CN101424919A - Rescheduling and deciding system for semi-conductor manufacturing system - Google Patents
Rescheduling and deciding system for semi-conductor manufacturing system Download PDFInfo
- Publication number
- CN101424919A CN101424919A CNA2008102030133A CN200810203013A CN101424919A CN 101424919 A CN101424919 A CN 101424919A CN A2008102030133 A CNA2008102030133 A CN A2008102030133A CN 200810203013 A CN200810203013 A CN 200810203013A CN 101424919 A CN101424919 A CN 101424919A
- Authority
- CN
- China
- Prior art keywords
- parameter
- module
- neural network
- output
- decision
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention relates to a rescheduling decision making system of a semiconductor manufacturing system in the semiconductor manufacturing field. In the invention, a GUI module is in charge of alternating with a user and other modules; a fuzzy neural network decision making module obtains data of production state and interfering information of the semiconductor manufacturing system to perform the rescheduling decision making treatment of the semiconductor manufacturing system, and outputs the information result performed by the rescheduling decision making treatment to a fuzzy neural network parameter training module and the GUI module. The fuzzy neural network parameter training module performs the training treatment to the central value of a fuzzification layer and the width value parameter as well as the connection weight parameter of the fuzzy neural network decision making module. The invention improves the accuracy of the rescheduling decision making of the semiconductor manufacturing system.
Description
Technical field
The present invention relates to a kind of information handling system of technical field of manufacturing semiconductors, specifically is a kind of reschedule decision system of semi-conductor manufacturing system.
Background technology
But semi-conductor manufacturing system has hyperpycnal inflow, single-piece and the coexistence of batch processing equipment group, apparatus of load singularity and the complicacy that is different from conventional production lines such as unbalanced.Uncertain factors such as process equipment random fault, workpiece are heavily processed, urgent order make semi-conductor manufacturing system be in the highly uncertain environment.For the interference to system responds fast, improve the production efficiency and the stability of system, need adopt reschedule decision-making means to come the influence of semi-conductor manufacturing system to be assessed, and select the reschedule control hierarchy of optimization according to disturbing.The reschedule decision-making of at present general semiconductor manufacturing shop is operated enforcement by experienced control engineer usually, decision process is usually in conjunction with working experience, adopt the method for heuristic rule to carry out the selection of reschedule control hierarchy, instruct the operation of the dynamic process and the production run of the various interference in workshop.
Find by prior art documents, " Modeling and rescheduling of a re-entrant wafer fabricationline involving machine unreliability " (reentried production line model building and the reschedule under the unreliable situation of consideration equipment) that R.Kumar etc. deliver on " International Journal ofProduction Research " (producing the research international magazine) (2004 42 21 phases of volume) 4431-4455 page or leaf, adopt the reschedule decision-making that is used for semiconductor production line based on the method for heuristic rule in this article, by to the equipment of system fault whether, whether faulty equipment is repaired, whether queuing workpiece quantity greater than setting value with whether arrive four rules such as average reschedule cycle and detect, and whether the decision-making manufacturing line of can reentrying carries out reschedule and the specifically selection of reschedule control method.Because the system factor that the reschedule decision process of the document is considered is less, do not consider the influence of common interference such as rush order is inserted, workpiece is done over again to semiconductor production line; The rule of decision-making is too simple, so adaptability and intelligent relatively poor; The index of Decision-Making Evaluation is mainly considered delivery date and is dragged current cost simultaneously, does not consider the performance index such as robustness of reschedule decision-making technique.Therefore this method is only applicable to simple, the situation of problem on a small scale through simplifying, and is not suitable for the reschedule decision-making of the semiconductor fabrication lines under complexity, the extensive and uncertain condition.
Summary of the invention
The present invention is directed to the deficiencies in the prior art and defective, a kind of reschedule decision system of semi-conductor manufacturing system is proposed, it is had according to present dynamic equipment failure, wafer do over again and the reschedule control hierarchy optimized and the function of method are selected in the influence of interference such as rush order insertion fast, improved the reschedule decision-making capability of semiconductor production line significantly.
