[人工智能] lgb训练模型后使用AutoML-NNI对注塑工艺参数调参优化

开发: C++知识库 Java知识库 JavaScript Python PHP知识库人工智能区块链大数据移动开发嵌入式开发工具数据结构与算法开发测试游戏开发网络协议系统运维
教程: HTML教程 CSS教程 JavaScript教程 Go语言教程 JQuery教程 VUE教程 VUE3教程 Bootstrap教程 SQL数据库教程 C语言教程 C++教程 Java教程 Python教程 Python3教程 C#教程
数码: 电脑笔记本显卡显示器固态硬盘硬盘耳机手机 iphone vivo oppo 小米华为单反装机图拉丁

-> 人工智能 -> lgb训练模型后使用AutoML-NNI对注塑工艺参数调参优化 -> 正文阅读

[人工智能]lgb训练模型后使用AutoML-NNI对注塑工艺参数调参优化

1、摘要

本文主要讲解：lgb训练模型后使用AutoML-NNI对注塑工艺参数调参优化
主要思路：

使用lgb训练注塑品的三维尺寸预测模型，保存模型
将训练特征作为需要调优的参数，使用保存的模型对这个参数进行预测
将预测的结果作为nni的最佳指标，当最佳指标为最优的时候，即为最优的工艺参数

2、数据介绍

a) 传感器高频数据：该数据来自于模温机及模具传感器采集的数据，文件夹内每一个模次对应一个csv文件，单个模次时长为40~43s，采样频率根据阶段有20Hz和50Hz两种，含有24个传感器采集的数据；
b) 成型机状态数据（data_spc）：该数据来自成型机机台，均为表征成型过程中的一些状态数据，每一行对应一个模次，数据维度为86维；
c) 机台工艺设定参数（data_set）：文件夹中含有注塑成型的81种工艺设定参数；
d) 产品测量尺寸(size)：文件夹内含有每个模次产品的3维尺寸；
e) 现场巡检数据（初赛不提供此类数据）。各数据字段含义可在数据下载包中获得。
数据下载链接

3、相关技术

注塑工艺参数优化
工艺参数优化指的是对生产制造领域所需的工艺参数进行优化，在大部分制造业都存在工艺参数优化问题，例如注塑行业需要对保压压力、时间、温度等参数进行优化，以提高注塑品的质量，防止注塑品异常。

NNI (Neural Network Intelligence) 是一个轻量但强大的工具包，帮助用户自动的进行特征工程，神经网络架构搜索，超参调优以及模型压缩。

NNI 管理自动机器学习 (AutoML) 的 Experiment，调度运行由调优算法生成的 Trial 任务来找到最好的神经网络架构和/或超参。

4、完整代码和步骤

代码输出如下：
在这里插入图片描述
当最佳指标达到300时，说明已找到最佳的工艺生产参数

训练模型并保存

'''
This project is for automatically tuning parameters for GBDT.

Trial（尝试） 是将一组参数组合（例如，超参）在模型上独立的一次尝试。
定义 NNI 的 Trial，需要首先定义参数组（例如，搜索空间），并更新模型代码。
nnictl create --config config.yml -p 8888 --debug
'''
import logging

import lightgbm as lgb
import nni
import pandas as pd
from sklearn.metrics import mean_squared_error
import os

from sklearn.model_selection import train_test_split

from my_utils.data_utils import min_max_select

os.chdir(r'H:\工业大数据\工业大数据创新竞赛-注塑成型完整解决方案\data\Train')

LOG = logging.getLogger('auto-gbdt')

def get_default_parameters():
    params = {
        'boosting_type': 'gbdt',
        'objective': 'regression',
        'metric': {'l2', 'auc'},
        'num_leaves': 28,
        'learning_rate': 0.2,
        'feature_fraction': 0.930,
        'bagging_fraction': 0.7656,
        'bagging_freq': 4,
        'verbose': 0
    }
    # The rmse of prediction is: 0.419561
    return params


def load_data():
    df_TRAIN = pd.read_csv('df_TRAIN.csv')
    size_train = pd.read_csv('Size.csv')
    X_col = [i for i in df_TRAIN.columns if not i in ['Id', 'spcTime']]
    i = 'size3'
    X_train = df_TRAIN[X_col]
    y_train = size_train[i]
    X_train,X_eval,y_train ,y_eval = train_test_split(X_train,y_train, test_size=0.3, random_state=0)
    X_eval,X_test,y_eval ,y_test = train_test_split(X_eval,y_eval, test_size=0.6, random_state=0)

    lgb_train = lgb.Dataset(X_train, y_train)
    lgb_eval = lgb.Dataset(X_eval, y_eval, reference=lgb_train)
    return lgb_train, lgb_eval, X_test, y_test

def run(lgb_train, lgb_eval, params, X_test, y_test):
    print('Start training...')

