Topic 1
Basics of Python and Scientific Computing
1.1 Getting Started and Installing Python
1.1.1 Python background and its application in meteorology
1.1.2 Anaconda interpretation and installation and Jupyter configuration 1.1.3 Python basic syntax
1.2 Basic library for scientific data processing
1.2.1 Numpy library 1.2.2 Pandas library 1.2.3 Scipy library
1.2.4 Matplotlib and Cartopy Libraries
1.2.5 IO of common data formats
Master the configuration of python environment based on Anaconda, and use Jupyterlab to develop and debug code. After understanding the basic syntax of python, learn commonly used scientific computing and visualization libraries, such as Numpy, Pandas, and Matplotlib. After laying a good foundation, machine learning data preprocessing and result display will be at your fingertips.
2.1 Basic Theory of Machine Learning and Deep Learning
2.1.1 Basic Theory of Machine Learning
Basic knowledge of basic learning, such as error backpropagation, gradient descent method, and the entire general process of machine learning.
2.1.2 Basic Theory of Deep Learning
Basic theory of deep learning, such as convolutional neural network CNN, recurrent neural network RNN and generative adversarial network.
2.1.3 Application of machine learning and deep learning in meteorology
Topic 2
Basic theory and practice of machine learning and deep learning
The application of AI in weather model correction, short-term forecast, climate prediction and other scenarios, as well as the introduction of deep learning downscaling and PINN embedded physical neural network.
2.1 sklearn and pytorch libraries
2.1.1 Introduction to sklearn, common functions and machine learning methods
Learn the common functions of the classic machine learning library sklearn, such as the acquisition of public data sets such as iris flowers and handwritten fonts, the division of training sets and test sets, model building and model verification, etc.
2.1.2 Introduction to pytorch and model building
Learn the current popular deep learning framework pytorch, understand tensor tensor, automatic derivation, gradient promotion, etc., take BP neural network learning sin function as an example, master how to build single-layer and multi-layer neural networks, and how to use GPU for model calculations .
3.1 Correction of wind speed forecast by GFS numerical model
3.1.1 Random forest selection of important features
3.1.2 K-nearest neighbor and decision tree model to correct wind speed
3.1.3 Gradient boosting decision tree GBDT to correct wind speed 3.1.4 Model evaluation and comparison
3.2 Intelligent correction of typhoon forecast data
3.2.1 Introduction and preprocessing of CMA typhoon forecast dataset
3.2.2 Random forest model correction typhoon forecast
Topic Three
Examples of Machine Learning Applications in Meteorology
3.1.1 XGBoost model corrected typhoon forecast
3.1.2 Typhoon "Fireworks" Forecast Effect Test
3.1 Machine learning predicts wind power of wind farms
3.1.1 LightGBM model predicts wind power
3.1.2 Tool for parameter adjustment—GridSearch for K-fold verification
In this topic, on the basis of explaining in detail two types of ensemble learning algorithms commonly used in machine learning, Bagging and Boosting, on the basis of in-depth explanations of the two types of algorithms and their commonly used representative models, combined with three learning examples, and talking about some common machine learning techniques in series, Combine theory with practice.
Topic 4
Examples of Deep Learning Applications in Meteorology
Deep learning predicts shallow water equation patterns
4.1.1 Shallow water model introduction and data acquisition
4.1.2 Traditional Neural Network ANN Learning Shallow Water Equation
4.1.3 Physical Constraint Network PINN Learning Shallow Water Equations
4.2 LSTM method to predict ENSO
4.2.1 ENSO Introduction and Data Introduction
4.2.2 Introduction to the principle of LSTM method
4.2.3 LSTM method for predicting meteorological sequence data
4.3 Deep Learning—Convolutional Networks
4.3.1 Introduction to Convolutional Neural Networks
4.3.2 Radar echo prediction by Unet
In this topic, on the basis of learning to use ANN to predict shallow water equations, further master how to use PINN method to add dynamic equations to the model to alleviate the problem of poor physical interpretation of deep learning. In addition, meteorological data is typical spatiotemporal data, learn the classic time series forecasting method LSTM, and the spatial convolution algorithm UNET.
Original text of the tutorial:
Practical technical application of Python artificial intelligence in meteorology