CN112308298B - Multi-scenario performance index prediction method and system for semiconductor production line - Google Patents

Multi-scenario performance index prediction method and system for semiconductor production line Download PDF

Info

Publication number
CN112308298B
CN112308298B CN202011108406.3A CN202011108406A CN112308298B CN 112308298 B CN112308298 B CN 112308298B CN 202011108406 A CN202011108406 A CN 202011108406A CN 112308298 B CN112308298 B CN 112308298B
Authority
CN
China
Prior art keywords
scene
prediction
load
production line
under
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.)
Active
Application number
CN202011108406.3A
Other languages
Chinese (zh)
Other versions
CN112308298A (en
Inventor
乔非
高陈媛
刘鹃
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tongji University
Original Assignee
Tongji University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tongji University filed Critical Tongji University
Priority to CN202011108406.3A priority Critical patent/CN112308298B/en
Publication of CN112308298A publication Critical patent/CN112308298A/en
Application granted granted Critical
Publication of CN112308298B publication Critical patent/CN112308298B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Abstract

The invention relates to a multi-scenario performance index prediction method and a multi-scenario performance index prediction system for a semiconductor production line, wherein the method comprises the following steps: the production scene quantitative division module is driven by data, quantitatively maps the product value, the average product processing period and the utilization rate of each device of the production line, and divides the production line into three scenes of light load, normal load and heavy load; the main prediction network construction module is used for combining a deep neural network algorithm and semiconductor production line performance prediction by taking the divided normal load data as sample data to construct a prediction network under a normal load scene; and the multi-scene prediction model building module is used for applying the idea of transfer learning to production line prediction and building networks under light load and heavy load scenes according to the built main prediction network so as to form the multi-scene prediction model. Compared with the prior art, the method quantitatively divides the production line scene, can more accurately predict the performance indexes of a plurality of production lines under different load scenes, and can be used on line.