The present invention is achieved by the following technical solutions, the present invention includes: GUI (graphic user interface) module, fuzzy neural network decision-making module and fuzzy neural network parameter training module, wherein:
The GUI module is the employed graphic user interface of client, be responsible for the man-machine interactive operation of user and modules such as fuzzy neural network decision-making module and fuzzy neural network parameter training module, this module receives historical data, semi-conductor manufacturing system production status and the interfere information of the reschedule decision-making of importing in the man-machine reciprocal process, the fuzzy neural network parameter training module of importing in the reception man-machine interactive operation process and the initiation parameter information of fuzzy neural network decision-making module, and show the intermediate result of each module and last decision-making output result;
The fuzzy neural network decision-making module obtains the production status and the interfere information data of semi-conductor manufacturing system on the one hand from the GUI module, obtain the connection weight value parameter of the central value parameter of obfuscation layer and width value parameter information, output layer on the other hand from fuzzy neural network parameter training module, carry out the reschedule decision-making treatment of semi-conductor manufacturing system, and the information result of reschedule decision-making treatment is outputed to fuzzy neural network parameter training module and GUI module;
Fuzzy neural network parameter training module receives the output information of the reschedule decision-making treatment of fuzzy neural network decision-making module on the one hand, obtain the historical data of fuzzy neural network training initiation parameter set point information and reschedule decision-making on the one hand from the GUI module, central value to the obfuscation layer of fuzzy neural network decision-making module, the connection weight value parameter of width value parameter and output layer is carried out training managing, and with the central value of the obfuscation layer optimized, the connection weight value parameter of width value parameter and output layer is transferred to the fuzzy neural network decision-making module, and is transferred to the GUI module.
Described fuzzy neural network decision-making module comprises: input layer submodule, obfuscation layer processing sub, rules layer processing sub, normalization layer processing sub, output layer processing sub and reschedule decision-making output sub-module, wherein:
The input layer submodule is responsible for obtaining m production status and the interfere information that receives semi-conductor manufacturing system from the GUI module, specifying information comprises: the stable β c of the equipment group average queue length L that is disturbed, semi-conductor manufacturing system scheduling, the equipment group of being disturbed average load η, the equipment group workpiece postorder average relaxation time ts that is disturbed and the interference parameter T of semi-conductor manufacturing system etc. relatively, and export to obfuscation layer processing sub;
Obfuscation layer processing sub obtains production status and interfere information from input layer submodule module on the one hand, obtain the central value parameter c and the width value parameter σ of Gaussian function on the one hand from fuzzy neural network parameter training module, production status and interfere information at each input, central value parameter c and width value parameter σ according to 3 groups of correspondences, adopt 3 Gaussian functions to carry out blurring mapping respectively and handle, obtain 3m Fuzzy processing output data and output to the rules layer processing sub;
The rules layer processing sub is responsible for handling carrying out regular product from the data of obfuscation layer processing sub input, at 3m Fuzzy processing data that obtain from the Fuzzy processing submodule, at first according to the rule of from each production status and pairing 3 the Fuzzy processing data of interfere information, getting data, select m Fuzzy processing data, then m the data of selecting being carried out product handles, according to this picking rule, the rules layer processing sub forms 3 altogether
mThe output data that individual regular product is handled also outputs to normalization layer processing sub;
Normalization layer processing sub is responsible for the data of rules layer processing sub output are carried out normalized, at 3 of rules layer processing sub output
mThe output data that individual regular product is handled, at first to this 3
mIndividual output data is sued for peace, and the value that obtains of will suing for peace is then removed each output data, obtains 3 of normalization layer processing sub
mIndividual normalized data output also is input to the output layer processing sub;
The output layer processing sub obtains the normalization output data from normalization layer processing sub on the one hand, obtains output layer weighting parameter w data from fuzzy neural network parameter training module on the one hand, to 3 of normalization layer processing sub output
mIndividual normalized data, it is multiplied each other with the output layer training weighting parameter w that obtains from fuzzy neural network parameter training module and be weighted, the result of weighting adopts 3 tansig functions to carry out the output layer functional transformation and handles, obtain the processing output variable data of 3 output layer processing sub, the data variable of handling is outputed to fuzzy neural network parameter training module and reschedule decision-making output sub-module;
After reschedule decision-making output sub-module obtains 3 output variable data from the output layer processing sub, the variable data of absolute value maximum wherein as the reschedule decision-making output of optimizing, is outputed to the GUI module.
