多元异构数据处理先进技术
（英文版）
Advanced Techniques for Processing Multi-Source
Heterogeneous Data
王迪 李方昱 著
清华大学出版社
北京
内容简介
本书构建了多元异构数据介绍、预处理、表征学习、融合、系统建模的分析体系，并针
对特定工业场景聚焦多元异构数据的系统状态预测、故障诊断、具有定性和定量因素的系统
优化问题，为读者提供包含多元异构数据的复杂系统实际应用的见解 , 提供全面而系统的理
论知识与实践指南，使读者能够掌握该领域的核心技术。
本书可供数据科学、机器学习、人工智能、工业大数据等领域的高校师生和科研院所研
究人员及相关技术人员阅读参考。
版权所有，侵权必究。举报：010-62782989, beiqinquan@tup.tsinghua.edu.cn。
图书在版编目（CIP）数据
多元异构数据处理先进技术 : 英文 /王迪,李方昱著. --北京 :清华大学出版社,
2026. 6. -- ISBN 978-7-302-71981-6
Ⅰ. TP274
中国国家版本馆 CIP数据核字第 2026QZ7387号
责任编辑：孙亚楠
封面设计：常雪影
责任校对：欧洋
责任印制：丛怀宇
出版发行：清华大学出版社
网址：https://www.tup.com.cn, https://www.wqxuetang.com
地址：北京清华大学学研大厦A座邮编：100084
社总机： 010-83470000邮购：010-62786544
投稿与读者服务：010-62776969, c-service@tup.tsinghua.edu.cn
质量反馈：010-62772015, zhiliang@tup.tsinghua.edu.cn
印装者：天津鑫丰华印务有限公司
经销：全国新华书店
开本：170mm.240mm 印张：17插页：3 字数：421千字
版次：2026年 6月第 1版印次：2026年 6月第 1次印刷
定价：89.00元
产品编号： 108174-01
Acronyms
AAIL
adaptation-aware
interactive
learning
ACO
ant
colony
optimization
AE-MSEN
autoencoder
multiscale
ensemble
ANN
artificial
neural
network
ASS
adaptive
sensor
selection
BDL
Bayesian
deep
learning
BMU
best
matching
unit
CCA
canonical
correlation
analysis
CDF
cumulative
distribution
function
CFA
confirmatory
factor
analysis
CGD
constraint
gradient
descent
CLIP
contrastive
language-image
pre-training
CMFI
cross-modal
fusion
and
integration
CML
cross-modal
learning
CNNs
convolutional
neural
networks
CORAL
correlation
alignment
DANN
domain-adversarial
neural
networks
DBN
deep
belief
network
DBSCAN
density-based
spatial
clustering
of
applications
with
noise
DL-heuristic
deep
learning
benchmark
constructs
HI
based
on
multiple
sensor
signals
by
heuristic
metrics
DL-nonseparate
deep
learning
benchmark
model
makes
prognostics
under
a
nonseparated
framework
by
integrating
two
deep
learning
models
for
HI
construction
and
degradation
modeling
DL-separate
deep
learning
benchmark
models
the
HI
and
degradation
status
in
two
separate
steps
DL-single
deep
learning-based
fusion
model
under
a
single
OC
DLF
deep
learning
fusion
Acronyms
DNNs
deep
neural
networks
DSOM
distributed
self-organizing
modeling
DT
decision
tree
Eclat
equivalence
class
clustering
and
bottom-up
lattice
traversal
EFA
exploratory
factor
analysis
EHR
electronic
health
records
EM
expectation-maximization
ETL
extract,
transform,
load
FM
failure
mode
FP-Growth
frequent
pattern
growth
FP-Tree
frequent
pattern
tree
GAE
graph
autoencoder
GANs
generative
adversarial
networks
GATs
graph
attention
networks
GCN
graph
convolutional
networks
GDPR
general
data
protection
regulation
GMM
Gaussian
mixture
model
GNN
graph
neural
network
GRL
graph-based
representation
learning
HDFS
Hadoop
distributed
file
system
HI
health
index
HI-Quadratic
HI-based
linear
fusion
model
with
quadratic
degradation
function
HI-statistical
HI-based
statistical
model
HIN
heterogeneous
information
networks
HKG
heterogeneous
knowledge
graphs
HMMs
hidden
Markov
models
HPC
high-pressure
compressor
IF
isolation
forest
IGD
indirect
gradient
descent
IoT
internet
of
things
IQR
