A Production Scheduling Method and System for RV Conversion Based on Process Control
By introducing an improved genetic algorithm and GRU model with Henon chaotic mapping initialization, the system can perceive the process status in real time and realize dynamic rescheduling of the RV conversion production scheduling system. This solves the information blind spots and passive response problems of the scheduling system in the existing technology, and improves the timeliness and reliability of production scheduling.
Patent Information
- Application Number
- CN202611282953.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
The RV conversion production scheduling system lacks real-time perception of process execution status, lacks a dynamic rescheduling trigger mechanism based on execution feedback, suffers from poor initial population quality in the standard genetic algorithm which is prone to premature convergence and insufficient local search capability, and lacks the ability to predict process time deviations in advance, and can only respond passively after delays occur.
An improved genetic algorithm based on Henon chaotic mapping initialization is adopted, combined with the GRU model for process scheduling. The process status is perceived in real time. Through a dual verification mechanism of predicted actual time deviation and post-process time deviation, the process scheduling scheme is dynamically adjusted to achieve proactive early warning and dynamic rescheduling.
It significantly extends the response window for rescheduling, improves the on-time delivery rate of orders, reduces the false trigger rate, and enhances the reliability of scheduling decisions and solution efficiency.
Smart Images

Figure CN123113735A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of production scheduling technology, specifically relating to a method and system for production scheduling of RV conversion based on process control. Background Technology
[0002] RV conversion is a typical customized, multi-variety, small-batch production model. Its production scheduling involves multiple processes with strict process sequence constraints and resource competition. The process composition and time quotas of different orders vary significantly, which brings great complexity to production scheduling.
[0003] Currently, large-scale RV manufacturers have introduced commercial manufacturing operations management systems (MAS) that utilize built-in Advanced Planning and Scheduling (APS) modules for automated scheduling. However, these commercial systems employ generalized scheduling logic, essentially static scheduling schemes generated only when an order is placed. They lack real-time awareness of the execution status of processes on the shop floor and cannot automatically trigger dynamic adjustments based on execution deviations. Both commercial APS systems and static scheduling methods based on genetic algorithms remain in a passive mode, only knowing the actual working hours after a process is completed. This means they cannot predict whether a process will ultimately be delayed during execution and can only passively initiate rescheduling after a delay occurs. Furthermore, static scheduling methods based on standard genetic algorithms also have inherent flaws: the initial population of standard genetic algorithms relies on random generation, resulting in inconsistent population quality and affecting solution efficiency and stability; the algorithm tends to converge prematurely to local optima, making it difficult to find the optimal scheduling scheme globally; and its local search capability is insufficient, making it difficult to accurately locate a better solution when close to the optimal one.
[0004] This reveals the following problems with existing RV conversion production scheduling technology: 1. Lack of real-time perception of process execution status, resulting in information blind spots between the scheduling system and the workshop; 2. Existing scheduling schemes are all open-loop static scheduling, lacking a dynamic rescheduling trigger mechanism based on execution feedback; 3. Standard genetic algorithms suffer from poor initial population quality, premature convergence, and insufficient local search capabilities; 4. Lack of the ability to predict process time deviations in advance, only able to passively respond after delays occur, unable to proactively warn during process execution. Summary of the Invention
[0005] (a) Technical problems to be solved
[0006] To address the problems in existing RV conversion production scheduling systems, such as lack of real-time perception of process execution status, lack of dynamic rescheduling triggering mechanism based on execution feedback, poor initial population quality of standard genetic algorithms leading to premature convergence and insufficient local search capabilities, and lack of ability to predict process time deviations in advance, which can only be responded to passively after delays occur, this invention provides an RV conversion production scheduling method and system based on process control.
[0007] (II) Technical Solution
[0008] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:
[0009] S1. Obtain RV modification orders, perform requirement analysis on RV modification orders, and obtain order task parameters;
[0010] S2. Match the RV modification process database to obtain the multiple modification processes and standard working hours of each process corresponding to the order task parameters;
[0011] The improved genetic algorithm is obtained by initializing the machine selection code based on Henon chaotic mapping and combining it with adaptive crossover mutation probability.
[0012] An improved genetic algorithm is used to perform static scheduling of the multiple modification processes to generate a process scheduling scheme.
[0013] S3. Execute the process scheduling plan;
[0014] During the execution of the process scheduling scheme, the start time, completion time and quality inspection results of each process are sensed and collected in real time to obtain a real-time status dataset of the process.
[0015] S4. Input the consumed working hours and the proportion of completed processes from the real-time status dataset of the process into the GRU model to obtain the predicted actual working hours.
[0016] The predicted actual working hours are compared with the standard working hours to obtain the predicted working hour deviation value;
[0017] If the predicted time deviation value is greater than the preset predicted deviation threshold, an advance rescheduling trigger signal is generated.
[0018] The actual working hours are calculated based on the start time and completion time in the real-time status dataset of the process.
[0019] The actual working hours are compared with the standard working hours to obtain the post-event working hour deviation value;
[0020] If the quality inspection result is qualified and the subsequent time deviation value is less than or equal to the preset time deviation threshold, then continue to execute the current process scheduling plan and return to S3;
[0021] Otherwise, return to S2 and call the improved genetic algorithm to generate an updated process scheduling scheme that only applies to incomplete processes;
[0022] Preferably, step S1 includes the following steps:
[0023] S11. Obtain the original data of RV conversion orders through the enterprise resource planning system interface or order management platform;
[0024] S12. Perform structured parsing on the raw data to obtain order task parameters that include vehicle model information, modification requirement list, delivery deadline, and customer special requirements;
[0025] By structuring and parsing the raw data of RV conversion orders, unstructured order information is transformed into order task parameters that can be directly read by the scheduling engine, providing standardized data input for subsequent process matching and scheduling.
[0026] Preferably, step S2 includes the following steps:
[0027] S21. Based on the vehicle model information and modification requirement list in the order task parameters, query the RV modification process database to obtain multiple modification processes and the standard working hours for each process.
[0028] S22. The machine selection encoding is initialized based on Henon chaotic mapping, and the genetic algorithm is improved by combining adaptive crossover mutation probability to obtain the improved genetic algorithm.
[0029] S23. With minimizing the maximum completion time and minimizing the total delay as the optimization objectives, and with the process sequence constraints of each process and the available equipment resources of each process as constraints, the improved genetic algorithm is used to perform static scheduling of the multiple modification processes to obtain the process scheduling scheme.
[0030] Preferably, the improved genetic algorithm used in S22 includes the following steps:
[0031] S221. Initialize the process code using a strategy based on the earliest delivery date to obtain an initialized process code sequence;
[0032] S222. Initialize the machine selection code using Henon chaotic mapping to obtain the initialized machine selection code sequence;
[0033] S223. The elite retention strategy is adopted to retain the individual with the highest fitness value in each generation of the population, thus obtaining the optimal individual to be retained.
[0034] S224. By dynamically adjusting the crossover probability and mutation probability according to the current iteration number and the population fitness value during the iteration process, the adjusted crossover probability and mutation probability are obtained.
