Measured Value Processing Method and Measured Value Processing System

Abstract

The present invention comprises: a reference value searching step for searching for a reference value R from among N measurement values; a selection step for selecting M measurement values by excluding, from the estimation target, N-M measurement values, as outliers, in descending order of deviation from the reference value R, among the N measurement values, on the basis of the reference value R searched for in the reference value searching step; and an estimation value calculation step for setting an average value of the M measurement values selected in the selection step as an estimation value of a true value. The reference value searching step includes a loop repeating from a state n=N until n=1, a process for determining the median or average value of the n measurement values, excluding one measurement value with the largest deviation from the median value or the average value among the n measurement values, and setting n-1 measurement values as candidates for the reference value R, and after the loop ends, one last remaining measurement value as the candidate is set as the reference value R.

Description

Technical Field

[0001] The present invention relates to a measured value processing method and a measured value processing system.

Background Art

[0002] In a pretreatment apparatus or the like that performs pretreatment of a sample to be analyzed, the liquid level height in a container that houses the sample or reagent may be measured to monitor the remaining amount of the sample or reagent. The liquid level height in the container can be measured with a non-contact liquid level sensor such as an ultrasonic sensor. However, the liquid level sensor itself includes a measurement error, and an accurate liquid level height may not be obtained by a single measurement. Therefore, it is preferable to measure the liquid level height a plurality of times and obtain the average value of these measured values as an estimated value of the true value.

Summary of the Invention

Problems to be Solved by the Invention

[0003] When measuring the liquid level height in a container installed in an apparatus such as a pretreatment apparatus, the liquid level in the container sways due to vibrations caused by the operation of a motor mounted on the apparatus, etc., and as a result, some measured values deviate from the true value and become outlier values. Furthermore, it has been found that the distribution of the measured values becomes a non-normal distribution. The average value of such non-normal distributed measured values tends to be a value with a large difference from the true value, and there is a risk that the error between the obtained average value and the true value exceeds the allowable error for the measurement of the liquid level height. Therefore, it is preferable to exclude outlier values from the calculation target of the average value before taking the average of the measured values. However, since the distribution of the measured values is a non-normal distribution, it is difficult to set a criterion for determining outlier values, and there is a possibility that outlier values cannot be accurately excluded.

[0004] The present invention has been made in view of the above problems, and an object thereof is to accurately exclude outlier values even when the distribution of measured values is a non-normal distribution so as to obtain an estimated value close to the true value.

Means for Solving the Problems

[0005] The measurement value processing method according to the present invention is a measurement value processing method for estimating the true value of a measurement object by using N measurement values obtained by performing a large number of N measurements on a measurement object whose measurement value distribution is a non-normal distribution, and the measurement value processing method includes: a reference value search step of searching for a reference value R from among the N measurement values; a selection step of selecting M measurement values by excluding, as outliers, N-M measurement values in order from the ones with larger deviations from the reference value R among the N measurement values, based on the reference value R searched in the reference value search step; an estimated value calculation step of using the average value of the M measurement values selected in the selection step as the estimated value of the true value, and the reference value search step includes a loop that repeats, from the state where n = N until n = 1, a process of obtaining the median or average value of the n measurement values, excluding one measurement value with the largest deviation from the median or the average value among the n measurement values, and using the n-1 measurement values as candidates for the reference value R, and after the loop ends, using the one measurement value that remains last as the reference value R.

[0006] The measurement value processing system according to the present invention includes an information storage unit that stores N measurement values obtained by performing a large number of N measurements on a measurement object whose measurement value distribution is a non-normal distribution, and an arithmetic unit that performs an estimation process of the true value of the measurement object by using the N measurement values stored in the information storage unit, and in the estimation process, the arithmetic unit includes a reference value search step of searching for a reference value R from among the N measurement values; a selection step of selecting M measurement values by excluding, as outliers, N-M measurement values in order from the ones with larger deviations from the reference value R among the N measurement values, based on the reference value R searched in the reference value search step; An estimated value calculation step of using the average value of the M measured values selected in the selection step as an estimated value of the true value, and is configured to execute, and In the reference value search step, the arithmetic unit obtains the median or average value of the n measured values, and excludes one measured value having the largest deviation from the median or the average value among the n measured values, and uses the n - 1 measured values as candidates for the reference value R. The arithmetic unit executes a loop that repeats the process from the state where n = N to the state where n = 1, and is configured to use the one measured value remaining last after executing the loop as the reference value R.

