WO2020093644A1 - 一种基于小数据动态预测模型的疲劳预警方法 - Google Patents

一种基于小数据动态预测模型的疲劳预警方法 Download PDF

Info

Publication number
WO2020093644A1
WO2020093644A1 PCT/CN2019/079213 CN2019079213W WO2020093644A1 WO 2020093644 A1 WO2020093644 A1 WO 2020093644A1 CN 2019079213 W CN2019079213 W CN 2019079213W WO 2020093644 A1 WO2020093644 A1 WO 2020093644A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
prediction model
model
method based
fatigue
Prior art date
Application number
PCT/CN2019/079213
Other languages
English (en)
French (fr)
Inventor
陈梅芬
仵博
陈锐浩
Original Assignee
深圳职业技术学院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳职业技术学院 filed Critical 深圳职业技术学院
Publication of WO2020093644A1 publication Critical patent/WO2020093644A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment

Definitions

  • the invention relates to the field of fatigue early warning methods, in particular to a fatigue early warning method based on a small data dynamic prediction model.
  • the main purpose of the present invention is to establish a dynamic prediction model based on small data to get rid of dependence on network communication. Realize the monitoring of driving fatigue, learning fatigue and sports fatigue.
  • the data in step 1) is small data, which is 3-scatter data.
  • the prediction model includes an exponential model, a linear model, a logarithmic model, a polynomial model and a power model.
  • the high-fit prediction model is matched with the low-fit model to determine the predicted trend range, that is, the corresponding R2 value is used as the predicted trend range.
  • step 3 if the collected data reaches or exceeds a preset threshold, an early warning is issued.
  • step 4 if both the collected data and the predicted trend range reach or exceed a preset threshold, a continuous warning is continued.
  • the data is blink data, blood pressure data, pulse data, skin temperature data or electromyography data.
  • the data is blink data;
  • the high-fit prediction model is a polynomial quadratic function model.
  • the threshold is set to 4 times the average of the initial three eye blinks.
  • the present invention has the following beneficial effects:
  • the present invention is based on experimental data, by analyzing the prediction model R2 value (R2 value represents the fitting index, the value is less than or equal to 1, the closer to 1, indicating that the model fits higher), to determine the range of forecast trends, to achieve the forecast model Of the establishment. After the model is established, it is necessary to regularly collect data to modify the model. Then, the range of the predicted trend is compared with the fatigue threshold for fatigue warning. The data volume is moderate, the prediction accuracy is high, and the applicable data range is wide.
  • Figure 1 Predictive model flow chart
  • Figure 2 selects 3 commonly used models for scatter data modeling
  • a fatigue early warning method based on a small data dynamic prediction model includes the following steps:
  • the prediction models include exponential model, linear model, logarithmic model, polynomial model and power model.
  • the current data required to build a prediction model is small data, including 2-5 scattered data.
  • R2 value ( 1) the high-fit prediction model, and let the R2 value the smallest be the low-fit prediction model.
  • the calculation method of R2 value in this step includes the following:
  • R2 value regression sum of squares (ssreg) / total sum of squares (sstotal)
  • the regression sum of squares total sum of squares-residual sum of squares (ssresid).
  • RSQ RSQ
  • SPSS Statistical Product and Service Solutions
  • Statistical Product and Service Solution software
  • Step 3 Determine the predicted trend range based on the high-fit prediction model and the low-fit model, that is, use the corresponding R2 value as the predicted trend range. After determining the predicted trend range, collect data at a certain time interval to detect the predicted trend range and judge Whether the collected data is within the predicted trend range, if so, there is no need to modify the predicted trend range, enter 4), if not, return to step 1), re-establish a prediction model based on the collected data, and modify the predicted trend range.
  • step 3 if the collected data reaches or exceeds a preset threshold, an early warning is issued.
  • step 4 if both the collected data and the predicted trend range reach or exceed a preset threshold, a continuous warning is continued.
  • the data of the present invention is blink data, blood pressure data, pulse data, skin temperature data or electromyography data or other physiological data.
  • Data collection may not be uniform, especially for long-term predictions such as fatigue driving, the initial collection interval should be short, which is convenient to determine the threshold and prediction model as soon as possible.