[0035] S225. Construct an initial population based on the initialization process encoding sequence and the initialization machine selection encoding sequence to obtain the initial population;
[0036] S226. Genetic iteration is performed based on the initial population, the retained optimal individual, the adjusted crossover probability, and the adjusted mutation probability until the number of iterations reaches the preset maximum number of iterations or the fitness value converges to the convergence judgment threshold. The process sequence and resource allocation scheme corresponding to the current optimal individual are output to obtain the process scheduling scheme or the updated process scheduling scheme.
[0037] Preferably, the initialization of the process code based on the earliest delivery date strategy in S221 includes the following steps:
[0038] S2211. Obtain the vehicle delivery deadline of the RV conversion order, calculate the difference between the vehicle delivery deadline and the current time, and obtain the remaining delivery buffer time for the vehicle.
[0039] The reciprocal of the remaining delivery buffer time is used to obtain the roulette wheel weight for each modification process;
[0040] S2212. The probability of roulette wheel betting is allocated according to the weight of each modification process. The process with the shorter remaining delivery buffer time of the whole vehicle receives a higher probability of roulette wheel betting.
[0041] S2213. Based on the roulette wheel probability, select each modification process by roulette wheel, and generate a process code sequence according to the order of selection to obtain the initial process code sequence;
[0042] By initializing the process codes based on the earliest delivery date, processes with shorter delivery buffer times have a higher probability of being selected. This prioritizes processes with higher urgency in the initial population, improving the quality of the initial population and the solution efficiency.
[0043] Preferably, in S223, retaining the individual with the highest fitness value in each generation of the population by employing an elite retention strategy includes:
[0044] S2231. Calculate the fitness value of each process scheduling scheme in the current population;
[0045] S2232. Sort the scheduling schemes of each process in the current population according to the fitness value from high to low to obtain the sorted population sequence.
[0046] S2233. Extract the process scheduling schemes that are in the top 20% of the sorted population sequence to obtain the elite retention scheme set;
[0047] The elite retention scheme set is directly copied into the next generation population;
[0048] S2234. Extract the process scheduling scheme with the highest fitness value from the elite retention scheme set, and use it as the optimal individual to be retained;
[0049] By employing an elite retention strategy, the individual with the highest fitness value in each generation is directly retained into the next generation, ensuring that the optimal solution is not lost during the iteration process and accelerating the convergence process of the algorithm.
[0050] Preferably, step S224, which dynamically adjusts the crossover and mutation probabilities based on the current iteration number and the population fitness value during the iteration process, includes the following steps:
[0051] S2241. Calculate the variance of the fitness values of each process scheduling scheme in the current population to obtain the population fitness variance value.
[0052] S2242. Compare the population fitness variance value with a preset variance threshold;
[0053] S2243. When the population fitness variance value is less than the preset variance threshold, it is determined that the current population tends to a local optimum, the mutation probability is increased and the crossover probability is decreased, and the first adjusted crossover probability and the first adjusted mutation probability are obtained.
[0054] S2244. When the population fitness variance value is greater than the preset variance threshold, it is determined that the current population is too dispersed, the crossover probability is increased and the mutation probability is decreased to obtain the second adjusted crossover probability and the second adjusted mutation probability.
[0055] S2245. The first adjusted crossover probability and the first adjusted mutation probability or the second adjusted crossover probability and the second adjusted mutation probability are used as the adjusted crossover probability and the adjusted mutation probability.
[0056] The population state is judged in real time by the population fitness variance value. When the population tends to a local optimum, the mutation probability is increased to escape the local optimum. When the population is too scattered, the crossover probability is increased to accelerate convergence. This achieves adaptive dynamic adjustment of crossover probability and mutation probability, balancing global search and local development capabilities.
[0057] Preferably, step S3 includes the following steps:
[0058] S31. According to the process sequence and resource allocation set in the process scheduling scheme, start the execution of each modification process in sequence;
[0059] S32. At the start and end points of each process, the start and end times of the process are recorded by scanning a code with a mobile terminal or by automatic reporting from the equipment data acquisition terminal.
[0060] S33. After each process is completed, the quality inspection results of that process are obtained by inputting through the quality inspection terminal or by automatic output from the quality inspection equipment.
[0061] S34. Organize the start time, completion time and quality inspection results of each process according to the process number and collection time to generate the real-time status dataset of the process.
[0062] By using mobile terminal scanning and automatic reporting from equipment data acquisition terminals, real-time perception and collection of multi-source data during process execution are achieved, enabling the scheduling system to obtain the actual execution status of the workshop in a timely manner, providing a data foundation for subsequent time prediction, deviation detection and dynamic rescheduling.
[0063] Preferably, step S4, which involves inputting the consumed time and the proportion of completed processes from the real-time status dataset of the process into the GRU model to obtain the predicted actual working time, includes the following steps:
[0064] S41. Extract the start time and current time of the current process from the real-time status dataset of the process, calculate the difference between the start time and the current time, and obtain the consumed working hours.
[0065] S42. Extract the percentage of completed processes for the current process from the real-time status dataset of the process.
[0066] S43. Align the consumed working hours and the proportion of completed processes according to timestamps to form an input feature vector;
[0067] S44. Input the input feature vector into the GRU model, and output the predicted actual working hours through the GRU model; the training dataset of the GRU model comes from the historical working hour data of the same type of process stored in the RV modification process database;
[0068] By inputting a feature vector composed of consumed time and the proportion of completed processes into a GRU model, and leveraging the GRU model's ability to remember time series data, the dynamic relationship between consumed time and the proportion of completed processes during process execution is captured, thus enabling advance prediction of the final actual time of the current process.
[0069] Preferably, the RV conversion production scheduling system based on process control is used to implement the RV conversion production scheduling method based on process control. The system includes an order parsing module, a scheduling engine module, a process perception module, a deviation detection module, and a dynamic rescheduling module.
[0070] The order parsing module is used to obtain RV conversion orders and perform requirement parsing on the RV conversion orders to obtain order task parameters;
[0071] The scheduling engine module is used to match the RV modification process database to obtain the multiple modification processes and standard working hours of each process corresponding to the order task parameters, and to perform static scheduling of the multiple modification processes through an improved genetic algorithm to generate a process scheduling scheme.
[0072] The process awareness module is used to sense and collect the start time, consumed time, completed process ratio, completion time and quality inspection results of each process in real time during the execution of the process scheduling plan, so as to obtain the real-time status dataset of the process.
[0073] The deviation detection module is used to calculate the post-process time deviation value and trigger the advance rescheduling trigger signal based on the real-time status dataset of the process, and to determine whether to trigger dynamic rescheduling based on the quality inspection results, the post-process time deviation value and the advance rescheduling trigger signal.
[0074] The dynamic rescheduling module is used to input the processes with unrecorded completion times and the current occupancy status of each piece of equipment into the scheduling engine module when dynamic rescheduling is triggered. The scheduling engine module then calls the improved genetic algorithm to generate an updated process scheduling scheme that only applies to the incomplete processes.