inter-quartile
range
JSON
JavaScript
object
notation
KL
Kullback-Leibler
KNN
K-nearest
neighbors
KNR
K-neighbor
regressor
LCM
lightweight
collaborative
modeling
LDA
linear
discriminant
analysis
LiDAR
light
detection
and
ranging
Acronyms
t-SNE
t-distributed
stochastic
neighbor
embedding
UMAP
uniform
manifold
approximation
and
projection
VAEs
variational
auto-encoders
WOL
Web
ontology
language
XML
extensible
markup
language
Contents
Chapter 1 Introduction to Multi-source Heterogeneous Data 1
1.1 Overview of MSHD 1
1.2 Significance of MSHD 2
1.3 Key Challenges in MSHD Processing 3
1.4 Advanced Techniques and Methodologies for Processing MSHD 4
1.5 Applications and Use Cases of MSHD 6
1.6 Organization of the Book 8
References 9
Part I Fundamental Methodologies for
Processing Multi-source Heterogeneous Data
Chapter 2 Preprocessing of Multi-source Heterogeneous Data 13
2.1 Introduction to MSHD Preprocessing 13
2.2 Data Types and Quality Assessments 14
2.2.1 Types and Sources of Heterogeneous Data 15
2.2.2 Metrics and Methods for Data Quality Assessments 16
2.3 Data Cleaning 19
2.3.1 Introduction to Data Cleaning 19
2.3.2 Challenges and Solutions in Cleaning Heterogeneous Data 21
2.3.3 Methods for Data Cleaning 22
2.4 Data Transformation 24
2.4.1 Concepts of Data Transformation 24
2.4.2 Methods for Data Normalization 25
2.4.3 Challenges and Solutions in Data Transformation of Heterogeneous Data 27
2.5 Handling Missing Data 28
2.5.1 Types and Causes of Missing Data 28
2.5.2 Methods of Handling Missing Data 29
2.6 Handling Anomalies 32
2.6.1 Types of Anomaly Data 32
2.6.2 Methods for Anomaly Detection 33
2.6.3 Strategies for Handling Anomalies in Heterogeneous Data 34
2.7 Summary 36
References 36
Chapter 3 Representation Learning for Multi-source Heterogeneous Data 39
3.1 Introduction to MSHD Representation Learning 39
3.2 Multi-view Learning for MSHD 41
3.2.1 Concepts of Multi-view Learning 41
3.2.2 Methods for Multi-view Representation Learning 42
3.3 Cross-modal Learning for MSHD 47
3.3.1 Concepts of Cross-modal Learning 47
3.3.2 Methods for Cross-modal Learning 48
3.4 Graph-based Representation Learning for MSHD 51
3.4.1 Graph Models for Heterogeneous Data 52
3.4.2 Techniques of GRL for Heterogeneous Data 54
3.5 Deep Learning for MSHD 56
3.5.1 Overview of Deep Learning for Heterogeneous Data 57
3.5.2 Transfer Learning and Domain Adaptation 59
3.6 Summary 62
References 62
Chapter 4 Integration and Fusion for Multi-source Heterogeneous Data 66
4.1 Introduction to MSHD Integration and Fusion 66
4.1.1 Introduction to MSHD Integration 66
4.1.2 Introduction to MSHD Fusion 67
4.2 Data Integration Methods and Techniques 69
4.2.1 Schema Matching 69
4.2.2 Data Mapping and Transformation 70
4.2.3 Semantic Integration and Ontology-based Approaches 72
4.3 Data Fusion Strategies and Algorithms 74
4.3.1 Approaches for Combining and Fusing Information from Different Data
Sources 74
4.3.2 Ensemble Learning 76
4.3.3 Statistical and Probabilistic Models 79
4.4 Architectures and Systems for MSHD Integration and Fusion 81