Advantages of the Invention

[0007] According to the measured value processing method and the measured value processing system according to the present invention, the median or average value of n measured values is obtained, and one measured value having the largest deviation from the median or the average value among the n measured values is excluded, and the n - 1 measured values are used as candidates for the reference value R. The process is repeated from the state where n = N to the state where n = 1, and after the loop ends, the one measured value remaining last as the candidate is used as the reference value R for excluding N - M outliers from the N measured values. Therefore, the reference value R can be made closer to the true value. As a result, even when the distribution of the measured values is a non-normal distribution, the outliers can be accurately excluded and an estimated value close to the true value can be obtained.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Embodiments for Carrying Out the Invention

[0009] Hereinafter, embodiments of the measurement value processing method and the measurement value processing system according to the present invention will be described with reference to the drawings. In the following, an example in which the liquid level height in a container is the measurement target will be described, but the present invention is not limited to this, and it can be applied to any measurement target as long as the distribution of the measurement values obtained by performing multiple measurements is a non-normal distribution.

[0010] As shown in FIG. 1, the measurement value processing system 2 includes an information storage unit 4 and an arithmetic unit 6. A large number of measurement value data regarding the liquid level height in the container 10 acquired by the liquid level sensor 8 are taken into the measurement value processing system 2. The liquid level sensor 8 is, for example, an ultrasonic sensor. The large number of measurement value data taken into the measurement value processing system 2 are stored in the information storage unit 4. The arithmetic unit 6 is configured to perform an estimation process of the true value of the liquid level height in the container 10 using the large number of measurement value data stored in the information storage unit 4.

[0011] Here, the measurement value processing system 2 is a computer device such as a personal computer in which software for processing measurement data is incorporated. The information storage unit 4 is a function realized by a part of the storage area of the information storage device mounted on the computer device, and the arithmetic unit 6 is a function realized by a CPU (central processing unit) mounted on the computer device executing a predetermined program.

[0012] An example of the estimation process of the true value of the liquid level height executed by the arithmetic unit 6 will be described using the flowchart in FIG. 2. Here, the number of measurement values used for the estimation process is set to N (N>3).

[0013] First, search for the measured value to be used as the reference value R among the N target measured values (step 101). The reference value R is a value that serves as a criterion for selecting M (M < N) measured values described later, and is different from simple median or average values. The search procedure for the reference value R will be described later.

[0014] When the reference value R is determined, select M measured values from the N measured values to be used for calculating the estimated value of the true value using the reference value R (step 102). The selection of the M measured values is performed by excluding N - M measured values with large deviations from the reference value R among the N measured values as outliers from the calculation target of the estimated value.

[0015] Here, when there are multiple measured values with the same deviation from the reference value R, it is preferable to preferentially exclude the measured value with a larger value, that is, the measured value exceeding the reference value R, as an outlier. The ultrasonic sensor 8 measures the distance to the liquid surface based on the bounce-back time of ultrasonic waves from the liquid surface. However, when the liquid surface of the measurement target is fluctuating, the ultrasonic sensor 8 tends to measure the position closest to the ultrasonic sensor 8, that is, the highest position of the liquid surface. Therefore, the measured value of the liquid surface height in the container 10 is likely to be higher than the true value. Here, "the same deviation" includes not only the case where the deviations are exactly the same, but also the case where the deviations are approximated to the extent that they can be regarded as the same.

