A method and system for assembling a pole-mounted transformer package, and a storage medium

Abstract

The application provides a kind of pole-mounted transformer complete equipment assembly method, system and storage medium, including obtaining equipment workpiece information and physical, electrical safety constraints, establishing the evaluation model containing assembly time, operation difficulty, safety risk weighted sum, generate the initial assembly sequence conforming to constraint as current solution;Through preset neighborhood structure generates neighborhood solution set, adopts tabu search iteration optimization to termination condition, outputs optimal sequence;Tabu search takes the inverse operation of moving operation as the tabu object, establishes global memory matrix to record the adjacent process pairing frequency in historical optimal solution;Tabu step is used: when target improvement, step is positively correlated with improvement amount, space constraint level;When target deteriorates, step is negatively correlated with process pairing historical frequency;If the new process pairing frequency corresponding to tabu operation exceeds threshold value and solution is better than historical optimal, then activate special pardon criterion.

Description

Technical Field

[0001] This application belongs to the field of equipment assembly, and in particular relates to an assembly method, system and storage medium for a complete set of pole-mounted transformer equipment. Background Technology

[0002] Pole-mounted transformer assemblies are complex in structure, typically comprising the transformer body, high and low voltage switchgear, protection devices, measuring instruments, and numerous connecting cables and other components. During actual assembly, constraints exist not only between these components due to physical interference, spatial installation sequence, and support forces, but also for electrical safety constraints such as clearances, insulation protection, and safe operating sequences. Any reversal, omission, or improper arrangement of any assembly step can lead to insulation failure, short circuits, or other potential electrical faults. Current assembly sequence planning for pole-mounted transformer assemblies relies heavily on the experience and judgment of process engineers, lacking systematic theoretical support and a scientific planning mechanism. This results in low planning efficiency and difficulty in adapting to rapid product structure iteration and diversified customized production needs. While heuristic optimization algorithms have been applied to assembly sequence planning for some industrial products, existing methods struggle to simultaneously consider multiple dimensions of comprehensive evaluation indicators such as assembly time, operational difficulty, and safety risks when facing the multiple constraints of large power equipment. Furthermore, the search mechanism lacks flexibility, easily leading to search stagnation or getting trapped in local optima. Meanwhile, existing optimization processes typically lack a continuous memory and in-depth mining of historically successful assembly features, making it difficult to dynamically adjust exploration strategies based on the search status. This results in generated assembly sequences that fail to achieve optimal overall efficiency in actual production lines. Therefore, there is an urgent need to propose an intelligent assembly planning method that can integrate multiple constraints, dynamically adjust the search process, and fully utilize historically successful assembly features to achieve rapid generation of high-quality assembly sequences. Summary of the Invention

[0003] This invention proposes an assembly method for a complete pole-mounted transformer assembly, addressing the problem of the lack of a more intelligent and flexible assembly planning method capable of overcoming multiple constraint bottlenecks in existing technologies. The method includes: Acquire workpiece information for the equipment, as well as the physical and electrical safety constraints between workpieces; Based on the physical constraints and electrical safety constraints, an evaluation model is established with the weighted sum of total assembly time, operational difficulty, and safety risk as the objective function. An initial assembly sequence that conforms to the physical constraints and electrical safety constraints is generated as the current solution. Based on a preset neighborhood structure, the assembly process of the current solution is operated to generate a neighborhood solution set. The tabu search mechanism is used to iteratively search the neighborhood solution set until the termination condition is met, and the optimal assembly sequence is output. The tabu search mechanism sets the inverse operation of the move operation that causes the current solution to change as a tabu object, and establishes a global memory matrix to record the frequency of adjacent process pairings in the historical optimal solution. A taboo step size is set for the taboo object. When the objective function of the neighborhood solution improves relative to the current solution, the taboo step size is jointly determined by the baseline step size, the gain step size which is positively correlated with the absolute value of the improvement of the objective function, and the constraint step size which is positively correlated with the spatial constraint level of the workpiece involved. When the objective function deteriorates, the guiding factor is determined based on the historical frequency of the new process pairing in the global memory matrix. The taboo step size is determined by the baseline step size, the weight based on the workpiece attributes, and the guiding factor, and is negatively correlated with the historical frequency. When the movement operation of the neighborhood solution is in a taboo state, the amnesty criterion is activated. If the historical frequency of the new process pairing is higher than a preset threshold and the neighborhood solution is better than the historical best solution, the taboo state is amnesty.

[0004] Furthermore, the present invention also relates to an assembly system for a pole-mounted transformer assembly, comprising: The acquisition module is used to acquire workpiece information of the equipment and the physical and electrical safety constraints between workpieces; The generation module is used to establish an evaluation model based on the physical constraints and electrical safety constraints, with the weighted sum of total assembly time, operation difficulty and safety risk as the objective function, and generate an initial assembly sequence that meets the physical constraints and electrical safety constraints as the current solution; and to operate on the assembly process of the current solution based on a preset neighborhood structure to generate a neighborhood solution set. The recording module is used to perform iterative search in the neighborhood solution set using the tabu search mechanism until the termination condition is met, and output the optimal assembly sequence. The tabu search mechanism sets the inverse operation of the move operation that causes the current solution to change as a tabu object, establishes a global memory matrix, and records the frequency of adjacent process pairings in the historical optimal solution. The determination module is used to set a taboo step size for the taboo object. When the objective function of the neighborhood solution improves relative to the current solution, the taboo step size is jointly determined by the reference step size, the gain step size which is positively correlated with the absolute value of the improvement of the objective function, and the constraint step size which is positively correlated with the spatial constraint level of the workpiece involved. When the objective function deteriorates, the guiding factor is determined according to the historical frequency of the new process pairing in the global memory matrix. The taboo step size is determined by the reference step size, the weight based on the workpiece attribute, and the guiding factor and is negatively correlated with the historical frequency. When the movement operation of the neighborhood solution is in a taboo state, the amnesty criterion is activated. If the historical frequency of the new process pairing is higher than a preset threshold and the neighborhood solution is better than the historical best solution, the taboo state is amnesty.