4.4.1 Distributed and Parallel Computing Techniques for Scalable Data Integration
and Fusion 82
4.4.2 Integration Systems and Tools for Handling Large-scale Heterogeneous
Data 83
4.5 Summary 84
References 84
Chapter 5 Modeling of Systems with Multi-source Heterogeneous Data 88
5.1 Introduction to MSHD Modeling 88
5.2 Multi-scene Association Analysis 89
5.2.1 Importance of MSAA 90
5.2.2 Methods for MSAA 91
5.2.3 Application of MSAA 97
5.3 Distributed Self-organizing Modeling 98
5.3.1 Importance of DSOM 99
5.3.2 Methods for DSOM 99
5.3.3 Application of DSOM 103
5.4 Lightweight Collaborative Modeling 105
5.4.1 Importance of LCM 105
5.4.2 Methods for LCM 106
5.4.3 Application of LCM with MSHD 112
5.5 Summary 113
References 113
Part II Methodological Advances in Processing
Multi-source Heterogeneous Data via Case Studies
Chapter 6 Case Study 1: Fusion and Prognosis of Systems with Heterogeneous Degradation
Profiles 119
6.1 Introduction 120
6.2 Model Details 123
6.2.1 A Generic Indirect Deep Learning Approach for Multisensor Fusion and
Degradation Modeling 123
6.2.2 An Integrated Deep Learning-based Data Fusion and Degradation Modeling
Method for Improving Prognostics 134
6.3 Experimental Results of Model 1 145
6.3.1 Model Training Using Historical Units 146
6.3.2 RUL Prediction for In-service Units 147
6.4 Experiment Results of Model 2 149
6.4.1 Model Training Using Historical Units 150
6.4.2 RUL Prediction for In-service Units 153
6.5 Summary 158
References 158
Chapter 7 Case Study 2: Prognosis and Diagnosis of Systems Under Multiple Failure
Modes 162
7.1 Introduction 163
7.2 Model Details 167
7.2.1 Joint Learning of FM Recognition and Prognostics for Degradation
Processes 167
7.2.2 Multimodal Regression and Mode Recognition via An Integrated Deep Neural
Network 175
7.3 Experiment Results of Model 1 183
7.3.1 Degradation Feature Extraction 184
7.3.2 Hyperparameter Selection 186
7.3.3 Results and Comparison 188
7.4 Experiment Results of Model 2 193
7.4.1 Hyperparameter Selection 193
7.4.2 Results and Comparison 195
7.5 Summary 199
References 199
Chapter 8 Case Study 3: Prognosis and Diagnosis of Systems Under Multiple Operating
Conditions 203
8.1 Introduction 204
8.2 Model Details 207
8.2.1 An Adaptation-aware Interactive Learning Approach for Multiple OC-based
Degradation Modeling 207
8.2.2 Deep Learning-based Sensor Selection for Failure Mode Recognition and
Prognostics Under Time-varying Operating Conditions 217
8.3 Experiment Results of Model 1 225
8.3.1 Influence of Hyperparameters 226
8.3.2 Model Training 229
8.3.3 Results and Comparison 231
8.4 Experiment Results of Model 2 234
8.4.1 Hyperparameter Selection 235
8.4.2 Results and Comparison 237
8.4.3 Ablation Study 247
8.5 Summary 249
References 249
Chapter 9 Future of Advanced Techniques for Multi-source Heterogeneous Data
Processing 254
9.1 Summary of Key Findings and Contributions 254
9.2 Potential Impacts 256
9.3 Future Directions for MSHD 257
9.4 Conclusion and Remarks 259
