AI Time Series Control System Modelling

ISBN-13: 9789811945939 / Angielski / Twarda / 2022

Chuzo Ninagawa

Opis

This book describes the practical application of artificial intelligence (AI) methods using time series data in system control. This book consistently discusses the application of machine learning to the analysis and modelling of time series data of physical quantities to be controlled in the field of system control.

Since dynamic systems are not stable steady states but changing transient states, the changing transient states depend on the state history before the change. In other words, it is essential to predict the change from the present to the future based on the time history of each variable in the target system, and to manipulate the system to achieve the desired change. 

In short, time series is the key to the application of AI machine learning to system control. This is the philosophy of this book: "time series data" + "AI machine learning" = "new practical control methods".

This book can give my helps to undergradate or graduate students, institute researchers and senior engineers whose scientific background are engineering, mathematics, physics and other natural sciences.

Spis treści

Chapter 1   Introduction

1.1 Time Series

1.1.1 What is the "Time Series" Dealt with in this Book?

1.1.2 Time Series for Statistical Control

1.1.3 Dissemination of Time Series Data for Control

1.2 Time Series and Control Models

1.2.1 Control Modeling

1.2.2 Control Model Building Methods

1.3 Control Time Series and AI Methods

1.3.1 Control Model by Time Series Analysis

1.3.2 Control and AI Methods

Chapter 2   Linear Time Series Modeling

2.1 Linear Regression Models

2.1.1 Regression Model

2.1.2 Multiple Regression Model and its Parameter Estimation

2.2 Fundamentals of AR Models

2.2.1 Overview of the AR model

2.2.2 Yule-Walker Method (One Variable)

2.2.3 Yule-Walker Method (Multivariate)

2.3 Practical Example 1: Multiple Regression Model with Stable Interval

2.3.1 Air-conditioning stable power model

2.3.2 Selection of explanatory variables

2.3.3 Linear Multiple Regression Analysis

2.3.4 Model Evaluation and Validation

2.4 Practical Example 2: Step Response AR Model

2.4.1 Limit Control of Building Air-conditioning Power

2.4.2 Fitting the AR Equation Model

2.4.3 Model Identification from Measured Data

2.4.4 AR Model Identification Results

Chapter 3   Deep Learning AI Modeling

3.1 Fundamentals of Deep Learning

3.1.1 Fundamentals of Neural Networks

3.1.2 Principles of Deep Learning

3.1.3 Stacked Denoising Autoencoder Method

3.2 Time Series Data Deep Learning

3.2.1 Time Series Parallel Input Neural Network

3.2.2 Number of Layers for Time Series Deep Learning

3.2.3 Hyper parameters of Time Series Deep Learning

3.3 Practical Example 3: Step Response AR Neural Network

3.3.1 Step Response AR Neural Network

3.3.2 Training a Step Response Time Series Model

3.3.3 Evaluating the Step Response Time Series Model

3.4 Example 4: Deep Learning in Practice - Sudden Event Prediction Model

3.4.1 Example of a Sudden Event

3.4.2 Sudden Event Prediction Neural Network Model

3.4.3 Training a Sudden Event Prediction Neural Network Model

Chapter 4   LSTM AI Modeling

4.1 Fundamentals of LSTM Neural Networks

4.1.1 What is LSTM Neural Network?

4.1.2 LSTM Forward Propagation Calculation

4.1.3 LSTM Back Propagation Calculation

4.2 LSTM Time Series Models

4.2.1 Construction of LSTM Time Series Model

4.2.2 Predictive Performance Evaluation Method

4.2.3 Results of Predictive Performance Evaluation

4.2.4 Considerations for Applying the LSTM Predictive Model

4.3 Example 5: Electricity Wholesale Market LSTM Model

4.3.1 Prediction of Wholesale Electricity Prices

4.3.2 Electricity Wholesale Price LSTM Forecasting Model

4.3.3 Evaluation of the Wholesale Electricity Price LSTM Forecasting Model

4.4 Example 6: LSTM Model for Prediction of Equipment Occurrence Events

4.4.1 Example of a Time-Series Unexpected Event

4.4.2 Complexity of Equipment Maintenance Operation

4.4.3 Prediction Model for Equipment Maintenance Operation

4.4.4 Evaluation of Predictive Model for Equipment Maintenance Operation

Chapter 5   Optimal Control by Time-Series AI Model

5.1 Fundamentals of Optimal Search Control

5.1.1 Simulated Annealing Optimal Search Method

5.1.2 Principle of SA Optimal Search Algorithm

5.1.3 Example of Evaluation Function of SA Optimal Search Control

5.2 State Explosion and Parallel Search

5.2.1 Large-scale State Space to be Controlled

5.2.2 Parallel SA Search Algorithm

5.2.3 Trials of Large-scale Parallel Search

5.3 Example 7: Electricity Price Optimal Search Control

5.3.1 Real-Time Electricity Price

5.3.2 Optimal Control of Air-conditioning Power Rates

5.3.3 Actual Test of Optimal Search Control

5.4 Example 8: Practical Cessation of Large-Scale Search

5.4.1 Practicality of Optimal Search Control

5.4.2 Cessation of Large-Scale Search

Chapter 6   The Reality of Time Series Learning Data Collection

6.1 Practical Example 9: Generating Training Data with Pseudo-Step Response

6.1.1 Step Response Training Data

6.1.2 Break-point Step Response Extraction Method

6.1.3 Example of Break-point Method Training Data Collection

6.2 Practical Example 10: Artificial Augmentation of Training Data Collection

6.2.1 The Reality of Learning Data Collection

6.2.2 Artificial Augmentation of Training Data

6.2.3 The Reality of Artificial Training Data Augmentation

6.3 Example 11: Generating Training Data with Emulators

6.3.1 Baseline and Reproducibility

6.3.2 Baseline Emulator Training

6.3.3 Baseline Estimation Model Evaluation

Chapter 7   Practical Work on Time Series AI Modeling

7.1 IoT Time Series Data Collection Methods

7.1.1 IEEE1888 Standard for Time Series Data Collection

7.1.2 IEEE1888 Time Series Data Transmission Method

7.1.3 IEEE1888 Standard IoT Communication Software Implementation

7.2 Zone Selection for Time Series AI Training Data

7.2.1 Reality of Time Series Training Data Collection

7.2.2 Zone Selection of Time Series Learning Data

7.2.3 Practical Methods with Time Series Data Selection

7.3 Example Software for Time Series AI Modeling

7.3.1 Off-the-shelf Learning Tools and Home-grown Learning Software

7.3.2 Homemade Machine Learning Software

7.3.3 Visualization with Homemade Machine Learning Software

Appendix

Appendix A  Example Source Code of MLP Deep Learning Algorithm

Appendix B  Example Source Code of LSTM Neural Network Learning Algorithm

​

O autorze

Prof. Chuzo Ninagawa is CEO of N Laboratory, Inc. and Professor of Smart Grid Power Control Engineering Joint Research Laboratory¸ Gifu University, Gifu, Japan. He has been Executive Chief Engineer of Mitsubishi Heavy Industries, Ltd., which is one of the largest hi-tech manufacturers in Japan. His research interests span various topics of smart grid, with special focus on virtual power plant (VPP) with a large-scale aggregation of fast automated demand responses. He has published over 110 academic papers and three advanced research books.