Described fuzzy neural network parameter training module comprises: parameter input submodule, training parameter updating submodule and parameter output sub-module, wherein:
Parameter input submodule is responsible for obtaining the initial value of weighting parameter w of initial value, output layer of the central value parameter c of frequency of training parameter, initialization study rate parameter, obfuscation layer of fuzzy neural network training and width value parameter σ and the historical data of reschedule decision-making from the GUI module, obtain the output data of weighting function conversion process from the output layer processing sub of fuzzy neural network decision-making module, and be input to the training parameter updating submodule;
The training parameter updating submodule is responsible for 3 of 3m central value parameter c of the obfuscation layer in the fuzzy neural network decision-making module and width value parameter σ, output layer
M+1Individual weighting parameter w is optimized training managing, this submodule is according to current central value parameter c, width value parameter σ and the weighting parameter w of output data, fuzzy neural network decision-making module of the weighting function conversion process that obtains, employing has the method for the gradient decline optimizing of supervision to calculate 3m central value parameter c of obfuscation layer in the fuzzy neural network decision-making module and the width value parameter σ value along the rate of change of gradient direction respectively, calculates 3 of output layer in the fuzzy neural network decision-making module
M+1The value along the rate of change of gradient direction of individual weighting parameter w correspondence, and add up with current central value parameter c, width value parameter σ and weighting parameter w value, thereby obtain the value of central value parameter c, width value parameter σ and the weighting parameter w of training optimization, realize that by aforesaid operations the optimization training managing value of acquisition outputs to the parameter output sub-module to obfuscation layer in the fuzzy neural network decision-making module and the training managing of output layer Parameter Optimization;
The parameter output sub-module is exported to obfuscation layer processing sub and GUI module in the fuzzy neural network decision-making module from central value parameter c, the width value parameter σ of the obfuscation layer of training parameter updating submodule acquisition optimization; Output layer weighting parameter w from the acquisition of training parameter updating submodule is optimized exports to output layer processing sub and GUI module in the fuzzy neural network decision-making module.
In the operational process of the present invention in the training stage, the GUI module at first receives production status and interfere information, the initiation parameter set point information of fuzzy neural network training and the historical data of reschedule decision-making of the semi-conductor manufacturing system of user's input, outputs it to fuzzy neural network decision-making module and fuzzy neural network parameter training module respectively; The fuzzy neural network decision-making module obtains the production status and the interfere information of semi-conductor manufacturing system on the one hand from the GUI module, one side is from the central value of the obfuscation layer of fuzzy neural network parameter training module acquisition optimization and the weighting parameter information of width value parameter and output layer, carry out semi-conductor manufacturing system reschedule decision-making treatment, and the decision-making treatment result who optimizes is outputed to fuzzy neural network parameter training module and GUI module; Fuzzy neural network parameter training module receives the decision-making treatment object information from the fuzzy neural network module on the one hand, the GUI module receives the initializing set value information of fuzzy neural network training on the one hand, the central value of the obfuscation layer in the fuzzy neural network module and the weighting parameter of width value parameter and output layer are carried out training managing, the central value of the obfuscation layer optimized and the weighting parameter of width value parameter and output layer are transferred to the fuzzy neural network module, and be transferred to the GUI module, frequency of training arrival initial set value when fuzzy neural network parameter training module, then finish the training of fuzzy neural network, enter the decision phase of fuzzy neural network reschedule decision system.
In the decision phase, the GUI module at first receives the production status and the interfere information of the semi-conductor manufacturing system of user's input, outputs it to the fuzzy neural network decision-making module; The fuzzy neural network decision-making module obtains the production status and the interfere information of semi-conductor manufacturing system on the one hand from the GUI module, one side is from the central value of the obfuscation layer of fuzzy neural network parameter training module acquisition optimization and the weighting parameter information of width value parameter and output layer, carry out semi-conductor manufacturing system reschedule decision-making treatment, and the decision-making treatment result who optimizes is outputed to the GUI module.
Compared with prior art, the present invention has following beneficial effect:
The present invention is according to the various real-time production status of semi-conductor manufacturing system and interfere information (as equipment failure, do over again and rush order etc.), indexs such as robustness based on the reschedule decision-making, adopt a plurality of fuzzy neural network processing sub counterweight scheduling strategies to be optimized decision-making treatment, have adaptability and more excellent decision-making performance preferably.The actual production data of based semiconductor manufacturing enterprise carry out the reschedule decision analysis and with compare with traditional rule-based decision system, the accuracy of the reschedule decision-making of semi-conductor manufacturing system has improved 8%-10%, reaches 93%-98.3% (the relative error rate of exporting with desirable optimum reschedule decision-making is controlled in the 1.75%-5%).Therefore, the reschedule Control and Optimization to large-scale semi-conductor manufacturing system has the obvious practical meaning.The processing time of reschedule decision process of the present invention in second, is satisfied the demand of semi-conductor manufacturing system scheduling production run at 30-60 simultaneously.
Description of drawings
Fig. 1 is a system architecture diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing embodiments of the invention are elaborated: present embodiment is being to implement under the prerequisite with the technical solution of the present invention, provide detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.