[0016] FIG. 4 shows measurement data obtained by measuring the liquid surface height in the container 10 at 300 μs intervals with the ultrasonic sensor 8. The horizontal axis of this data is the number of measurements, and the vertical axis is the measured value (unit: 0.1 mm). The true value of the liquid surface height in the container 10 is 65.4 mm. As is clear from this data, almost all the measured values (outliers) deviated from the true value are much higher than the true value. Thus, when measuring the liquid surface height using a liquid surface sensor such as an ultrasonic sensor, since the outlier is mostly a value larger than the true value, when selecting outliers based on the deviation from the reference value R and there are multiple measured values with the same deviation from the reference value, by preferentially treating the measured value with a larger value as an outlier, the selection of outliers can be accurately performed.

[0017] Returning to FIG. 2 to continue the description. After excluding N - M outliers from the target of calculating the estimated value as described above, the estimated value is calculated by taking the average of the remaining M measured values (step 103). Note that M being 1 is not excluded. When M is 1, the reference value R becomes the estimated value of the true value.

[0018] Next, the procedure for determining the reference value R will be described using the flowchart of FIG. 3.

[0019] In the initial stage, all of the N measured values are candidates for the reference value R, and the number of candidates for the reference value R, n = N (step 201). The n measured values that are candidates are sorted by the magnitude of the values (step 202), and the median of the n measured values is obtained (step 203). When n is even, two measured values will correspond to the candidates for the median. However, as described above, considering that outliers almost always exceed the true value, the smaller measured value among the two measured values is taken as the median. Thereby, a measured value closer to the true value can be used as the median.

[0020] After obtaining the median, the measured value with the largest deviation from the median among the n measured values that are candidates is excluded from the candidates (step 204). At this time, if there are multiple measured values with the largest deviation from the median, for the same reason as the selection of outliers, the larger measured value is preferentially excluded from the candidates. The above steps 202 to 204 are repeatedly executed until the number n of the measured values remaining as candidates becomes 1 (step 205), and the last remaining one measured value as a candidate is set as the reference value R (step 206). In this way, after excluding the measured value with the largest deviation from the median from the candidates, by repeating the process of obtaining the median of the n measured values remaining as candidates, the median gradually approaches the true value, and finally a reference value R close to the true value is obtained. By selecting outliers using such a reference value R, the measured values deviating from the true value can be surely set as outliers, and the estimation accuracy of the true value is improved.

[0021] In the above description, the median of n measured values is obtained, and the measured value with the largest deviation from the median is excluded from the candidates for the reference value R. However, the present invention is not limited to this. The average value of n measured values may be obtained, and the measured value with the largest deviation from the average value may be excluded from the candidates for the reference value R. In that case, the step 201 of sorting the n measured values is unnecessary.

[0022] FIG. 5 shows the distribution of estimated values for each measurement obtained by taking the average of each measured value and the seven (total of 15) measured values before and after it based on the measurement data of FIG. 4. FIG. 6 shows the distribution of estimated values for each measurement obtained by removing 10 outliers from among each measured value and the seven (total of 15) measured values before and after it based on the measurement data of FIG. 4 and taking the average of the remaining five measured values. For example, the estimated value of the true value in the 50th measurement is calculated using the measured values from the 43rd to the 57th measurements.

[0023] As is clear from comparing FIG. 5 and FIG. 6, in the comparative example where the average of 15 measured values is taken as the estimated value, the estimated value often deviates upward from the true value (65.4 mm). On the other hand, in the example where the average is taken after removing outliers, an estimated value close to the true value is mostly obtained, indicating that the estimation accuracy of the true value has been improved. Also, in the measurement data of FIG. 4, an error of -0.2 mm to +1.9 mm occurs with respect to the true value. However, in the example of FIG. 6, an error of only -0.1 mm to +0.4 mm occurs with respect to the true value, indicating that the measurement error has been significantly reduced.

[0024] Note that the embodiments described above are merely examples of embodiments of the measured value processing method and the measured value processing system according to the present invention. The embodiments of the measured value processing method and the measured value processing system according to the present invention are as follows.