[0075] (III) Beneficial Effects
[0076] The present invention has the following beneficial effects:
[0077] This invention achieves advance prediction of actual working hours for a process by introducing a GRU model. By utilizing the GRU model's ability to remember time-series data, it captures the dynamic relationship between the time consumed and the proportion of completed processes during process execution. During process execution, it can output the predicted actual working hours and identify potential delay risks, generating an advance rescheduling trigger signal. This allows the system to proactively initiate the pre-scheduling process before a delay occurs. Compared to existing technologies that can only passively respond after a delay occurs, this significantly extends the response window for rescheduling.
[0078] This invention constructs a dual verification mechanism of prediction deviation and post-event deviation, which balances the timeliness of early warning and the accuracy of judgment. During the execution of the process, the prediction deviation is used to achieve proactive early warning, and the post-event deviation is used for final confirmation after the process is completed. The current plan will continue to be executed only when three conditions are met simultaneously: the quality inspection is qualified, the post-event deviation does not exceed the threshold, and no early trigger signal is generated. Otherwise, rescheduling is triggered, which avoids misjudgment or omission that may be caused by a single judgment criterion and improves the reliability of scheduling decisions.
[0079] This invention optimizes the initial population quality by using a strategy based on the earliest delivery date and Henon chaotic mapping, ensures that the optimal solution is not lost by using an elite retention strategy, and judges the population state in real time by using the population fitness variance value and adaptively adjusts the crossover probability and mutation probability. This invention overcomes the shortcomings of standard genetic algorithms, such as poor initial population quality, premature convergence, and insufficient local search ability, and improves the solution efficiency and stability.
[0080] To verify the technical effectiveness of this invention, a simulation comparison experiment was conducted using the actual production data of a motorhome modification company over six months. The data covered 123 orders and a total of 2826 processes. Compared with the commercial APS static scheduling scheme, this invention increased the on-time delivery rate from 78.3% to 94.7% and reduced the average delay time from 14.2 hours to 3.8 hours, verifying the performance improvement effect of the dynamic closed-loop scheduling mechanism based on execution feedback. Compared with the standard genetic algorithm, the improved genetic algorithm of this invention reduced the average number of convergence iterations from 156 generations to 87 generations and improved the quality of the optimal solution by 11.3%, verifying the dual improvement of solution efficiency and solution quality based on the earliest delivery date strategy, Henon chaotic mapping initialization method, and adaptive crossover mutation probability strategy. In terms of time deviation early warning, the dual verification mechanism reduced the false trigger rate from 12.7% to 2.3% compared with the single prediction trigger mechanism, verifying the effectiveness of the dual verification mechanism of prediction deviation and post-deviation in reducing invalid rescheduling.
[0081] 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
[0082] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.
[0083] Figure 1 This is a flowchart illustrating the RV conversion production scheduling method based on process control according to the present invention.
[0084] Figure 2 This is a hardware topology diagram of the RV conversion production scheduling method based on process control according to the present invention.
[0085] Figure 3 This is a schematic diagram of the module of the RV conversion production scheduling method based on process control according to the present invention. Detailed Implementation
[0086] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.
[0087] Example 1:
[0088] Please see Figure 1 This invention discloses a method comprising the following steps:
[0089] Please see Figure 1 This invention discloses a method for scheduling RV conversion production based on process control, comprising the following steps:
[0090] S1. Obtain RV modification orders, perform requirement analysis on RV modification orders, and obtain order task parameters;
[0091] S1 includes the following steps:
[0092] S11. Obtain the original data of RV conversion orders through the enterprise resource planning system interface or order management platform;
[0093] Specifically, in embodiment S11, the system obtains the original data of the RV modification order through the enterprise resource planning system interface or the order management platform; the original data includes vehicle information, modification requirement list, delivery deadline and customer special requirements; in this embodiment, the ordered vehicle is a self-propelled Class C RV, the modification requirement list includes four modification items: electrical system modification, water system installation, interior furniture customization and solar panel installation, and the delivery deadline is August 15, 2026;
[0094] S12. Perform structured parsing on the raw data to obtain order task parameters that include vehicle model information, modification requirement list, delivery deadline, and customer special requirements;
[0095] Specifically, in embodiment S12, the system performs structured parsing on the raw data obtained in S11, extracts the basic process route corresponding to the vehicle model, the process dependency relationship of each modification project and the bill of materials, and generates structured order task parameters; in this embodiment, the order task parameters obtained after parsing are: vehicle model C, 4 modification projects, delivery deadline August 15, 2026, and the process dependency relationship and bill of materials of each project have been stored in a structured manner.
[0096] By structuring and parsing the raw data of RV conversion orders, unstructured order information is transformed into order task parameters that can be directly read by the scheduling engine, providing standardized data input for subsequent process matching and scheduling;
[0097] S2. Match the RV modification process database to obtain the multiple modification processes and standard working hours of each process corresponding to the order task parameters;
[0098] The improved genetic algorithm is obtained by initializing the machine selection code based on Henon chaotic mapping and combining it with adaptive crossover mutation probability.
[0099] An improved genetic algorithm is used to perform static scheduling of the multiple modification processes to generate a process scheduling scheme.
[0100] S2 includes the following steps:
[0101] S21. Based on the vehicle model information and modification requirement list in the order task parameters, query the RV modification process database to obtain multiple modification processes and the standard working hours for each process.
[0102] Specifically, in embodiment S21, the system queries the RV modification process database based on the vehicle model information and modification requirement list in the order task parameters; the RV modification process database stores the standard modification procedures and standard working hours for each vehicle model; in this embodiment, a total of 23 modification procedures are matched for 4 modification projects, namely, circuit system modification, water system installation, interior furniture customization and solar panel installation, and the standard working hours for each procedure are 210 hours in total;
[0103] S22. The machine selection encoding is initialized based on Henon chaotic mapping, and the genetic algorithm is improved by combining adaptive crossover mutation probability to obtain the improved genetic algorithm.
[0104] The improved genetic algorithm obtained in S22 includes the following steps:
[0105] S221. Initialize the process code using a strategy based on the earliest delivery date to obtain an initialized process code sequence;
[0106] The process coding initialization based on the earliest delivery date strategy in S221 includes the following steps:
[0107] S2211. Obtain the vehicle delivery deadline of the RV conversion order, calculate the difference between the vehicle delivery deadline and the current time, and obtain the remaining delivery buffer time for the vehicle.
[0108] Specifically, in embodiment S2211, the system obtains the vehicle delivery deadline of the RV conversion order, calculates the difference between the vehicle delivery deadline and the current time, obtains the remaining delivery buffer time of the vehicle, and uses the reciprocal of the remaining delivery buffer time of the vehicle as the roulette weight of each conversion process; in this embodiment, the vehicle delivery deadline is August 15, 2026, the current time is July 1, 2026, the remaining delivery buffer time of the vehicle is 45 days, and the most urgent process gets the largest roulette weight;
[0109] The reciprocal of the remaining delivery buffer time is used to obtain the roulette wheel weight for each modification process;
[0110] S2212. The probability of roulette wheel betting is allocated according to the weight of each modification process. The process with the shorter remaining delivery buffer time of the whole vehicle receives a higher probability of roulette wheel betting.