The historical data of the production status of the semi-conductor manufacturing system that obtains in the present embodiment and interfere information and reschedule decision-making forms the training sample data of the semi-conductor manufacturing system reschedule decision-making of 90 time periods (3 hours/period), and is as shown in table 1.Wherein production status and interfere information number m=5, that is: the stable β c of the equipment group average queue length L that is disturbed, semi-conductor manufacturing system scheduling, the equipment group of being disturbed average load η, the equipment group workpiece postorder average relaxation time ts that is disturbed relatively and the information such as interference parameter T of semi-conductor manufacturing system.As following table: the training data table of table 1 reschedule decision-making:
As shown in Figure 1, present embodiment comprises: GUI module, fuzzy neural network decision-making module and fuzzy neural network parameter training module, wherein:
The GUI module is the employed graphic user interface of client, is responsible for the man-machine interactive operation of user and modules such as fuzzy neural network decision-making module and fuzzy neural network parameter training module.This module receives historical data, semi-conductor manufacturing system production status and the interfere information of the reschedule decision-making of importing in the man-machine reciprocal process; The fuzzy neural network parameter training module of importing in the reception man-machine interactive operation process and the initiation parameter information of fuzzy neural network decision-making module; And show the intermediate result and the last decision-making output result of each module;
The fuzzy neural network decision-making module obtains the production status and the interfere information data of semi-conductor manufacturing system on the one hand from the GUI module, obtain the connection weight value parameter of the central value parameter of obfuscation layer and width value parameter information, output layer on the other hand from fuzzy neural network parameter training module, carry out the reschedule decision-making treatment of semi-conductor manufacturing system, and the information result of reschedule decision-making treatment is outputed to fuzzy neural network parameter training module and GUI module;
Fuzzy neural network parameter training module receives the output information of the reschedule decision-making treatment of fuzzy neural network decision-making module on the one hand, obtain the historical data of fuzzy neural network training initiation parameter set point information and reschedule decision-making on the one hand from the GUI module, central value to the obfuscation layer of fuzzy neural network decision-making module, the connection weight value parameter of width value parameter and output layer is carried out training managing, and with the central value of the obfuscation layer optimized, the connection weight value parameter of width value parameter and output layer is transferred to the fuzzy neural network decision-making module, and narrates the module to GUI.
Described fuzzy neural network decision-making module comprises: input layer submodule, obfuscation layer processing sub, rules layer processing sub, normalization layer processing sub, output layer processing sub and reschedule decision-making output sub-module, wherein:
The input layer submodule is responsible for obtaining m=5 production status and the interfere information that receives semi-conductor manufacturing system from the GUI module, and exports to obfuscation layer processing sub;
Obfuscation layer processing sub is responsible for 5 production statuses of input and interfere information are carried out the obfuscation conversion process.Obfuscation layer processing sub obtains production status and interfere information from the input layer submodule on the one hand, obtains the central value parameter c and the width value parameter σ of Gaussian function on the one hand from fuzzy neural network parameter training module.At the production status and the interfere information of each input, central value parameter c and width value parameter σ according to 3 groups of correspondences adopt 3 Gaussian functions respectively
Carry out blurring mapping and handle, obtain 15 Fuzzy processing output datas and output to the rules layer processing sub, realize the Gaussian function Fuzzy processing of obfuscation layer processing sub by above method;
The rules layer processing sub is responsible for handling carrying out regular product from the data of obfuscation layer processing sub input, at 15 Fuzzy processing data that obtain from the Fuzzy processing submodule, at first according to the rule of from each production status and pairing 3 the Fuzzy processing data of interfere information, getting data, select 5 Fuzzy processing data, 5 data will selecting are then carried out product and are handled.According to this picking rule, the rules layer processing sub forms 3 altogether
5The output data that individual regular product is handled also outputs to normalization layer processing sub, handles by the regular product of above method implementation rule layer processing sub;
Normalization layer processing sub is responsible for the data of rules layer processing sub output are carried out normalized, at 3 of rules layer processing sub output
5The output data that individual regular product is handled, at first to this 3
5Individual output data is sued for peace, and the value that obtains of will suing for peace is then removed each output data, obtains 3 of normalization layer processing sub
5Individual normalized data output also is input to the output layer processing sub, realizes the normalized of normalization layer processing sub by above method;
The output layer processing sub is responsible for output data with normalization layer processing sub and further is weighted with functional transformation and handles.The output layer processing sub is imported the normalization output data of normalization layer processing sub on the one hand, obtains output layer weighting parameter w data from fuzzy neural network parameter training module on the one hand.At 3 of normalization layer processing sub output
5Individual normalized data, it is multiplied each other with the output layer training weighting parameter w that obtains from fuzzy neural network parameter training module and be weighted, the result of weighting adopts 3 tansig functions to carry out the output layer functional transformation and handles, obtain the processing output variable data of 3 output layer processing sub, the data variable of handling is outputed to fuzzy neural network parameter training module and reschedule decision-making output sub-module; Realize the weighted sum functional transformation processing of output layer processing sub by above method;
After reschedule decision-making output sub-module obtains 3 output variable data from the output layer processing sub, and wherein the variable data of absolute value maximum is exported as the reschedule decision-making of optimizing, and outputs to the GUI module.