[0005] This invention utilizes the physical and electrical safety constraints of the workpiece to establish an evaluation model centered on the weighted sum of total assembly time, operational difficulty, and safety risks. By employing an improved tabu search mechanism and establishing a global memory matrix, it explores and records the frequency patterns of adjacent process pairings in historical optimal solutions, using best practices to guide the search direction and improve the algorithm's optimization efficiency. A multi-dimensional adjustment mechanism is set for the tabu step size. When improving a solution, the step size is set according to the target change and spatial constraint level; when a solution deteriorates, a historical frequency guiding factor is used to intervene, expanding the global search scope. Combined with a pardon criterion based on historical frequency and solution quality, high-potential excellent solutions with false tabus are retained, preventing the omission of potential optimal assembly schemes, thereby ensuring that the output assembly sequence has high global optimality and engineering practical value. Attached Figure Description

[0006] Figure 1 A flowchart of the assembly method for a complete set of pole-mounted transformer equipment; Figure 2 A schematic diagram of the physical and electrical constraint matrix for assembling a complete pole-mounted transformer system. Figure 3 A comparative diagram showing the convergence curves of the objective function for planning the assembly sequence of the pole-mounted transformer. Detailed Implementation

[0007] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0008] Example 1 In Embodiment 1 of the present invention, as Figure 1 As shown, an assembly method for a pole-mounted transformer assembly includes: S1, obtain the workpiece information of the equipment and the physical and electrical safety constraints between the workpieces.

[0009] Using the OpenCASCADE open-source geometry kernel library, the STEP format 3D computer-aided design model file of the pole-mounted transformer assembly is parsed to extract the geometric topology information of all assembled parts. For workpiece mass and center of gravity coordinates, if the STEP format model file already contains material density, mass attribute, or center of gravity attribute fields, the corresponding attributes are directly read. If the STEP format model file does not contain mass attributes, the mass calibration value, material density, or material type information is read from the bill of materials, product lifecycle management database, or manufacturing execution system database based on the workpiece number. When a mass calibration value exists, it is directly used as the workpiece mass, and the corresponding center of gravity calibration value is read from the bill of materials, product lifecycle management database, or manufacturing execution system database. If no center of gravity calibration value exists, the centroid of the geometric model is used as an approximation of the center of gravity coordinates. When no mass calibration value exists but material density exists, the workpiece mass is calculated based on the solid volume and material density obtained from the geometric model, and the center of gravity coordinates are calculated based on the centroid or mass attribute of the geometric model.

[0010] Based on the extracted 3D geometric information, the GJK collision detection algorithm is used to determine collisions or calculate minimum distances between workpieces on the assembly path, constructing a Boolean interference matrix. When the GJK collision detection result indicates that two workpieces collide, the geometric Boolean intersection algorithm or the bounding box overlap volume estimation method is called to calculate the corresponding interference volume. The interference volume is used to characterize the spatial constraint level and is not used as the direct output of the GJK algorithm.

[0011] By parsing the bill of materials and process documents stored in comma-separated value format, the rated voltage level and insulation medium parameters can be extracted.

[0012] A constraint graph data structure is established using the NetworkX graph computation library. Voltage levels are mapped to minimum air insulation gaps according to power industry standards, generating a numerical electrical safety constraint distance matrix. In this embodiment, assembly sequence optimization uses assembly processes as the sorting object. Workpieces represent the physical components being installed, and processes represent installation, connection, fixing, inspection, or protection operations performed on one or more workpieces; each process is associated with at least one workpiece through a process-workpiece mapping table. Therefore, subsequent topology sorting, neighborhood operations, global memory matrices, and taboo objects are all indexed by process numbers, while parameters such as the spatial constraint level, mass, and volume of workpieces are read from the process-workpiece mapping table.

[0013] In some implementations, obtaining the workpiece information of the device and the physical and electrical safety constraints between workpieces includes: The weight, dimensions, and interface type number of each workpiece can be extracted by scanning the QR code on the device. Extract the physical constraint relationship model between the workpieces, which includes the three-dimensional bounding box collision detection matrix and the support force relationship matrix of each workpiece; Extract the electrical safety constraint relationship model between workpieces, the electrical safety constraint relationship model includes the electrical clearance setting values ​​between the high voltage side and the low voltage side and the installation sequence of the insulating baffle; Based on the extracted workpiece information, the physical constraint relationship model, and the electrical safety constraint relationship model, an association graph data structure is constructed.

[0014] When acquiring workpiece information and the physical and electrical safety constraints between workpieces, an industrial-grade barcode scanner is used to optically scan the QR codes on each component of the pole-mounted transformer assembly to extract the specific parameters of the workpiece. For example, the weight of the transformer body is extracted as 350 kg, its dimensions as 800 mm long × 600 mm wide × 1000 mm high, and its interface type number as T-01.

[0015] The constraint relationship model is extracted from the database of the background manufacturing execution system. This model is represented by a predefined three-dimensional bounding box collision detection matrix. The three-dimensional bounding box collision detection matrix is ​​a Boolean matrix of size N×N, where an element value of 1 indicates that there is spatial interference between two workpieces in the current assembly pose, and simultaneous operation should be avoided. At the same time, the support force relationship matrix is ​​extracted, such as a force directed graph, indicating that the high-voltage surge arrester can only be installed under load after the transformer crossarm is completely fixed and the torque reaches 45 N·m.

[0016] An electrical safety constraint model was invoked, which explicitly stipulates that the phase-to-phase and phase-to-ground electrical clearances on the 10kV high-voltage side must be greater than or equal to 125mm, and the 0.4kV electrical clearance on the low-voltage side must be greater than or equal to 20mm. The installation sequence of insulation baffles was also specified. Based on the above multidimensional data, a correlation graph data structure was constructed.

[0017] In this diagram, the nodes of the workpiece constraint diagram represent physical workpieces such as the transformer body, surge arrester, crossarm, and fuse, along with their attributes. Edges are used to record spatial interference, support forces, electrical clearances, or the timing relationships of insulating baffles between workpieces. The nodes of the process constraint diagram represent specific assembly processes, and edges represent the sequential constraints between processes. The edges in the process constraint diagram are obtained by combining the workpiece constraint diagram with a process-workpiece mapping table. The weights and labels on the edges store the corresponding physical constraint type, electrical constraint threshold, and constraint source, respectively. This graph model structure highly structures scattered engineering parameters into a low-level framework that can be traversed and calculated by algorithms, ensuring the generated assembly sequence is safe and feasible at both physical and electrical levels. A schematic diagram of the physical and electrical constraint matrix for the assembly of a pole-mounted transformer assembly is shown below. Figure 2 As shown.

[0018] S2. Based on the physical constraints and electrical safety constraints, establish an evaluation model with the weighted sum of total assembly time, operational difficulty, and safety risk as the objective function, and generate an initial assembly sequence that conforms to the physical constraints and electrical safety constraints as the current solution.