    params['num_leaves'] = int(params['num_leaves'])

    # train
    gbm = lgb.train(params,
                    lgb_train,
                    num_boost_round=20,
                    valid_sets=lgb_eval,
                    early_stopping_rounds=5)

    print('Start predicting...')
    # predict
    y_pred = gbm.predict(X_test, num_iteration=gbm.best_iteration)
    gbm.save_model('gbm_model_size3.txt')

    # eval
    rmse = mean_squared_error(y_test, y_pred) ** 0.5
    print('The rmse of prediction is:', rmse)

    nni.report_final_result(rmse)

if __name__ == '__main__':
    lgb_train, lgb_eval, X_test, y_test = load_data()

    try:
        # get parameters from tuner
        RECEIVED_PARAMS = nni.get_next_parameter()
        LOG.debug(RECEIVED_PARAMS)
        PARAMS = get_default_parameters()
        PARAMS.update(RECEIVED_PARAMS)
        LOG.debug(PARAMS)

        # train
        run(lgb_train, lgb_eval, PARAMS, X_test, y_test)
    except Exception as exception:
        LOG.exception(exception)
        raise

主运行程序入口
命令
nnictl create --config config.yml -p 8882 --debug

import logging

import lightgbm as lgb
import nni
import numpy as np
import os

os.chdir(r'H:\工业大数据\工业大数据创新竞赛-注塑成型完整解决方案\data\Train')

LOG = logging.getLogger('auto-gbdt')
'''
目前用机器学习模型拟合工艺过程，并用NNI求优，但一方面调优算法具有随机性，
每次寻优结果都不同，另一方面无法解决通过时序性的产品尺寸（即输出）来优化工艺参数（即输入）。

根据y（即输出）来寻找x（即输入），解一般都是很多组，很难找到最优解，加额外的约束条件。
'''