Described fuzzy neural network parameter training module comprises: parameter input submodule, training parameter updating submodule and parameter output sub-module, wherein:
Parameter input submodule is responsible for obtaining the initial value of weighting parameter w of initial value, output layer of the central value parameter c of frequency of training parameter, initialization study rate parameter, obfuscation layer of fuzzy neural network training process and width value parameter σ and the historical data of reschedule decision-making from the GUI module, obtain the output data of weighting function conversion process from the output layer processing sub of fuzzy neural network decision-making module, and be input to the training parameter updating submodule;
The training parameter updating submodule is responsible for 3 of 15 central value parameter c of the obfuscation layer in the fuzzy neural network decision-making module and width value parameter σ, output layer
6Individual weighting parameter w is optimized training managing, this submodule is according to current central value parameter c, width value parameter σ and the weighting parameter w of output data, fuzzy neural network decision-making module of the weighting function conversion process that obtains, employing has the method for the gradient decline optimizing of supervision to calculate 15 central value parameter c of obfuscation layer in the fuzzy neural network decision-making module and the width value parameter σ value along the rate of change of gradient direction respectively, calculates 3 of output layer in the fuzzy neural network decision-making module
6The value along the rate of change of gradient direction of individual weighting parameter w correspondence, and add up with current central value parameter c, width value parameter σ and weighting parameter w value, thereby obtain the value of central value parameter c, width value parameter σ and the weighting parameter w of training optimization, realize that by above process to obfuscation layer in the fuzzy neural network decision-making module and the training managing of output layer Parameter Optimization, the optimization training managing value of acquisition outputs to the parameter output sub-module;
The parameter output sub-module is exported to obfuscation layer processing sub and GUI module in the fuzzy neural network decision-making module from central value parameter c, the width value parameter σ of the obfuscation layer of training parameter updating submodule acquisition optimization; Output layer weighting parameter w from the acquisition of training parameter updating submodule is optimized exports to output layer processing sub and GUI module in the fuzzy neural network decision-making module.
Described fuzzy neural network parameter training module, 15 central value parameter c of the obscuring layer processing sub in the fuzzy neural network decision-making module of its training and the initial value of width value parameter σ are chosen as shown in table 2,3 in the output layer submodule
6The initial value of individual weighting parameter w is set at 0.8, initialization study rate parameter γ=0.01; The frequency of training parameter N
Max=10
4As following table: the initial value of table 2 central value parameter and width value parameter:
Sequence number | c ijParameter name | Initial value | σ ijParameter name | Initial value |
1 | C 11 | 1 | σ 11 | 2.0 |
2 | C 12 | 4 | σ 12 | 2.0 |
3 | C 13 | 7 | σ 13 | 2.0 |
4 | C 21 | 2 | σ 21 | 2.0 |
5 | C 22 | 2.5 | σ 22 | 2.0 |
6 | C 23 | 3.5 | σ 23 | 2.0 |
7 | C 31 | 0.6 | σ 31 | 2.0 |
8 | C 32 | 0.7 | σ 32 | 2.0 |
9 | C 33 | 0.8 | σ 33 | 2.0 |
10 | C 41 | 2 | σ 41 | 2.0 |
11 | C 42 | 3.5 | σ 42 | 2.0 |
12 | C 43 | 5 | σ 43 | 2.0 |
13 | C 51 | 1.5 | σ 51 | 2.0 |
14 | C 52 | 6 | σ 52 | 2.0 |
15 | C 53 | 9 | σ 53 | 2.0 |
Status information and interfere information in the present embodiment in the fuzzy neural network decision-making module based semiconductor Production Line process carry out the reschedule decision-making, and the reschedule result of decision is as shown in the table, the reschedule decision-making output of table 3 fuzzy neural network decision-making module:
The present embodiment system carries out the reschedule decision-making to the interference of semiconductor production line, according to the equipment failure of semiconductor production line, do over again and influence and production status information that rush order etc. is disturbed, consider the performance index such as robustness of reschedule decision-making output, adopt a plurality of fuzzy neural network processing sub counterweight scheduling strategies to be optimized decision-making treatment, make the reschedule decision-making of semiconductor production line have adaptability and more excellent decision-making performance preferably, improved the efficient of semiconductor production line reschedule control under the disturbed condition.The accuracy of the reschedule decision-making of the semi-conductor manufacturing system that the present invention obtains has also improved 8%-10%, reaches 93%-98.3% (the relative error rate of exporting with desirable optimum reschedule decision-making is controlled in the 1.75%-5%).