[0019] In the Python environment, the evaluation model matrix equation is established using the NumPy numerical computing library. The objective function is set as the sum of the dot product of the total assembly time, the operation difficulty index, and the safety risk coefficient.

[0020] The weights of each indicator are calculated using the Analytic Hierarchy Process (AHP). In this embodiment, the weights of total assembly time, operational difficulty, and safety risk are determined to be 0.5, 0.3, and 0.2, respectively, using AHP and expert calibration results. In other embodiments, these weights can be recalibrated according to equipment model, assembly process requirements, or safety level. Before constructing the initial assembly sequence, the interference matrix and electrical distance matrix are not directly used as adjacency matrices for topological sorting. Instead, physical constraints and electrical safety constraints are converted into directed sequential constraints between processes based on the support force relationship, insulation baffle installation sequence, electrical clearance release conditions, assembly accessibility conditions, and the process-workpiece mapping table. For matrix elements that only indicate symmetrical interference or insufficient distance between two workpieces, if the corresponding process has engineering rules such as support, obstruction, installation before avoidance, or installation of insulating parts before installation of live parts, the direction of the directed edge is determined according to these engineering rules. If the direction cannot be determined, it is treated as a constraint condition that cannot be operated simultaneously or requires manual confirmation and is not directly involved in topological sorting.

[0021] After the directed edges are generated, the directed sequential constraint edges are deduplicated and loop detected. If a loop is detected, electrical safety constraints and support force constraints are retained according to preset priorities, low-priority process preference edges are automatically removed or downgraded, and loop detection is performed again until a directed acyclic graph of assembly constraints is formed. If a loop still exists after retaining only electrical safety constraints and support force constraints, it is determined that the current constraint set cannot generate a feasible assembly sequence, and a constraint conflict prompt is output.

[0022] By utilizing a random perturbation mechanism and the built-in random module in Python, the positions of processes that do not have constraints are randomly swapped in the topology sorting results. After physical and electrical safety constraints are verified, multiple feasible sequences are generated, and the objective function values ​​of each sequence are calculated.

[0023] Select the sequence with the minimum objective function value and initialize it as the current solution and the historical best solution.

[0024] In some implementations, the evaluation model established based on the physical constraints and electrical safety constraints, with the weighted sum of total assembly time, operational difficulty, and safety risks as the objective function, includes: The total assembly time is calculated by adding the standardized operation time of each process to the movement time corresponding to the position movement distance involved in the assembly process. Obtain the required spatial line-of-sight obstruction rate for assembly and multiply it by the square root of the component mass value to calculate the operational difficulty score. When two adjacent assembly processes have hazardous source parameters marked for insulation medium puncture, the assembly sequence is discarded as an invalid solution that does not comply with electrical safety constraints. If there are no hazard source parameter markings, the basic safety risk score will be used as the safety risk indicator score. After dimensionless processing of the total assembly time, operation difficulty index score, and safety risk index score, the dimensionless total assembly time is multiplied by 0.5, the operation difficulty by 0.3, and the safety risk by 0.2, and the three are summed to generate the evaluation model.

[0025] In establishing the multi-objective weighted evaluation model, time statistics are performed on all processes in the target assembly sequence. For any process, its standardized operation time is read from the standard time library. For example, the time for crimping transformer leads is 15 minutes. The movement time is obtained by dividing the distance between the corresponding operation positions of two adjacent processes in the current assembly sequence by the reference moving speed, which is set to 0.5 m / s. The time components of all processes are added together to obtain the total assembly time of the entire sequence.

[0026] When assessing operational difficulty, the spatial line-of-sight occlusion rate at the current workstation is calculated using a ray tracing algorithm in the virtual assembly engine. The value ranges from [0,1], for example, 0.6 represents 60% line-of-sight occlusion. Simultaneously, the mass value of the components is acquired; for example, a 25kg surge arrester. The operational difficulty score for a single process or component is calculated by multiplying the occlusion rate by the square root of the component's mass value. For example, for the aforementioned surge arrester, the score would be 3.0. For a complete assembly sequence, the operational difficulty score for each process is calculated sequentially, and the total operational difficulty score is obtained by summing the results. When balancing different sequence lengths, the total operational difficulty score can be divided by the number of processes to obtain the average operational difficulty. Before being included in the operational difficulty calculation, the component mass value can be normalized based on the baseline or maximum mass in the current assembly task to ensure a uniform dimension for the operational difficulty score.

[0027] In terms of safety risk, electrical safety constraint checks are performed. For each current process in the assembly sequence, the set of workpieces that have been assembled before the current process is executed is read, and it is determined whether there are hazard source parameter markers between the current process and the set of assembled workpieces, such as insulation medium puncture, insufficient insulation gap, or missing installation sequence of insulation protection components. If such hazard source parameter markers exist, the assembly sequence is determined to be an invalid solution and is removed from the algorithm. If the invalid solution removal condition is not triggered, the safety risk index score is calculated based on the minimum electrical clearance margin corresponding to each process in the current assembly sequence, the exposure duration during the period when the insulation baffle is not installed, the number of operations near live parts, and the degree of adjacency of high and low voltage connection processes.

[0028] A maximum-minimum normalization method is used to perform dimensionless mapping on the total assembly time, the total score of sequence operation difficulty, and the score of safety risk index. The maximum and minimum values ​​are determined based on the initial feasible sequence set, the preset boundary of the process database, or the calibration boundary of historical samples, and remain unchanged in the same assembly sequence optimization task. During normalization calculation, a preset small positive number is added to the denominator to avoid division by zero when the maximum and minimum values ​​are equal. According to the preset expert weights, the dimensionless total assembly time is multiplied by 0.5, the operation difficulty by 0.3, and the safety risk by 0.2. The sum of the three products is calculated to obtain the comprehensive evaluation objective function value, thereby pursuing assembly efficiency while taking into account human-machine ergonomics and electrical safety.

[0029] S3, based on the preset neighborhood structure, performs the assembly process of the current solution to generate a neighborhood solution set.

[0030] Three neighborhood generation operators are defined, including a two-point exchange operator, a single-point insertion operator, and a subsequence inversion operator based on the 2opt algorithm.

[0031] For the current solution sequence, generate all valid process position index pairs in the current assembly sequence using the Itertools standard library, and then apply the above three operators in sequence to generate alternative sequences.

