CN113361189B - Chip performance degradation trend prediction method based on multi-step robust prediction learning machine - Google Patents
Chip performance degradation trend prediction method based on multi-step robust prediction learning machine Download PDFInfo
- Publication number
- CN113361189B CN113361189B CN202110514211.7A CN202110514211A CN113361189B CN 113361189 B CN113361189 B CN 113361189B CN 202110514211 A CN202110514211 A CN 202110514211A CN 113361189 B CN113361189 B CN 113361189B
- Authority
- CN
- China
- Prior art keywords
- prediction
- neural network
- network module
- data
- time
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Abstract
The invention discloses a chip performance degradation trend prediction method based on a multi-step robust prediction learning machine, which combines an extreme learning machine and a recurrent neural network, has extremely high information fusion capability and information rapid processing capability, simultaneously constructs similarity based on relevant entropy by establishing an error code book, establishes real-time prediction model updating according to chip degradation diversity and dynamics, and overcomes the influence of interference on a prediction result. Therefore, compared with the existing method, the method has higher online prediction precision, and the multi-step prediction result is more accurate compared with the existing method.
Description
Technical Field
The invention belongs to the technical field of electronic device health management and machine learning, and particularly relates to a chip performance degradation trend prediction method based on a multi-step robust prediction learning machine.
Background
Along with the rapidity of chip technology development and the diversity of chip working environments, the existing chip performance degradation trend prediction method faces the challenges of time complexity and technical flexibility. On one hand, along with the improvement of the chip information interaction speed, the estimation quantity of the health state needs to be quickly given by a prediction algorithm; on the other hand, due to the complex and various chip application environments, the prediction algorithm needs to flexibly evaluate the health state according to the actual operation condition of the chip. However, the existing chip performance degradation trend prediction methods usually employ an offline health state model based on physical mechanism modeling or data driving. The degradation trend prediction method is highly dependent on experimental data and empirical data, and the operation complexity caused by training is high. Therefore, the existing degradation trend prediction method cannot meet new requirements on rapidity and flexibility of prediction.
In order to improve the rapidity and flexibility of the degradation trend prediction method, an online degradation trend prediction method is gradually started. The degradation trend prediction method uses real-time data and historical data in the chip work to evaluate the future chip work condition according to the chip operation state. The online degradation trend prediction method can adjust the model according to the real-time response of the electronic element, so that the method has strong flexibility in practical application, and the prediction precision is improved to a certain extent compared with an offline prediction method. However, the existing online degradation trend prediction method still has the following problems in engineering practice: firstly, the existing online degradation trend prediction method cannot ensure that effective information in historical data is efficiently transmitted in a multi-dimensional prediction process, so that part of the effective information cannot play a role in actual prediction; secondly, although the online pre-degradation trend measurement method can make corresponding adjustment according to real-time data information, the parameter adjustment needs iterative operation or re-checking of part of models, and cannot adapt to the running speed requirement of related elements; thirdly, since most models assume that the degradation trend of elements meets the wiener process, the models are easily interfered by non-gaussian noise and singular points in data, and the prediction result is greatly influenced.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a chip performance degradation trend prediction method based on a multi-step robust prediction learning machine so as to improve the information fusion capability and the information rapid processing capability and overcome the influence of interference on a prediction result.