def get_default_parameters():
    params = {"Sensor1_mean": 26312.942, "Sensor2_mean": 8837.184, "Sensor3_mean": 31569.637, "IJ_mean": 40236.952, "Sensor5_mean": 189215.722, "Sensor6_mean": 154451.291, "MouldTemp1_mean": 23400.0, "MouldTemp2_mean": 23400.0, "MouldTemp3_mean": 23400.0, "MouldTemp4_mean": 23400.0, "MouldTemp5_mean": 23400.0, "MouldTemp9_mean": 121115.999, "MouldTemp10_mean": 126094.039, "MouldTemp11_mean": 123119.346, "MouldTemp12_mean": 24315.0, "MouldTemp13_mean": 122862.835, "MouldTemp14_mean": 124401.039, "Sensor8_mean": -57212.977, "MouldFlow1_mean": 7550.0, "MouldFlow2_mean": 7550.0, "MouldFlow3_mean": 7550.0, "SP_mean": 81659.177, "EL_CYC_TIME": 9.238, "EL_IV_END_TIME": 0.0, "EL_PLST_TIME": 0.0, "EL_LAST_COOL_TIME": 0.0, "EL_MC_TIME": 0.406, "EL_MO_TIME": 0.0, "EL_EF_TIME": 0.0, "EL_EB_TIME": 0.0, "EL_IDLE_TIME": 0.0, "EL_CORE_IN_TIME_0": 0.0, "EL_CORE_IN_TIME_1": 0.0, "EL_CORE_IN_TIME_2": 0.0, "EL_CORE_IN_TIME_3": 0.0, "EL_CORE_OUT_TIME_0": 0.0, "EL_CORE_OUT_TIME_1": 0.0, "EL_CORE_OUT_TIME_2": 0.0, "EL_CORE_OUT_TIME_3": 0.0, "EL_IV_TIME": 1.522, "EL_PP_TIME": 2.927, "EL_SB1_TIME": 0.0, "EL_SB2_TIME": 0.0, "EL_CF_TIME": 0.0, "EL_CB_TIME": 0.0, "EL_IV_STG_TIME_0": 0.311, "EL_IV_STG_TIME_1": 0.346, "EL_IV_STG_TIME_2": 0.527, "EL_IV_STG_TIME_3": 0.106, "EL_IV_STG_TIME_4": 0.048, "op_mode": 2.0, "ext_mold_time_flt_0": 0.0, "EL_VPS_TIME": 0.0, "ext_mold_time_flt_1": 0.0, "ext_mold_time_flt_2": 0.0, "ext_mold_time_flt_3": 0.0, "ext_mold_time_flt_4": 0.0, "EL_GET_PART_TIME": 0.0, "EL_NZL_MEAN": 248.00876, "EL_BARREL_MEAN_1": 258.09712, "EL_BARREL_MEAN_2": 248.4281, "EL_BARREL_MEAN_3": 238.21912, "EL_BARREL_MEAN_4": 211.84534, "EL_BARREL_MEAN_5": 15.0, "EL_BARREL_MEAN_6": 15.0, "EL_BARREL_MEAN_7": 15.0, "EL_HTRN_TI_0": 15.0, "EL_HTRN_TI_1": 15.0, "EL_HTRN_TI_2": 15.0, "EL_HTRN_TI_3": 15.0, "EL_HTRN_TI_4": 15.0, "EL_HTRN_TI_5": 15.0, "EL_HTRN_TI_6": 15.0, "EL_HTRN_TI_7": 15.0, "ext_mold_temp_flt_0": 15.0, "ext_mold_temp_flt_1": 15.0, "ext_mold_temp_flt_2": 15.0, "ext_mold_temp_flt_3": 15.0, "ext_mold_temp_flt_4": 0.0, "ext_mold_temp_flt_5": 0.0, "ext_mold_temp_flt_6": 0.0, "ext_mold_temp_flt_7": 0.0, "rob_enbl": 1.0, "ext_mold_flow_flt_0": 5.0, "EL_MOP": 10.0, "EL_PLST_END_ERR": 10.0, "EL_INJ_START_STR": 104.02236, "EL_IV_END_STR": 39.980728, "EL_VPS_END_STR": 16.880154, "EL_MIN_INJ_STR": 16.598343, "EL_EF_END_POS": 48.023884, "EL_PP_STG_POS_0": 21.992069, "EL_PP_STG_POS_1": 21.335789, "EL_PP_STG_POS_2": 16.876579, "EL_PP_STG_POS_3": 10.0, "EL_PP_STG_POS_4": 10.0, "EL_CLAMP_PRESS": 53.827827, "EL_MAX_INJ_PRESS": 62.843433, "EL_IV_END_PRESS": 41.594925, "EL_CF_END_PRESS": 5.0, "EL_PLST_BP_0": 5.0, "EL_PLST_BP_1": 5.0, "EL_PLST_BP_2": 5.0, "EL_PLST_BP_3": 5.0, "EL_PLST_BP_4": 5.0}
    return params


def run(params):
    print('Start training...')
    X_test = np.array(list(params.values())).reshape(1, 105)

    print('Start predicting...')
    # gbm = lgb.Booster(model_file='gbm_model_size1.txt')
    gbm = lgb.Booster(model_file='gbm_model_size3.txt')
    # predict
    y_pred = gbm.predict(X_test, num_iteration=gbm.best_iteration)

    nni.report_final_result(round(y_pred[0], 3))


if __name__ == '__main__':
    try:
        # get parameters from tuner
        RECEIVED_PARAMS = nni.get_next_parameter()
        LOG.debug(RECEIVED_PARAMS)
        PARAMS = get_default_parameters()
        PARAMS.update(RECEIVED_PARAMS)
        LOG.debug(PARAMS)

        # train
        run(PARAMS)
    except Exception as exception:
        LOG.exception(exception)
        raise

config.yml

authorName: default
experimentName: zhusu-gbdt
trialConcurrency: 1
maxExecDuration: 10h
maxTrialNum: 100
#choice: local, remote, pai
trainingServicePlatform: local
searchSpacePath: search_space1.json
#choice: true, false
useAnnotation: false
tuner:
  #choice: TPE, Random, Anneal, Evolution, BatchTuner, MetisTuner, GPTuner
  #SMAC (SMAC should be installed through nnictl)
  builtinTunerName: Anneal
  classArgs:
    #choice: maximize, minimize
    optimize_mode: minimize
trial:
  command: python main1.py
  codeDir: .
  gpuNum: 0