  • the intermediate collection interval should be long to avoid data redundancy. Once the threshold is approached, the collection interval should be gradually shortened, and the forecast trend should be gradually shortened accordingly, in order to improve the forecast accuracy.
  • the setting of the preset threshold value is slightly different for different data types, and should be based on experimental tests.
  • the R2 value of all models is 1, which cannot be predicted.
  • the predicted trend range is 11.35 to 16, which is fitted with the fourth actual value of 13.
  • the predicted trend range is 11 to 13.26, which does not fit the fifth actual value of 8.
  • the predicted trend range can be determined.
  • the forecast trend range After the forecast trend range is determined, data is collected at certain intervals between the forecast periods to detect the forecast range. When the collected data falls within the predicted trend range, there is no need to modify the predicted range. When the collected data does not fall within the predicted trend range, the model needs to be re-established based on the collected data to correct the predicted trend range.
  • the threshold range of the average number of eye blinks (12 to 15 times per minute) should also be referenced to 12 * 4 to 15 * 4, that is, 48 to 60.
  • 48 to 60 4 times the average of the initial 3 data is taken as the threshold.
  • the lower limit of 48 is taken as the threshold.
  • the upper limit of 60 is taken as the threshold value.
  • eye blink data as an example to illustrate the whole process of building a model to make predictions.
  • Figure 3 General diagram of eye blink scatter coordinates. In order to describe the detection and prediction range of the collected data, the collected data is every 5 minutes. So the horizontal coordinate takes 5 minutes as the basic unit. Each modeling will reconstruct the origin of the horizontal coordinate.
  • Figure 4 is a prediction model of scatter 1 to 3 scatters of eye blinking, with a predicted trend range of 6.74 to 26.00, and the actual value of the fourth point of the general coordinate map is 11, fitting.
  • the predicted trend range is 7.01 ⁇ 20.93, and the actual value of the 5th point of the general coordinate map is 13, fitting.
  • Figure 5 Blinking scatter 3 to 5 scatter coordinates (including detection points), the predicted trend range is 12.90 to 16.01, and the actual value of the 6th point of the general coordinate map is 27. It is not fitted and needs to be corrected.
  • the forecast trend range is 24.78 ⁇ 53, and the actual value of the 7th point of the coordinate total map is 18. Does not fit and needs to be corrected.
  • the correction method is the same as 3 to 5.
  • the predicted trend range is -14 to 27.50, and eye blinking cannot be negative, so the predicted trend range is 0 to 27.50, the actual value of the eighth point of the coordinate total map is 27, and the fit.
  • the predicted trend range is 24.00 ⁇ 54.00, and the actual value of the 9th point of the general coordinate map is 19, which is not fitted and needs to be corrected.
  • the correction method is the same as 3 to 5. During the correction process, when the revised predicted trend range reaches or exceeds the threshold of 29.33, an early warning is required, otherwise the early warning is lifted.
  • the predicted trend range is -6.01 to 22.33, and eye blinking cannot be negative, so the predicted trend range is 0 to 22.33, and the actual value of the 10th point of the coordinate total map is 12, fitting.
  • the forecast trend range is 6.01 ⁇ 11.84, and the actual value of the 11th point of the general coordinate map is 40. It is not fitted and needs to be corrected.
  • the correction method is the same as 3 to 5. During the correction process, when the collected test data reaches or exceeds the threshold of 29.33, an early warning is required, otherwise the early warning is lifted.
  • the predicted trend range is 27.31 ⁇ 103.01, and the actual value of the 12th point of the general coordinate map is 42. It fits and reaches the threshold, and needs continuous warning.
  • the actual value of the 13th point of the inspection coordinate map is 35, which has also reached the threshold value and requires early warning. Explain that the prediction is accurate.
  • Eye blink 1 time / 15 minutes, 120 minutes in total
  • Eye blink 2 time / 20 minutes, 240 minutes in total
  • the invention provides a fatigue early warning method based on a small data dynamic prediction model.
  • the range of the prediction trend is determined to realize the establishment of the prediction model.
  • fatigue warning is carried out.
  • the invention has moderate data volume, high prediction accuracy, wide applicable data range and good industrial applicability.