Description

Multi-scenario performance index prediction method and system for semiconductor production line
Technical Field
The invention relates to the technical field of intelligent manufacturing, in particular to a multi-scenario performance index prediction method and system for a semiconductor production line.
Background
The manufacturing industry is the main body of the national economy, is an engine for the high-speed growth of the economy and is the important embodiment of the comprehensive national force. In order to ensure competitiveness, manufacturing enterprises need to perform dynamic scheduling management quickly and reasonably allocate resources. The real-time performance index prediction result can provide a basis for production decision and evaluation, so that the performance index of the production line which is achieved under a certain scheduling mode is required to be known quickly.
At present, three methods, namely mathematical modeling, simulation modeling and machine learning modeling, are mainly used as prediction model modeling methods. Firstly, the semiconductor manufacturing is a complex process with high nonlinearity, multivariable coupling and random uncertainty, and the establishment of a mathematical model is extremely difficult; secondly, the traditional simulation modeling method needs to invest a large amount of capital, is long in prediction time, and is poor in capability of adapting to a dynamic environment. In order to overcome the limitations of mathematical models and traditional simulation prediction models, expert scholars introduce machine learning techniques to study data-driven production line prediction methods.
Chinese patent "Performance prediction method for semiconductor production line dynamic scheduling" (patent publication No. CN 103310285A) invented a semiconductor production line scheduling system based on extreme learning machine, the method collects semiconductor production line historical data to establish sample set and test sample set; constructing a prediction model by adopting an extreme learning machine method; and testing the network performance of the prediction model by using the test sample, and outputting the prediction result after normalizing. Chinese patent "a production line spare part damage rate prediction system based on LR" (patent publication No. CN 108898254A) invented a production line spare part damage rate prediction system, and this method collects the operation record of the equipment through the sensor, the current per hour, the average voltage value, the length of operation and other data; manually acquiring the time for installation of the spare part for replacement; obtaining a model through LR training; and predicting the current loss rate of the equipment in the corresponding category through the model.
It can be seen that most of the existing methods only aim at a single production scene, a semiconductor production line has a plurality of uncertain dynamic production factors such as feeding change, production line load change and the like, the existing identification and prediction methods are difficult to adapt to a dynamic production environment, most production line performance predictions focus on single performance indexes such as productivity and the like, and meanwhile, the prediction of multiple performance indexes for predicting equipment utilization rate and productivity is less. Therefore, a timely performance index prediction method capable of adapting to various production line states is needed, and the method comprises three components of scene division, construction of a main prediction network and construction of each scene network to form a new model, so that the accuracy and adaptability of production line performance prediction are improved, and more powerful support is provided for scheduling decisions. There are no documents and patents related to the above-mentioned production line performance prediction method.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a multi-scenario performance index prediction method and system for a semiconductor production line.
The purpose of the invention can be realized by the following technical scheme:
a multi-scenario performance index prediction method for a semiconductor production line comprises the following steps:
step 1: carrying out quantitative mapping on historical data of a production line, and carrying out data division according to scene quantification respectively corresponding to light load scenes, normal load scenes and heavy load scenes;
step 2: taking the divided data under the normal load scene as sample data, and combining a deep neural network algorithm with the performance prediction of a semiconductor production line to construct a prediction network under the normal load scene;
and step 3: based on the prediction network under the normal load scene, the prediction networks under the light load scene and the heavy load scene are further constructed, and the prediction networks under the light load scene, the normal load scene and the heavy load scene form a multi-scene prediction model;
and 4, step 4: and after the online data of the production line are divided into different scene results according to the threshold value, selecting a corresponding network from the multi-scene prediction model for prediction to obtain a performance index prediction result.
Further, the production line historical data includes input quantity and output quantity, wherein the input quantity includes: sampling days, average processing period of products, average daily moving steps, total number of products in process, queue length of each buffer area, total queue length of the buffer areas and total moving steps; the output quantity comprises: average processing period of products and utilization rate of each device.
Further, the process of constructing the prediction network under the normal load scenario in step 2 includes the following sub-steps:
step 201: coding the scheduling rule symbols in the sample data, taking the coding as input to carry out normalization processing, and limiting the number to [0, 1%]Within the interval, the normalization formula is [ x-min (x)i)]/[max(xi)-min(xi)]Wherein x refers to an input variable, and i is a sample variable;
step 202: adopting a deep neural network to construct a prediction network under a normal load scene; combining a deep learning algorithm with the performance prediction of a production line, and obtaining the number of proper hidden layers, the number of neurons of each hidden layer and excitation functions of each layer by adopting a grid search method;