[0111] S2213. Based on the roulette wheel probability, select each modification process by roulette wheel, and generate a process code sequence according to the order of selection to obtain the initial process code sequence;
[0112] S222. Initialize the machine selection code using Henon chaotic mapping to obtain the initialized machine selection code sequence;
[0113] The system sets the control parameters for the Henon chaotic mapping to a = 1.4 and b = 0.3, and the initial values to x0 = 0.1 and y0 = 0.1. The control parameters a = 1.4 and b = 0.3 are a classic combination of parameters for the Henon mapping, where a controls the nonlinearity of the system and b controls the dissipation of the system. When a = 1.4 and b = 0.3, the system is in a chaotic state, and the generated chaotic sequence is uniformly distributed in the solution space. The system uses the iterative formula of the Henon chaotic mapping... A chaotic sequence is generated, and the initial chaotic sequence is normalized to map each chaotic value to the interval [0,1]. Then, each value in the normalized chaotic sequence is proportionally mapped to the range of values for the selectable machine numbers of each process, thus obtaining the initial machine selection encoding sequence. The specific mapping rule for proportionally mapping each value in the normalized chaotic sequence to the range of values for the selectable machine numbers of each process is as follows: For the i-th process, its set of selectable machine numbers is {1,2,…,M}. i}, M i The total number of optional equipment for this process; the value c at the corresponding position in the normalized chaotic sequence. i The machine number j mapped to this process is represented by the mapping method: j = ceil(c i ×M i ), that is, when c i ∈((j-1) / M i ,j / M iWhen the mapping is performed, it is mapped to machine number j; the encoded value obtained after mapping is the machine number corresponding to the i-th process in the initial machine selection encoding sequence; in this embodiment, the 23 modification processes involve a total of 6 types of equipment resources, and the number of selectable equipment for each process varies from 2 to 4. The initial machine selection encoding sequence generated after chaotic mapping is evenly distributed in the solution space, which improves the diversity of the initial population.
[0114] S223. The elite retention strategy is adopted to retain the individual with the highest fitness value in each generation of the population, thus obtaining the optimal individual to be retained.
[0115] In S223, the individual with the highest fitness value in each generation of the population is retained by employing an elite retention strategy, including:
[0116] S2231. Calculate the fitness value of each process scheduling scheme in the current population;
[0117] Specifically, in embodiment S2231, the system first normalizes the maximum completion time and total delay in the current population. The actual value of each scheme is subtracted from the minimum value in the current population and then divided by the range in the current population, mapped to the [0,1] interval to obtain the normalized maximum completion time and normalized total delay. With minimizing the weighted objective value as the optimization direction, the fitness value is defined as the degree of deviation of the weighted sum of the normalized maximum completion time and the normalized total delay from the theoretical optimal value. The formula for calculating the fitness value F is:
[0118]
[0119] in, This represents the normalized maximum completion time. This represents the normalized total delay. 0.7 and 0.3 are the weighting coefficients for the maximum completion time and the total delay, respectively, and their sum is 1. These weighting coefficients are determined using the analytic hierarchy process (AHP). To prevent numerical overflow due to excessively small numerator and denominator values, a lower bound protection constant ε is introduced when calculating the fitness value. The calculation formula is as follows:
[0120] Fitness value = 1 / (0.7 × maximum completion time + 0.3 × total delay + ε);
[0121] Where ε is a very small positive number, its value is less than the objective function value of any feasible scheduling scheme, and in this embodiment ε is taken as 10. -6To ensure the denominator remains positive in extreme cases and avoid computational overflow, a hierarchical model is established with the scheduling scheme's merits as the target layer and the maximum completion time and total delay as the criterion layer. Domain experts compare and quantify the importance of each indicator pairwise based on the rigid requirements of the RV modification industry for timely order delivery. The maximum eigenvalue and its corresponding eigenvector of the judgment matrix are calculated, and the weights of each indicator are obtained after normalization. In this embodiment, the weight of the maximum completion time is determined to be 0.7, and the weight of the total delay is determined to be 0.3. The maximum completion time and the total delay are then compared... Before performing weighted summation, the system normalizes the maximum completion time and the total delay, mapping them to the same order of magnitude before weighted calculation to avoid weight invalidation due to differences in numerical magnitude. The normalization process obtains the maximum and minimum values of the maximum completion time and the maximum and minimum values of the total delay in the current population, subtracts the minimum value from the actual value of each scheme, divides by the range, and maps them to the [0,1] interval. When multiple RV modification orders exist in the workshop at the same time, the processes of each order share the same roulette wheel probability allocation space.
[0122] The system incorporates all unstarted processes of all orders into the same coding sequence for unified coding. The remaining delivery buffer time for each process is the difference between the delivery deadline of the order to which the process belongs and the current time. The process weight of different orders is uniformly calculated based on their respective remaining delivery buffer time. The process with the shorter remaining delivery buffer time receives a higher weight and is given priority in the unified coding sequence, regardless of which order it belongs to. This formula transforms the minimization problem into a maximization problem. The shorter the maximum completion time and the less the total delay, the higher the fitness value and the greater the probability that the scheduling scheme will be selected for the next generation.
[0123] S2232. Sort the scheduling schemes of each process in the current population according to the fitness value from high to low to obtain the sorted population sequence.
[0124] Specifically, in embodiment S2232, the system sorts the scheduling schemes of each process in the current population according to the fitness value from high to low, and obtains the sorted population sequence;
[0125] S2233. Extract the process scheduling schemes that are in the top 20% of the sorted population sequence to obtain the elite retention scheme set;
[0126] The elite retention scheme set is directly copied into the next generation population;
[0127] Specifically, in embodiment S2233, the system extracts the top 20% of the process scheduling schemes in the sorted population sequence to obtain the elite retention scheme set; in this embodiment, the population size is 100, and the elite retention scheme set contains the 20 process scheduling schemes with the highest fitness; the 20 process scheduling schemes in the elite retention scheme set are directly copied into the next generation population to ensure that the current optimal solution is not lost during the iteration process;
[0128] The specific method for directly copying the elite retention scheme set into the next generation population is as follows: all process scheduling schemes in the elite retention scheme set are directly copied and used to replace an equal number of individuals with the lowest fitness values among all new individuals generated by the crossover and mutation operation in the next generation population. Individuals in the elite retention scheme set do not participate in the crossover and mutation operation in this round and directly enter the next generation population. The crossover and mutation operation is performed on the remaining individuals after removing the elite retention scheme set to generate new individuals to fill the remaining positions in the next generation population.
[0129] S2234. Extract the process scheduling scheme with the highest fitness value from the elite retention scheme set, and use it as the optimal individual to be retained.
[0130] Specifically, in embodiment S2234, the system extracts the process scheduling scheme with the highest fitness value from the elite retention scheme set as the optimal individual to be retained, and this optimal individual is continuously retained in subsequent iterations; the remaining 80 individuals in the next generation population are filled by new individuals generated by crossover mutation operations to maintain the diversity of the population; the proportion of elites in the standard elite retention strategy is usually set in the range of 5% to 30%, and the 20% elite proportion in this embodiment can both ensure the retention of excellent genes and provide sufficient search space for crossover mutation;
[0131] S224. By dynamically adjusting the crossover probability and mutation probability according to the current iteration number and the population fitness value during the iteration process, the adjusted crossover probability and mutation probability are obtained.