Landscapes

  • Medical Informatics (AREA)
  • Engineering & Computer Science (AREA)
  • Public Health (AREA)
  • Health & Medical Sciences (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

一种基于小数据动态预测模型的疲劳预警方法,通过分析预测模型R2值,确定预测趋势的范围,实现预测模型的建立,所述R2值表示拟合指数,值小于等于1,越接近1,说明模型拟合度越高。模型建立后,还要定期采集数据对模型进行修正。再对预测趋势的范围与疲劳阈值进行比较,进行疲劳预警。上述方案数据量适中,预测准确度高,适用数据范围广。

Description

一种基于小数据动态预测模型的疲劳预警方法 技术领域
本发明涉及疲劳预警方法领域,特别是一种基于小数据动态预测模型的疲劳预警方法。
背景技术
当前是大数据的时代,很多预测都是通过大数据运算实现。大数据后台运算的前提必须有良好的网络为基础,确保数据的有效传输。但由于环境的复杂性,有些时候无法保证网络通信的畅通。如网络覆盖不好的偏僻山区,自然灾害,太阳黑子活动等。一旦出现这些情况,依赖大数据通信的预测将无法正常工作。
现有的时间序列模型预测方法,对初始数据量要求比较多,要进行数据预处理,相关性分析,回归分析等。过程复杂繁琐,比较适合大数据运算实现。
发明内容
本发明的主要目的在于建立基于小数据的动态预测模型,摆脱对网络通信的依赖。实现对驾驶疲劳,学习疲劳,运动疲劳的监测。
本发明采用如下技术方案:
一种基于小数据动态预测模型的疲劳预警方法,其特征在于,包括如下步骤:
1)引入当前数据分别建立若干个预测模型;
2)计算各个预测模型的R2值,令R2值等于1的作为高拟合预测模型;令R2值最小的作为低拟合预测模型;
3)根据高拟合预测模型和低拟合模型确定预测趋势范围,采集数据,并判断该数据是否属于预测趋势范围,若是,则进入4),若否,回到步骤1);
4)判断预测趋势范围是否达到或超过预设阈值,若是,则预警;若否,则回到步骤1)。
优选的,在步骤1)中所述数据为小数据,其为3散点数据。
优选的,所述预测模型包括指数模型、线性模型、对数模型、多项式模型和乘幂模型。
优选的,所述将高拟合预测模型搭配低拟合模型确定预测趋势范围,即将其对应的R2值作为预测趋势范围。
优选的,所述步骤3)中,若采集的数据达到或超过预设阈值,则预警。
优选的,所述步骤4)中,若采集的数据及预测趋势范围均达到或超过预设阈值,则持续预警。
优选的,所述数据为眨眼数据、血压数据、脉搏数据、皮温数据或肌电数据。
一个实施例中,所述数据为眨眼数据;所述高拟合预测模型为多项式的二次函数模型。
一个实施例中,所述阈值设定为初始3个眼眨动次数的平均值的4倍。
由上述对本发明的描述可知,与现有技术相比,本发明具有如下有益效果:
本发明是以实验数据为依据,通过分析预测模型R2值(R2值表示拟合指数,值小于等于1,越接近1,说明模型拟合度越高),确定预测趋势的范围,实现预测模型的建立。模型建立后,还要定期采集数据对模型进行修正。再对预测趋势的范围与疲劳阈值进行比较,进行疲劳预警。数据量适中,预测准确度高,适用数据范围广。
附图说明
图1预测模型流程图;
图2选择3散点数据建模的常用模型;
图3眼眨动散点坐标总图;
图4眼眨动散点1~3散点预测模型图;
图5眼眨动散点3~5散点坐标(含检测点)图;
图6眼眨动散点3~5散点预测模型图;
图7眼眨动散点3~5散点预测模型修正图1;
图8眼眨动散点3~5散点预测模型修正图2;
具体实施方式
以下通过具体实施方式对本发明作进一步的描述。
参见图1,一种基于小数据动态预测模型的疲劳预警方法,包括如下步骤:
1)引入当前数据分别建立若干个预测模型,该若干个预测模型包 括指数模型、线性模型、对数模型、多项式模型和乘幂模型等。建立预测模型所需的当前数据为小数据,包括2-5散点数据。
2)计算各个预测模型的R2值,令R2值等于1的作为高拟合预测模型,令R2值最小的作为低拟合预测模型。该步骤中R2值的计算方式包括如下:
在统计学中,R2值的计算方法如下:
R2值=回归平方和(ssreg)/总平方和(sstotal)
其中回归平方和=总平方和-残差平方和(ssresid)。总平方和:Const(常量)参数为True(真)的情况下,总平方和=y的实际值与平均值的平方差之和;Const(常量)参数为False(假)的情况下,总平方和=y的实际值的平方和。残差平方和:残差平方和=y的估计值与y的实际值的平方差之和。
在回归分析中,可以使用RSQ函数计算R2值。RSQ函数语法为RSQ(known_y's,known_x's),将源数据中的y轴数据和x轴数据分别代入,就可以求得其趋势线的R2值。