Claims (3)
1, a kind of reschedule decision system of semi-conductor manufacturing system is characterized in that, comprising: GUI module, fuzzy neural network decision-making module and fuzzy neural network parameter training module, wherein:
The GUI module is the employed graphic user interface of client, be responsible for the man-machine interactive operation of user and modules such as fuzzy neural network decision-making module and fuzzy neural network parameter training module, this module receives historical data, semi-conductor manufacturing system production status and the interfere information of the reschedule decision-making of importing in the man-machine reciprocal process, the fuzzy neural network parameter training module of importing in the reception man-machine interactive operation process and the initiation parameter information of fuzzy neural network decision-making module, and show the intermediate result of each module and last decision-making output result;
The fuzzy neural network decision-making module obtains the production status and the interfere information data of semi-conductor manufacturing system on the one hand from the GUI module, obtain the connection weight value parameter of the central value parameter of obfuscation layer and width value parameter information, output layer on the other hand from fuzzy neural network parameter training module, carry out the reschedule decision-making treatment of semi-conductor manufacturing system, and the information result of reschedule decision-making treatment is outputed to fuzzy neural network parameter training module and GUI module;
Fuzzy neural network parameter training module receives the output information of the reschedule decision-making treatment of fuzzy neural network decision-making module on the one hand, obtain the historical data of fuzzy neural network training initiation parameter set point information and reschedule decision-making on the one hand from the GUI module, central value to the obfuscation layer of fuzzy neural network decision-making module, the connection weight value parameter of width value parameter and output layer is carried out training managing, and with the central value of the obfuscation layer optimized, the connection weight value parameter of width value parameter and output layer is transferred to the fuzzy neural network decision-making module, and is transferred to the GUI module.
2, the reschedule decision system of semi-conductor manufacturing system according to claim 1, it is characterized in that, described fuzzy neural network decision-making module, comprise: input layer submodule, obfuscation layer processing sub, rules layer processing sub, normalization layer processing sub, output layer processing sub and reschedule decision-making output sub-module, wherein:
The input layer submodule is responsible for obtaining m production status and the interfere information that receives semi-conductor manufacturing system from the GUI module, specifying information comprises: the stable β c of the equipment group average queue length L that is disturbed, semi-conductor manufacturing system scheduling, the equipment group of being disturbed average load η, the equipment group workpiece postorder average relaxation time ts that is disturbed and the interference parameter T of semi-conductor manufacturing system etc. relatively, and export to obfuscation layer processing sub;
Obfuscation layer processing sub obtains production status and interfere information from the input layer submodule on the one hand, obtain the central value parameter c and the width value parameter σ of Gaussian function on the one hand from fuzzy neural network parameter training module, production status and interfere information at each input, central value parameter c and width value parameter σ according to 3 groups of correspondences, adopt 3 Gaussian functions to carry out blurring mapping respectively and handle, obtain 3m Fuzzy processing output data and output to the rules layer processing sub;
The rules layer processing sub is responsible for handling carrying out regular product from the data of obfuscation layer processing sub input, at 3m Fuzzy processing data that obtain from the Fuzzy processing submodule, at first according to the rule of from each production status and pairing 3 the Fuzzy processing data of interfere information, getting data, select m Fuzzy processing data, then m the data of selecting being carried out product handles, according to this picking rule, the rules layer processing sub forms 3 altogether
mThe output data that individual regular product is handled also outputs to normalization layer processing sub;
Normalization layer processing sub is responsible for the data of rules layer processing sub output are carried out normalized, at 3 of rules layer processing sub output
mThe output data that individual regular product is handled is at first sued for peace to this 3m output data, and the value that obtains of will suing for peace is then removed each output data, obtains 3 of normalization layer processing sub
mIndividual normalized data output also is input to the output layer processing sub;
The output layer processing sub obtains the normalization output data from normalization layer processing sub on the one hand, obtains output layer weighting parameter w data from fuzzy neural network parameter training module on the one hand, to 3 of normalization layer processing sub output
mIndividual normalized data, it is multiplied each other with the output layer training weighting parameter w that obtains from fuzzy neural network parameter training module and be weighted, the result of weighting adopts 3 tansig functions to carry out the output layer functional transformation and handles, obtain the processing output variable data of 3 output layer processing sub, the data variable of handling is outputed to fuzzy neural network parameter training module and reschedule decision-making output sub-module;
After reschedule decision-making output sub-module obtains 3 output variable data from the output layer processing sub, the variable data of absolute value maximum wherein as the reschedule decision-making output of optimizing, is outputed to the GUI module.
