# FastLink

**Repository Path**: ponspc/fastlink

## Basic Information

- **Project Name**: FastLink
- **Description**: FastLink: a machine learning and GPU based fast phase association method
基于GPU的快速震相关联算法
- **Primary Language**: Python
- **License**: GPL-3.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 2
- **Created**: 2023-04-06
- **Last Updated**: 2023-04-06

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# FastLink

#### 介绍
FastLink: a machine learning and GPU based fast phase association method

基于GPU的快速震相关联算法

![关联过程图](data/link.png)

#### 软件架构
- ckpt:预训练模型
- config:设置文件
- fastlinker.py:关联主文件
- reallinker.py:real关联工具


#### 安装教程

依赖库
1.  pytorch
2.  obspy 
3.  tqdm 

#### 使用说明

##### 训练
稍后添加
##### 推断
直接运行
```bash 
python fastlinker.py -o odata/2020.1107.200km.txt -i ckpt/new.200km.wave.txt -s station 
```


参数依次为
- -o输出文件夹，文件夹应当存在
- -i输入拾取文件，拾取文件为lppnpicker格式
- -s台站文件。如果为null，即不制作台站文件

震相文件格式
```text 
#单分量数据位置
震相,相对时间（秒）,置信度,绝对时间（格式%Y-%m-%d %H:%M:%S.%f）,信噪比,前后200个采样点振幅均值,台站名,前95%分位数,后95%分位数,最大值,标准差,峰值
```

在config/fastlink.py需要调整更多拾取参数：
```
    # GPU parameters
    win_length = 25.0   # seconds
    win_stirde = 1      # seconds 
    ngrid = 200 
    lonrange = [] # range of longtitude
    latrange = []
    locpad = 0.5 # if the range of lat and lon is not defined. the range is calculated by the range of stations with pad. 
    datadir = "fastdata"
    saveitv = 86400 # interval to save temp files. 
    
    # model
    modeldir = "ckpt/link32.ckpt"
    basetime = "2020-01-01" # initial time, which doesn't need to be setted. 

    # parameters
    np = 2 # number of P phase 
    ns = 4 # number of S phase
    nps = 8 # number of P+S phase
    nboth = 2 # number of both P and S phase
```

关联结果结构：
```text 
#EVENT,位置编码,关联时间,经度,纬度
PHASE,震相时刻,关联类型,台站名,原始类型
```

#### 引用文献
Yu ZY, Wang WT. FastLink: a machine learning and GPU based fast phase association method and its application to Yangbi Ms 6.4 aftershock sequences. 2022. GJI.


#### License 
[GPLV3](LICENSE)