step 203: testing the network performance of the prediction network by adopting a test sample, comparing an output value corresponding to a prediction result obtained by the test sample after reverse normalization processing with an output value of the test sample, and judging whether the accuracy requirement is met;
step 204: if the prediction accuracy of the test result can meet the accuracy requirement, the prediction network under the normal load scene is successfully established, if the prediction accuracy of the test result cannot meet the accuracy requirement, the step 202 is returned, the number of hidden layers, the number of neurons of each hidden layer and the excitation function of each layer are reselected, and the model is trained again.
Further, the scheduling rule symbol in step 201 is encoded in a one-hot encoding manner, so that it can be received by the network.
Further, the inputs of the deep learning algorithm in step 202 are:
for a given training set of k different samples:
Xk＝{Sk,Dk,Ruk,Pk|Sk∈R∝,Dk∈Rβ,Ruk∈Rλ,Pk∈Rm}
wherein the content of the first and second substances,
Figure BDA0002727737330000031
the state is the state of an intelligent workshop system;
Figure BDA0002727737330000032
feeding information for an intelligent workshop;
Figure BDA0002727737330000033
scheduling rules for the intelligent workshop;
Figure BDA0002727737330000034
Figure BDA0002727737330000035
for the current state S of the intelligent workshop systemkCurrent feeding information DkCurrent scheduling rule RukPerformance index after 1 day under the circumstances;
the output of the deep learning algorithm in step 202 is oiWhere i is 1, …, m1, m1 indicates the dimension of the output value o, i.e. there are m1 outputs.
Further, the accuracy requirement in step 204 is based on the root mean square error, which is described by the formula:
Figure BDA0002727737330000036
where m is the number of output variables, P is the input performance index, Y is the predicted performance index,
Figure BDA0002727737330000037
is the output variable sample number.
Further, the process of constructing the prediction network under the light load and heavy load scenarios in step 3 includes the following sub-steps:
step 301: assigning the number of neurons of an input layer of the prediction network under a normal load scene to the prediction network under light load and heavy load scenes;
step 302: assigning the number of neurons of each hidden layer except the last hidden layer of the prediction network under the normal load scene to the prediction network under the light load scene and the heavy load scene as the number of neurons of the corresponding front n-1 layer;
step 303: assigning the weight threshold information of each hidden layer except the last hidden layer of the prediction network under the normal load scene to the prediction network under the light load scene and the heavy load scene as initialization data;
step 304: assigning excitation functions of all hidden layers except the last hidden layer of the prediction network under a normal load scene to the prediction network under light load and heavy load scenes;
step 305: and (3) adjusting the number of neurons of the last layer of hidden layer by a network search method, inputting the data under the light load and heavy load scene divided in the step (1) into a prediction network under the light load and heavy load scene for training, and obtaining the prediction network under the light load and heavy load scene after the training is finished.
Further, the multi-scenario prediction model building module comprises:
constructing a light-load scene prediction network, namely changing the number of neurons of the last layer of hidden layer on the basis of a main prediction network by taking light-load scene data as a sample, and constructing the light-load scene prediction network;
constructing a heavy-load scene prediction network, namely changing the number of neurons of a last layer of hidden layer on the basis of a main prediction network by taking heavy-load scene data as a sample, and constructing the heavy-load scene prediction network;
and (4) network combination, namely combining the three prediction networks. And forming a multi-scene prediction model.
Further, the step 1 comprises the following sub-steps:
step 101: collecting historical data of a production line for quantitative mapping under the same sampling days, and drawing a line graph;
step 102: and carrying out data division according to curve change rules in each line graph and scene quantification corresponding to the light load scene, the normal load scene and the heavy load scene respectively.
Further, in the light load scenario in step 102, in the corresponding curve, the starting point is a point corresponding to the total in-process value when the equipment utilization rate starts to enter the stationary stage, and the end point is a point corresponding to the total in-process value when the slope of the average processing cycle curve of the product is maximum;
under a normal load scene, in the corresponding curve, the starting point is the point corresponding to the product value when the slope of the average processing period curve of the product is maximum, and the end point is the point corresponding to the product value when the slope of the average processing period curve of the product is minimum;
under a heavy load scene, the starting point in the corresponding curve is the point corresponding to the total product value when the slope of the average processing cycle curve of the product is minimum.
The invention also provides a system adopting the semiconductor production line-oriented multi-scenario performance index prediction method, which comprises the following steps:
the main prediction network construction module is used for combining a deep neural network algorithm and semiconductor production line performance prediction by taking the divided data under the normal load scene as sample data to construct a prediction network under the normal load scene;
the multi-scene prediction model construction module is used for further constructing prediction networks under light load and heavy load scenes based on the prediction networks under normal load scenes, and forming the prediction networks under the light load scenes, the normal load scenes and the heavy load scenes into a multi-scene prediction model;
and the production scene quantitative dividing module is used for carrying out quantitative mapping on the production line historical data and carrying out data division according to the scene quantification respectively corresponding to the light load scene, the normal load scene and the heavy load scene.
