Master-slave relation-oriented multi-stack-position steel plate stack transferring method for intelligent unmanned crown block
By optimizing the steel plate stacking decision through a neural network ensemble model and a generative decoding rule set, the problem of predicting the master-slave relationship constraint between steel plates was solved, and efficient and smooth stacking operation of the intelligent unmanned crane was realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN UNIV OF SCI & TECH
- Filing Date
- 2026-01-09
- Publication Date
- 2026-04-21
AI Technical Summary
Existing technologies lack embedded prediction and collaborative decision-making mechanisms that constrain the master-slave relationship between steel plates, making it difficult to avoid deadlock and redundancy in the stacking step sequence, and thus failing to meet the high-efficiency operation requirements of intelligent unmanned overhead cranes.
A neural network ensemble model is used to extract the digital features of stacking, and combined with the rules for balancing temporary stack loads and the generative decoding rule set, the optimal stacking step sequence is generated through a two-stage decoding logic, including the rules for yielding, moving away, and moving back, to optimize the stacking decision of the master-slave relationship between steel plates.
By accurately matching the steel plate ownership and master-slave relationship, steel plate obstruction and repeated handling are avoided, improving the continuity and efficiency of operations, shortening the optimization time, and ensuring that the stacking process is free of jams and redundancy.
Smart Images
Figure CN121901641A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of multi-stack steel plate stacking technology, specifically to an intelligent unmanned overhead crane method for multi-stack steel plate stacking oriented towards a master-slave relationship. Background Technology
[0002] The essence of stacking operations is to utilize limited stacking resources, typically one initial stacking position and multiple temporary stacking positions, to reorganize a randomly stacked stack of steel plates on the initial stacking position into an ordered sequence of steel plates on multiple temporary stacking positions by means of overhead cranes, while minimizing the total number of handling operations.
[0003] Early research focused on establishing mathematical models for the stacking problem. For example, some studies reduced it to a sequence-dependent generalized assignment problem and used specific encoding methods, such as SCALING encoding based on steel plate serial numbers, to characterize the solution. Other studies proposed a variety of heuristic adjustment strategies to minimize the number of stacking steps, given the rolling schedule and real-time inventory.
[0004] Based on the above findings, the limitations of existing technologies include at least the following problems: Firstly, the lack of an embedded predictive and collaborative decision-making mechanism for the master-slave relationship constraints between steel plates makes it difficult to proactively identify the pressure conflicts caused by the placement of the main board on future slave boards with larger plate numbers when formulating the stacking sequence. This makes it difficult to avoid invalid stacking steps due to forced unblocking later, and it easily leads to operational deadlock when dealing with complex resource constraints of multiple temporary stacking positions. Secondly, because the initial stacking clearing process and the temporary stacking arrangement process are not globally collaboratively modeled, it is difficult to achieve the globally optimal decision goal of the total number of stacking steps while satisfying the rigid constraint of the final order of each stacking position. Consequently, the generated stacking scheme has decision blind spots, making it difficult to meet the precise requirements of intelligent unmanned overhead cranes for efficient operation. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a method for stacking steel plates in multiple positions using an intelligent unmanned overhead crane with a master-slave relationship. This method solves the problems of deadlock and redundant steps in decision-making caused by the lack of master-slave relationship prediction in existing technologies.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for multi-stacking steel plate reloading of an intelligent unmanned overhead crane with a master-slave relationship, comprising the following steps: acquiring reloading task data of the intelligent unmanned overhead crane, wherein the reloading task data includes an initial disordered sequence of steel plates at each stack position and the number of temporary stack positions; performing feature extraction and optimization processing on the reloading task data of the intelligent unmanned overhead crane based on a pre-trained neural network ensemble model to generate a recommended steel plate encoding set for the intelligent unmanned overhead crane; and generating an initial population, including several [unclear text - possibly related to a specific set of rules or parameters], based on a preset rule for balancing the load of temporary stack positions and in conjunction with the recommended steel plate encoding set of the intelligent unmanned overhead crane. A recommended steel plate encoding set is selected. Based on a preset production decoding rule set, which includes yielding rules, removal rules, and return rules based on master-slave relationship constraints between steel plates, each recommended steel plate encoding set in the initial population is decoded to obtain the stacking step sequence and total stacking steps corresponding to each recommended steel plate encoding set in the initial population. Based on an intelligent optimization algorithm, the stacking step sequence and total stacking steps corresponding to each recommended steel plate encoding set in the initial population are iteratively optimized to generate the optimal stacking step sequence. Based on the optimal stacking step sequence, the intelligent unmanned crane performs multi-stack steel plate stacking processing.
[0007] Furthermore, the initial stacking position steel plate disordered sequence specifically refers to several steel plates on the initial stacking position. The neural network ensemble model includes several heterogeneous perceptrons and a random forest model. The specific steps for generating the recommended steel plate encoding set for the intelligent unmanned crane are as follows: inputting the stacking task data of the intelligent unmanned crane into the neural network ensemble model; the neural network ensemble model extracts features from the initial stacking position steel plate disordered sequence of the intelligent unmanned crane based on preset rules to generate the stacking digital feature set of the intelligent unmanned crane; each heterogeneous perceptron and random forest model in the neural network ensemble model performs independent attribution prediction processing based on the stacking digital feature set to generate several candidate initial steel plate encoding sets; and performs fusion decision processing on each candidate initial steel plate encoding set to extract and generate the recommended steel plate encoding set for the intelligent unmanned crane.
[0008] Further, the specific steps for generating the stacking digital feature set of the intelligent unmanned overhead crane are as follows: For each steel plate in the initial stacking steel plate disordered sequence of the intelligent unmanned overhead crane, determine the order position and extract the position features of the corresponding steel plate; perform statistical processing on the initial stacking steel plate disordered sequence of the intelligent unmanned overhead crane to extract the minimum and maximum number pairs of the corresponding steel plates; and extract locally ordered segments from the initial stacking steel plate disordered sequence of the intelligent unmanned overhead crane to extract the starting and ending data chains of the corresponding steel plates; based on the position features, minimum and maximum number pairs, starting and ending data chains of each steel plate, construct the stacking digital feature set of the intelligent unmanned overhead crane.