[0025] One embodiment of the measured value processing method according to the present invention is a measured value processing method for estimating the true value of a measurement object using N measured values obtained by performing a large number of N measurements on a measurement object whose measured value distribution is a non-normal distribution, wherein The measurement value processing method is as follows: a reference value search step of searching for a reference value R from among the N measurement values; a selection step of selecting M measurement values by excluding N-M measurement values from among the N measurement values in order from the ones with larger deviations from the reference value R as outliers from the target of the estimation, based on the reference value R searched in the reference value search step; an estimated value calculation step of using the average value of the M measurement values selected in the selection step as the estimated value of the true value, and the reference value search step includes a loop of obtaining the median or average value of n measurement values, excluding one measurement value with the largest deviation from the median or the average value among the n measurement values, and using the remaining n-1 measurement values as candidates for the reference value R, repeating this process from the state where n = N until n = 1, and after the loop ends, using the one measurement value finally remaining as the candidate as the reference value R.

[0026] In the first aspect of the above embodiment, in the loop of the reference value search step, the median of the n measurement values is obtained, and one measurement value with the largest deviation from the median among the n measurement values is excluded, and the remaining measurement values are used as candidates for the reference value R.

[0027] In the second aspect of the above embodiment, the measurement target is the liquid level height in the container, the measurement values are obtained by an ultrasonic sensor, and when there are a plurality of measurement values with equal deviations from the reference value R when excluding one measurement value with the largest deviation from the median or the average value in the reference value search step, the measurement value with the larger value is preferentially excluded.

[0028] In the third aspect of the above embodiment, the measurement target is the liquid level height in the container, the measurement values are obtained by an ultrasonic sensor, and when there are a plurality of measurement values with equal deviations from the reference value R when excluding N-M measurement values as outliers from the target of the estimation in the selection step, the measurement value with the larger value is preferentially excluded from the target of the estimation.

[0029] In one embodiment of the measurement value processing system according to the present invention, an information storage unit stores N measurement values obtained by performing a large number of N measurements on a measurement object whose measurement value distribution is a non-normal distribution, and an arithmetic unit that performs an estimation process of the true value of the measurement object using the N measurement values stored in the information storage unit. In the estimation process, the arithmetic unit performs a reference value search step of searching for a reference value R from among the N measurement values, a selection step of selecting M measurement values by excluding N-M measurement values from among the N measurement values in order from the one with the largest deviation from the reference value R as outliers from the object of the estimation, with reference to the reference value R searched in the reference value search step, and an estimated value calculation step of setting the average value of the M measurement values selected in the selection step as the estimated value of the true value, and in the reference value search step, the arithmetic unit calculates the median or average value of the n measurement values, and repeats a loop of excluding the one measurement value with the largest deviation from the median or average value among the n measurement values and using the remaining n-1 measurement values as candidates for the reference value R until n=N becomes n=1. After the execution of the loop is completed, the arithmetic unit is configured to set the one measurement value that remains as the candidate finally as the reference value R.

[0030] In the first aspect of the above embodiment, in the loop of the reference value search step, the arithmetic unit calculates the median of the n measurement values, and is configured to exclude the one measurement value with the largest deviation from the median among the n measurement values and use the remaining measurement values as candidates for the reference value R.

[0031] In the first aspect of the above embodiment, the measurement object is the liquid level height in a container, and the measurement value is obtained by an ultrasonic sensor. When excluding one measurement value with the largest deviation from the median or the average value in the reference value search step, if there are multiple measurement values with the same deviation from the reference value R, the arithmetic unit is configured to preferentially exclude the measurement value with the larger value.

[0032] In the third aspect of the above embodiment, the measurement target is the liquid level height in the container, and the measurement value is obtained by an ultrasonic sensor. When excluding N-M measurement values as outliers from the estimation target in the selection step, if there are multiple measurement values with the same deviation from the reference value R, the arithmetic unit is configured to preferentially exclude the measurement value with the larger value from the estimation target.