[0132] S224, which involves dynamically adjusting the crossover and mutation probabilities based on the current iteration number and the population fitness value during the iteration process, includes the following steps:
[0133] S2241. Calculate the variance of the fitness values of each process scheduling scheme in the current population to obtain the population fitness variance value.
[0134] Specifically, in embodiment S2241, the system calculates the variance of the fitness values of each process scheduling scheme in the current population to obtain the population fitness variance value, which is used to determine whether the population is currently in a local optimum or a dispersed state.
[0135] S2242. Compare the population fitness variance value with a preset variance threshold;
[0136] S2243. When the population fitness variance value is less than the preset variance threshold, it is determined that the current population tends to a local optimum, the mutation probability is increased and the crossover probability is decreased, and the first adjusted crossover probability and the first adjusted mutation probability are obtained.
[0137] S2244. When the population fitness variance value is greater than the preset variance threshold, it is determined that the current population is too dispersed, the crossover probability is increased and the mutation probability is decreased to obtain the second adjusted crossover probability and the second adjusted mutation probability.
[0138] S2245. The first adjusted crossover probability and the first adjusted mutation probability or the second adjusted crossover probability and the second adjusted mutation probability are used as the adjusted crossover probability and the adjusted mutation probability.
[0139] The first adjusted crossover probability and the first adjusted mutation probability or the second adjusted crossover probability and the second adjusted mutation probability are used as the adjusted crossover probability and the adjusted mutation probability; further, the adjusted crossover probability and the adjusted mutation probability are also corrected according to the current iteration number: a higher crossover probability and a lower mutation probability are used in the early stage of the iteration to enhance the global search capability, and a lower crossover probability and a higher mutation probability are used in the later stage of the iteration to enhance the local search capability and avoid premature convergence;
[0140] S225. Construct an initial population based on the initialization process encoding sequence and the initialization machine selection encoding sequence to obtain the initial population;
[0141] Specifically, in embodiment S225, the system combines the initialization process encoding sequence with the initialization machine selection encoding sequence to generate a complete process scheduling scheme individual; by executing the initialization operations of S221 and S222 multiple times, a population-sized number of individuals are generated to form an initial population; in this embodiment, the population size is 100, and the system generates 100 process scheduling scheme individuals as the initial population; the population size is a preset value, which is adaptively determined by the system based on the total number of processes in the multiple modification processes;
[0142] The adaptive population size determination rule is based on the following: Too small a population size leads to insufficient search space coverage, making the algorithm prone to local optima; too large a population size significantly increases the computational cost per iteration, reducing scheduling efficiency; when the total number of processes does not exceed 5, the solution space is relatively small, and a population size of 50 is sufficient to cover the main feasible solution region; when the total number of processes is between 6 and 20, a population size of 10 times the total number of processes is used as the population size, ensuring a linear growth relationship between the population size and the solution space, achieving a balance between search capability and computational efficiency; when the total number of processes exceeds 20, the solution space increases dramatically, and further increasing the population size has limited impact on search capability, while computational cost increases linearly, therefore the upper limit of the population size is truncated to 200; the selection of the coefficient 10 and the truncated value 200 is based on the RV modification industry... The distribution range of the total number of processes in typical orders in the industry is verified and determined by the conventional parameter configuration of the genetic algorithm in the job shop scheduling problem; the maximum number of iterations is adaptively determined by the system according to the population size, and the adaptive determination rule is: maximum number of iterations = max(50, 5 × population size / 10), and 200 is taken when the calculated value exceeds 200; the basis for setting it is: the maximum number of iterations should ensure that the algorithm has enough iteration rounds to make the fitness value converge, while avoiding unnecessary computational waste; too few iterations will cause the algorithm to terminate before sufficient search, and too many iterations will waste computational resources and reduce the benefits; a linear mapping is performed based on the population size, and the larger the population size, the more iterations are required. The values of coefficient 5 and upper limit 200 are determined with reference to the common ratio relationship between the number of iterations and the population size in the field of job shop scheduling of genetic algorithms;
[0143] The construction process includes: combining the initialization process coding sequence with the initialization machine selection coding sequence, verifying whether there is a device conflict at the same time in the combination; if there is a device capacity conflict, adjusting the machine selection coding of the conflicting process to allocate it to available equipment to complete the feasibility repair; after verifying and repairing all individuals, the initial population is formed.
[0144] The specific rules for feasibility repair are as follows: After combining the initialization process code sequence with the initialization machine selection code sequence, check in sequence according to the process code order whether there is a capacity conflict between the equipment allocated to each process at the planned start time. The capacity conflict is when the same equipment is occupied by multiple processes at the same time. If the equipment allocated to the current process is already occupied by other processes at the planned start time, check in sequence in the set of optional equipment for that process whether there is any equipment available at the planned start time. If so, modify the machine selection code of the current process to the number of the available equipment. If all optional equipment for the current process is occupied at the planned start time, postpone the planned start time of the process to the earliest time when any optional equipment becomes available, and update the machine selection code with the number of that equipment. After completing the conflict check and repair of all processes in sequence according to the process code order, the individual after feasibility repair is obtained. After performing the same repair operation on all initial individuals, the initial population is formed.
[0145] S226. Genetic iteration is performed based on the initial population, the retained optimal individual, the adjusted crossover probability, and the adjusted mutation probability until the number of iterations reaches the preset maximum number of iterations or the fitness value converges to the convergence threshold; the process sequence and resource allocation scheme corresponding to the current optimal individual are output to obtain the process scheduling scheme or the updated process scheduling scheme.
[0146] Specifically, in embodiment S226, the system starts with the initial population obtained in S225, and performs genetic iteration processing using the best individuals retained in S2234 and the adjusted crossover and mutation probabilities obtained in S2245. Selection, crossover, and mutation operations are performed in each iteration. When the number of iterations reaches the preset maximum of 100 generations or the fitness value converges to the convergence threshold of 0.95, the system outputs the process sequence and resource allocation scheme corresponding to the current best individual. The convergence threshold of 0.95 is set based on the following: a fitness value reaching this threshold indicates that the current scheduling scheme has approached the theoretical optimal solution; further iterations will improve the result by less than 5% while the computation time increases exponentially. This threshold is a commonly used engineering convergence standard for genetic algorithms in the field of production scheduling. In this embodiment, after 87 iterations, the fitness value converges to 0.95, and the system outputs the process sequence and resource allocation scheme corresponding to the current best individual as the process scheduling scheme; in the rescheduling scenario, this serves as the updated process scheduling scheme.
[0147] When the number of iterations reaches the preset maximum number of iterations (100 generations) or the fitness value converges to the convergence threshold of 0.95, the iteration terminates and the current best individual is output. If the fitness convergence condition is met before reaching the maximum number of iterations, the iteration terminates early and the search is stopped to reduce unnecessary computational overhead.