In order to achieve the purpose, the chip performance degradation trend prediction method based on the multi-step robust prediction learning machine is characterized by comprising the following steps of:
(1) initializing the prediction model
1.1), setting the maximum prediction step number K, and importing training historical data Xhisto={x1,x2,…,xt0In which xtThe chip performance characterization parameter data (such as saturation voltage drop of an insulated gate bipolar transistor) recorded by the sensor at the t-th moment is represented by t, 1,2, …, t0 and t0, which are the number of historical data;
structure of initializing recurrent neural network M ═ M1,M2,M3,…,MK},Mk={Xk,Whk,βk,Ek,σkThe neural network module for the k prediction, where XkFor neural network module MkInput of (1), WhkFor neural network module MkHidden layer weight of (1), betakFor neural network module MkOutput layer weight of EkFor neural network module MkOf the code book, sigmakFor neural network module MkThe estimated variance under the codebook record of (1);
obtaining optimal prediction dimension m and delay coefficient t of model input sample by utilizing optimal entropy rate methodd;
Initializing WhkIs a two-dimensional random matrix of m × N, betakIs an N-dimensional zero vector;
1.2), calculating the k step neural network module MkInput vector X ofk:
Setting a k step neural network module MkThe true outputs of (c) are:
1.3) generating a recurrent neural network under offline training as a prediction model
1.3.1), compute hidden layer input Hk:
Wherein the content of the first and second substances,is the k step neural network module MkT 1,2, …, t0-k-mtd:
Wherein f (.) is a Sigmoid activation function;
1.3.2) generating weights β for output layersk:
1.4) generating an error code book in an off-line training state
1.4.2), estimating neural network module MkEstimated variance under codebook record of (1) Wherein med (.) is the median calculation;
(2) real-time multi-step prediction
2.1) initializing the running time ts of the prediction model to be t0+ 1;
2.2) obtaining chip performance characterization parameter data x recorded by the sensorts
2.3), judging whether the prediction program is manually stopped or whether the running time ts reaches the set upper limit value tmaxIf yes, ending the real-time prediction program; if not, entering the step 2.4);
2.4), updating the prediction model and generating real-time prediction data
2.4.1), initializing the prediction step number k to be 1;
2.4.2), judging whether the predicted step number K is larger than the maximum predicted step number K, if so, entering the step 2.5), and if not, entering the step 2.4.3);
2.4.3), generating neural network module MkInput vector of (2):
2.4.6), updating codebook information
2.4.6.2), estimating neural network module MkEstimated variance under codebook record of (1) Wherein med (.) is the median calculation;
2.4.7), update prediction module output weightsWherein ΛkFor the correlation entropy diagonal matrix:
2.4.8), update k ═ k +1, return to step 2.4.2);
2.5) obtaining K-step prediction data of operation time tsAnd (5) updating ts to be ts +1, and returning to the step 2.2), and predicting the K-step prediction data of the next operation time in real time.
The invention aims to realize the following steps:
the chip performance degradation trend prediction method based on the multi-step robust prediction learning machine combines the extreme learning machine and the recurrent neural network, has extremely high information fusion capability and information rapid processing capability, simultaneously constructs similarity based on relevant entropy by establishing an error code book, establishes real-time prediction model updating according to chip degradation diversity and dynamics, and overcomes the influence of interference on prediction results. Therefore, compared with the existing method, the method has higher online prediction precision, and the multi-step prediction result is more accurate compared with the existing method.
Drawings
FIG. 1 is a flowchart of an embodiment of a method for predicting degradation trend of chip performance based on a multi-step robust prediction learning machine according to the present invention;
FIG. 2 is a flowchart of an embodiment of the steps shown in FIG. 1 for updating a predictive model and generating real-time predictive data.
Detailed Description
The following description of the embodiments of the present invention is provided in order to better understand the present invention for those skilled in the art with reference to the accompanying drawings. It is to be expressly noted that in the following description, a detailed description of known functions and designs will be omitted when it may obscure the subject matter of the present invention.
FIG. 1 is a flowchart of an embodiment of a method for predicting a chip performance degradation trend based on a multi-step robust prediction learning machine according to the present invention.
In this embodiment, as shown in fig. 1, the method for predicting the chip performance degradation trend based on the multi-step robust prediction learning machine of the present invention includes the following steps:
step S1: initializing a predictive model
Step S1.1: initializing model parameters
Setting the maximum predicted step number K, and importing training historical data Xhisto={x1,x2,…,xt0In which xtSetting maximum prediction step number K for chip performance characterization parameter data (such as saturation voltage drop of insulated gate bipolar transistor) recorded by the sensor at the t-th moment, and importing training historical data Xhisto={x1,x2,…,xt0In which xtFor the chip performance characterization parameter data (such as saturation voltage drop of the insulated gate bipolar transistor) recorded by the sensor at the t-th moment, t is 1,2, …, t0, and t0 is the number of historical data.
Structure of initializing recurrent neural network M ═ M1,M2,M3,…,MK},Mk={Xk,Whk,βk,Ek,σkThe neural network module for the k prediction, where XkFor neural network module MkInput of (1), WhkFor neural network module MkHidden layer weight of (1), betakFor neural network module MkOutput layer weight of EkFor neural network module MkOf the code book, sigmakFor neural network module MkThe estimated variance under the codebook record.