也可通过相关统计学软件直接得到,如SPSS(Statistical Product and Service Solutions),“统计产品与服务解决方案”软件。
3)根据高拟合预测模型和低拟合模型确定预测趋势范围,即将其对应的R2值作为预测趋势范围,确定预测趋势范围后,间隔一定的时间采集数据,以检测该预测趋势范围,判断该采集的数据是否处于预测趋势范围,若是,则无需修正预测趋势范围,进入4),若否,回到步骤1),根据采集的数据重新建立预测模型,修正预测趋势范围。
4)判断预测趋势范围是否达到或超过预设阈值,若是,则预警;若否,则回到步骤1)。
另外,在步骤3)中,若采集的数据达到或超过预设阈值,则预警。步骤4)中,若采集的数据及预测趋势范围均达到或超过预设阈值,则持续预警。
本发明的数据为眨眼数据、血压数据、脉搏数据、皮温数据或肌电数据或其他生理数据。数据采集可以不是均匀的,特别是像疲劳驾驶这种长时间预测,初始时采集间隔要短,便于尽快确定阈值和预测模型。中间采集间隔要长,避免数据冗余。一旦临近阈值,采集间隔要逐渐缩短,预测趋势也相应逐渐缩短,以便提高预测精度。
预设阈值的设定,不同的数据种类略有不同,要以实验测试为基准。
例如对于眼眨数据,假设以3个数据建模,预测第四个数据。所以取初始3个数据的平均值的4倍为阈值比较合理。还要参考一般眼眨动次数(每分钟12~15次)的阈值范围12*4~15*4,即48~60。低于48时,取初始3个数据的平均值的4倍为阈值。在48~60范围,取48为阈值。高于60取60为阈值。
应用举例:
通过对实验数据模型分析发现,如果利用2点建模,数据太少,只能建立单向模型,所有R2值为1,显然不适用。如果利用4点或4点以上建模,数据处理比较繁琐,采样周期偏长。优选利用3点建模,数据量适中,预测准确度高,适用数据范围广。
选择常用模型(指数、线性、对数、多项式、乘幂),以下面一组数 据为例说明选择个数不同数据点的区别:
表1眼眨动1被试2数据表
Figure PCTCN2019079213-appb-000001
◆选择初始的2散点数据常用模型:
Y1=6.125e0.1335x,R 2=1;Y2=x+6,R 2=1;
Y3=1.4427ln(x)+7,R 2=1;Y4=7x0.1926,R 2=1
所有模型的R2值为1,无法预测。
◆选择初始的3散点数据常用模型(R2值降序排列):
Y1=x2-2x+8,R 2=1;Y2=5.4146e0.226x,R 2=0.9472;
Y3=2x+4.6667,R 2=0.9231;Y4=6.7493x0.3879,R 2=0.861;
Y5=3.4042ln(x)+6.6335,R 2=0.8254
选择最大值R 2=1与最小值R 2=0.8254的模型,即选择预测Y在下一个周期的最大值与最小值。
Y1=x2-2x+8;Y5=3.4042ln(x)+6.6335
当x=4时,即下一个周期的时间,代入模型:
Y1=42-2*4+8=16;Y2=3.4042*ln(4)+6.6335=11.35
预测趋势范围11.35~16,与第4个实际值13是拟合的。
◆选择初始的4散点数据常用模型:
选择最大值R 2=1与最小值R 2=0.8826的模型,
Y1=-0.5x3+4x2-7.5x+11,R 2=1
Y2=4.3033ln(x)+6.3309,R 2=0.8826
当x=5时,即下一个周期的时间,代入模型:
Y1=-0.5*53+4*52-7.5*5+11=11
Y2=4.3033*ln(5)+6.3309=13.26
预测趋势范围11~13.26,与第5个实际值8是不拟合的。
因此选择3散点数据建立模型最理想。并且以多项式的二次函数模型为高拟合模型,搭配其他函数模型为低拟合模型,就可以确定预测趋势范围。
确定预测趋势范围后,在预测周期之间,间隔一定的时间采集数据检测预测范围。当采集的数据属于预测趋势范围内时,不需要修正预测范围。当采集的数据不属于预测趋势范围内时,需要根据采集的数据重新建立模型,修正预测趋势范围。
阈值设定:根据实验分析,由于选择3散点数据建立模型,用于预测下一周期的趋势范围。因此取初始3个眼眨动次数的平均值的4倍为阈值,比较合理。表1被试2的阈值为((7+8+11)/3)*4=34.67。
阈值设定除了按照建立模型方式设定,还要参考平均眼眨动次数(每分钟12~15次)的阈值范围12*4~15*4,即48~60。低于48时,取初始3个数据的平均值的4倍为阈值。在48~60范围,取48下限为阈值。因为每分钟眼眨动超过60次是异常的,所以高于60取上限60为阈值。
利用眼眨动数据为例说明建立模型进行预测的全过程。
表2眼眨动2被试5数据表
Figure PCTCN2019079213-appb-000002