[0032] After each operation, the preset constraint verification module is invoked to verify whether the relevant processes in the new sequence cause workpiece collision, unmet support conditions, or unreachable assembly paths based on the interference matrix, support force relationship matrix, and assembly path information. At the same time, the module verifies whether the new sequence violates electrical safety clearance or insulation protection sequence based on the electrical safety constraint distance matrix and insulation baffle installation sequence, filtering out all illegal sequences that violate constraints.

[0033] The remaining valid assembly sequences that have passed verification are summarized to obtain the neighborhood solution set for the current iteration.

[0034] S4. Using the tabu search mechanism, iterative search is performed in the neighborhood solution set until the termination condition is met, and the optimal assembly sequence is output. The tabu search mechanism sets the inverse operation of the move operation that causes the current solution to change as a tabu object, establishes a global memory matrix, and records the frequency of adjacent process pairings in the historical optimal solution.

[0035] Declare a two-dimensional NumPy integer array in memory as a global memory matrix. The rows and columns of the matrix correspond to all assembly process numbers of the pole-mounted transformer. Initialize all elements of the matrix to zero.

[0036] When updating the historical optimal solution in each iteration, all adjacent process pairs in the sequence are extracted, and the count value at the intersection of the corresponding row and column in the memory matrix is ​​incremented by 1.

[0037] For move operations that cause sequence changes, the process number and action type of the participating operation are extracted. The completely reversed process is instantiated as a taboo object and pushed into a taboo list implemented using the deque double-ended queue data structure in Python's built-in collections module. Each taboo object corresponds to a taboo record, which includes the move operation type, participating process number, position before move, position after move, reverse operation identifier, and remaining taboo lifetime. After each iteration, the remaining taboo lifetime of each taboo record in the taboo list is decremented by one, and taboo records with a remaining taboo lifetime less than or equal to zero are deleted. Before a candidate move action enters the solution pool for selection, it is first matched with the reverse operation identifier and participating process number in the taboo list. If a match is successful and the remaining taboo lifetime is greater than zero, the candidate move action is determined to be in a taboo state. If the taboo step size calculation requires workpiece attributes, the process-workpiece mapping table is queried based on the participating process number to read the corresponding workpiece's mass, volume, and spatial constraint parameters.

[0038] In some implementations, establishing a global memory matrix to record the frequency of adjacent process pairings in historical optimal solutions includes: A two-dimensional integer matrix is ​​set as the global memory matrix, with the row index and column index corresponding to all assembly process numbers, respectively. Set all initial values ​​in the two-dimensional integer matrix to 0; After each optimization process, determine whether the content value of the saved current iteration history optimal solution has been updated; When it is determined that the content value of the historical optimal solution has been updated, the coordinate pairing of the two adjacent process numbers in the historical optimal solution sequence is traversed, and the value of the cell where the coordinate pair is located in the two-dimensional integer matrix is ​​added by 1.

[0039] Construct a two-dimensional integer matrix of dimension N×N to serve as the global memory matrix, where N represents the total number of assembly steps in the assembly task of the pole-mounted transformer assembly equipment, for example, including crossarm installation, fuse installation, etc., totaling N=30 steps. The row index i and column index j of the matrix map to the corresponding step numbers.

[0040] At the start of the algorithm, an initialization operation is performed, setting all element cell values ​​in the two-dimensional integer matrix to 0 to establish an initial frequency record. At the end of each single-step optimization or neighborhood exploration loop of the tabu search algorithm, the objective function value is compared to determine if a better solution has replaced the previously stored historical optimal solution. Once an update to the historical optimal solution is confirmed, the new solution sequence, presented as an ordered array vector, is extracted, and a sequential scan from beginning to end begins.

[0041] The process involves extracting two adjacent process numbers from the sequence to form ordered coordinate pairs, locating the corresponding cell in the global memory matrix, and incrementing its current value by 1. Through this update counting mechanism, the memory matrix can detect and represent process features that improve assembly efficiency.

[0042] S5, set a taboo step size for the taboo object. When the objective function of the neighborhood solution improves relative to the current solution, the taboo step size is jointly determined by the reference step size, the gain step size which is positively correlated with the absolute value of the improvement of the objective function, and the constraint step size which is positively correlated with the spatial constraint level of the workpiece involved. When the objective function deteriorates, the guiding factor is determined according to the historical frequency of the new process pairing in the global memory matrix. The taboo step size is determined by the reference step size, the weight based on the workpiece attribute, and the guiding factor and is negatively correlated with the historical frequency. When the movement operation of the neighborhood solution is in a taboo state, the amnesty criterion is activated. If the historical frequency of the new process pairing is higher than the preset threshold and the neighborhood solution is better than the historical best solution, the taboo state is amnesty.

[0043] Using the Math math library, the difference between the neighborhood solution and the objective function value of the current solution is calculated as the improvement factor. , The objective function value is obtained for the neighborhood. ΔF represents the objective function value of the current solution. When ΔF is less than zero, it indicates that the objective function value of the neighboring solution decreases, i.e., the objective function improves. When ΔF is greater than zero, it indicates that the objective function value of the neighboring solution increases, i.e., the objective function deteriorates.

[0044] When ΔF is less than zero, i.e. the objective function is improved, the preset amplification factor is extracted and multiplied by the absolute value of the improvement to calculate the gain step size; the spatial analysis function is called to count the total number of mating constraint surfaces or constraint levels of the parts involved in the process of moving operation in the spatial assembly relationship matrix, and the constraint step size is calculated; it is added to the preset basic constant step size, and the calculated taboo step size is rounded down to be used as the survival iteration number of the object.

[0045] When ΔF is greater than zero, indicating a deterioration of the objective function, the global memory matrix is ​​queried to obtain the historical count values ​​of the newly formed adjacent process pairings after the move. A guiding factor inversely proportional to the historical count value is calculated according to a preset reciprocal frequency decay rule. The taboo step size is calculated by combining the baseline step size and the workpiece mass and volume attribute weights. Specifically, the weights based on workpiece attributes are determined by the normalized mass and normalized volume of the workpieces corresponding to the processes involved in the move operation. This is achieved by first reading the mass and volume values ​​of the relevant workpieces, normalizing them using the maximum mass and maximum volume of all workpieces in the current assembly task, and then averaging the normalized mass mean and normalized volume mean to obtain the attribute composite value. This composite value is then multiplied by a preset amplification factor, rounded up, and limited to an integer range of 1 to 3 to obtain the weights based on workpiece attributes.