Obtaining optimal prediction dimension m and delay coefficient t of model input sample by utilizing optimal entropy rate methodd。
Initializing WhkIs a two-dimensional random matrix of m × N, betakIs an N-dimensional zero vector.
Step S1.2: computing an input vector
Calculating k step neural network module MkInput vector X ofk:
Setting a k step neural network module MkThe true outputs of (c) are:
step S1.3: generating a recurrent neural network under offline training as a prediction model
Step S1.3.1: computing hidden layer input Hk:
Wherein the content of the first and second substances,is the k step neural network module MkT-th input of (2) corresponds to a hidden layer input of (1, 2, …, t0-k-mtd:
Wherein f (.) is a Sigmoid activation function;
step S1.3.2: generating weights beta of output layersk:
Step S1.4: generating an error codebook in an offline training state
Step S1.4.2: estimating neural network module MkEstimated variance under codebook record of (1) Wherein med (.) is the median calculation;
step S2: real-time multi-step prediction
Step S2.1: initializing the running time ts of the prediction model as t0+ 1;
step S2.2: obtaining chip performance characterization parameter data x recorded by sensorts
Step S2.3: judging whether the prediction program is manually stopped or whether the running time ts reaches the set upper limit value tmaxIf yes, ending the real-time prediction program; if not, the step S2.4 is carried out: (ii) a
Step S2.4: updating a prediction model and generating real-time prediction data
As shown in fig. 2, this step includes:
step S2.4.1: initializing the prediction step number k to be 1;
step S2.4.2: judging whether the predicted step number K is larger than the maximum predicted step number K, if so, entering step S2.5, and if not, entering step S2.4.3;
step S2.4.3 generating neural network module MkInput vector of (2):
Step S2.4.6 updating codebook information
Step S2.4.6.2 estimating neural network Module MkEstimated variance under codebook record of (1) Wherein med (.) is the median calculation;
step S2.4.7, update the output weight of the prediction moduleWherein ΛkFor the correlation entropy diagonal matrix:
step S2.4.8, updating k to k +1, and returning to step S2.4.2;
s2.5, obtaining K-step prediction data of the running time tsAnd (5) updating ts to ts +1, returning to the step S2.2, and predicting the K-step prediction data of the next operation time in real time.
Examples of the invention
To illustrate the technical effects of the present invention, the present invention is now verified by taking the prediction of the accelerated degradation saturation voltage drop of the insulated gate bipolar transistor as an example. The accelerated degradation saturation voltage drop of the insulated gate bipolar transistor can effectively reflect the health state of the device. In order to verify the effectiveness of the method, the implemented saturation voltage drop of the insulated gate bipolar transistor under the accelerated degradation experiment is predicted through the prediction model established by the method.
Meanwhile, the online prediction precision is shown in table 1 by comparing the method with a gated cyclic unit network (GRU), a long and short memory network (LSTM), an Extreme Learning Machine (ELM) and a cyclic extreme learning machine (RNN-ELM).
TABLE 1
As can be seen from Table 1, the method of the present invention has higher online prediction accuracy compared to the existing method, and the multi-step prediction result is more accurate compared to the existing method.
Although illustrative embodiments of the present invention have been described above to facilitate the understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, and various changes may be made apparent to those skilled in the art as long as they are within the spirit and scope of the present invention as defined and defined by the appended claims, and all matters of the invention which utilize the inventive concepts are protected.
