JPS60238975A - Correcting method of time-series data - Google Patents

Abstract

PURPOSE:To increase the stability of a data input to variation and to speed up the response by using a moving average method and an index smoothing method in combination. CONSTITUTION:An index smoothed average value is calculated by the index smoothing method between the up-to-date moving average value and the preceding index average value after the moving averaging arithmetic of data inputs in time series is performed for every specified numbers constant all the time, and this is used as a data correction output. Namely, a variable C corresponding to the counted number of data inputs is set to ''0'' in a step 1 and a data input value D is read in the next step 2. Further, the moving average value is calculated in stpes 3-5, the index smoothed average value is calculated in steps 6-9, and the index smoothed average value Xe calculated in the step 8 or 9 is displayed on a display device. A regression to the step 2 is made after the display and a next data input value D is red and operations in the step 3 and succeeding steps are repeated.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for correcting measurement data when processing time-series measurement data from various measuring devices such as a fish finder, a seabed depth measuring device, and a tidal current meter.

Generally, the measurement data output from the detection section of a measurement device contains electrical noise and systematic errors caused by disturbances, so it is necessary to first correct this output and send it to the next processing device, transmitter, or display device. Conventionally, either the moving average method or the exponential smoothing method has been adopted as this correction method.

Needless to say, the moving average method calculates the average value between a predetermined number of latest time series data, and has excellent stability. However, since a predetermined number of time series data are always stored, Not only does this require a corresponding shift register, but it also has the disadvantage of poor ability to follow changes in time-series data and slow response. For example, as shown in Figure 1, if data labor suddenly drops at time Ta,
The moving average value Xm falls in a gentle downward line,
At time point TI), which is a considerable amount of time after time point Ta,
Data manpower reaches a depressed level. In this example, the number of data input values to be moving averaged is 28.

On the other hand, the exponential smoothing method weights and averages the current data input value and the previous data input value (calculated as the previous exponential smoothing average value), and the exponential smoothing average value
The calculation formula for eo is given by the following formula: Lo Xeo = [(R1)
is the previous exponential smoothed average value, and Xn is the current data input value. As can be seen from this equation, the weight added to the current data input Xn is 1/R, the weight to the previous data input is (R-1)/R2, and the weight to the two previous data inputs is ( R-1)2/R3, and its weight decreases sequentially. This exponential smoothing averaging method does not require many shift registers; all that is required is a shift register that stores the previous exponential smoothing average value Xeo' and the current data input value X; Even if the data processing rate suddenly decreases, the exponential smoothed average value Xeo (proportionality constant R = 4
(calculated as ) reaches a depressed level of data input almost instantaneously, and is quick to respond to changes in data input, but suffers from a lack of stability.

In view of the drawbacks mentioned above, the present invention aims to provide a data correction method that responds quickly to fluctuations in data input and has excellent stability. , after always performing a moving average calculation on a fixed number of time-series data inputs, the exponential smoothing average value is calculated using the exponential smoothing method between the latest moving average value and the previous exponential smoothing average value. The main purpose is to use this as a data correction output.

Therefore, the formula for calculating the moving average value Xm is as follows, assuming that n data input values Xo, X, , X2, . . . , Xn- are taken in.

Xm=(Xo+X++X2+--+X
n -1) / n・・・・・・・・・・・・(b) In addition, the above formula for calculating the exponential smoothed average value Xe is as follows: It is given by the following formula, where the moving average value is Xm.

Xe=[:(R-1)Xe'+Xm)/R...l
c) Hereinafter, the present invention will be explained in detail based on embodiments shown in the drawings. FIG. 2 is a flowchart showing each process of the method of the present invention, in which steps 3 to 5 are the steps for calculating the moving average value, and steps 6 to 9 are the steps for calculating the exponential smoothed average value = 3 - The steps to calculate are shown below. Therefore, in step 1, the variable C is set to O. This variable C corresponds to the count number of data input, and becomes a proportionality constant for calculating the exponential smoothing average value in the subsequent step 8. Next step 2
Read the data input value with . In step 3, which is a DO loop, the 11 data input values outside the moving average are shifted one by one. That is, in formula (b), Xo becomes x1, x1 becomes X2t, X2 becomes X3,
・= -X n-2 is replaced by Xn-, l, and so on. And in step 4, the calculation formula (b)
Input the data input value already read in step 2 into Xo. As a result, n data input values, including the latest data input value, are available to be subjected to the moving average. For these n data input values,
In step 5, a moving average value Xm is calculated based on formula (b).

The following is an exponential smoothing average value × based on this moving average value
When calculating e, 1 is added to variable C in step 6. In step 7, this variable C is compared with a proportionality constant R that is set in advance, and if the variable C is smaller than the proportionality constant R, the process moves to step 8.
The latest moving average value Xm with variable C as a proportionality constant
An exponentially smoothed average value Xe is calculated using the previous exponentially smoothed average value Xe'. The calculation formula is obtained by substituting D for R in the calculation formula (c). That is, Xe = ((D-1)Xe' +
Alternatively, if it becomes larger than the proportionality constant R, the process moves to Stellar 7'9, and the exponential smoothed average value Xe is calculated based on the calculation formula (c) with R as the proportionality constant. Step 10 is Step 8
Alternatively, the exponential smoothed average value Xe determined in step 9 is displayed on the display.

When the display is completed, the process returns to step 2, reads the next data input value, and repeats the operations from step 3 described above. Each time, the latest data input value is added to the moving averaged data input value, and the oldest value is removed. Also, the variable C increases one by one.

FIG. 1 shows the waveform of the output value, ie, the exponentially smoothed average value Xe, when corrected by the method of the present invention. Here, the number n of data input values subjected to the moving average is 7, and the proportionality constant R of exponential smoothing is 4. As can be seen from this waveform, the response of the output value Xe is slow at first, just like the output value Xm of the conventional moving average, but the response gradually accelerates and is almost the same as the output value Xeo of the conventional exponential smoothing. At this point, the level of the data input value is reached and stabilized.

- As mentioned above, the method of the present invention is a combination of the conventional moving average method and exponential smoothing method so as to take advantage of their respective advantages, and is highly stable against fluctuations in data input. , and the response is fast.

In addition, since the method of the present invention performs data correction twice, at the stage of calculating the moving average value, it is only necessary to perform the moving average on a much smaller number of data input values than in the conventional moving average method. It does not require a large number of shift registers to store data input values for averaging, and combined with the fact that the data processing procedure is relatively simple, it has the advantage of being easy and inexpensive to implement.

[Brief explanation of drawings]

FIG. 1 is a waveform diagram showing a comparison between the output of the conventional correction method and the output of the method of the present invention, and FIG. 2 is a waveform diagram showing the output of the method of the present invention.
0-Chart. Xm...Corrected output value (moving average value) by conventional moving average, Xeo...Corrected output value by conventional exponential smoothing (
(exponential smoothed average value), Xe...Corrected output value of the method of the present invention. Applicant Furuno Electric Co., Ltd. Agent Patent Attorney Oka 1) Kazuhide 8-

Claims (1)

[Claims] (1) After always performing a moving average calculation on a fixed number of time-series data inputs, the exponential smoothing average value is calculated using the exponential smoothing method between the latest moving average value and the previous exponential smoothing average value. A time-series data correction method characterized in that the data is used as a data correction output.