[0009] Furthermore, the specific steps for generating the initial population are as follows: based on the recommended steel plate code set of the intelligent unmanned crane, the steel plate allocation sequence is statistically analyzed; based on the preset balanced temporary stack load rules, the steel plate allocation sequence is balanced and adjusted to generate the initial population.
[0010] Furthermore, the specific steps of the pre-defined production decoding rule set are as follows: obtain the stacking sample dataset, and perform analytical processing on the stacking sample dataset based on the pre-established mixed integer linear programming model to generate the optimal sample solution set; analyze the decision rules constrained by the master-slave relationship between steel plates in the optimal sample solution set; and solidify the decision rules into a production decoding rule set that includes yielding rules, moving rules, and moving back rules.
[0011] Furthermore, the specific steps for obtaining the stacking step sequence and total stacking steps corresponding to each recommended steel plate code set in the initial population are as follows: Based on the production decoding rule set, each recommended steel plate code set in the initial population is decoded in two stages according to the preset two-stage decoding; the two-stage decoding includes: the first stage decoding, which is used to process each recommended steel plate code set in the initial population according to the production decoding rule set until the preset clearing condition is met; the second stage decoding, which is used to sort each recommended steel plate code set in order according to the production decoding rule set until the target sorting direction is met;
[0012] Statistical processing was performed on each recommended steel plate code set in the initial population after two-stage decoding to obtain the stacking step sequence and total stacking steps corresponding to each recommended steel plate code set in the initial population.
[0013] Furthermore, the specific steps of the first stage of decoding are as follows: Based on each recommended steel plate encoding set in the initial population, determine whether it meets the preset first condition; if it meets the preset first condition, then execute the yield rule and determine whether it meets the preset second condition; if it does not meet the preset second condition, then execute the relocation rule and the withdrawal rule in sequence; if it meets the preset second condition, then execute the withdrawal rule; if it does not meet the preset first condition, then determine whether it meets the preset second condition; if it meets the preset second condition, then execute the withdrawal rule.
[0014] Furthermore, the specific steps of the second stage decoding are as follows: For each recommended steel plate encoding set in the initial population after the first stage decoding, determine whether the preset third condition is met. If the preset third condition is met, determine whether the preset fourth condition is met. If the preset fourth condition is met, execute the withdrawal rule and generate the corresponding stacking step sequence. If the preset fourth condition is not met, execute the removal rule and withdrawal rule in sequence and generate the corresponding stacking step sequence. If the preset third condition is not met, repeat the determination of the preset first and second conditions until the corresponding stacking step sequence is generated.
[0015] Furthermore, the specific steps for generating the optimal stacking step sequence are as follows: Initialize the initial population to generate the current population and the current optimal solution; Perform crossover and mutation operations and neighborhood search on the current population in sequence, and update the current optimal solution; Determine whether the algorithm termination condition is met; If it is met, output the optimal stacking step sequence; If it is not met, continue iterative optimization.
[0016] Furthermore, the specific steps of the initialization process are as follows: mark the initial population as the current population; based on the total number of stacking steps corresponding to each recommended steel plate code set in the current population, count the recommended steel plate code set with the smallest total stacking steps, and mark it as the current optimal solution.
[0017] The present invention has the following beneficial effects:
[0018] (1) The intelligent unmanned crane multi-stack steel plate reloading method for master-slave relationship integrates the master-slave relationship between steel plates into the core of reloading decision by pre-setting a generative decoding rule set including yielding rules, removal rules, and return rules. During the decoding process, the system proceeds in an orderly manner according to the recommended steel plate encoding set and two-stage decoding logic: the first stage focuses on clearing the initial stack position, and automatically allocates the temporary placement of the main board and clears obstacles of the assigned stack position through rule judgment; the second stage focuses on the orderly arrangement of the temporary stack position to ensure that the steel plates are arranged in the target direction. The three rules work together to ensure that each reloading operation accurately matches the steel plate's ownership and master-slave relationship, which not only avoids mutual obstruction when the steel plates are returned to their positions, but also reduces the invalid actions of repeated reloading, ensuring that the reloading process is smooth and free of redundancy, and greatly improving the continuity and efficiency of the intelligent unmanned crane operation.
[0019] (2) The intelligent unmanned crane multi-stack steel plate re-stacking method for master-slave relationship constructs an initial population through neural network ensemble optimization and load balancing rules, providing a high-quality starting point for intelligent optimization algorithm. The neural network ensemble model first extracts the re-stacking digital feature set, and generates a recommended steel plate code set through multi-model independent prediction and fusion decision. Then, combined with the load balancing rules of temporary stacks, it ensures that the steel plate distribution of each stack is balanced. The proportion of high-potential candidate solutions in the high-quality initial population is significantly increased. With the iterative optimization process of crossover, mutation, and neighborhood search, the algorithm does not need to waste resources in low-quality solutions, can quickly focus on the optimal solution range, greatly shorten the optimization time, and meet the needs of real-time on-site operation.
[0020] (3) The intelligent unmanned crane multi-stack steel plate reloading method for master-slave relationship provides a unified constraint framework by relying on the mixed integer linear programming model to ensure that the reloading operation conforms to the basic rules; the neural network ensemble model is trained with multiple sets of samples and can accurately adapt to different combinations of initial steel plate disordered sequences and temporary stacking positions; the algorithm termination condition supports custom operation time and can balance decision speed and solution quality according to on-site needs. All output optimal reloading step sequences have been iterated and verified multiple times to ensure that the process is feasible and the number of steps is optimal. Whether it is a regular scale task or a complex scenario with multiple steel plates and multiple stacking positions, it can stably output a reloading solution without human intervention, reduce the difficulty of on-site operation, and give full play to the operation value of intelligent unmanned crane.