Compared with the prior art, the invention has the following advantages:
1. the invention divides the production environment into three scenes of light load, normal load and heavy load by quantitatively dividing the production scene, so that the prediction method is suitable for the dynamic production environment;
2. the invention combines the deep learning algorithm with the semiconductor production line prediction, and establishes the neural network prediction model by respectively taking the divided light load, normal load and heavy load data as sample data, thereby being capable of responding to the real-time change in the system in time and reducing the need of rescheduling. The system is regarded as a black box, and the provided modeling method excavates useful knowledge in historical data and off-line data which can be obtained by using a production line, so that real-time on-line optimization control is realized;
3. the invention applies the transfer learning idea to the training process of the multi-scenario network, and constructs the network under the light load and heavy load scenarios according to the constructed main prediction network, thereby forming a multi-scenario prediction model, reducing the network training time and improving the accuracy of the production linear performance prediction;
4. the method can predict the performance indexes of a plurality of production lines one day later and provide more data support for production decision.
Drawings
FIG. 1 is a schematic structural view of the present invention;
FIG. 2 is a diagram of a MiniFAB model structure according to an embodiment of the present invention;
fig. 3 is a schematic diagram of the average processing cycles of EDD, SRPT, and EDD products respectively according to the scheduling rules of the sampling Ma to Me devices on the 26 th day in the embodiment of the present invention;
fig. 4 is a diagram illustrating quantities of work in process of EDD, SRPT, EDD, respectively, of the device scheduling rules from Ma to Me sampled on day 26 in the embodiment of the present invention;
fig. 5 is a schematic diagram illustrating utilization rates of EDD, SRPT, EDD, and EDD of the scheduling rules of the equipment from Ma to Me sampled on the 26 th day in the embodiment of the present invention;
FIG. 6 is a schematic diagram illustrating a comparison of root mean square errors of scene models, non-divided scene models, and multi-scene multi-performance index prediction models in terms of productivity according to an embodiment of the present invention;
FIG. 7 is a schematic diagram illustrating a comparison of root mean square errors of scene models, non-divided scene models, and multi-scene multi-performance index prediction models in the Ma device utilization rate according to an embodiment of the present invention;
FIG. 8 is a schematic diagram illustrating a comparison of root mean square errors of scene models, non-divided scene models, and multi-scene multi-performance index prediction models in the Mb device utilization according to an embodiment of the present invention;
FIG. 9 is a schematic diagram illustrating a comparison of root mean square errors of each scene model, an undivided scene model and a multi-scene multi-performance index prediction model in the embodiment of the present invention in the utilization rate of the Mc device;
FIG. 10 is a schematic diagram illustrating a comparison of root mean square errors of scene models, non-divided scene models, and multi-scene multi-performance index prediction models in the Md device utilization according to an embodiment of the present invention;
fig. 11 is a schematic diagram of comparing root mean square errors of each scene model, an undivided scene model and a multi-scene multi-performance index prediction model in the Me device utilization rate in the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.
DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION
As shown in FIG. 1, the invention provides a multi-scenario multi-performance index prediction method for a semiconductor production line, which comprises a production scenario quantitative division module, a main prediction network construction module and a multi-scenario prediction model construction module. The production scene quantitative dividing module is driven by data, quantitatively maps the product value of the production line, the average processing period of products and the utilization rate of each device, and divides the production line into three scenes of light load, normal load and heavy load; the main prediction network construction module is used for combining a deep neural network algorithm and semiconductor production line performance prediction by taking the divided normal load data as sample data to construct a prediction network under a normal load scene; the multi-scene prediction model building module comprises: and applying the idea of transfer learning to production line prediction, and constructing a network under light load and heavy load scenes according to the constructed main prediction network so as to form a multi-scene prediction model.
This example is described by taking the MiniFab model as an example. The MiniFab is an experimental simulation model of a semiconductor production process proposed by doctor Karl Kempf, a chief scientist of the manufacturing system of Intel corporation, and consists of 5 devices and 6 procedures, and the model structure is shown in fig. 2.
The method for predicting the multi-scene multi-performance index for the semiconductor production line applies a MiniFab model as an implementation object, and comprises the following working processes:
step 1, defining sample information Xk＝{Sk,Dk,Ruk,Pk|Sk∈R∝,Dk∈Rβ,Ruk∈Rλ,Pk∈RmTherein of
Figure BDA0002727737330000071
Describing the state of the intelligent workshop system;
Figure BDA0002727737330000072
describing feeding information of the intelligent workshop;
Figure BDA0002727737330000073
describing an intelligent workshop scheduling rule;
Figure BDA0002727737330000074
Figure BDA0002727737330000075
described in the current state SkCurrent feeding mode DkCurrent scheduling rule RukPerformance index after the next 1 day. Selecting workshop productivity PROD and equipment Ma utilization rate UaMb utilization U of equipmentbUtilization rate U of Mc of equipmentcMd utilization rate U of equipmentdMe utilization rate U of equipmenteThe six performance indexes are used as prediction targets, and the production line system state S in the sample informationkAnd feeding information DkScheduling rules RukThe definitions are given in tables 1, 2 and 3. Selecting 10% as a test sample set;