Claims (1)
1. A chip performance degradation trend prediction method based on a multi-step robust prediction learning machine is characterized by comprising the following steps:
(1) initializing the prediction model
1.1), setting the maximum prediction step number K, and importing training historical data Xhisto={x1,x2,…,xt0In which xtThe chip performance characterization parameter data recorded by the sensor at the t-th moment is t 1,2, …, t0 and t0, wherein t is the number of historical data;
structure of initializing recurrent neural network M ═ M1,M2,M3,…,MK},Mk={Xk,Whk,βk,Ek,σkThe neural network module for the k prediction, where XkFor neural network module MkInput of (1), WhkFor neural network module MkHidden layer weight of (1), betakFor neural network module MkOutput layer weight of EkFor neural network module MkOf the code book, sigmakFor neural network module MkThe estimated variance under the codebook record of (1);
obtaining optimal prediction dimension m and delay coefficient t of model input sample by utilizing optimal entropy rate methodd;
Initializing WhkIs a two-dimensional random matrix of m × N, betakIs an N-dimensional zero vector;
1.2), calculating the k step neural network module MkInput vector X ofk:
Setting a k step neural network module MkThe true outputs of (c) are:
1.3) generating a recurrent neural network under offline training as a prediction model
1.3.1), compute hidden layer input Hk:
Wherein the content of the first and second substances,is the k step neural network module MkT 1,2, …, t0-k-mtd:
Wherein f (.) is a Sigmoid activation function;
1.3.2) generating weights β for output layersk:
1.4) generating an error code book in an off-line training state
1.4.2), estimating neural network module MkEstimated variance under codebook record of (1) Wherein med (.) is the median calculation;
(2) real-time multi-step prediction
2.1) initializing the running time ts of the prediction model to be t0+ 1;
2.2) obtaining chip performance characterization parameter data x recorded by the sensorts
2.3), judging whether the prediction program is manually stopped or whether the running time ts reaches the set upper limit value tmaxIf yes, ending the real-time prediction program; if not, entering the step 2.4);
2.4), updating the prediction model and generating real-time prediction data
2.4.1), initializing the prediction step number k to be 1;
2.4.2), judging whether the predicted step number K is larger than the maximum predicted step number K, if so, entering the step 2.5), and if not, entering the step 2.4.3);
2.4.3), generating neural network module MkInput vector of (2):
2.4.6), updating codebook information
2.4.6.2), estimating neural network module MkEstimated variance under codebook record of (1) Wherein med (.) is the median calculation;
2.4.7), update prediction module output weightsWherein ΛkFor the correlation entropy diagonal matrix:
2.4.8), update k ═ k +1, return to step 2.4.2);
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110514211.7A CN113361189B (en) | 2021-05-12 | 2021-05-12 | Chip performance degradation trend prediction method based on multi-step robust prediction learning machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110514211.7A CN113361189B (en) | 2021-05-12 | 2021-05-12 | Chip performance degradation trend prediction method based on multi-step robust prediction learning machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113361189A CN113361189A (en) | 2021-09-07 |
CN113361189B true CN113361189B (en) | 2022-04-19 |
Family
ID=77526176
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110514211.7A Active CN113361189B (en) | 2021-05-12 | 2021-05-12 | Chip performance degradation trend prediction method based on multi-step robust prediction learning machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113361189B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114646891B (en) * | 2022-03-10 | 2023-05-30 | 电子科技大学 | Residual life prediction method combining LSTM network and wiener process |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108645615A (en) * | 2018-04-08 | 2018-10-12 | 太原科技大学 | A kind of Adaptive Fuzzy Neural-network gear method for predicting residual useful life |
CN110220725A (en) * | 2019-05-30 | 2019-09-10 | 河海大学 | A kind of wheel for metro vehicle health status prediction technique integrated based on deep learning and BP |
CN110458278A (en) * | 2019-04-19 | 2019-11-15 | 四川大学 | Degradation prediction method based on quantum attention loop coding decoding neural network |
CN110598334A (en) * | 2019-09-17 | 2019-12-20 | 电子科技大学 | Performance degradation trend prediction method based on collaborative derivation related entropy extreme learning machine |
CN110633516A (en) * | 2019-08-30 | 2019-12-31 | 电子科技大学 | Method for predicting performance degradation trend of electronic device |
CN111931413A (en) * | 2020-06-22 | 2020-11-13 | 中国人民解放军陆军军事交通学院 | Method for predicting performance of high-altitude diesel engine of optimized extreme learning machine based on chaotic particle swarm with extreme disturbance |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010011918A2 (en) * | 2008-07-24 | 2010-01-28 | University Of Cincinnati | Methods for prognosing mechanical systems |