[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description
[0022] Figure 1 This is a flowchart of a multi-position steel plate stacking method for an intelligent unmanned overhead crane with a master-slave relationship according to the present invention.
[0023] Figure 2 This is a flowchart illustrating the specific steps involved in generating a recommended steel plate code set for an intelligent unmanned overhead crane in a master-slave relationship-oriented multi-stack steel plate stacking method according to the present invention.
[0024] Figure 3 This is a flowchart illustrating the specific steps of the pre-set production decoding rule set in the intelligent unmanned overhead crane multi-stack steel plate stacking method for master-slave relationship according to the present invention. Detailed Implementation
[0025] Please see Figure 1This invention provides a technical solution: a method for multi-stacking steel plates using an intelligent unmanned overhead crane with a master-slave relationship, comprising the following steps: acquiring stacking task data of the intelligent unmanned overhead crane, the stacking task data including an initial disordered sequence of steel plates in the stacking positions and the number of temporary stacking positions (including the stacking position number of each temporary stacking position); performing feature extraction and optimization processing on the stacking task data of the intelligent unmanned overhead crane based on a pre-trained neural network ensemble model to generate a recommended steel plate encoding set for the intelligent unmanned overhead crane; generating an initial population, including several recommended steel plate encoding sets, based on a preset balanced temporary stacking position load rule and combined with the recommended steel plate encoding set of the intelligent unmanned overhead crane; and generating an initial population, including several recommended steel plate encoding sets, based on a preset production decoding rule set, the production decoding rule set including rules based on the master-slave relationship between steel plates. Based on the yielding, moving, and returning rules constrained by relationships, each recommended steel plate code set in the initial population is decoded to obtain the stacking step sequence and total stacking steps corresponding to each recommended steel plate code set in the initial population. Then, based on an intelligent optimization algorithm, the stacking step sequence and total stacking steps corresponding to each recommended steel plate code set in the initial population are iteratively optimized to generate the optimal stacking step sequence. Finally, based on the optimal stacking step sequence, the intelligent unmanned overhead crane performs multi-stack steel plate stacking processing, that is, converting the optimal stacking step sequence into a specific instruction set that the intelligent unmanned overhead crane control system can recognize and execute. Each step in this sequence is usually represented in the form of [source stacking position number, target stacking position number], clearly indicating the start and end points of each handling operation.
[0026] Specifically, such as Figure 2-3 As shown, the neural network ensemble model includes several heterogeneous perceptrons and a random forest model. The initial stacked steel plate disordered sequence specifically refers to several steel plates on the initial stack (it should be noted that the steel plates are stacked in random order, and each steel plate has a unique plate number). The specific steps for generating the recommended steel plate encoding set for the intelligent unmanned crane are as follows:
[0027] Input the stacking task data of the intelligent unmanned overhead crane into the neural network ensemble model;
[0028] The neural network ensemble model extracts features from the initial disordered sequence of steel plates in the stacking position of the intelligent unmanned crane based on preset rules, and generates a stacking digital feature set of the intelligent unmanned crane.
[0029] Each heterogeneous perceptron and random forest model in the neural network ensemble model performs independent attribution bit prediction processing based on the stacking digital feature set, generating several candidate initial steel plate coding sets, specifically as follows:
[0030] Based on the preset master-slave relationship definition rules, each steel plate on the initial stack of the intelligent unmanned crane is defined and processed. The steel plate at the top of the initial stack and currently to be moved is defined as the master plate, and the steel plate below the master plate and to be moved later is defined as the slave plate.
[0031] Each heterogeneous perceptron contains multiple fully connected layers and nonlinear activation functions. Through forward propagation, the input digital features are mapped to a predicted probability distribution of the assigned stack position for each steel plate within the range of temporary stack position numbers. The stack position with the highest probability is selected as the predicted assigned position of the steel plate, thus forming a candidate initial steel plate encoding set output by each heterogeneous perceptron.
[0032] For the random forest model, it contains a large number (e.g., 500) decision trees. Each decision tree makes a decision independently based on the input features. Finally, by statistically analyzing the prediction results of all decision trees in the forest for the assigned stack position of each steel plate, the predicted assigned position of the steel plate is determined by majority vote, thus forming a candidate initial steel plate encoding set output by the model.
[0033] For each candidate initial steel plate coding set, a fusion decision process is performed to extract and generate a recommended steel plate coding set for the intelligent unmanned crane. Specifically, for each steel plate in the initial stack position disordered sequence, the following operations are performed:
[0034] Voting statistics: Traverse all candidate code sets and count the number of stack positions assigned to the specific steel plate in each candidate scheme and the number of times it appears (i.e., the number of votes).
[0035] Decision generation: Compare the number of votes for each stack position number above, and determine the stack position number with the highest number of votes as the position of the specific steel plate in the final recommended scheme. If there is a tie in the number of votes, a decision can be made according to preset rules (such as random selection or priority selection of the stack position with the smaller number).
[0036] Repeat the above steps until a final recommended assignment position is determined for each steel plate in the sequence; arrange the final recommended assignment positions determined for all steel plates according to the order of the corresponding steel plates in the initial sequence to generate the recommended steel plate code set for the intelligent unmanned crane.
[0037] The training steps for the neural network ensemble model are as follows:
[0038] Construct a training dataset: Collect multiple sets of steel plate stacking sample data of different sizes. Each set of samples includes an initial disordered steel plate sequence, the number of available temporary stacking positions, the steel plate attribution rules, and the corresponding optimal steel plate code (as a label). Extract the position features, number pair features, and data chain features of the disordered steel plate sequence in each set of samples to form a stacking digital feature set, which is combined with the initial steel plate code set to form a training sample pair.