[0046] Simultaneously, a pardon condition checker is configured: when a candidate operation matches a forbidden object in the double-ended queue, it compares whether the historical frequency of the current new process pairing in the memory matrix is ​​greater than a pre-defined empirical statistical threshold, and calls the evaluation model to assess whether the neighborhood solution is less than the target value of the historical best solution stored in the global variable; if both conditions are met, the forbidden object is removed from the double-ended queue, and the corresponding assembly sequence movement operation is executed. Wherein, if the neighborhood solution generated by the forbidden candidate operation is already better than the historical best solution, the pardon criterion can be triggered independently; the condition that the historical frequency of the new process pairing is higher than the preset threshold is used to further confirm that the process connection relationship corresponding to the forbidden operation has historical stability and engineering reusability value. Figure 3 As shown, the iterative optimization effects of the three algorithms are compared.

[0047] The search stops when the iteration counter reaches the set maximum number of iterations; the globally optimal assembly sequence is exported as a standard industrial data table using the Pandas library.

[0048] In some implementations, the reference step size includes: Read the current iteration count and determine the current search stage based on the ratio of the current iteration count to the preset total maximum iteration count; During the global divergence exploration phase in the first half of the iteration cycle, the base constant value of the baseline step size is set to 7. In the local depth mining phase during the second half of the iteration cycle, the basic constant value of the baseline step size is reduced to 4. Obtain the type feature identifier corresponding to the pre-stored operation action in the tabu search mechanism, query the offset value corresponding to the type feature identifier in the preset operation action attribute table, and add the queried offset value to the current base constant value to obtain the baseline step size.

[0049] When the algorithm is running, the program controller reads the current iteration count variable, such as the 250th iteration, and extracts the preset total maximum iteration count constant, which is set to 500. The current search phase is defined by calculating the ratio between the two, such as 0.5.

[0050] When the ratio is less than or equal to 0.5, it is determined that the algorithm is in the "global divergence exploration phase" of the first half of the iteration cycle, and the basic constant value of the baseline step size is set to a larger value of 7. When the ratio is greater than 0.5, it is determined that the algorithm has entered the "local depth mining phase" of the second half of the cycle. At this time, it is necessary to enhance the search capability near the current solution, so the basic constant value is reduced to the value of 4.

[0051] After establishing the basic constants for different stages, extract the specific operation that triggers the change in solution structure in the tabu mechanism, such as the swap operation or the reversal operation, and read the type feature identifier corresponding to the operation, for example, the identifier for a two-point swap is ID_01.

[0052] The key-value matching mechanism retrieves the offset value mapped to the identifier from a preset operation action attribute table. For example, if a two-point swap has a large range of variation, its offset value is set to +2. The calculation module adds the retrieved offset value to the base constant value determined for the current stage. For example, in the exploration stage, the base constant is 7, and a two-point swap operation is used, so the offset value is +2, generating a baseline step size of 9. This calculation method ensures reasonable control over the duration of taboos in different regions of the search space.

[0053] In some implementations, when the neighboring solution improves the objective function relative to the current solution, the tabu step size is jointly determined by the baseline step size, the gain step size positively correlated with the absolute value of the improvement in the objective function, and the constraint step size positively correlated with the spatial constraint level of the workpiece, including: Obtain a defined baseline step size; Obtain the absolute value of the improvement of the objective function and multiply it by a preset amplification factor. Use the method of rounding down to obtain the integer improvement step size, and set the result of adding 1 to the integer improvement step size as the gain step size. The total number of mating constraint surfaces in the spatial assembly relationship matrix of the parts involved in the two processes related to the movement operation is counted, and the total number of mating constraint surfaces is set as the constraint step size. Add the base step size, the gain step size, and the constraint step size to the base step size, and set the result of the cumulative calculation as the specific value of the taboo step size.

[0054] In the scenario where a better solution is found through neighborhood search, triggering positive feedback, a multi-dimensional fusion calculation with a tabu step size is performed. The baseline step size parameter determined in the previous process is called and set to the value 9.

[0055] Calculate the absolute value of the objective function improvement between the new neighborhood solution and the current solution. Assuming the absolute value is 0.035, multiply it by a preset amplification factor, such as 100 on the order of fitness, i.e. 3.5. Use a floor function to obtain the integer improvement step size, i.e. 3, and add a constant value of 1 to ensure the gain limit. The result is a gain step size of 4, which allows for a longer protection period to be applied to the move operation with a larger improvement.

[0056] Meanwhile, for the two processes that caused the sequence change, the 3D model data of the involved components, such as the transformer body and the crossarm fasteners, were analyzed. The spatial assembly relationship matrix was traversed to count the total number of mating constraint surfaces where the two come into contact. For example, this includes the support bottom surface, the lateral positioning surface, and the two bolt hole cylindrical surfaces, totaling 4 constraint surfaces. This total number of 4 was assigned to the constraint step size, indicating the degree of correlation in the workpiece installation.

[0057] The central computing unit performs algebraic accumulation, summing the obtained baseline step size 9, gain step size 4, and constraint step size 4 to obtain a taboo step size of 17 iterations. This indicates that the operation with high assembly quality improvement and strong correlation will be prohibited from being reversed in the next 17 iterations.

[0058] In some implementations, determining the guiding factor based on the historical frequency of new process pairings in the global memory matrix when the objective function deteriorates includes: Read the historical cumulative total number of newly generated process pairs stored in the corresponding coordinate cells of the global memory matrix; Add a constant 1 to the total number of historical occurrences to obtain the base multiplier; Divide the base constant 8 by the base multiplier and round down to obtain the integer result of the guiding parameter; The integer result of the guiding parameter is added to a constant 2 to form a guiding factor, and the guiding factor is used in the calculation of the taboo step size.

[0059] If, during the iteration process, the objective function value of the new neighborhood solution generated by a certain move operation deteriorates compared to the current solution, an anti-stagnation mechanism is immediately activated. All adjacent process pairs newly added due to the ranking change caused by this move are located, forming a new pair set. For example, a single-point insertion operation may simultaneously change multiple adjacent pairs before and after the insertion position, and a 2opt subsequence reversal operation may simultaneously change adjacent pairs within the subsequence and the boundary pairs at both ends of the subsequence. The algorithm accesses the coordinate grid corresponding to each new pair in the global memory matrix, reads the historical cumulative occurrence count of each pair in previous best solutions, and performs averaging, weighted averaging, or minimum value processing on the multiple historical cumulative occurrence counts to obtain the comprehensive historical frequency of the move operation. Then, a constant 1 is added to the comprehensive historical frequency to obtain the base multiplier.