CN103581188B (en) * | 2013-11-05 | 2016-08-03 | 中国科学院计算技术研究所 | A kind of network security situation prediction method and system |
JP6384065B2 (en) * | 2014-03-04 | 2018-09-05 | 日本電気株式会社 | Information processing apparatus, learning method, and program |
CN111737911B (en) * | 2020-06-15 | 2022-09-27 | 大连理工大学 | Bearing recession trend prediction method for deep confidence network and extreme learning machine |
-
2021
- 2021-05-12 CN CN202110514211.7A patent/CN113361189B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108645615A (en) * | 2018-04-08 | 2018-10-12 | 太原科技大学 | A kind of Adaptive Fuzzy Neural-network gear method for predicting residual useful life |
CN110458278A (en) * | 2019-04-19 | 2019-11-15 | 四川大学 | Degradation prediction method based on quantum attention loop coding decoding neural network |
CN110220725A (en) * | 2019-05-30 | 2019-09-10 | 河海大学 | A kind of wheel for metro vehicle health status prediction technique integrated based on deep learning and BP |
CN110633516A (en) * | 2019-08-30 | 2019-12-31 | 电子科技大学 | Method for predicting performance degradation trend of electronic device |
CN110598334A (en) * | 2019-09-17 | 2019-12-20 | 电子科技大学 | Performance degradation trend prediction method based on collaborative derivation related entropy extreme learning machine |
CN111931413A (en) * | 2020-06-22 | 2020-11-13 | 中国人民解放军陆军军事交通学院 | Method for predicting performance of high-altitude diesel engine of optimized extreme learning machine based on chaotic particle swarm with extreme disturbance |
Non-Patent Citations (6)
Title |
---|
Application of Gaussian Process Regression for Bearing Degradation Assessment;Sheng Hong等;《2012 6th International Conference on New Trends in Information Science, Service Science and Data Mining (ISSDM2012)》;20130613;644-648 * |
Evolved-Cooperative Correntropy-Based Extreme Learning Machine for Robust Prediction;Wenjuan Mei等;《Entropy》;20190919;1-23 * |
区域地下水埋深的数据驱动预报模型研究;郭田丽;《中国优秀博硕士学位论文全文数据库(硕士)基础科学辑》;20210215(第2期);A011-691 * |
基于动态长短期记忆网络的设备性能退化预测方法;卫炳坤等;《北京化工大学学报( 自然科学版)》;20201120;第47卷(第6期);92-99 * |
基于在线相关熵极限学习机的器件退化趋势实时流预测方法;梅文娟等;《仪器仪表学报》;20191130;第40卷(第11期);212-224 * |
基于循环神经网络的轴承性能退化趋势预测方法研究;周佑光;《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》;20200615(第6期);C029-126 * |
Also Published As
Publication number | Publication date |
---|---|
CN113361189A (en) | 2021-09-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Espinoza et al. | Kernel based partially linear models and nonlinear identification | |
Jwo et al. | Neural network-based GPS GDOP approximation and classification | |
CN106462803B (en) | Enhancing neural networks with external memory | |
Guo et al. | Toward the optimal design and FPGA implementation of spiking neural networks | |
KR20190138712A (en) | Batch normalization layers | |
CN111753952A (en) | Learning parameters of a probabilistic model including a Gaussian process | |
CN111433689B (en) | Generation of control systems for target systems | |
CN113361189B (en) | Chip performance degradation trend prediction method based on multi-step robust prediction learning machine | |
CN110555526B (en) | Neural network model training method, image recognition method and device | |
WO2022156182A1 (en) | Methods and apparatuses for constructing vehicle dynamics model and for predicting vehicle state information | |
CN109033021B (en) | Design method of linear equation solver based on variable parameter convergence neural network | |
CN111461445A (en) | Short-term wind speed prediction method and device, computer equipment and storage medium | |
Heravi et al. | Long term trajectory prediction of moving objects using gaussian process | |
Bhalaji | Cloud load estimation with deep logarithmic network for workload and time series optimization | |
CN112633503B (en) | Tool variable generation and handwritten number recognition method and device based on neural network | |
US20230025626A1 (en) | Method and apparatus for generating process simulation models | |
CN110874553A (en) | Recognition model training method and device | |
Jin et al. | Expectation-Maximization Algorithm Based Identification of Hammerstein Nonlinear ARMAX Systems with Gaussian Mixture Noises | |
Mirza | Online additive updates with FFT-IFFT operator on the GRU neural networks | |
CN111416595B (en) | Big data filtering method based on multi-core fusion | |
Xia et al. | A conditional classification recurrent RBM for improved series mid-term forecasting | |
US20200219166A1 (en) | Apparatus and method for providing fashion coordination knowledge based on neural network having explicit memory | |
Chen et al. | Adaptive Machine Learning-Based Proactive Thermal Management for NoC Systems | |
KR20150064673A (en) | Method and device for determining a gradient of a data-based function model | |
Kocijan et al. | System identification with GP models |
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 |