[0148] The crossover operation in the genetic iteration process includes two parts: process code crossover and machine selection code crossover. The process code crossover uses a process priority-based crossover operator. Specifically, two process positions are randomly selected as crossover points. The process code segment in parent generation 1 located between the two crossover points is copied to the same position in child generation 1. Processes in parent generation 2 that are not in this segment are filled into the remaining positions of child generation 1 according to their relative order in parent generation 2, thus completing the generation of child generation 1. Child generation 2 is generated from parent generation 2 and parent generation 1 in a symmetrical manner. The machine selection code crossover uses a uniform crossover operator. Specifically, for the machine selection code of each process, the machine numbers of parent generation 1 and parent generation 2 at that position are swapped with a probability of 0.5 and retained to the child generation.
[0149] The mutation operation in the genetic iteration process includes two parts: process code mutation and machine selection code mutation. The process code mutation uses a swap mutation operator, specifically: two different positions are randomly selected in the process code sequence, and the process numbers at these two positions are swapped to obtain the mutated process code sequence. The machine selection code mutation uses a random replacement mutation operator, specifically: a position is randomly selected in the machine selection code sequence, and the machine number at that position is replaced with a random available device number from the set of available devices for that process that is different from the current number, to obtain the mutated machine selection code sequence.
[0150] S23. With minimizing the maximum completion time and minimizing the total delay as the optimization objectives, and with the process sequence constraints of each process and the available equipment resources of each process as constraints, the improved genetic algorithm is used to perform static scheduling of the multiple modification processes to obtain the process scheduling scheme.
[0151] The above steps optimize the initialization of process coding and machine selection coding by using the earliest delivery date-based strategy and chaotic mapping, respectively. Combined with the elite retention strategy and adaptive crossover mutation probability, the quality of the initial population and the search efficiency of the algorithm are improved.
[0152] S3. Execute the process scheduling plan;
[0153] During the execution of the process scheduling scheme, the start time, completion time and quality inspection results of each process are sensed and collected in real time to obtain a real-time status dataset of the process.
[0154] S3 includes the following steps:
[0155] S31. According to the process sequence and resource allocation set in the process scheduling scheme, start the execution of each modification process in sequence;
[0156] Specifically, in embodiment S31, the system sends the process scheduling plan generated in S22 to the workshop site, and each process starts and executes in sequence according to the process order and resource allocation set in the scheduling plan; in this embodiment, the first process is the wiring harness laying for the circuit system modification, which is assigned to electrician team 1, the equipment resource is the wiring harness processing table, the planned start time is 8:00 on July 1, 2026, and the planned completion time is 17:00 on July 3, 2026;
[0157] S32. At the start and end points of each process, the start and end times of the process are recorded by scanning a code with a mobile terminal or by automatic reporting from the equipment data acquisition terminal.
[0158] Specifically, in embodiment S32, data collection points are set at the start and end points of each process. When a process starts, the operator scans the process QR code with a mobile terminal, and the system automatically records the current time as the start time. When a process is completed, the operator scans the process QR code again, and the system automatically records the current time as the completion time. In this embodiment, the actual start time of the wire harness laying process is 8:05 on July 1, 2026, and the actual completion time is 18:00 on July 3, 2026.
[0159] S33. After each process is completed, the quality inspection results of that process are obtained by inputting through the quality inspection terminal or by automatic output from the quality inspection equipment.
[0160] Specifically, in embodiment S33, after each process is completed, the quality inspector enters the quality inspection results through the quality inspection terminal, or the quality inspection equipment automatically outputs the quality inspection results to the system; the quality inspection results include two states: qualified and unqualified; in this embodiment, the quality inspection result of the wire harness laying process is qualified.
[0161] S34. Organize the start time, completion time and quality inspection results of each process according to the process number and collection time to generate the real-time status dataset of the process.
[0162] Specifically, in embodiment S34, the system organizes the start time and completion time collected in S32 and the quality inspection results obtained in S33 according to the process number and collection time to generate a real-time status dataset of the process; in this embodiment, the real-time status data of the wire harness laying process is: process number P001, start time 2026-07-01 08:05, completion time 2026-07-03 18:00, actual working hours 34 hours, quality inspection results qualified, equipment resource wire harness processing table;
[0163] The above embodiments realize the real-time perception and collection of start time, completion time and quality inspection results during the process by scanning codes with mobile terminals and automatic reporting by equipment data acquisition terminals. This enables the scheduling system to obtain the actual execution status of the workshop in a timely manner, providing a data foundation for subsequent deviation detection and dynamic rescheduling.
[0164] S4. Input the consumed working hours and the proportion of completed processes from the real-time status dataset of the process into the GRU model to obtain the predicted actual working hours.
[0165] The predicted actual working hours are compared with the standard working hours to obtain the predicted working hour deviation value;
[0166] If the predicted time deviation value is greater than the preset predicted deviation threshold, an advance rescheduling trigger signal is generated.
[0167] The actual working hours are calculated based on the start time and completion time in the real-time status dataset of the process.
[0168] The actual working hours are compared with the standard working hours to obtain the post-event working hour deviation value;
[0169] If the quality inspection result is qualified and the subsequent time deviation value is less than or equal to the preset time deviation threshold, then continue to execute the current process scheduling plan and return to S3;
[0170] Otherwise, return to S2 and call the improved genetic algorithm to generate an updated process scheduling scheme that only applies to incomplete processes;
[0171] S4 includes the following steps:
[0172] The process of obtaining the predicted actual working hours in S4 includes the following steps:
[0173] S41. During the execution of the process, the consumed working hours and the proportion of completed processes in the real-time status dataset of the process are input into the GRU model, and the actual working hours are predicted through the GRU model.
[0174] Specifically, in embodiment S41, the system extracts the start time of the current process from the real-time status dataset of the process generated in S34, obtains the current system time, subtracts the start time from the current time to obtain the consumed working hours; in this embodiment, the start time of the wire harness laying process is 8:05 on July 1, 2026, the current time is 14:00 on July 2, 2026, and the consumed working hours are 30 hours when the start time is subtracted from the current time.
[0175] S42. Compare the predicted actual working hours with the standard working hours to obtain the predicted working hour deviation value;
[0176] Specifically, in embodiment S42, the system extracts the percentage of completed processes for the current process from the real-time process status dataset generated in S34; in this embodiment, the percentage of completed processes for the wire harness laying process is 85%.
[0177] S43. Align the consumed working hours and the proportion of completed processes according to timestamps to form an input feature vector;
[0178] Specifically, in embodiment S43, the system aligns the consumed working hours and the completed process ratio obtained in S41 with the same timestamp to form a two-dimensional input feature vector; in this embodiment, the consumed working hours of the wire harness laying process are 30 hours, the completed process ratio is 85%, and after aligning with the timestamp 2026-07-02 14:00, the input feature vector formed is [30, 0.85];
[0179] S44. Input the input feature vector into the GRU model, and output the predicted actual working hours through the GRU model;
[0180] Specifically, in embodiment S44, the system inputs the input feature vector composed in S43 into the pre-trained GRU model. After forward propagation calculation, the GRU model outputs the predicted actual working time of the current process. In this embodiment, after the input feature vector [30, 0.85] is input into the GRU model, the GRU model outputs a predicted actual working time of 34.5 hours.