[0060] The algorithm engine utilizes a preset baseline constant of 8. Dividing this constant by the baseline multiplier of 4 yields a quotient of 2.0. This quotient is then rounded down to obtain an integer result of 2 for the guiding parameter. Adding this integer result to the constant 2 generates a guiding factor of 4. This guiding factor is used as a multiplication coefficient in the calculation of the taboo step size under the current deterioration state. Further, the normalized mass and normalized volume of the workpieces associated with the grounding connection and testing processes are read, and a weight based on workpiece attributes is calculated. For example, if the weight is 2, the actual taboo step size equals the product of the baseline step size, the weight based on workpiece attributes, and the guiding factor. That is, the initial value of the actual taboo step size = baseline step size × weight based on workpiece attributes × guiding factor. Subsequently, the initial value of the actual taboo step size is limited between a preset minimum taboo step size and a preset maximum taboo step size to obtain the final taboo lifecycle. Specifically: if the initial value of the actual taboo step size is less than the preset minimum taboo step size, then the preset minimum taboo step size is used; if the initial value of the actual taboo step size is greater than the preset maximum taboo step size, then the preset maximum taboo step size is used.

[0061] This mechanism embodies heuristic guidance: if the process pairings that lead to solution deterioration occur frequently in history, the guidance factor converges to approximately 2, maintaining a low penalty; if they are unprecedented, the guidance factor will reach a maximum of 10, imposing a high taboo penalty on the exploration path. Since the guidance factor decreases with increasing historical frequency, the taboo step size when the objective function deteriorates is generally negatively correlated with historical frequency.

[0062] In some implementations, the amnesty criterion is activated when the neighboring solution's move operation is in a taboo state. If the historical frequency of new process pairing is higher than a preset threshold and the neighboring solution is better than the historical best solution, then the taboo state is amended, including: In each round of the cyclic solution pool selection, check whether the taboo lifecycle tag value of the candidate movement action in the attribute record entry is greater than 0; When the taboo lifecycle tag value is greater than 0, an inspection command is initiated to read the preceding node number and following node number of the new connection pairing formed by the action information; By querying the corresponding rows and columns of the global memory matrix, the corresponding historical frequency integers are extracted, and the difference is calculated by subtracting the set static threshold constant from the historical frequency integers. When the result is found that the difference is greater than 0 and the neighborhood solution is better than the historical best solution, the existing prohibition instruction of the operation is ignored, and the corresponding taboo lifecycle label value is modified and assigned to 0 to remove the taboo state.

[0063] During the algorithm's cyclic solution pool selection, the control program performs a taboo state scan on all current candidate movement actions. For any candidate action, it retrieves the taboo lifetime tag value residing in memory. If the tag value is determined to be greater than 0, for example, displayed as 5, it indicates that there are still 5 taboo freeze periods remaining, triggering a check instruction for the amnesty criterion. At this time, it extracts the new connection pair generated by the taboo action, for example, if process node 5 is placed before node 9, it reads the preceding node number 5 and the following node number 9 respectively.

[0064] Using this set of numbers as row and column indices, a query is initiated into the global memory matrix to extract the corresponding historical frequency integer number accumulated for this process combination, assumed to be 15 times. A pre-configured static threshold constant is read, setting the medium-to-high frequency standard to 10. The difference between the historical frequency number 15 and the static threshold constant 10 is calculated to be +5. At this point, system verification reveals that this difference is greater than 0, and synchronous calculation shows that the objective function value of the neighborhood solution derived from this prohibited action is better than the currently recorded global historical optimal solution.

[0065] When both conditions are met, the amnesty mechanism intervenes. The system ignores the existing prohibition instructions for the action, accepts the move, and modifies the corresponding taboo lifecycle tag value to 0, thereby unlocking the state and ensuring that potential assembly operations are not blocked.

[0066] Example 2 Embodiment 2 of the present invention proposes an assembly system for a complete pole-mounted transformer assembly, comprising the following modules: The acquisition module is used to acquire workpiece information of the equipment and the physical and electrical safety constraints between workpieces; The generation module is used to establish an evaluation model based on the physical constraints and electrical safety constraints, with the weighted sum of total assembly time, operation difficulty and safety risk as the objective function, and generate an initial assembly sequence that meets the physical constraints and electrical safety constraints as the current solution; and to operate on the assembly process of the current solution based on a preset neighborhood structure to generate a neighborhood solution set. The recording module is used to perform iterative search in the neighborhood solution set using the tabu search mechanism until the termination condition is met, and output the optimal assembly sequence. The tabu search mechanism sets the inverse operation of the move operation that causes the current solution to change as a tabu object, establishes a global memory matrix, and records the frequency of adjacent process pairings in the historical optimal solution. The determination module is used to set a taboo step size for the taboo object. When the objective function of the neighborhood solution improves relative to the current solution, the taboo step size is jointly determined by the reference step size, the gain step size which is positively correlated with the absolute value of the improvement of the objective function, and the constraint step size which is positively correlated with the spatial constraint level of the workpiece involved. When the objective function deteriorates, the guiding factor is determined according to the historical frequency of the new process pairing in the global memory matrix. The taboo step size is determined by the reference step size, the weight based on the workpiece attribute, and the guiding factor and is negatively correlated with the historical frequency. When the movement operation of the neighborhood solution is in a taboo state, the amnesty criterion is activated. If the historical frequency of the new process pairing is higher than a preset threshold and the neighborhood solution is better than the historical best solution, the taboo state is amnesty.

[0067] In some implementations, obtaining the workpiece information of the device and the physical and electrical safety constraints between workpieces includes: The weight, dimensions, and interface type number of each workpiece can be extracted by scanning the QR code on the device. Extract the physical constraint relationship model between the workpieces, which includes the three-dimensional bounding box collision detection matrix and the support force relationship matrix of each workpiece; Extract the electrical safety constraint relationship model between workpieces, the electrical safety constraint relationship model includes the electrical clearance setting values ​​between the high voltage side and the low voltage side and the installation sequence of the insulating baffle; Based on the extracted workpiece information, the physical constraint relationship model, and the electrical safety constraint relationship model, an association graph data structure is constructed.