search_space1.json

{"Sensor1_mean": {"_type": "uniform", "_value": [26312.942000000003, 75080.232]}, "Sensor2_mean": {"_type": "uniform", "_value": [8837.184000000001, 199116.303]}, "Sensor3_mean": {"_type": "uniform", "_value": [31569.637, 80692.242]}, "IJ_mean": {"_type": "uniform", "_value": [40236.95200000001, 89024.65400000002]}, "Sensor5_mean": {"_type": "uniform", "_value": [189215.722, 212030.38]}, "Sensor6_mean": {"_type": "uniform", "_value": [154451.29100000006, 219932.88300000003]}, "MouldTemp1_mean": {"_type": "uniform", "_value": [23400.0, 145170.95300000007]}, "MouldTemp2_mean": {"_type": "uniform", "_value": [23400.0, 95793.20000000004]}, "MouldTemp3_mean": {"_type": "uniform", "_value": [23400.0, 112825.865]}, "MouldTemp4_mean": {"_type": "uniform", "_value": [23400.0, 164148.8]}, "MouldTemp5_mean": {"_type": "uniform", "_value": [23400.0, 144523.1]}, "MouldTemp9_mean": {"_type": "uniform", "_value": [121115.99900000001, 159108.548]}, "MouldTemp10_mean": {"_type": "uniform", "_value": [126094.039, 167141.36099999998]}, "MouldTemp11_mean": {"_type": "uniform", "_value": [123119.346, 1381420.0000000005]}, "MouldTemp12_mean": {"_type": "uniform", "_value": [24315.0, 1315599.4000000004]}, "MouldTemp13_mean": {"_type": "uniform", "_value": [122862.835, 1315599.4000000004]}, "MouldTemp14_mean": {"_type": "uniform", "_value": [124401.03900000002, 1315599.4000000004]}, "Sensor8_mean": {"_type": "uniform", "_value": [-57212.977000000006, 31930.982000000004]}, "MouldFlow1_mean": {"_type": "uniform", "_value": [7550.0, 36424.32]}, "MouldFlow2_mean": {"_type": "uniform", "_value": [7550.0, 8500.0]}, "MouldFlow3_mean": {"_type": "uniform", "_value": [7550.0, 8500.0]}, "SP_mean": {"_type": "uniform", "_value": [81659.177, 125933.012]}, "EL_CYC_TIME": {"_type": "uniform", "_value": [9.238, 42.073]}, "EL_IV_END_TIME": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_PLST_TIME": {"_type": "uniform", "_value": [0.0, 9.933]}, "EL_LAST_COOL_TIME": {"_type": "uniform", "_value": [0.0, 9.21]}, "EL_MC_TIME": {"_type": "uniform", "_value": [0.406, 3.276]}, "EL_MO_TIME": {"_type": "uniform", "_value": [0.0, 2.141]}, "EL_EF_TIME": {"_type": "uniform", "_value": [0.0, 2.092]}, "EL_EB_TIME": {"_type": "uniform", "_value": [0.0, 3.127]}, "EL_IDLE_TIME": {"_type": "uniform", "_value": [0.0, 4.412]}, "EL_CORE_IN_TIME_0": {"_type": "uniform", "_value": [0.0, 1.1]}, "EL_CORE_IN_TIME_1": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_CORE_IN_TIME_2": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_CORE_IN_TIME_3": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_CORE_OUT_TIME_0": {"_type": "uniform", "_value": [0.0, 1.1]}, "EL_CORE_OUT_TIME_1": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_CORE_OUT_TIME_2": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_CORE_OUT_TIME_3": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_IV_TIME": {"_type": "uniform", "_value": [1.5219999999999998, 2.128]}, "EL_PP_TIME": {"_type": "uniform", "_value": [2.927, 13.502]}, "EL_SB1_TIME": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_SB2_TIME": {"_type": "uniform", "_value": [0.0, 1.497]}, "EL_CF_TIME": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_CB_TIME": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_IV_STG_TIME_0": {"_type": "uniform", "_value": [0.311, 0.358]}, "EL_IV_STG_TIME_1": {"_type": "uniform", "_value": [0.34600000000000003, 0.914]}, "EL_IV_STG_TIME_2": {"_type": "uniform", "_value": [0.527, 0.7440000000000001]}, "EL_IV_STG_TIME_3": {"_type": "uniform", "_value": [0.106, 0.229]}, "EL_IV_STG_TIME_4": {"_type": "uniform", "_value": [0.048, 0.353]}, "op_mode": {"_type": "uniform", "_value": [2.0, 