3, the reschedule decision system of semi-conductor manufacturing system according to claim 1 is characterized in that, described fuzzy neural network parameter training module comprises: parameter input submodule, training parameter updating submodule and parameter output sub-module, wherein:
Parameter input submodule is responsible for obtaining the initial value of weighting parameter w of initial value, output layer of the central value parameter c of frequency of training parameter, initialization study rate parameter, obfuscation layer of fuzzy neural network training and width value parameter σ and the historical data of reschedule decision-making from the GUI module, obtain the output data of weighting function conversion process from the output layer processing sub of fuzzy neural network decision-making module, and be input to the training parameter updating submodule;
The training parameter updating submodule is responsible for 3 of 3m central value parameter c of the obfuscation layer in the fuzzy neural network decision-making module and width value parameter σ, output layer
M+1Individual weighting parameter w is optimized training managing, this submodule is according to current central value parameter c, width value parameter σ and the weighting parameter w of output data, fuzzy neural network decision-making module of the weighting function conversion process that obtains, employing has the method for the gradient decline optimizing of supervision to calculate 3m central value parameter c of obfuscation layer in the fuzzy neural network decision-making module and the width value parameter σ value along the rate of change of gradient direction respectively, calculates 3 of output layer in the fuzzy neural network decision-making module
M+1The value along the rate of change of gradient direction of individual weighting parameter w correspondence, and add up with current central value parameter c, width value parameter σ and weighting parameter w value, thereby obtain the value of central value parameter c, width value parameter σ and the weighting parameter w of training optimization, realize that by aforesaid operations the optimization training managing value of acquisition outputs to the parameter output sub-module to obfuscation layer in the fuzzy neural network decision-making module and the training managing of output layer Parameter Optimization;
The parameter output sub-module is exported to obfuscation layer processing sub and GUI module in the fuzzy neural network decision-making module from central value parameter c, the width value parameter σ of the obfuscation layer of training parameter updating submodule acquisition optimization; Output layer weighting parameter w from the acquisition of training parameter updating submodule is optimized exports to output layer processing sub and GUI module in the fuzzy neural network decision-making module.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN200810203013A CN100595700C (en) | 2008-11-20 | 2008-11-20 | Reschedule decision making system of semiconductor manufacturing system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN200810203013A CN100595700C (en) | 2008-11-20 | 2008-11-20 | Reschedule decision making system of semiconductor manufacturing system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101424919A true CN101424919A (en) | 2009-05-06 |
CN100595700C CN100595700C (en) | 2010-03-24 |
Family
ID=40615575
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN200810203013A Expired - Fee Related CN100595700C (en) | 2008-11-20 | 2008-11-20 | Reschedule decision making system of semiconductor manufacturing system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN100595700C (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101893888A (en) * | 2010-07-30 | 2010-11-24 | 上海交通大学 | Performance parameter information detection system of complicated manufacturing system |
CN101604409B (en) * | 2009-07-15 | 2011-11-09 | 北京化工大学 | Dispatching method applied to multi-enterable complex manufacturing system |
CN102393687A (en) * | 2011-09-15 | 2012-03-28 | 中国科学院沈阳自动化研究所 | Method for limiting distribution and scheduling for solving machine changing problem |
CN106444649A (en) * | 2016-07-05 | 2017-02-22 | 同济大学 | Semiconductor production line closed-loop scheduling control method |
CN110675005A (en) * | 2019-10-15 | 2020-01-10 | 深圳埃克斯工业自动化有限公司 | Intelligent decision-making method based on artificial intelligence technology and ROPN technology |
CN111539556A (en) * | 2020-04-02 | 2020-08-14 | 北京北方华创微电子装备有限公司 | Wafer scheduling method and device |
CN112650179A (en) * | 2020-12-23 | 2021-04-13 | 同济大学 | Dynamic scheduling method of semiconductor manufacturing system |
CN114047706A (en) * | 2022-01-11 | 2022-02-15 | 北京航空航天大学 | Fuzzy dynamic integral sliding mode control method for reentrant industrial manufacturing system |
CN115097737A (en) * | 2022-08-24 | 2022-09-23 | 北京航空航天大学 | Multi-level regulation and control method capable of being re-entered into manufacturing system |
-
2008
- 2008-11-20 CN CN200810203013A patent/CN100595700C/en not_active Expired - Fee Related