Explanation of Reference Signs

[0033] 2 Measurement value processing system 4 Information storage unit 6 Arithmetic unit 8 Ultrasonic sensor 10 Container

Claims

1. A measured value processing method for estimating the true value of a measurement target by using N measured values obtained by performing a large number of N measurements on a measurement target whose distribution of measured values is a non-normal distribution, The measured value processing method includes: A reference value search step of searching for a reference value R from among the N measured values; A selection step of selecting M measured values by excluding N−M measured values from the N measured values in order from the one with the largest deviation from the reference value R as outliers from the object of the estimation, based on the reference value R searched in the reference value search step; An estimated value calculation step of using the average value of the M measured values selected in the selection step as the estimated value of the true value, and The reference value search step includes a loop of obtaining the median or average value of n measured values, excluding one measured value with the largest deviation from the median or average value among the n measured values, and using the remaining n−1 measured values as candidates for the reference value R, repeating the process from the state of n = N until n = 1, and after the loop ends, using the one measured value finally remaining as the candidate as the reference value R. Measured value processing method.

2. In the loop of the reference value search step, obtain the median of n measured values, and exclude one measured value with the largest deviation from the median among the n measured values, and use the remaining measured values as candidates for the reference value R. The measured value processing method according to claim 1.

3. The measurement target is the liquid level height in a container, the measured values are obtained by an ultrasonic sensor, When there are a plurality of measured values having the same deviation from the reference value R when excluding one measured value with the largest deviation from the median or average value in the reference value search step, preferentially exclude the measured value with the larger value. The measured value processing method according to claim 1.

4. The measurement target is the liquid level height in a container, the measured values are obtained by an ultrasonic sensor, When excluding N - M of the measured values as outliers from the target of the estimation in the selection step, if there are a plurality of measured values having the same deviation from the reference value R, the measured value with the larger value is preferentially excluded from the target of the estimation. The measured value processing method according to claim 1.

5. An information storage unit that stores N measured values obtained by performing a large number of N measurements on a measurement target whose measured value distribution is a non - normal distribution; An arithmetic unit that performs an estimation process of the true value of the measurement target using the N measured values stored in the information storage unit, comprising: In the estimation process, the arithmetic unit A reference value search step of searching for a reference value R from among the N measured values; A selection step of selecting M measured values by excluding N - M measured values in order from the ones with larger deviations from the reference value R among the N measured values as outliers from the target of the estimation, with the reference value R searched in the reference value search step as a reference; An estimated value calculation step of setting the average value of the M measured values selected in the selection step as the estimated value of the true value, and In the reference value search step, the arithmetic unit executes a loop that repeatedly performs a process of obtaining the median or average value of n measured values, excluding the one measured value with the largest deviation from the median or average value among the n measured values, and using the remaining n - 1 measured values as candidates for the reference value R, starting from the state where n = N until n = 1. After the execution of the loop is completed, the arithmetic unit is configured to set the one measured value that remains as the candidate finally as the reference value R. A measured value processing system.

6. In the loop of the reference value search step, the arithmetic unit is configured to obtain the median of n measured values and exclude the one measured value with the largest deviation from the median among the n measured values, and use the remaining measured values as candidates for the reference value R. The measured value processing system according to claim 5.

7. The measurement target is the liquid level height in the container, and the measurement value is obtained by an ultrasonic sensor. When there are a plurality of measurement values having the same deviation from the reference value R when the arithmetic unit excludes one measurement value having the largest deviation from the median value or the average value in the reference value search step, the arithmetic unit is configured to preferentially exclude the measurement value having a larger value. The measurement value processing system according to claim 5.

8. The measurement target is the liquid level height in the container, and the measurement value is obtained by an ultrasonic sensor. When there are a plurality of measurement values having the same deviation from the reference value R when the arithmetic unit excludes N - M of the measurement values as the outliers from the estimation target in the selection step, the arithmetic unit is configured to preferentially exclude the measurement value having a larger value from the estimation target. The measurement value processing system according to claim 5.

Images