[0039] Dataset partitioning: Divide the training sample pairs into a training set and a validation set in an 8:2 ratio. The validation set is used to determine the convergence status of the model.
[0040] Model structure construction: Three structurally heterogeneous multilayer perceptrons (MLPs) were built: the first MLP has 3 hidden layers with 16, 16 and 32 neurons respectively; the second MLP has 2 hidden layers with 16 neurons each; the third MLP has 4 hidden layers with 8, 8, 16 and 16 neurons respectively; each MLP uses the tanh function as the nonlinear activation function.
[0041] Build a random forest model: containing 500 decision trees, with the maximum depth of a single decision tree set to 8 and the minimum number of split samples set to 5;
[0042] Model training configuration: Set the maximum number of training iterations to 2000, use the cross-entropy loss function to calculate the prediction loss of each model, and update the network parameters of the multilayer perceptron through backpropagation; the random forest model completes the construction and training of decision trees based on the features and labels of the training set samples.
[0043] Convergence determination: After each training round, the prediction accuracy of each model is calculated using the validation set; if the accuracy of the validation set does not improve for ten consecutive generations, the model training is determined to be converged and training is stopped.
[0044] Integration training complete: Save the parameters of the three converged multilayer perceptrons and random forest models to form a pre-trained neural network ensemble model, which will be used for the subsequent generation of the recommended steel plate coding set.
[0045] The specific steps for generating the stacking digital feature set of the intelligent unmanned overhead crane are as follows: For each steel plate in the initial stacking steel plate disordered sequence of the intelligent unmanned overhead crane, determine its sequential position in the sequence and extract the position features of the corresponding steel plate. Specifically, the initial stacking steel plate disordered sequence is defined from top to bottom as sequential positions 1, 2, N, where N is the total number of steel plates; for each steel plate in the sequence, record its sequential position as the position feature of the steel plate.
[0046] The initial disordered sequence of steel plates in the stacking position of the intelligent unmanned overhead crane is statistically processed to extract the minimum and maximum number pairs of the corresponding steel plates. Specifically, for each target steel plate in the initial disordered sequence of steel plates in the stacking position, the following operations are performed:
[0047] Minimum number pair extraction: Iterate through all steel plates following the target steel plate and count the number of steel plates with plate numbers greater than the target steel plate number. This number is the minimum number pair for the target steel plate.
[0048] Maximum number pair extraction: Traverse all steel plates preceding the target steel plate and count the number of steel plates with plate numbers less than the target steel plate number. This number is the maximum number pair of the target steel plate.
[0049] The system extracts locally ordered segments from the initial disordered sequence of steel plates in the intelligent unmanned overhead crane, extracting the starting and ending data chains of the corresponding steel plates. Specifically, starting from the position of the steel plate in the sequence, the plate numbers of subsequent steel plates are checked sequentially downwards (towards the end of the sequence). If the plate numbers of subsequent steel plates maintain a strict increment, they are included in the chain until the plate numbers no longer meet the increment condition. The total number of steel plates contained in the chain is counted, and this number is the starting data chain of the target steel plate.
[0050] End of data chain extraction: Starting from the position of the steel plate in the sequence, check the plate number of the preceding steel plate sequentially upwards (towards the start of the sequence); if the plate number of the preceding steel plate remains strictly decreasing, include it in the chain until the plate number no longer meets the decreasing condition; count the total number of steel plates contained in the chain, and this number is the end of the data chain for the target steel plate.
[0051] Based on the positional characteristics of each steel plate, the minimum and maximum set of numbers, the starting and ending data chains, a stacking digital feature set is constructed.
[0052] For example, given an initial randomized sequence of steel plates in a stack, with plate numbers from top to bottom as 3, 6, 1, 7, 8, 10, 2, 4, 9, 5, taking plate number 7 in the initial randomized sequence as an example, the features are defined as follows: Position feature: 4th from top to bottom; Minimum number pair feature: all steel plate numbers greater than 7 in the sequence below plate number 7, i.e., [7,8][7,9][7,10]; Maximum number pair feature: all steel plate numbers less than 7 in the sequence above plate number 7, [1,7][3,7][6,7]; Starting data chain feature: the sequence of consecutive steel plate numbers greater than 7 below plate number 7, [7→8→10]; Ending data chain feature: the sequence of consecutive steel plate numbers less than 7 above plate number 7, [3←6←7].
[0053] The specific steps for generating the initial population are as follows:
[0054] Based on the recommended steel plate code set of the intelligent unmanned crane, the steel plate allocation sequence is statistically analyzed. Specifically, the belonging position of each steel plate in the recommended steel plate code set is analyzed one by one to clarify the temporary stack position to which each steel plate is allocated. According to the top-down order of the initial stack position steel plate disorder sequence, the correspondence between the steel plate identifier and the temporary stack position number of each steel plate is recorded in turn to form an ordered steel plate allocation sequence. At the same time, the total number of steel plates corresponding to each temporary stack position in the allocation sequence is counted.
[0055] Based on preset rules for balancing temporary stack loads, the steel plate allocation sequence is adjusted to generate an initial population. Specifically, the theoretical average load for each temporary stack is calculated based on the total number of temporary stacks and the total number of steel plates (i.e., the total number of steel plates divided by the total number of temporary stacks, rounded up). A load balancing threshold is set to the theoretical average load ±1; exceeding this range indicates a load imbalance. For the recommended steel plate code set (baseline code) corresponding to the steel plate allocation sequence, if there are temporary stacks with loads exceeding the balancing threshold, steel plates with smaller plate numbers and lower impact on subsequent sorting are prioritized, and their assigned codes are adjusted to temporary stacks with loads below the balancing threshold. This ensures that the loads of all temporary stacks are within the balancing threshold range after adjustment, generating one set of balanced loads. To ensure the diversity of the initial population, the baseline variant coding is balanced. Multiple independent small perturbation adjustments are performed based on the balanced baseline variant coding: each time, only two steel plates from different temporary stacks are randomly selected, and their assigned position codes are swapped. After the swap, the load of each temporary stack is re-verified to ensure that the balance threshold is not exceeded. This operation is repeated at least 3 times to generate at least 3 sets of differentiated balanced variant codes. The recommended steel plate code set (baseline code), the balanced baseline variant code, and all differentiated balanced variant codes are collected to form the initial population. This initial population contains several recommended steel plate code sets that meet the load balancing requirements. This not only ensures a high-quality foundation for the population but also enriches the population diversity through variant coding, providing sufficient high-quality candidate solutions for subsequent iterative optimization.