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101604409B (en) * | 2009-07-15 | 2011-11-09 | 北京化工大学 | Dispatching method applied to multi-enterable complex manufacturing system |
CN101893888A (en) * | 2010-07-30 | 2010-11-24 | 上海交通大学 | Performance parameter information detection system of complicated manufacturing system |
CN101893888B (en) * | 2010-07-30 | 2012-05-30 | 上海交通大学 | Performance parameter information detection system of complicated manufacturing system |
CN102393687A (en) * | 2011-09-15 | 2012-03-28 | 中国科学院沈阳自动化研究所 | Method for limiting distribution and scheduling for solving machine changing problem |
CN102393687B (en) * | 2011-09-15 | 2013-05-22 | 中国科学院沈阳自动化研究所 | Method for limiting distribution and scheduling for solving machine changing problem |
CN106444649B (en) * | 2016-07-05 | 2019-01-11 | 同济大学 | A kind of semiconductor production line closed loop dispatch control method |
CN106444649A (en) * | 2016-07-05 | 2017-02-22 | 同济大学 | Semiconductor production line closed-loop scheduling control method |
CN110675005A (en) * | 2019-10-15 | 2020-01-10 | 深圳埃克斯工业自动化有限公司 | Intelligent decision-making method based on artificial intelligence technology and ROPN technology |
CN110675005B (en) * | 2019-10-15 | 2021-12-07 | 埃克斯工业(广东)有限公司 | Intelligent decision-making method based on artificial intelligence technology and ROPN technology |
CN111539556A (en) * | 2020-04-02 | 2020-08-14 | 北京北方华创微电子装备有限公司 | Wafer scheduling method and device |
CN112650179A (en) * | 2020-12-23 | 2021-04-13 | 同济大学 | Dynamic scheduling method of semiconductor manufacturing system |
CN114047706A (en) * | 2022-01-11 | 2022-02-15 | 北京航空航天大学 | Fuzzy dynamic integral sliding mode control method for reentrant industrial manufacturing system |
CN114047706B (en) * | 2022-01-11 | 2022-04-12 | 北京航空航天大学 | Fuzzy dynamic integral sliding mode control method for reentrant industrial manufacturing system |
CN115097737A (en) * | 2022-08-24 | 2022-09-23 | 北京航空航天大学 | Multi-level regulation and control method capable of being re-entered into manufacturing system |
CN115097737B (en) * | 2022-08-24 | 2022-11-08 | 北京航空航天大学 | Multi-level regulation and control method capable of being re-entered into manufacturing system |
Also Published As
Publication number | Publication date |
---|---|
CN100595700C (en) | 2010-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN100595700C (en) | Reschedule decision making system of semiconductor manufacturing system | |
CN107767022B (en) | Production data driven dynamic job shop scheduling rule intelligent selection method | |
Wang et al. | Secondary forecasting based on deviation analysis for short-term load forecasting | |
Li et al. | Adaptive dispatching rule for semiconductor wafer fabrication facility | |
WO2016169287A1 (en) | Productivity allocation method for mixed production line | |
CN102819772B (en) | Power matching network builds material requirements Forecasting Methodology and device | |
CN107203891A (en) | A kind of automatic many threshold values characteristic filter method and devices | |
CN104376371B (en) | A kind of distribution based on topology is layered load forecasting method | |
CN105654240A (en) | Machine tool manufacturing system energy efficiency analysis method | |
CN109325880A (en) | A kind of Mid-long term load forecasting method based on Verhulst-SVM | |
CN107918368A (en) | The dynamic prediction method and equipment of iron and steel enterprise's coal gas yield and consumption | |
CN103853939A (en) | Combined forecasting method for monthly load of power system based on social economic factor influence | |
CN110941909A (en) | Method and system for identifying and calculating dominance of parameters of photovoltaic power station and wind power plant | |
CN114140024A (en) | Power plant loss analysis system based on power utilization model | |
CN104299043A (en) | Ultra-short-term load prediction method of extreme learning machine | |
CN112736902B (en) | STL decomposition-based time series short-term power load prediction method | |
Kaku et al. | A mathematical model for converting conveyor assembly line to cellular manufacturing | |
CN104698838B (en) | Based on the fuzzy scheduling rule digging method that domain dynamic is divided and learnt | |
CN106327053B (en) | Construction method of weaving process recommendation model based on multi-mode set | |
CN104462329A (en) | Operation process digging method suitable for diversified environment | |
Xu et al. | Accelerated and adaptive power scheduling for more electric aircraft via hybrid learning | |
CN106033217B (en) | Dispatching system and work dispatching method | |
CN113705923A (en) | Load prediction method based on intelligent algorithm | |
Li et al. | Scheduling strategy of semiconductor production lines with remaining cycle time prediction | |
CN112836876A (en) | Power distribution network line load prediction method based on deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
C17 | Cessation of patent right | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20100324 Termination date: 20121120 |