4.0]}, "ext_mold_time_flt_0": {"_type": "uniform", "_value": [0.0, 28000.0]}, "EL_VPS_TIME": {"_type": "uniform", "_value": [0.0, 0.0]}, "ext_mold_time_flt_1": {"_type": "uniform", "_value": [0.0, 8000.0]}, "ext_mold_time_flt_2": {"_type": "uniform", "_value": [0.0, 20000.0]}, "ext_mold_time_flt_3": {"_type": "uniform", "_value": [0.0, 2000.0]}, "ext_mold_time_flt_4": {"_type": "uniform", "_value": [0.0, 2000.0]}, "EL_GET_PART_TIME": {"_type": "uniform", "_value": [0.0, 0.0]}, "EL_NZL_MEAN": {"_type": "uniform", "_value": [248.00876, 252.9367]}, "EL_BARREL_MEAN_1": {"_type": "uniform", "_value": [258.09712, 270.70891]}, "EL_BARREL_MEAN_2": {"_type": "uniform", "_value": [248.4281, 258.29715]}, "EL_BARREL_MEAN_3": {"_type": "uniform", "_value": [238.21912000000003, 246.75429]}, "EL_BARREL_MEAN_4": {"_type": "uniform", "_value": [211.84534, 219.92403]}, "EL_BARREL_MEAN_5": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_BARREL_MEAN_6": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_BARREL_MEAN_7": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_0": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_1": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_2": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_3": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_4": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_5": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_6": {"_type": "uniform", "_value": [15.0, 15.0]}, "EL_HTRN_TI_7": {"_type": "uniform", "_value": [15.0, 15.0]}, "ext_mold_temp_flt_0": {"_type": "uniform", "_value": [15.0, 68.2]}, "ext_mold_temp_flt_1": {"_type": "uniform", "_value": [15.0, 58.9]}, "ext_mold_temp_flt_2": {"_type": "uniform", "_value": [15.0, 31.7]}, "ext_mold_temp_flt_3": {"_type": "uniform", "_value": [15.0, 103.5]}, "ext_mold_temp_flt_4": {"_type": "uniform", "_value": [0.0, 0.0]}, "ext_mold_temp_flt_5": {"_type": "uniform", "_value": [0.0, 0.0]}, "ext_mold_temp_flt_6": {"_type": "uniform", "_value": [0.0, 0.0]}, "ext_mold_temp_flt_7": {"_type": "uniform", "_value": [0.0, 0.0]}, "rob_enbl": {"_type": "uniform", "_value": [1.0, 1.0]}, "ext_mold_flow_flt_0": {"_type": "uniform", "_value": [5.0, 18.084266]}, "EL_MOP": {"_type": "uniform", "_value": [10.0, 289.28607]}, "EL_PLST_END_ERR": {"_type": "uniform", "_value": [10.0, 95.41502]}, "EL_INJ_START_STR": {"_type": "uniform", "_value": [104.02236, 106.61102]}, "EL_IV_END_STR": {"_type": "uniform", "_value": [39.980728000000006, 42.0]}, "EL_VPS_END_STR": {"_type": "uniform", "_value": [16.880153999999994, 41.98508]}, "EL_MIN_INJ_STR": {"_type": "uniform", "_value": [16.598343, 39.419006]}, "EL_EF_END_POS": {"_type": "uniform", "_value": [48.023884, 50.703327]}, "EL_PP_STG_POS_0": {"_type": "uniform", "_value": [21.992069, 44.527035]}, "EL_PP_STG_POS_1": {"_type": "uniform", "_value": [21.335789000000002, 44.223236]}, "EL_PP_STG_POS_2": {"_type": "uniform", "_value": [16.876579, 36.093275]}, "EL_PP_STG_POS_3": {"_type": "uniform", "_value": [10.0, 10.0]}, "EL_PP_STG_POS_4": {"_type": "uniform", "_value": [10.0, 10.0]}, "EL_CLAMP_PRESS": {"_type": "uniform", "_value": [53.827827, 133.01634]}, "EL_MAX_INJ_PRESS": {"_type": "uniform", "_value": [62.843433, 136.74251]}, "EL_IV_END_PRESS": {"_type": "uniform", "_value": [41.594925, 58.31493]}, "EL_CF_END_PRESS": {"_type": "uniform", "_value": [5.0, 5.0]}, "EL_PLST_BP_0": {"_type": "uniform", "_value": [5.0, 11.872882]}, "EL_PLST_BP_1": {"_type": "uniform", "_value": [5.0, 11.8771534]}, "EL_PLST_BP_2": {"_type": "uniform", "_value": [5.0, 5.0]}, "EL_PLST_BP_3": {"_type": "uniform", "_value": [5.0, 5.0]}, "EL_PLST_BP_4": {"_type": "uniform", "_value": [5.0, 5.0]}}