阈值设定:该被试的阈值为(8+4+10)*4/3=29.33。
图3眼眨动散点坐标总图,为了描述采集数据检测预测范围,采集数据每5分钟1次。因此水平坐标取5分钟为基本单位。每次建模都将重建水平坐标原点。
◆眼眨动散点1~3建模:
Y1=0.3125x2-2.875x+10.563,R 2=1
Y2=6.9375x-0.011,R 2=0.0007
图4为眼眨动散点1~3散点预测模型图,预测趋势范围6.74~26.00,坐标总图第4个点实际值11,拟合。
◆眼眨动散点2~4建模:方法与眼眨动散点1~3建模相同。
Y1=-0.1562x2+2.4375x+1.7188,R 2=1
Y2=4.0417e0.1265x,R 2=0.82
预测趋势范围7.01~20.93,坐标总图第5个点实际值13,拟合。
◆眼眨动散点3~5建模:
Y1=0.0313x2+0.0625x+9.9063,R 2=1
Y2=1.2078ln(x)+9.8008,R 2=0.8089
图5眼眨动散点3~5散点坐标(含检测点)图,预测趋势范围12.90~16.01,坐标总图第6个点实际值27,不拟合,需要修正。
图6眼眨动散点3~5散点预测模型图,间隔5分钟采集一次检测数据,当x=10时,预测趋势范围12.58~13.66,采集数据为15,需要修正。用散点4、5和第一次采集点建模,修正后的模型为:
Y1=0.3x2-1.3x+12,R 2=1
Y2=1.8428ln(x)+10.911,R 2=0.8254
图7眼眨动散点3~5散点预测模型修正图1
当x=7时,预测趋势范围14.50~17.60,采集数据为18,需要修正。用散点5和第一、二次采集点建模,修正后的模型为:
Y1=0.5x2+0.5x+12,R 2=1
Y2=4.3718ln(x)+12.722,R 2=0.9314
图8眼眨动散点3~5散点预测模型修正图2
当x=4时,预测趋势范围18.78~22,采集数据为20,不需要修正。
当x=5时,预测趋势范围19.76~27,坐标总图第6个点实际值27,拟合。修正完成。
◆眼眨动散点4~6建模:
Y1=0.375x2-1.75x+12.375,R 2=1
Y2=6.0028ln(x)+9.3831,R 2=0.6135
预测趋势范围24.78~53,坐标总图第7个点实际值18。不拟合,需要修正。修正方法与3~5相同。
◆眼眨动散点5~7建模:
Y1=-0.6875x2+7.625x+6.0625,R 2=1
Y2=14.848e0.0474x,R 2=0.2695
预测趋势范围-14~27.50,眼眨动不可能为负,因此预测趋势范围0~27.50,坐标总图第8个点实际值27,拟合。
眼眨动散点6~8建模:
Y1=0.5625x2-5.625x+32.063,R 2=1
Y2=24,R 2=0
预测趋势范围24.00~54.00,坐标总图第9个点实际值19,不拟合,需要修正。修正方法与3~5相同。修正过程中,当修正后的预测趋势范围达到或超过阈值29.33时,需要预警,反之解除预警。
◆眼眨动散点7~9建模:
Y1=-0.5313x2+5.4375x+13.094,R 2=1
Y2=0.125x+20.708,R 2=0.0103
预测趋势范围-6.01~22.33,眼眨动不可能为负,因此预测趋势范围0~22.33,坐标总图第10个点实际值12,拟合。
◆眼眨动散点8~10建模:
Y1=0.0313x2-2.1875x+29.156,R 2=1
Y2=28.112x-0.337,R 2=0.8893
预测趋势范围6.01~11.84,坐标总图第11个点实际值40,不拟合,需要修正。修正方法与3~5相同。修正过程中,当采集的检测数据达到或超过阈值29.33时,需要预警,反之解除预警。
◆眼眨动散点9~11建模:
Y1=1.0938x2-8.3125x+26.219,R 2=1
Y2=16.075x0.2066,R 2=0.1496
预测趋势范围27.31~103.01,坐标总图第12个点实际值42,拟合,达到阈值,需要持续预警。
检验坐标总图第13个点实际值35,也达到阈值,需要预警。说明预测准确。
实验数据
眼眨动1:次/15分钟,共120分钟
Figure PCTCN2019079213-appb-000003
眼眨动2:次/20分钟,共240分钟
Figure PCTCN2019079213-appb-000004
上述仅为本发明的具体实施方式,但本发明的设计构思并不局限于此,凡利用此构思对本发明进行非实质性的改动,均应属于侵犯本发明保护范围的行为。
工业实用性
本发明一种基于小数据动态预测模型的疲劳预警方法,通过分析预测模型R2值,确定预测趋势的范围,实现预测模型的建立。通过对预测趋势的范围与疲劳阈值进行比较,进行疲劳预警。本发明数据量适中,预测准确度高,适用数据范围广,具有良好的工业使用性。