[0056] In this implementation plan, through scientific model training and rule extraction, the prediction of steel plate location and the decision-making of stacking are made more accurate and reliable. The neural network ensemble model combines three heterogeneous multilayer perceptrons and random forests, and through training with multiple sets of samples and fusion with hard voting, it can output a high-quality recommended steel plate encoding set. The mixed integer linear programming model provides the benchmark for the optimal solution, and the three production decoding rules extracted from it accurately grasp the core constraints of the master-slave relationship. The whole process not only uses model learning to capture complex features, but also simplifies the decision-making logic through rule solidification, avoiding blind operation, so that the stacking scheme not only conforms to the optimal law, but also has executability, providing a good foundation for subsequent iterative optimization.
[0057] Specifically, the steps to obtain the stacking step sequence and total stacking steps corresponding to each recommended steel plate code set in the initial population are as follows: Based on the production decoding rule set, each recommended steel plate code set in the initial population is processed according to the preset two-stage decoding; the two-stage decoding includes: the first stage decoding, which is used to process each recommended steel plate code set in the initial population based on the production decoding rule set (that is, to move each steel plate in each recommended steel plate code set from the initial stacking position to the temporary stacking position in sequence) until the preset clearing condition is met (that is, the initial stacking position is cleared).
[0058] The second stage of decoding is used to organize each recommended steel plate code set (after the first stage of decoding) in an ordered manner based on the production decoding rule set until the (steel plates in each stack position) meet the target sorting direction; and to perform statistical processing on each recommended steel plate code set in the initial population after the two stages of decoding to obtain the stacking step sequence and the total number of stacking steps corresponding to each recommended steel plate code set in the initial population (that is, to record every steel plate handling operation performed by the crane in the entire decoding process, including the initial stack position clearing stage and the temporary stack position sorting stage, to form a complete stacking step sequence in sequence, and at the same time, to count the total number of handling operations contained in the sequence, that is, to obtain the total number of stacking steps corresponding to the recommended steel plate code set).
[0059] The specific steps of the first stage of decoding are as follows: Based on each recommended steel plate code set in the initial population, determine whether the preset first condition is met (i.e., the main board and the slave board belong to the same group).
[0060] If the preset first condition is met, the yielding rule is executed (according to the yielding rule, to prevent subsequent slave boards with larger board numbers from being unable to be placed, the main board needs to be temporarily placed on a temporary stack location other than its assigned location). After executing the yielding rule, it is determined whether the preset second condition is met (i.e., the main board's assigned location is empty). If the preset second condition is not met, the removal rule and the withdrawal rule are executed in sequence (first, according to the removal rule, the obstructing steel plate occupying the main board's assigned location is moved away; if the obstructing steel plate's own assigned location is also occupied, its assigned location is recursively cleared according to the withdrawal rule, and finally the obstructing steel plate is moved back to its correct position). If the preset second condition is met, the withdrawal rule is executed (i.e., there is no need to perform the removal operation, and the action of placing the main board into its assigned stack location can be directly executed).
[0061] If the first preset condition is not met (then the motherboard is directly placed in its assigned position), then it is determined whether the second preset condition is met; if the second preset condition is met, then the withdrawal rule is executed.
[0062] The detailed process is as follows:
[0063] Step 1: Determine the master-slave relationship (first condition):
[0064] First, determine whether the current motherboard and its key slave board below it meet the first condition, that is, whether their assigned positions are the same (whether they are pre-assigned to the same temporary stack position).
[0065] If the first condition is met (the motherboard and the slave board belong to the same location), it indicates that there is a slave board competing with the motherboard for the same stack position. In order to prevent the slave board with the same location and a larger board number from being unable to be placed due to being pressed down, the yielding rule must be triggered first. According to the yielding rule, the algorithm will not put the motherboard directly into its location, but will temporarily place it on another available temporary stack position that is not its location, to make way for the slave board.
[0066] If the first condition (the motherboard and the slave board have different affiliations) is not met, it means that the slave board will not directly compete for the motherboard's affiliation position. The motherboard does not need to give way and can directly proceed to judge its affiliation position status.
[0067] Step Two: Determine the motherboard's home bit status (second condition) and execute the corresponding rules:
[0068] After completing the first condition judgment and possible yielding operations, the next step is to determine whether the second condition is met, namely whether the temporary stacking position to which the motherboard belongs is currently empty.
[0069] For the case where the first condition (giving way has been performed):
[0070] If the second condition is not met (the motherboard's assigned position is not empty, i.e., it is occupied by steel plates from other stacks), then the removal and relocation rules must be triggered sequentially. According to the removal rule, the obstructing steel plate occupying the motherboard's assigned position must be removed. If the assigned position of the obstructing steel plate is also occupied by other steel plates, then according to the relocation rule, all steel plates occupying its assigned position must be recursively cleared to form a clearing chain. Only then can the obstructing steel plate be accurately moved back to its correct assigned position. After the obstruction is cleared, the motherboard can be placed from its temporary clearing position (or directly) to its vacated assigned position.