[0068] In some implementations, the evaluation model established based on the physical constraints and electrical safety constraints, with the weighted sum of total assembly time, operational difficulty, and safety risks as the objective function, includes: The total assembly time is calculated by adding the standardized operation time of each process to the movement time corresponding to the position movement distance involved in the assembly process. Obtain the required spatial line-of-sight obstruction rate for assembly and multiply it by the square root of the component mass value to calculate the operational difficulty score. When two adjacent assembly processes have hazardous source parameters marked for insulation medium puncture, the assembly sequence is discarded as an invalid solution that does not comply with electrical safety constraints. If there are no hazard source parameter markings, the basic safety risk score will be used as the safety risk indicator score. After dimensionless processing of the total assembly time, operation difficulty index score, and safety risk index score, the dimensionless total assembly time is multiplied by 0.5, the operation difficulty by 0.3, and the safety risk by 0.2, and the three are summed to generate the evaluation model.

[0069] In some implementations, establishing a global memory matrix to record the frequency of adjacent process pairings in historical optimal solutions includes: A two-dimensional integer matrix is ​​set as the global memory matrix, with the row index and column index corresponding to all assembly process numbers, respectively. Set all initial values ​​in the two-dimensional integer matrix to 0; After each optimization process, determine whether the content value of the saved current iteration history optimal solution has been updated; When it is determined that the content value of the historical optimal solution has been updated, the coordinate pairing of the two adjacent process numbers in the historical optimal solution sequence is traversed, and the value of the cell where the coordinate pair is located in the two-dimensional integer matrix is ​​added by 1.

[0070] In some implementations, the reference step size includes: Read the current iteration count and determine the current search stage based on the ratio of the current iteration count to the preset total maximum iteration count; During the global divergence exploration phase in the first half of the iteration cycle, the base constant value of the baseline step size is set to 7. In the local depth mining phase during the second half of the iteration cycle, the basic constant value of the baseline step size is reduced to 4. Obtain the type feature identifier corresponding to the pre-stored operation action in the tabu search mechanism, query the offset value corresponding to the type feature identifier in the preset operation action attribute table, and add the queried offset value to the current base constant value to obtain the baseline step size.

[0071] In some implementations, when the neighboring solution improves the objective function relative to the current solution, the tabu step size is jointly determined by the baseline step size, the gain step size positively correlated with the absolute value of the improvement in the objective function, and the constraint step size positively correlated with the spatial constraint level of the workpiece, including: Obtain a defined baseline step size; Obtain the absolute value of the improvement of the objective function and multiply it by a preset amplification factor. Use the method of rounding down to obtain the integer improvement step size, and set the result of adding 1 to the integer improvement step size as the gain step size. The total number of mating constraint surfaces in the spatial assembly relationship matrix of the parts involved in the two processes related to the movement operation is counted, and the total number of mating constraint surfaces is set as the constraint step size. Add the base step size, the gain step size, and the constraint step size to the base step size, and set the result of the cumulative calculation as the specific value of the taboo step size.

[0072] In some implementations, determining the guiding factor based on the historical frequency of new process pairings in the global memory matrix when the objective function deteriorates includes: Read the historical cumulative total number of newly generated process pairs stored in the corresponding coordinate cells of the global memory matrix; Add a constant 1 to the total number of historical occurrences to obtain the base multiplier; Divide the base constant 8 by the base multiplier and round down to obtain the integer result of the guiding parameter; The integer result of the guiding parameter is added to a constant 2 to form a guiding factor, and the guiding factor is used in the calculation of the taboo step size.

[0073] In some implementations, the amnesty criterion is activated when the neighboring solution's move operation is in a taboo state. If the historical frequency of new process pairing is higher than a preset threshold and the neighboring solution is better than the historical best solution, then the taboo state is amended, including: In each round of the cyclic solution pool selection, check whether the taboo lifecycle tag value of the candidate movement action in the attribute record entry is greater than 0; When the taboo lifecycle tag value is greater than 0, an inspection command is initiated to read the preceding node number and following node number of the new connection pairing formed by the action information; By querying the corresponding rows and columns of the global memory matrix, the corresponding historical frequency integers are extracted, and the difference is calculated by subtracting the set static threshold constant from the historical frequency integers. When the result is found that the difference is greater than 0 and the neighborhood solution is better than the historical best solution, the existing prohibition instruction of the operation is ignored, and the corresponding taboo lifecycle label value is modified and assigned to 0 to remove the taboo state.

[0074] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0075] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An assembly method for a complete set of pole-mounted transformer equipment, characterized in that, include: Acquire workpiece information for the equipment, as well as the physical and electrical safety constraints between workpieces; Based on the physical constraints and electrical safety constraints, an evaluation model is established with the weighted sum of total assembly time, operational difficulty, and safety risks as the objective function, and an initial assembly sequence that conforms to the physical constraints and electrical safety constraints is generated as the current solution. Based on the preset neighborhood structure, the assembly process of the current solution is operated to generate a neighborhood solution set; The tabu search mechanism is used to iteratively search the neighborhood solution set until the termination condition is met, and the optimal assembly sequence is output. The tabu search mechanism sets the inverse operation of the move operation that causes the current solution to change as a tabu object, and establishes a global memory matrix to record the frequency of adjacent process pairings in the historical optimal solution. A taboo step size is set for the taboo object. When the objective function of the neighborhood solution improves relative to the current solution, the taboo step size is jointly determined by the baseline step size, the gain step size which is positively correlated with the absolute value of the improvement of the objective function, and the constraint step size which is positively correlated with the spatial constraint level of the workpiece involved. When the objective function deteriorates, the guiding factor is determined based on the historical frequency of the new process pairing in the global memory matrix. The taboo step size is determined by the baseline step size, the weight based on the workpiece attributes, and the guiding factor, and is negatively correlated with the historical frequency. When the movement operation of the neighborhood solution is in a taboo state, the amnesty criterion is activated. If the historical frequency of the new process pairing is higher than a preset threshold and the neighborhood solution is better than the historical best solution, the taboo state is amnesty.

2. The method according to claim 1, characterized in that, The acquisition of workpiece information and physical and electrical safety constraints between workpieces includes: The weight, dimensions, and interface type number of each workpiece can be extracted by scanning the QR code on the device. Extract the physical constraint relationship model between the workpieces, which includes the three-dimensional bounding box collision detection matrix and the support force relationship matrix of each workpiece; Extract the electrical safety constraint relationship model between workpieces, the electrical safety constraint relationship model includes the electrical clearance setting values ​​of the high voltage side and the low voltage side and the installation sequence of the insulating baffle; Based on the extracted workpiece information, the physical constraint relationship model, and the electrical safety constraint relationship model, an association graph data structure is constructed.