Claims (9)

  1. 一种基于小数据动态预测模型的疲劳预警方法,其特征在于,包括如下步骤:
    1)引入当前数据分别建立若干个预测模型;
    2)计算各个预测模型的R2值,令R2值等于1的作为高拟合预测模型;令R2值最小的作为低拟合预测模型;
    3)根据高拟合预测模型和低拟合模型确定预测趋势范围,采集数据,并判断该数据是否属于预测趋势范围,若是,则进入4),若否,回到步骤1);
    4)判断预测趋势范围是否达到或超过预设阈值,若是,则预警;若否,则回到步骤1)。
  2. 如权利要求1所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,在步骤1)中所述数据为小数据,其为3散点数据。
  3. 如权利要求1所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,所述预测模型包括指数模型、线性模型、对数模型、多项式模型和乘幂模型。
  4. 如权利要求1所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,所述将高拟合预测模型搭配低拟合模型确定预测趋势范围,即将其对应的R2值作为预测趋势范围。
  5. 如权利要求1所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,所述步骤3)中,若采集的数据达到或超过 预设阈值,则预警。
  6. 如权利要求1所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,所述步骤4)中,若采集的数据及预测趋势范围均达到或超过预设阈值,则持续预警。
  7. 如权利要求1所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,所述数据为眨眼数据、血压数据、脉搏数据、皮温数据或肌电数据。
  8. 如权利要求2或7所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,所述数据为眨眼数据;所述高拟合预测模型为多项式的二次函数模型。
  9. 如权利要求8所述的一种基于小数据动态预测模型的疲劳预警方法,其特征在于,所述阈值设定为初始3个眼眨动次数的平均值的4倍。
PCT/CN2019/079213 2018-11-09 2019-03-22 一种基于小数据动态预测模型的疲劳预警方法 WO2020093644A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811330724.7 2018-11-09
CN201811330724.7A CN109616201B (zh) 2018-11-09 2018-11-09 一种基于小数据动态预测模型的疲劳预警方法

Publications (1)

Publication Number Publication Date
WO2020093644A1 true WO2020093644A1 (zh) 2020-05-14

Family

ID=66003710

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/079213 WO2020093644A1 (zh) 2018-11-09 2019-03-22 一种基于小数据动态预测模型的疲劳预警方法

Country Status (2)