[0071] If the second condition is met (the motherboard's assigned position is empty), it means that the motherboard can be returned to its assigned position directly without needing to be moved. The motherboard can be placed directly into its assigned stack position.
[0072] For situations where the first condition (failure to give way) is not met:
[0073] Directly check the second condition (whether the motherboard location bit is empty);
[0074] If the second condition is met (the home slot is empty), then the motherboard is placed directly into its home slot.
[0075] If the second condition (the assigned position is not empty and there is an obstruction steel plate) is not met, then the logic is the same as above. The move-away rule and the move-back rule need to be triggered in sequence. After clearing the obstruction steel plate on the assigned position, the motherboard is placed in its assigned position.
[0076] This set of conditional judgment logic runs through the first stage of the decoding process (clearing the initial stack position), ensuring that every motherboard handling decision can proactively consider master-slave constraints and properly handle stack position conflicts, avoiding invalid stacking and deadlock.
[0077] The specific steps for the second stage of decoding are as follows:
[0078] For each recommended steel plate encoding set in the initial population after the first segment of decoding, determine whether the preset third condition (i.e., the initial stack is empty) is met. If the preset third condition is met, determine whether the preset fourth condition (i.e., the largest number belongs to an empty stack) is met. If the preset fourth condition is met, execute the withdrawal rule and generate the corresponding stacking step sequence. If the preset fourth condition is not met, execute the removal rule and withdrawal rule in sequence and generate the corresponding stacking step sequence.
[0079] If the preset third condition is not met, then (indicating that there are still motherboards to be processed, a new cycle begins at this time, based on the updated stacking status, a new motherboard is taken out, and) the preset first and second conditions are repeatedly judged until the corresponding stacking step sequence is generated.
[0080] The detailed process is as follows:
[0081] Step 1: Determine the stage transition condition (third condition). First, determine whether the third condition is met, that is, whether all steel plates on the initial stack have been moved (the initial stack is empty).
[0082] If the third condition is met (the initial stack is empty), it means that all steel plates have been moved out of the initial stack and the first stage of decoding is completed. At this time, the decoding logic automatically switches to the second stage, which is to sort and finally sort the steel plates that are distributed on each temporary stack and have not yet been arranged in the final target order.
[0083] If the third condition (the initial stack still contains steel plates) is not met, it indicates that the first stage of decoding has not yet ended. Based on the latest stacking status, the next steel plate will be taken from the top of the initial stack as the new motherboard. Then, the first stage of decoding logic will be returned and repeated (i.e., the first and second conditions will be re-evaluated, and the yielding, removal, and return rules will be executed accordingly) until the initial stack is cleared and the third condition is met.
[0084] Step Two: Core Processing Logic of the Second Stage: Once the second stage is confirmed, the following operations will be performed repeatedly until all steel plates are in order:
[0085] Determine the current processing target: At the beginning of each processing cycle, the system scans all temporary stacks and selects the steel plate with the largest plate number from all steel plates that have not yet been placed in their final ordered position (i.e., not arranged in an ordered manner), and determines it as the target steel plate that needs to be processed now.
[0086] Determine the status of the target steel plate's assigned position (fourth condition): After determining the target steel plate, determine whether the fourth condition is met, that is, whether the temporary stacking position to which the target steel plate is assigned is currently empty (meaning that there is no steel plate on the top layer of the stacking position, or the existing top layer steel plate does not affect the target steel plate being placed in the final order).
[0087] The corresponding rule will be executed according to the fourth condition:
[0088] If the fourth condition is met (the target steel plate's assigned position is empty or can be placed directly), the simplified execution of the relocation rule is triggered. In this case, there is no need to move other steel plates, and the largest steel plate can be placed directly into the correct position of its assigned temporary stack.
[0089] If the fourth condition (the target steel plate's designated location is not empty, i.e., it is occupied by obstructing steel plates in other stacks) is not met, then the removal and relocation rules need to be triggered sequentially. First, according to the removal rule, the obstructing steel plate occupying the target steel plate's designated location must be moved away. Next, according to the relocation rule, if the obstructing steel plate's designated location is also occupied by other steel plates, then all steel plates occupying its designated location need to be recursively cleared to ensure that the obstructing steel plate can be moved back to its correct designated location. Only after this can the current target steel plate (the largest steel plate) be placed in its vacated designated location.
[0090] Generation steps and loop: Record the specific handling steps generated by the above operations (such as moving from stack X to stack Y), update the global stack position status, and after completing the processing of the largest steel plate, re-determine whether there are still steel plates that are not arranged in an orderly manner. If so, return to point 1 of this step and continue to process the next current largest steel plate; if not, it indicates that all steel plates have formed an orderly sequence on their respective stack positions, and the second stage of decoding and the entire decoding process are completed.
[0091] In this implementation plan, a two-stage decoding and clear condition judgment ensure that the stacking operation is orderly and free of redundancy. The first stage focuses on clearing the initial stack positions and flexibly executes three rules based on the master-slave relationship and the status of the assigned position to avoid conflicts in advance. The second stage focuses on sorting the temporary stack positions, prioritizing the largest steel plate and gradually achieving orderly arrangement. The entire process is logically coherent, and each operation has a clear judgment basis, which not only ensures that the initial stack positions are cleared smoothly, but also efficiently completes the sorting of temporary stack positions. At the same time, it accurately counts the stacking steps and the total number of steps, providing a clear and reliable evaluation basis for subsequent iterative optimization.
[0092] Specifically, the steps for generating the optimal stacking step sequence are as follows: Initialize the initial population, generate the current population and the current optimal solution. The specific steps for initialization are as follows:
[0093] The initial population is marked as the current population; based on the total number of stacking steps corresponding to each recommended steel plate code set in the current population, the recommended steel plate code set with the smallest total number of stacking steps is counted and marked as the current optimal solution; crossover and mutation operations and neighborhood search are performed sequentially on the current population, and the current optimal solution is updated, specifically as follows:
[0094] Crossover and mutation operations: Apply crossover operators (such as single-point crossover, two-point crossover, etc.) and mutation operators (such as basic position mutation) of genetic algorithms to individuals in the current population to generate a new set of candidate individuals, called the offspring population. This process aims to generate new solutions with diversity by combining and randomly perturbing the genes of the parent individuals (i.e., steel plate stacking position encoding).