[0181] Furthermore, the GRU model is a recurrent neural network model built on gated recurrent units (ROUs). A gated recurrent unit is a variant of an RNU that controls information flow through a gating mechanism. The GRU model uses two gating structures—a reset gate and an update gate—to model the long-short-term dependencies of time-series data. The reset gate controls the influence of the previous hidden state on the current candidate hidden state, while the update gate controls the proportion of the previous hidden state retained in the current time step. The network architecture of the GRU model includes an input layer, a GRU hidden layer, and a fully connected output layer. The input layer has two neurons, corresponding to the two feature dimensions of consumed time and completed process proportion in the input feature vector. The number of neurons in the GRU hidden layer... The GRU hidden layer has 64 parameters. It contains two gated structures: a reset gate and an update gate. The reset gate controls the influence of the previous hidden state on the current candidate hidden state, while the update gate controls the proportion of the previous hidden state retained in the current time step. The activation functions for both the reset and update gates are Sigmoid functions, and the activation function for the candidate hidden state is the Tanh function. The fully connected output layer has one neuron and maps the 64-dimensional feature vector output by the GRU hidden layer to a one-dimensional predicted value, i.e., the predicted actual working time. The fully connected output layer does not contain an activation function. The GRU model has a total of 13,440 trainable parameters, including weight parameters from the input layer to the hidden layer, cyclic weight parameters from the hidden layer to the hidden layer, and bias parameters. The GRU model is trained using the time backpropagation algorithm.
[0182] The training process of the GRU model is as follows: historical work time data of the same type of process is extracted from the RV modification process database. The historical work time data includes the sequence of consumed work time, the sequence of completed process proportions and the corresponding actual completion time of each process recorded at fixed time intervals during the historical execution process. The fixed time interval is recorded once every 30 minutes. The input of each training sample is a two-dimensional feature vector composed of the consumed work time and the proportion of completed process at a certain moment. The output label is the actual work time when the process is finally completed.
[0183] In this embodiment, a total of 1200 historical execution records of the same type of wire harness laying process were obtained, and divided into a training set of 960 records, a validation set of 120 records, and a test set of 120 records in an 8:1:1 ratio. The consumed working hours and the proportion of completed processes in the training dataset were normalized respectively. After normalization, all input features and labels were at the same order of magnitude. The training hyperparameters of the GRU model include learning rate, batch size, maximum number of iteration rounds, and number of early stop rounds. In this embodiment, the learning rate is set to 0.001, the batch size is set to 32, the maximum number of iteration rounds is set to 200 rounds, and the number of early stop rounds is set to 20 rounds.
[0184] During training, the goal is to minimize the mean squared error. The formula for calculating the mean squared error loss function is as follows:
[0185] ;
[0186] Where n is the batch size, For the predicted actual working hours of the i-th sample, Let be the actual completion time label of the i-th sample; the adaptive moment estimation optimizer is used to iteratively optimize the network parameters of the GRU model. In each iteration, the training set is divided into several batches of 32 according to the batch size and input into the model in sequence for forward propagation to calculate the loss function value. Then, the gradient of each network parameter is calculated by the time backpropagation algorithm. Finally, the adaptive moment estimation optimizer updates the network parameters according to the gradient to complete the training of one batch. After all batches are completed, one iteration is completed.
[0187] After each iteration, the validation set is input into the current GRU model to calculate the mean squared error loss value on the validation set. Training stops when the mean squared error loss value on the validation set does not decrease for 20 consecutive iterations. The network parameters corresponding to the minimum validation set loss value are taken as the parameters of the trained GRU model.
[0188] By combining the consumed time and the proportion of completed processes into an input feature vector and inputting it into the GRU model, the dynamic relationship between the consumed time and the proportion of completed processes during the execution of the process can be captured by utilizing the GRU model's ability to remember time series data, thus enabling advance prediction of the final actual time of the current process.
[0189] Example 2:
[0190] Please see Figure 2 and Figure 3 The RV conversion production scheduling system based on process control is used to implement the above-mentioned RV conversion production scheduling method based on process control, including an order parsing module, a scheduling engine module, a process perception module, a deviation detection module, and a dynamic rescheduling module.
[0191] The order parsing module is used to obtain RV conversion orders and perform requirement parsing on the RV conversion orders to obtain order task parameters;
[0192] The scheduling engine module is used to match the RV modification process database to obtain the multiple modification processes and standard working hours of each process corresponding to the order task parameters, and to perform static scheduling of the multiple modification processes through an improved genetic algorithm to generate a process scheduling scheme.
[0193] The process awareness module is used to sense and collect the start time, consumed time, completed process ratio, completion time and quality inspection results of each process in real time during the execution of the process scheduling plan, so as to obtain the real-time status dataset of the process.
[0194] The deviation detection module is used to calculate the post-process time deviation value and trigger the advance rescheduling trigger signal based on the real-time status dataset of the process, and to determine whether to trigger dynamic rescheduling based on the quality inspection results, the post-process time deviation value and the advance rescheduling trigger signal.
[0195] The dynamic rescheduling module is used to input the processes with unrecorded completion times and the current occupancy status of each piece of equipment into the scheduling engine module when dynamic rescheduling is triggered. The scheduling engine module then calls the improved genetic algorithm to generate an updated process scheduling scheme that only applies to the unfinished processes.
[0196] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0197] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.
Claims
1. A method for scheduling RV conversion production based on process control, characterized in that, Includes the following steps: S1. Obtain RV modification orders, perform requirement analysis on RV modification orders, and obtain order task parameters; S2. Match the RV modification process database to obtain the multiple modification processes and standard working hours of each process corresponding to the order task parameters; The improved genetic algorithm is obtained by initializing the machine selection code based on Henon chaotic mapping and combining it with adaptive crossover mutation probability. An improved genetic algorithm is used to perform static scheduling of the multiple modification processes to generate a process scheduling scheme. S3. Execute the process scheduling plan; During the execution of the process scheduling scheme, the start time, completion time and quality inspection results of each process are sensed and collected in real time to obtain a real-time status dataset of the process. S4. Input the consumed working hours and the proportion of completed processes from the real-time status dataset of the process into the GRU model to obtain the predicted actual working hours. The predicted actual working hours are compared with the standard working hours to obtain the predicted working hour deviation value; If the predicted time deviation value is greater than the preset predicted deviation threshold, an advance rescheduling trigger signal is generated. The actual working hours are calculated based on the start time and completion time in the real-time status dataset of the process. The actual working hours are compared with the standard working hours to obtain the post-event working hour deviation value; If the quality inspection result is qualified and the subsequent time deviation value is less than or equal to the preset time deviation threshold, then continue to execute the current process scheduling plan and return to S3; Otherwise, return to S2 and call the improved genetic algorithm to generate an updated process scheduling scheme that only applies to incomplete processes.
2. The RV conversion production scheduling method based on process control according to claim 1, characterized in that, S1 includes the following steps: S11. Obtain the original data of RV conversion orders through the enterprise resource planning system interface or order management platform; S12. Perform structured parsing on the raw data to obtain order task parameters that include vehicle model information, modification requirement list, delivery deadline, and customer special requirements.