TABLE 1 System State S in MiniFab sample informationkDefinition of
Name of field Description of the invention Data type
Day Number of days of sampling Shaping machine
MCT Average processing cycle of product Floating point type
MDayMOV Average daily moving step number Shaping machine
WIP Number of products being processed Shaping machine
Mab_Queue MaMb buffer Captain Shaping machine
Mcd_Queue McMd buffer Captain Shaping machine
Me_Queue Me buffer Captain Shaping machine
Total_Queue Total queue length of buffer Shaping machine
Total_MOV Total number of moving steps Shaping machine
TABLE 2 feeding information D in MiniFab sample informationkDefinition of
Name of field Description of the invention Data type
x_Release Daily average charge of product x Shaping machine
y_Release Daily average charge of product y Shaping machine
z_Release Daily average charge of product z Shaping machine
TABLE 3 scheduling rules Ru in MiniFab sample informationkDefinition of
Name of field Description of the invention Data type
Ma_Rule Scheduling rules for device Ma Character type
Mb_Rule Scheduling rules for device Mb Character type
Mc_Rule Scheduling rules for device Mc Character type
Md_Rule Scheduling rules for device Md Character type
Me_Rule Scheduling rules of device Me Character type
In this embodiment, the number of sampling days is 1 to 30, the daily average feeding of the product a, the product b, and the product c is increased from 1 to 19, the Ma and Mb scheduling rules select EDD or SRPT, the Mc and Md scheduling rules select EDD or SRPT or CR, the Me scheduling rules select EDD or SRPT, and the utilization rate and the productivity performance index of each device after 1 day are obtained.
Step 2, averaging the product processing period MCT, the product number WIP and the equipment Ma utilization rate UaMb utilization U of equipmentbUtilization rate U of Mc of equipmentcMd utilization rate U of equipmentdMe utilization rate U of equipmenteThe average total material feeding amount of the product day is mapped through the sampling days, and each material feeding amount in the graph is observedAnd quantitatively dividing light load scenes, normal load scenes and heavy load scenes of the production line according to the corresponding relation of curve change rules.
Scene 1: and (5) carrying light scenes. Under the scene, the utilization rate U of the equipment MaaMb utilization U of equipmentbUtilization rate U of Mc of equipmentcMd utilization rate U of equipmentdMe utilization rate U of equipmenteAnd the light load scene is quantitatively divided according to the WIP curve corresponding to the starting point and the end point of the scene.
Scene 2: and (5) normal scenes. Under the scene, the utilization rate of each device is stable, the MCT fluctuates, the starting point of the scene is the maximum slope point of the MCT curve, the end point of the scene is the minimum slope point of the MCT curve, and the normal scene is divided quantitatively according to the WIP curve corresponding to the starting point and the end point of the scene.
Scene 3: and (5) overloading the scene. Under the scene, the utilization rate of each device is stable, the MCT curve is always raised without descending, the WIP curve is raised, the starting point of the scene is the minimum point of the slope of the MCT curve, and the heavy load scene is divided quantitatively according to the WIP curve corresponding to the starting point and the end point of the scene.
And quantitatively obtaining WIP values of the three scene partitions according to the initial abscissa corresponding to the three scenes, and laying a foundation for the subsequent online data to partition the scenes according to the specific WIP values.
1a) Drawing a mapping chart: under the same sampling days, selecting the average processing period MCT, the WIP and the equipment Ma utilization rate U of the products under a plurality of scheduling rulesaMb utilization U of equipmentbUtilization rate U of Mc of equipmentcMd utilization rate U of equipmentdMe utilization rate U of equipmenteDrawing a line graph. Here, the EDD, SRPT, EDD, and EDD are drawn by taking the sampling days of 26, Ma, Mb, Mc, Md, and Me device scheduling rules as examples, respectively, fig. 3, 4, and 5;
1b) and (3) quantitative scene division: when the sample id is 6, the equipment utilization rate of fig. 5 tends to be stable, and the WIP value of fig. 4 is approximately 60; when the sample id is 8, the slope of the MCT waving phase of fig. 3 is maximum, when the WIP of fig. 4 waves up and down at 200; when the sample id is 16, MCT only rises and does not fall, and WIP fluctuates at 750. The scene division definition index is WIP, and therefore, it is determined as a light-load scene when WIP is [60,200 ], as a normal scene when WIP is [200,750), and as a heavy-load scene when WIP is [750, ∞).
And 3, coding the scheduling rule symbol to enable the scheduling rule symbol to be received by the network, carrying out normalization processing on the original data through min-max, removing the dimension of the original data, and changing the original data into a decimal between [0,1 ]. The coding table is shown in table 4.
TABLE 4 scheduling rules coding Table
Scheduling rules Encoding
EDD [0,1]
SRPT [1,0]
CR [1,1]
And 4, adopting a DNN algorithm to input the data divided into the normal load scene, taking the feeding information D, MaMb scheduling rule, the McMd scheduling rule, the Me scheduling rule, the number of products in process in the system state S, the MaMb buffer area queue length, the McMd buffer area queue length, the Me buffer area queue length, the total buffer area queue length and the total moving step number as input, taking P representing the performance index as output, and obtaining the matching relation between the feeding information, the scheduling rule and the system state and the performance index, namely a main prediction model.
The specific process of learning and obtaining the main prediction model by adopting the deep neural network algorithm comprises the following steps:
2a) weighting matrix W of networkηV and threshold matrix thetaqGamma initialization, with the value of [ -1,1 [ ]]A random number in between. A training number counter h is set.
2b) Selecting the kth pair of data samples, and calculating the output result of each hidden layer:
Figure BDA0002727737330000091
2c) determining an output result of an output layer:
Figure BDA0002727737330000092
2d) calculating correction error of each neuron of output layer and hidden layer