3. The method according to claim 1, characterized in that, The evaluation model established based on the physical constraints and electrical safety constraints, with the weighted sum of total assembly time, operational difficulty, and safety risks as the objective function, includes: The total assembly time is calculated by adding the standardized operation time of each process to the movement time corresponding to the position movement distance involved in the assembly process. Obtain the required spatial line-of-sight obstruction rate for assembly and multiply it by the square root of the component mass value to calculate the operational difficulty score. When two adjacent assembly processes have hazardous source parameters marked for insulation medium puncture, the assembly sequence is discarded as an invalid solution that does not comply with electrical safety constraints. If there are no hazard source parameter markings, the basic safety risk score will be used as the safety risk indicator score. After dimensionless processing of the total assembly time, operation difficulty index score, and safety risk index score, the dimensionless total assembly time is multiplied by 0.5, the operation difficulty by 0.3, and the safety risk by 0.2, and the three are summed to generate the evaluation model.

4. The method according to claim 3, characterized in that, The establishment of a global memory matrix to record the frequency of adjacent process pairings in historical optimal solutions includes: A two-dimensional integer matrix is ​​set as the global memory matrix, with the row index and column index corresponding to all assembly process numbers, respectively. Set all initial values ​​in the two-dimensional integer matrix to 0; After each optimization process, determine whether the content value of the saved current iteration history optimal solution has been updated; When it is determined that the content value of the historical optimal solution has been updated, the coordinate pairing of the two adjacent process numbers in the historical optimal solution sequence is traversed, and the value of the cell where the coordinate pair is located in the two-dimensional integer matrix is ​​added by 1.

5. The method according to claim 1 or 2, characterized in that, The reference step size includes: Read the current iteration count and determine the current search stage based on the ratio of the current iteration count to the preset total maximum iteration count; During the global divergence exploration phase in the first half of the iteration cycle, the base constant value of the baseline step size is set to 7. In the local depth mining phase during the second half of the iteration cycle, the basic constant value of the baseline step size is reduced to 4. Obtain the type feature identifier corresponding to the pre-stored operation action in the tabu search mechanism, query the offset value corresponding to the type feature identifier in the preset operation action attribute table, and add the queried offset value to the current base constant value to obtain the baseline step size.

6. The method according to claim 1, characterized in that, When the neighboring solution improves the objective function relative to the current solution, the tabu step size is jointly determined by the baseline step size, the gain step size positively correlated with the absolute value of the improvement in the objective function, and the constraint step size positively correlated with the spatial constraint level of the workpiece involved, including: Obtain a defined baseline step size; Obtain the absolute value of the improvement of the objective function and multiply it by a preset amplification factor. Use the method of rounding down to obtain the integer improvement step size, and set the result of adding 1 to the integer improvement step size as the gain step size. The total number of mating constraint surfaces in the spatial assembly relationship matrix of the parts involved in the two processes related to the movement operation is counted, and the total number of mating constraint surfaces is set as the constraint step size. Add the base step size, the gain step size, and the constraint step size to the base step size, and set the result of the cumulative calculation as the specific value of the taboo step size.

7. The method according to claim 1, characterized in that, When the objective function deteriorates, the guiding factor is determined based on the historical frequency of the new process pairing in the global memory matrix, including: Read the historical cumulative total number of newly generated process pairs stored in the corresponding coordinate cells of the global memory matrix; Add a constant 1 to the total number of historical occurrences to obtain the base multiplier; Divide the base constant 8 by the base multiplier and round down to obtain the integer result of the guiding parameter; The integer result of the guiding parameter is added to a constant 2 to form a guiding factor, and the guiding factor is used in the calculation of the taboo step size.

8. The method according to claim 1, characterized in that, The amnesty criterion is activated when the neighboring solution's move operation is in a taboo state. If the historical frequency of the new process pairing is higher than a preset threshold and the neighboring solution is better than the historical best solution, then the taboo state is amended, including: In each round of the cyclic solution pool selection, check whether the taboo lifecycle tag value of the candidate movement action in the attribute record entry is greater than 0; When the taboo lifecycle tag value is greater than 0, an inspection command is initiated to read the preceding node number and following node number of the new connection pairing formed by the action information; By querying the corresponding rows and columns of the global memory matrix, the corresponding historical frequency integers are extracted, and the difference is calculated by subtracting the set static threshold constant from the historical frequency integers. When the result is found that the difference is greater than 0 and the neighborhood solution is better than the historical best solution, the existing prohibition instruction of the operation is ignored, and the corresponding taboo lifecycle label value is modified and assigned to 0 to remove the taboo state.

9. An assembly system for a complete pole-mounted transformer assembly, characterized in that, include: The acquisition module is used to acquire workpiece information of the equipment and the physical and electrical safety constraints between workpieces; The generation module is used to establish an evaluation model based on the physical constraints and electrical safety constraints, with the weighted sum of total assembly time, operation difficulty and safety risk as the objective function, and to generate an initial assembly sequence that conforms to the physical constraints and electrical safety constraints as the current solution; Based on the preset neighborhood structure, the assembly process of the current solution is operated to generate a neighborhood solution set; The recording module is used to perform iterative search in the neighborhood solution set using the tabu search mechanism until the termination condition is met, and output the optimal assembly sequence. The tabu search mechanism sets the inverse operation of the move operation that causes the current solution to change as a tabu object, establishes a global memory matrix, and records the frequency of adjacent process pairings in the historical optimal solution. The determination module is used to set a taboo step size for the taboo object. When the objective function of the neighborhood solution improves relative to the current solution, the taboo step size is jointly determined by the reference step size, the gain step size which is positively correlated with the absolute value of the improvement of the objective function, and the constraint step size which is positively correlated with the spatial constraint level of the workpiece involved. When the objective function deteriorates, the guiding factor is determined according to the historical frequency of the new process pairing in the global memory matrix. The taboo step size is determined by the reference step size, the weight based on the workpiece attribute, and the guiding factor and is negatively correlated with the historical frequency. When the movement operation of the neighborhood solution is in a taboo state, the amnesty criterion is activated. If the historical frequency of the new process pairing is higher than a preset threshold and the neighborhood solution is better than the historical best solution, the taboo state is amnesty.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when executed by a processor, implements the method as described in any one of claims 1-8.

Images