3. The RV conversion production scheduling method based on process control according to claim 1, characterized in that, S2 includes the following steps: S21. Based on the vehicle model information and modification requirement list in the order task parameters, query the RV modification process database to obtain multiple modification processes and the standard working hours for each process. S22. The machine selection encoding is initialized based on Henon chaotic mapping, and the genetic algorithm is improved by combining adaptive crossover mutation probability to obtain the improved genetic algorithm. S23. With minimizing the maximum completion time and minimizing the total delay as the optimization objectives, and with the process sequence constraints of each process and the available equipment resources of each process as constraints, the improved genetic algorithm is used to perform static scheduling of the multiple modification processes to obtain the process scheduling scheme.
4. The RV conversion production scheduling method based on process control according to claim 3, characterized in that, The improved genetic algorithm used in S22 includes the following steps: S221. Initialize the process code using a strategy based on the earliest delivery date to obtain an initialized process code sequence; S222. Initialize the machine selection code using Henon chaotic mapping to obtain the initialized machine selection code sequence; S223. The elite retention strategy is adopted to retain the individual with the highest fitness value in each generation of the population, thus obtaining the optimal individual to be retained. S224. By dynamically adjusting the crossover probability and mutation probability according to the current iteration number and the population fitness value during the iteration process, the adjusted crossover probability and mutation probability are obtained. S225. Construct an initial population based on the initialization process encoding sequence and the initialization machine selection encoding sequence to obtain the initial population; S226. Genetic iteration is performed based on the initial population, the retained optimal individual, the adjusted crossover probability, and the adjusted mutation probability until the number of iterations reaches the preset maximum number of iterations or the fitness value converges to the convergence judgment threshold. The process sequence and resource allocation scheme corresponding to the current optimal individual are output to obtain the process scheduling scheme or the updated process scheduling scheme.
5. The RV conversion production scheduling method based on process control according to claim 4, characterized in that, The process coding initialization based on the earliest delivery date strategy in S221 includes the following steps: S2211. Obtain the vehicle delivery deadline of the RV conversion order, calculate the difference between the vehicle delivery deadline and the current time, and obtain the remaining delivery buffer time of the vehicle. The reciprocal of the remaining delivery buffer time is used to obtain the roulette wheel weight for each modification process; S2212. The probability of roulette wheel betting is allocated according to the weight of each modification process. The process with the shorter remaining delivery buffer time of the whole vehicle receives a higher probability of roulette wheel betting. S2213. Based on the roulette wheel probability, select each modification process by roulette wheel, and generate a process code sequence according to the order of selection to obtain the initial process code sequence.
6. The RV conversion production scheduling method based on process control according to claim 4, characterized in that, In S223, the individual with the highest fitness value in each generation of the population is retained by employing an elite retention strategy, including: S2231. Calculate the fitness value of each process scheduling scheme in the current population; S2232. Sort the scheduling schemes of each process in the current population according to the fitness value from high to low to obtain the sorted population sequence. S2233. Extract the process scheduling schemes that are in the top 20% of the sorted population sequence to obtain the elite retention scheme set; The elite retention scheme set is directly copied into the next generation population; S2234. Extract the process scheduling scheme with the highest fitness value from the elite retention scheme set, and use it as the optimal individual to be retained.
7. The RV conversion production scheduling method based on process control according to claim 4, characterized in that, S224, which involves dynamically adjusting the crossover and mutation probabilities based on the current iteration number and the population fitness value during the iteration process, includes the following steps: S2241. Calculate the variance of the fitness values of each process scheduling scheme in the current population to obtain the population fitness variance value. S2242. Compare the population fitness variance value with a preset variance threshold; S2243. When the population fitness variance value is less than the preset variance threshold, it is determined that the current population tends to a local optimum, the mutation probability is increased and the crossover probability is decreased, and the first adjusted crossover probability and the first adjusted mutation probability are obtained. S2244. When the population fitness variance value is greater than the preset variance threshold, it is determined that the current population is too dispersed, the crossover probability is increased and the mutation probability is decreased to obtain the second adjusted crossover probability and the second adjusted mutation probability. S2245. The first adjusted crossover probability and the first adjusted mutation probability or the second adjusted crossover probability and the second adjusted mutation probability are used as the adjusted crossover probability and the adjusted mutation probability.
8. The RV conversion production scheduling method based on process control according to claim 1, characterized in that, S3 includes the following steps: S31. According to the process sequence and resource allocation set in the process scheduling scheme, start the execution of each modification process in sequence; S32. At the start and end points of each process, the start and end times of the process are recorded by scanning a code with a mobile terminal or by automatic reporting from the equipment data acquisition terminal. S33. After each process is completed, the quality inspection results of that process are obtained by inputting through the quality inspection terminal or by automatic output from the quality inspection equipment. S34. Organize the start time, completion time and quality inspection results of each process according to the process number and collection time to generate the real-time status dataset of the process.
9. The RV conversion production scheduling method based on process control according to claim 1, characterized in that, In step S4, the steps of inputting the consumed time and the proportion of completed processes from the real-time status dataset of the process into the GRU model to obtain the predicted actual working time include: S41. Extract the start time and current time of the current process from the real-time status dataset of the process, calculate the difference between the start time and the current time, and obtain the consumed working hours. S42. Extract the percentage of completed processes for the current process from the real-time status dataset of the process. S43. Align the consumed working hours and the proportion of completed processes according to timestamps to form an input feature vector; S44. Input the input feature vector into the GRU model, and output the predicted actual working hours through the GRU model; the training dataset of the GRU model comes from the historical working hour data of the same type of process stored in the RV modification process database.
10. A RV conversion production scheduling system based on process control, characterized in that, The system is used to implement the RV conversion production scheduling method based on process control as described in any one of claims 1 to 9, the system comprising an order parsing module, a scheduling engine module, a process perception module, a deviation detection module, and a dynamic rescheduling module; The order parsing module is used to obtain RV conversion orders and parse the requirements of the RV conversion orders to obtain order task parameters; The scheduling engine module is used to match the RV modification process database to obtain the multiple modification processes and standard working hours of each process corresponding to the order task parameters, and to perform static scheduling of the multiple modification processes through an improved genetic algorithm to generate a process scheduling scheme. The process awareness module is used to sense and collect the start time, consumed time, completed process ratio, completion time and quality inspection results of each process in real time during the execution of the process scheduling plan, so as to obtain the real-time status dataset of the process. The deviation detection module is used to calculate the post-process time deviation value and trigger the advance rescheduling trigger signal based on the real-time status dataset of the process, and to determine whether to trigger dynamic rescheduling based on the quality inspection results, the post-process time deviation value and the advance rescheduling trigger signal. The dynamic rescheduling module is used to input the processes with unrecorded completion times and the current occupancy status of each piece of equipment into the scheduling engine module when dynamic rescheduling is triggered. The scheduling engine module then calls the improved genetic algorithm to generate an updated process scheduling scheme that only applies to the incomplete processes.