Figure BDA0002727737330000093
Figure BDA0002727737330000094
2e) Calculating layer errors of the eta layer hidden layer
Figure BDA0002727737330000095
Figure BDA0002727737330000096
2f) According to the selected gradient descent Optimizer as Adadelta Optimizer
2g) Setting a loss function:
Figure BDA0002727737330000097
where m is the number of output variables, P is the input performance index, Y is the predicted performance index,
Figure BDA0002727737330000098
is the output variable sample number.
2h) Setting the learning rate to be 0.05, and setting the network learning times to be 50000;
2i) and calculating an output total error according to a loss function defined in the network, jumping out of a training program if the loss function value is smaller than a set numerical value, and adjusting the weight and the threshold value of each layer to continue training if the loss function value is not satisfied.
2j) Checking whether a training frequency counter h reaches a network set value, if not, judging that the training frequency counter h + 1; and if so, jumping out of the program, and resetting the number of the hidden layers, the number of the neurons and the excitation function.
And finally, determining the number of the network layers of the network A to be 6 layers by a grid search method, wherein the network structure is 10 x 9 x 8 x 6, the excitation function of the hidden layers 1-4 is tanh, and the excitation function of the output layer is sigmoid.
And 5, constructing a light-load scene network and a heavy-load scene network by a transfer learning method according to the network information of the main prediction network, wherein the steps are as follows:
step K1: assigning the number of neurons in the input layer of the main prediction network to a new light load/heavy load network;
step K2: assigning the number of neurons of each hidden layer of the main prediction network except the last hidden layer to a new light load/heavy load network as the number of neurons of the first n-1 layers;
step K3: assigning the weight threshold information of each hidden layer of the main prediction network except the last hidden layer to a new light load/heavy load network as initialization data;
step K4: assigning the excitation function of each hidden layer of the main prediction network except the last hidden layer to a new light load/heavy load network;
step K5: and adjusting the number of neurons in the last layer of hidden layer by a grid search method, and inputting light load/heavy load scene data for training.
In the embodiment, the main prediction network A has 4 hidden layers, the first 3 hidden layers of the network A are transferred to the first 3 hidden layers of the network B and the network C, the network B and the network C are obtained through training by adjusting the number of neurons of the last hidden layer, the number of the network layers of the network B is finally determined to be 6, the network structure is 10 × 9 × 6, the excitation functions of the 1-4 hidden layers are tanh, and the excitation function of the output layer is sigmoid; the number of the network layers of the network C is 6, the network structure is 10 x 9 x 7 x 6, the excitation functions of the 1-4 layers of the hidden layers are tan h, and the excitation function of the output layer is sigmoid. The three networks are combined into a whole to form a multi-scene multi-index prediction model, and the model structure is shown in table 5.
TABLE 5 Multi-scenario Multi-index prediction model Structure
Figure BDA0002727737330000111
Before the online data enter the model, scene recognition is carried out according to a threshold value, and if the scene is recognized as a normal scene, a main prediction network in the model is selected for prediction; if the light-load scene is identified, selecting a light-load scene prediction network in the model for prediction; and if the model is identified as the heavy-load scene, selecting a heavy-load scene prediction network in the model for prediction.
And respectively applying the non-scene-division model, the single-scene model and the multi-scene multi-performance index prediction model to the MiniFAB model, and comparing the prediction results of the non-scene-division network, the single-scene model and the multi-scene multi-performance index prediction model.
The table for comparing the outputs of the non-scene model and the multi-scene multi-index prediction model with the average value of the test sample is shown in table 6.
TABLE 6 comparison table of average values of test samples and outputs of non-scene model and multi-scene multi-index prediction model
Figure BDA0002727737330000112
In the case of an undifferentiated scene model, namely, a whole network model is obtained by mixed training of data, the highest relative error reaches 3.4%, and the lowest error is 1.0%, in the case of a multi-scene multi-index prediction model, the highest relative error reaches 1.5%, and the lowest relative error is 0.15%, and from the perspective of the relative error of the output average value, the multi-scene multi-index prediction model is superior to the undifferentiated scene model.
The mean root mean square error output values of the single scene model, the non-divided scene model and the multi-scene multi-index prediction model are shown in table 7.
TABLE 7 RMS error mean value table for each model
Figure BDA0002727737330000121
On the whole, the multi-scene multi-index prediction model is higher in precision and better in quality.
The root mean square error of the productivity of the single scene model, the non-divided scene model and the multi-scene multi-index prediction model and the root mean square error of the utilization rate of each device are shown in fig. 6 to 11.
As can be seen from fig. 6 to 11, the light-load single scene model has the largest prediction error for the heavy-load scene and the smallest prediction error for the light-load scene; the prediction error of the heavy-load single scene model to the heavy-load scene is minimum, and the prediction error to the light-load scene is maximum; the prediction error of the normal single scene to the normal scene is minimum, and the prediction error to the heavy load scene is maximum; the non-divided scene model has the largest error for the heavy-load scene and the smallest prediction error for the normal scene; the multi-scene multi-index prediction model provided by the invention almost keeps the same prediction error for three scenes and is positioned at the optimal position of the four models.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (4)