Country Link
CN (1) CN109616201B (zh)
WO (1) WO2020093644A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112903001A (zh) * 2019-12-03 2021-06-04 财团法人纺织产业综合研究所 织物定型机的操作方法
CN118177807B (zh) * 2024-05-17 2024-09-17 苏州大学 基于压力传感器的疲劳监测控制方法和控制系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104598993A (zh) * 2015-01-14 2015-05-06 南京信息工程大学 基于眨眼次数的驾驶员疲劳驾驶预测方法
US20160236617A1 (en) * 2013-03-11 2016-08-18 U.S. Army Research Laboratory Attn: Rdrl-Loc-I Apparatus and method for estimating and using a predicted vehicle speed in an indirect vision driving task
CN107992958A (zh) * 2016-10-26 2018-05-04 航天信息股份有限公司 基于arma的人口超限预警方法
CN108245171A (zh) * 2017-12-28 2018-07-06 东软集团股份有限公司 获取参数模型的方法、疲劳检测方法及装置、介质及设备

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101339511B1 (ko) * 2011-01-26 2013-12-10 한국환경공단 수질 다 항목을 이용한 총 질소 예측 장치 및 방법
CN108021535A (zh) * 2016-11-03 2018-05-11 北京国双科技有限公司 数据拟合方法和装置
CN106534844A (zh) * 2016-11-15 2017-03-22 中国科学院软件研究所 基于心电和眼电信号的三维显示视疲劳评估系统和方法
CN106934368A (zh) * 2017-03-13 2017-07-07 长安大学 一种基于眼动指标数据的驾驶疲劳检测系统及识别方法
CN107368922A (zh) * 2017-07-20 2017-11-21 华中师范大学 基于夜间灯光强度的城市住宅平均房价预估方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160236617A1 (en) * 2013-03-11 2016-08-18 U.S. Army Research Laboratory Attn: Rdrl-Loc-I Apparatus and method for estimating and using a predicted vehicle speed in an indirect vision driving task
CN104598993A (zh) * 2015-01-14 2015-05-06 南京信息工程大学 基于眨眼次数的驾驶员疲劳驾驶预测方法
CN107992958A (zh) * 2016-10-26 2018-05-04 航天信息股份有限公司 基于arma的人口超限预警方法
CN108245171A (zh) * 2017-12-28 2018-07-06 东软集团股份有限公司 获取参数模型的方法、疲劳检测方法及装置、介质及设备

Also Published As

Publication number Publication date
CN109616201B (zh) 2023-10-31
CN109616201A (zh) 2019-04-12

Similar Documents

Publication Publication Date Title
CN108375715B (zh) 一种配电网线路故障风险日预测方法及系统
CN108375476B (zh) 一种水电机组健康评估方法
CN110702437B (zh) 一种水轮发电机组油压系统健康状况的评价方法
CN105354614A (zh) 一种基于大数据的电网信息运维主动预警方法
WO2020093644A1 (zh) 一种基于小数据动态预测模型的疲劳预警方法
CN109947064B (zh) 智能通水温度控制专家系统及硬件检测和数据监测方法
CN106940678B (zh) 一种系统实时健康度评估分析方法及装置
CN116151621A (zh) 一种基于数据分析的大气污染治理风险检测系统
WO2022083375A1 (zh) 人员位置异常监测方法及装置
CN106856442A (zh) 一种性能指标监控方法和装置
CN110716978A (zh) 一种山区水库式水源地智能管理方法
CN115760525A (zh) 一种基于物联网智慧养老系统及方法
CN116307886A (zh) 一种企业生产状态的实时监测方法及装置
CN105302476B (zh) 一种用于核电站设备的可靠性数据在线采集分析存储系统及其存储方法
CN114469084B (zh) 一种基于高精度adc的血氧监测系统
CN115062199A (zh) 基于数据可视化的水源地水质在线监测系统
CN114582478A (zh) 一种基于互联网的磁共振室检测管理系统
CN102682212A (zh) 机电类产品的一种可靠性度量方法
CN118643435A (zh) 一种基于人工智能的异常能耗识别方法
CN105678388A (zh) 一种基于基线的运维管理健康分析方法
CN108109675B (zh) 一种实验室质量控制数据管理系统
CN108469571A (zh) 基于趋势线拟合的多元试验数据分析的电缆状态评价方法
CN118364337A (zh) 一种能源数据处理评估管理系统及方法
CN115753906A (zh) 一种基于多数据融合的智能气体监测与预警系统及方法
CN108108864B (zh) 一种实验室质量控制数据管理方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19882286

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19882286

Country of ref document: EP

Kind code of ref document: A1