in quintuple Characterization, in which The hoisting sequence for this task. The coordinates of the work location, This refers to the duration of a single hoisting operation. To assist in the execution window of hoisting tasks, This is a continuous execution flag.

5. The method according to claim 1, characterized in that, Step S2 includes: According to the coordinates of the unloading point of each task micro-element Ascending order, same axis Press down on axis Axis ascending pair task infinitesimal set Sort to obtain an ordered task sequence ; Using the constrained diaphragm method to Divided into Each of the consecutive sub-segments is assigned to one of them. Taiwan cable machine, enumerate all partitioning schemes, among which This refers to the number of cable machines; For each partitioning scheme, the safety distance of the first task of adjacent cable cranes is checked, and schemes that do not meet the safety distance constraints are eliminated, while all feasible task allocation schemes are retained.

6. The method according to claim 5, characterized in that, Adjacent cable cars and The method for verifying the safe distance between them is as follows: Take the tasks assigned to each cable machine The task element with the smallest coordinates is taken as the first task of the cable machine; If a cable crane is assigned to an empty task group, its safety distance check is skipped. Record the two adjacent cable cranes and The first task The coordinates are respectively and ,like If the partitioning scheme satisfies the safety distance constraint, it is discarded; otherwise, it is discarded. This is the minimum safe distance required for operation between adjacent cable machines.

7. The method according to claim 5, characterized in that, Step S3 includes: The cable crane group scheduling problem is modeled as a Markov decision process, with the task completion status, the current operation status of each cable crane and the current system time constituting the system state, and the task micro-element combination of each cable crane to be executed next constituting the joint action. The initial state is defined as the condition where all task micro-elements are incomplete and all idle cable cranes are awaiting task assignment. Based on each feasible allocation scheme, the first task of each cable machine is combined into an initial joint action. After executing this action, the system... Transition to the corresponding new state To generate multiple parallel, independent search tree root nodes; For each independent root node of the search tree, the Monte Carlo tree search algorithm is used to perform four stages of iteration: selection, expansion, simulation, and backtracking, until the preset number of iterations is reached and then terminated. The task execution order corresponding to the leaf node path with the highest cumulative reward during the iteration process is taken as the optimal execution order of the scheme, and the corresponding completion time is recorded.

8. The method according to claim 7, characterized in that, The selection phase employs a confidence upper bound tree algorithm, starting from the root node and selecting child nodes layer by layer. Select child nodes At that time, calculate the values ​​of each candidate child node. value: ； in child node Cumulative rewards child node Number of visits, For its parent node Number of visits, To explore coefficients; Each time you choose The child node with the largest value expands downwards until it reaches a node that has not been fully expanded.

9. The method according to claim 7, characterized in that, In the expansion phase, at the current node, based on the continuity constraints of the unloading point, the time constraints of the billet layer coverage, the safety distance constraints between adjacent cable cranes, and the constraints of the continuous execution and window period of auxiliary tasks, the set of legal task micro-elements for each cable crane in the current state is selected, the Cartesian product of the legal actions of each cable crane is taken to construct the joint action space, and a child node corresponding to a joint action is randomly selected from it and added to the search tree. The simulation phase starts from the newly expanded node and proceeds according to the ordered task sequence. The predetermined sequence is rapidly deduced to the final outcome, for continuous execution of markers. The auxiliary task micro-element is forced to execute continuously. For time intervals caused by conflicts in the safety distance between adjacent cable machines, waiting actions are inserted. The simulated completion time is obtained after the simulation is completed. And use a nonlinear piecewise mapping function to Converted into reward value: ； in, As the base time, This is the operating condition factor, with a value ranging from 1.2 to 1.

3.

10. The method according to claim 7, characterized in that, The backtracking phase will simulate the obtained reward values. Propagate along the simulation path from leaf node to root node, node by node, for any node passed along the path. Update its access count and cumulative rewards After completing the update, return to the selection phase to begin the next iteration; After reaching the preset number of iterations, proceed along each layer. The largest child node greedily backtracks to the terminal node, and the sequence of joint actions along the path is the optimal task execution order for this feasible task allocation scheme.

Images