1. A multi-scenario performance index prediction method for a semiconductor production line is characterized by comprising the following steps:
step 1: carrying out quantitative mapping on historical data of a production line, and carrying out data division according to scene quantification respectively corresponding to light load scenes, normal load scenes and heavy load scenes;
step 2: taking the divided data under the normal load scene as sample data, and combining a deep neural network algorithm with the performance prediction of a semiconductor production line to construct a prediction network under the normal load scene;
and step 3: based on the prediction network under the normal load scene, the prediction networks under the light load scene and the heavy load scene are further constructed, and the prediction networks under the light load scene, the normal load scene and the heavy load scene form a multi-scene prediction model;
and 4, step 4: after dividing the online data of the production line into different scene results according to a threshold value, selecting a corresponding network from the multi-scene prediction model for prediction to obtain a performance index prediction result;
the process of constructing the prediction network under the normal load scene in the step 2 comprises the following sub-steps:
step 201: coding a scheduling rule symbol in the sample data, and performing normalization processing by taking the scheduling rule symbol as input;
step 202: adopting a deep neural network to construct a prediction network under a normal load scene; combining a deep learning algorithm with the performance prediction of a production line, and obtaining the number of proper hidden layers, the number of neurons of each hidden layer and excitation functions of each layer by adopting a grid search method;
step 203: testing the network performance of the prediction network by adopting a test sample, comparing an output value corresponding to a prediction result obtained by the test sample after reverse normalization processing with an output value of the test sample, and judging whether the accuracy requirement is met;
step 204: if the prediction precision of the test result can meet the precision requirement, the prediction network under the normal load scene is successfully established, if the prediction precision of the test result cannot meet the precision requirement, the step 202 is returned, the number of hidden layers, the number of neurons of each hidden layer and excitation functions of each layer are selected again, and the model is trained again;
the deep learning algorithm in step 202 has the following inputs:
for a given training set of k different samples:
Xk＝{Sk，Dk，Ruk，Pk|Sk∈R∝，Dk∈Rβ，Ruk∈Rλ，Pk∈Rm}
wherein the content of the first and second substances,
Figure FDA0003508999000000011
the state is the state of an intelligent workshop system;
Figure FDA0003508999000000012
feeding information for an intelligent workshop;
Figure FDA0003508999000000013
scheduling rules for the intelligent workshop;
Figure FDA0003508999000000014
Figure FDA0003508999000000021
for the current state S of the intelligent workshop systemkCurrent feeding information DkCurrent scheduling rule RukPerformance index after 1 day under the circumstances;
the output of the deep learning algorithm in step 202 is oiWhere i is 1, …, m1, m1 indicates the dimension of the output value o, i.e. there are m1 outputs;
the accuracy requirement in step 204 is based on the root mean square error, and the description formula is as follows:
Figure FDA0003508999000000022
where m is the number of output variables, P is the input performance index, Y is the predicted performance index,
Figure FDA0003508999000000023
is the output variable sample number;
the process of constructing the prediction network under the light load scene and the heavy load scene in the step 3 comprises the following sub-steps:
step 301: assigning the number of neurons of an input layer of the prediction network under a normal load scene to the prediction network under light load and heavy load scenes;
step 302: assigning the number of neurons of each hidden layer except the last hidden layer of the prediction network under the normal load scene to the prediction network under the light load scene and the heavy load scene as the number of neurons of the corresponding front n-1 layer;
step 303: assigning the weight threshold information of each hidden layer except the last hidden layer of the prediction network under the normal load scene to the prediction network under the light load scene and the heavy load scene as initialization data;
step 304: assigning excitation functions of all hidden layers except the last hidden layer of the prediction network under a normal load scene to the prediction network under light load and heavy load scenes;
step 305: adjusting the number of neurons in the last hidden layer by a network search method, inputting the data under the light load and heavy load scenes divided in the step 1 into a prediction network under the light load and heavy load scenes for training, and obtaining the prediction network under the light load and heavy load scenes after the training is finished;
the step 1 comprises the following sub-steps:
step 101: collecting historical data of a production line for quantitative mapping under the same sampling days, and drawing a line graph;
step 102: dividing data according to curve change rules in each line graph and scene quantification corresponding to light load scenes, normal load scenes and heavy load scenes respectively, wherein in the corresponding curves, a starting point is a point corresponding to a total product value when the equipment utilization rate starts to enter a stable stage, and an end point is a point corresponding to the total product value when the slope of the curve of the average processing period of the product is maximum;
under a normal load scene, in the corresponding curve, the starting point is the point corresponding to the product value when the slope of the average processing period curve of the product is maximum, and the end point is the point corresponding to the product value when the slope of the average processing period curve of the product is minimum;
under a heavy load scene, the starting point in the corresponding curve is the point corresponding to the total product value when the slope of the average processing cycle curve of the product is minimum.
2. The method as claimed in claim 1, wherein the production line history data includes input quantity and output quantity, wherein the input quantity includes: sampling days, average processing period of products, average daily moving steps, total number of products in process, queue length of each buffer area, total queue length of the buffer areas and total moving steps; the output quantity comprises: average processing period of products and utilization rate of each device.
3. The method as claimed in claim 1, wherein the scheduling rule symbol in step 201 is encoded by using a one-hot encoding method so that it can be received by a network.
4. A system using the semiconductor production line oriented multi-scenario performance index prediction method of claim 1, characterized in that the system comprises:
the main prediction network construction module is used for combining a deep neural network algorithm and semiconductor production line performance prediction by taking the divided data under the normal load scene as sample data to construct a prediction network under the normal load scene;
the multi-scene prediction model construction module is used for further constructing prediction networks under light load and heavy load scenes based on the prediction networks under normal load scenes, and forming the prediction networks under the light load scenes, the normal load scenes and the heavy load scenes into a multi-scene prediction model;
and the production scene quantitative division module is used for carrying out quantitative mapping on the historical data of the production line and carrying out data division according to the scene quantification respectively corresponding to the light load scene, the normal load scene and the heavy load scene.
CN202011108406.3A 2020-10-16 2020-10-16 Multi-scenario performance index prediction method and system for semiconductor production line Active CN112308298B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011108406.3A CN112308298B (en) 2020-10-16 2020-10-16 Multi-scenario performance index prediction method and system for semiconductor production line