[0095] Neighborhood search processing: To further improve the quality of the solution, for each individual in the offspring population, at least one of the following two neighborhood search operations is performed to conduct a fine search around it:
[0096] Neighborhood search based on 1-step perturbation: For each bit of an individual code, a random number in the range [0,1) is generated. If the random number is less than a preset first probability threshold (e.g., 0.1), the temporary stacking code corresponding to the bit is finely adjusted by "adding 1" or "subtracting 1" within the valid number range.
[0097] Neighborhood search for steel plate stack assignment within the same cluster: Analyze the initial disordered sequence of steel plates corresponding to the individual, and identify steel plate groups (steel plates within the same cluster) that are adjacent in position and have similar plate numbers. In the initial steel plate sequence, steel plates that are adjacent in position and have similar plate numbers are steel plates within the same cluster. The value of adjacent position is the largest of the two numbers rounded down from (the number of temporary stack positions divided by 2, and the number of initial steel plates divided by 10). The value of similar plate numbers is the largest of the two numbers rounded down from (the number of temporary stack positions divided by 3, and the number of initial steel plates divided by 10). If the value is less than 1, then the value of similar plate numbers is set to 1.
[0098] For example, 4 temporary stacking locations, 13 steel plates
[0099] Location proximity: Taking the larger of (4 / / 2=2, 13 / / 10=1), the location proximity value is 2.
[0100] For boards with similar numbers: take the larger of (4 / / 3=1, 13 / / 10=1), then the value for boards with similar numbers is 1.
[0101] During initialization, steel plates whose positions are within two steel plates and whose plate numbers differ by no more than 1 are assigned to the same temporary stacking position.
[0102] For all individuals in the offspring population after neighborhood search processing, the above-mentioned production decoding rule set is called again for decoding processing. The new total number of stacking steps for each individual is calculated. The new total number of stacking steps for all individuals in the offspring population is compared with the total number of stacking steps for the current optimal solution. If the total number of stacking steps for an individual is less than the total number of stacking steps for the current optimal solution, then the original current optimal solution is replaced by that individual, and its stacking step sequence is recorded.
[0103] To determine whether the algorithm's termination condition is met, the specific criteria are as follows: the algorithm's termination condition is one of the following three:
[0104] Iteration count limit reached: A maximum number of iterations is preset (e.g., 200 times), and the iterations will terminate when the actual number of iterations reaches this value;
[0105] Solution quality stagnation: The total number of stacking steps corresponding to the current optimal solution has not improved (i.e. has not decreased) in multiple consecutive generations (e.g., 10 generations).
[0106] Calculation time exhausted: The total calculation time has reached the preset threshold (e.g., 180 seconds in actual applications).
[0107] If satisfied, the optimal stacking step sequence (the updated stacking step sequence corresponding to the current optimal solution) is output. Specifically, after the entire iterative optimization process ends, the stacking step sequence corresponding to the current optimal solution is officially output as the optimal stacking step sequence of the intelligent unmanned crane.
[0108] If the conditions are not met, iterative optimization will continue. Specifically, the current population will be replaced by the offspring population that has been decoded and evaluated in this iteration, and used as the starting population for the next iteration. Then, the process will return to step two and start a new round of crossover mutation, neighborhood search, decoding evaluation, and updating of the optimal solution through iterative optimization until the algorithm termination condition is met.
[0109] In this implementation plan, the stacking scheme is continuously refined through multiple rounds of iterative optimization to ensure that the final output is the optimal solution. It starts with the encoding with the fewest steps in the initial population, enriches the candidate schemes through cross-mutation, and then expands the optimization range by finely adjusting the 1-step perturbation through two neighborhood search methods. The allocation of steel plates in the same cluster reduces redundant actions. After each round of optimization, the code is re-decoded and evaluated to update the optimal solution in a timely manner. With flexible termination conditions, it avoids invalid iterations and fully explores high-quality solutions. The whole process takes into account both the breadth and accuracy of optimization and gradually selects the stacking step sequence with the fewest steps and a smooth process to meet the high-efficiency operation requirements of intelligent unmanned overhead cranes.
[0110] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0111] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for stacking steel plates in multiple positions using an intelligent unmanned overhead crane with a master-slave relationship, characterized in that, Includes the following steps: Acquire the stacking task data of the intelligent unmanned overhead crane. The stacking task data includes the initial disordered sequence of steel plates at the stacking positions and the number of temporary stacking positions. Based on a pre-trained neural network ensemble model, feature extraction and optimization processing are performed on the stacking task data of the intelligent unmanned overhead crane to generate a recommended steel plate code set for the intelligent unmanned overhead crane. Based on the preset rules for balancing temporary stack loads and combined with the recommended steel plate code set of the intelligent unmanned crane, an initial population is generated, including several recommended steel plate code sets. Based on the preset production decoding rule set, which includes the yielding rule, moving rule, and moving back rule based on the master-slave relationship constraint between steel plates, the production decoding rule set is used to decode each recommended steel plate code set in the initial population to obtain the stacking step sequence and the total number of stacking steps corresponding to each recommended steel plate code set in the initial population. The intelligent optimization algorithm iteratively optimizes the stacking step sequence and the total number of stacking steps corresponding to each recommended steel plate coding set in the initial population to generate the optimal stacking step sequence; the intelligent unmanned crane performs multi-stack steel plate stacking processing based on the optimal stacking step sequence.
2. The intelligent unmanned overhead crane multi-position steel plate stacking method oriented towards master-slave relationship as described in claim 1, characterized in that, The initial disordered sequence of steel plates in the stacking position specifically refers to several steel plates in the initial stacking position. The neural network ensemble model includes several heterogeneous perceptrons and a random forest model. The specific steps for generating the recommended steel plate encoding set for the intelligent unmanned crane are as follows: Input the stacking task data of the intelligent unmanned overhead crane into the neural network ensemble model; The neural network ensemble model extracts features from the initial disordered sequence of steel plates in the stacking position of the intelligent unmanned crane based on preset rules, and generates a stacking digital feature set of the intelligent unmanned crane. Each heterogeneous perceptron and random forest model in the neural network ensemble model performs independent attribution bit prediction processing based on the stacked digital feature set to generate several candidate initial steel plate coding sets; For each candidate initial steel plate code set, a fusion decision process is performed to extract and generate a recommended steel plate code set for the intelligent unmanned crane.
3. The intelligent unmanned overhead crane multi-position steel plate stacking method oriented towards master-slave relationship as described in claim 2, characterized in that, The specific steps for generating the stacking digital feature set of the intelligent unmanned overhead crane are as follows: For each steel plate in the initial disordered sequence of stacked steel plates of the intelligent unmanned crane, determine its sequential position and extract the positional features of the corresponding steel plate; Statistical processing is performed on the initial disordered sequence of steel plates in the stacking position of the intelligent unmanned crane to extract the minimum and maximum number pairs of the corresponding steel plates; Furthermore, locally ordered segments are extracted from the initial disordered sequence of steel plates at the stacking position of the intelligent unmanned crane, and the starting and ending data chains of the corresponding steel plates are extracted. Based on the positional characteristics of each steel plate, the minimum and maximum set of digital pairs, the starting and ending data chains, a stacking digital feature set for the intelligent unmanned overhead crane is constructed.
4. The intelligent unmanned overhead crane multi-position steel plate stacking method oriented towards master-slave relationship as described in claim 2, characterized in that, The specific steps for generating the initial population are as follows: Based on the recommended steel plate coding set of the intelligent unmanned crane, the steel plate allocation sequence is statistically analyzed; Based on the preset balanced temporary stack load rules, the steel plate allocation sequence is balanced and adjusted to generate an initial population.
5. A method for multi-position steel plate stacking using an intelligent unmanned overhead crane with a master-slave relationship as described in claim 1, characterized in that, The specific steps of the preset production decoding rule set are as follows: Obtain the stack-over sample dataset and perform analytical processing on the stack-over sample dataset based on a pre-established mixed integer linear programming model to generate the optimal sample solution set; Analyze the decision-making patterns constrained by the master-slave relationship between steel plates in the optimal sample solution set; The decision-making rules are then solidified into a set of production decoding rules that includes rules for giving way, moving away, and moving back.
6. A method for stacking steel plates in multiple positions using an intelligent unmanned overhead crane with a master-slave relationship, as described in claim 5, is characterized in that... The specific steps to obtain the stacking step sequence and total stacking steps corresponding to each recommended steel plate code set in the initial population are as follows: Based on the generative decoding rule set, each recommended steel plate encoding set in the initial population is decoded in two stages according to the preset two-stage decoding; Two-stage decoding includes: The first stage of decoding is used to process each recommended steel plate code set in the initial population based on the production decoding rule set until the preset clearing condition is met. The second stage of decoding is used to organize each recommended steel plate code set in an ordered manner based on the generative decoding rule set until the target sorting direction is met. Statistical processing was performed on each recommended steel plate code set in the initial population after two-stage decoding to obtain the stacking step sequence and total stacking steps corresponding to each recommended steel plate code set in the initial population.
7. A method for stacking steel plates in multiple positions using an intelligent unmanned overhead crane with a master-slave relationship, as described in claim 6, is characterized in that... The specific steps of the first stage of decoding are as follows: Based on each recommended steel plate code set in the initial population, determine whether the preset first condition is met; If the preset first condition is met, the yield rule is executed, and it is determined whether the preset second condition is met. If the preset second condition is not met, the move rule and the withdrawal rule are executed in sequence. If the preset second condition is met, the withdrawal rule is executed. If the first preset condition is not met, then determine whether the second preset condition is met. If the preset second condition is met, the withdrawal rule will be executed.
8. A method for stacking steel plates in multiple positions using an intelligent unmanned overhead crane with a master-slave relationship, as described in claim 6, is characterized in that... The specific steps for the second stage of decoding are as follows: For each recommended steel plate encoding set in the initial population after the first segment of decoding, determine whether the preset third condition is met. If the preset third condition is met, determine whether the preset fourth condition is met. If the preset fourth condition is met, execute the withdrawal rule and generate the corresponding stacking step sequence. If the preset fourth condition is not met, execute the removal rule and withdrawal rule in sequence and generate the corresponding stacking step sequence. If the preset third condition is not met, the preset first and second conditions are repeatedly evaluated until the corresponding stacking step sequence is generated.
9. A method for stacking steel plates in multiple positions using an intelligent unmanned overhead crane with a master-slave relationship, as described in claim 1, is characterized in that... The specific steps for generating the optimal stacking sequence are as follows: The initial population is initialized to generate the current population and the current optimal solution. Perform crossover and mutation operations, as well as neighborhood search, on the current population in sequence, and update the current optimal solution; Determine whether the algorithm termination condition is met; If satisfied, output the optimal stacking step sequence; If the requirements are not met, iterative optimization will continue.
10. A method for stacking steel plates in multiple positions using an intelligent unmanned overhead crane with a master-slave relationship, as described in claim 9, is characterized in that... The specific steps of the initialization process are as follows: Mark the initial population as the current population; Based on the total number of stacking steps corresponding to each recommended steel plate code set in the current population, the recommended steel plate code set with the smallest total stacking steps is counted and marked as the current optimal solution.