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011108406.3A CN112308298B (en) 2020-10-16 2020-10-16 Multi-scenario performance index prediction method and system for semiconductor production line

Publications (2)

Publication Number Publication Date
CN112308298A CN112308298A (en) 2021-02-02
CN112308298B true CN112308298B (en) 2022-06-14

Family

ID=74327797

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011108406.3A Active CN112308298B (en) 2020-10-16 2020-10-16 Multi-scenario performance index prediction method and system for semiconductor production line

Country Status (1)

Country Link
CN (1) CN112308298B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115398251A (en) * 2021-04-30 2022-11-25 华为数字能源技术有限公司 Power semiconductor device's test system, cloud server and test machine
CN115049170B (en) * 2022-08-16 2022-10-25 启东先和丝杠制造有限公司 Method for debugging threading work of threading machine controller
CN116307405B (en) * 2023-05-25 2023-08-04 日照鲁光电子科技有限公司 Diode performance prediction method and system based on production data

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101931609A (en) * 2009-06-22 2010-12-29 Sap股份公司 Layout abiding service-level agreement for multiple-tenant database
USRE46430E1 (en) * 2007-04-20 2017-06-06 Cree, Inc. Illumination control network
CN110247299A (en) * 2019-05-22 2019-09-17 上海大学 A kind of parallel double light feedback semiconductor laser reserve pool computing systems

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
USRE46430E1 (en) * 2007-04-20 2017-06-06 Cree, Inc. Illumination control network
CN101931609A (en) * 2009-06-22 2010-12-29 Sap股份公司 Layout abiding service-level agreement for multiple-tenant database
CN110247299A (en) * 2019-05-22 2019-09-17 上海大学 A kind of parallel double light feedback semiconductor laser reserve pool computing systems

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
"IGBT模块结温预测BP神经网络模型";姚芳;《西北大学学报（自然科学版）》;20170630;第367-374页 *

Also Published As

Publication number Publication date
CN112308298A (en) 2021-02-02

Similar Documents

Publication Publication Date Title
CN112308298B (en) Multi-scenario performance index prediction method and system for semiconductor production line
CN112561156A (en) Short-term power load prediction method based on user load mode classification
CN110163429B (en) Short-term load prediction method based on similarity day optimization screening
CN110674999A (en) Cell load prediction method based on improved clustering and long-short term memory deep learning
CN110135635B (en) Regional power saturated load prediction method and system
CN110571792A (en) Analysis and evaluation method and system for operation state of power grid regulation and control system
CN111008504B (en) Wind power prediction error modeling method based on meteorological pattern recognition
CN110689162B (en) Bus load prediction method, device and system based on user side classification
CN111008726B (en) Class picture conversion method in power load prediction
CN113193551B (en) Short-term power load prediction method based on multi-factor and improved feature screening strategy
CN112947300A (en) Virtual measuring method, system, medium and equipment for processing quality
CN112990587A (en) Method, system, equipment and medium for accurately predicting power consumption of transformer area
CN113255900A (en) Impulse load prediction method considering improved spectral clustering and Bi-LSTM neural network
CN111985845B (en) Node priority optimization method of heterogeneous Spark cluster
CN110782083B (en) Aero-engine standby demand prediction method based on deep Croston method
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
CN116520772A (en) Sheet part numerical control machining quality prediction method based on multitasking transfer learning
CN112508286A (en) Short-term load prediction method based on Kmeans-BilSTM-DMD model
CN114819395A (en) Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model
CN116826745B (en) Layered and partitioned short-term load prediction method and system in power system background
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
Wang et al. Cloud computing and extreme learning machine for a distributed energy consumption forecasting in equipment-manufacturing enterprises
CN116663419A (en) Sensorless equipment fault prediction method based on optimized Elman neural network
CN116703644A (en) Attention-RNN-based short-term power load prediction method
CN101334636A (en) Flattening machine technological parameter selection method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant