Timestamp alignment method for conjoint analysis of high-frequency data in industrial field

Abstract

The invention discloses a timestamp alignment method for high-frequency data conjoint analysis in the industrial field. According to the method, a time sequence consists of a time window length, a timestamp, a numerical value and a data acquisition frequency; calculating the arrival rate of each data item in the time window; calculating the minimum arrival rate of all the data items, comparing the minimum arrival rate with a minimum arrival rate threshold value, and determining whether subsequent analysis is carried out or not; calculating a reference data frequency, and selecting a least common multiple or a largest common divisor according to an application scene; calculating a sequence of reference timestamps, each timestamp being an integral multiple of the reference data frequency; and calculating a numerical sequence corresponding to the reference timestamp sequence of each data item by adopting a nearest neighbor principle to obtain a final time sequence. According to the method, a proper reference time sequence is found based on an application scene, so that the curve form and distribution of the aligned data are kept unchanged, the characteristics of original data are reserved, the error rate of the data is reduced, and the accuracy and credibility of multi-data conjoint analysis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data application and analysis, and in particular to a timestamp alignment method for joint analysis of high-frequency data in an industrial field. Background Art

[0002] In data application and analysis research, a time series is an ordered, time-stamped collection of observations, each of which is a numerical scalar obtained from the same measurement process. In the industrial field, with the improvement of intelligent equipment maintenance requirements, higher requirements are put forward for the automation, intelligence, and real-time monitoring of equipment status. Usually, joint comparative analysis of multiple time series data items is required to more accurately reflect the operating status of the equipment.

[0003] For process-oriented production lines, data collection has the characteristics of many sensor types and high collection frequency, usually reaching the millisecond level, such as 10 milliseconds, 50 milliseconds or 100 milliseconds, and different collection frequencies corresponding to different process stages. In addition, transmission bandwidth, network delay, point loss, etc., will also cause the frequency of data collection to fail to reach the ideal state, resulting in inconsistent timestamps of data items. Inconsistent timestamps bring difficulties to the joint comparative analysis of multiple data items.

[0004] In the traditional timestamp alignment method, in order to make the two data items have the conditions for comparative analysis at the same time, the main thing is to confirm the benchmark time series. One way is to retain all non-repeating timestamps of the two data items, and fill in the missing parts of each data item after comparing it with the benchmark time series by using the mean, median, mode, etc.; another way is to only retain the timestamps shared by the two data items, and delete the redundant parts of each data item after comparing it with the benchmark time series.

[0005] In actual production, the probability of the timestamps of two data items coinciding at the same time is very low. Using the point-adding method will double the amount of data in the aligned time series, which is particularly evident in the alignment of high-frequency data, because its timestamps usually have slight differences at the millisecond level. Using the point-deleting method will cause a sharp decrease in the amount of data in the aligned time series, especially for low-frequency data. This method will eliminate sparse samples, lose important features, destroy data distribution, and reduce data quality and reliability. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a timestamp alignment method for joint analysis of high-frequency data in the industrial field. The method aligns the timestamps of multiple high-frequency data items within the same time window without changing the curve shape and distribution of the data, thereby improving the accuracy and credibility of the joint analysis of the data, and improving the data processing efficiency as well as the data quality and reliability.

[0007] In order to solve the above technical problems, the timestamp alignment method for joint analysis of high-frequency data in the industrial field of the present invention comprises the following steps:

[0008] Step 1: For the time series data item X within the time window length window i , which is timestamped by The corresponding value And data frequency step i composition,

[0009]

[0010] Where i = 1, 2, L, m, where m represents the time series data item X i The number of i Represents a time series data item X i The sequence length, time series t i The format of each timestamp is yyyy-MM-dd HH:mm:ss.SSS, which means year-month-day hour:minute:second.millisecond. Step means the data frequency, that is, how long it takes to collect a data point, in milliseconds.

[0011] Step 2: Calculate the data item X within the time window length window i The arrival rate i , get the data points in the time interval [start, end], the arrival rate rate i It is equal to the ratio of the actual points received to the expected points received.

[0012]

[0013] end=start+window (3)

[0014] Among them, start is the start timestamp of this data analysis, and end is the end timestamp of this data analysis;

[0015] Step 3: Calculate the minimum arrival rate among m data items min ,

[0016] rate min =min(rate 1 ,rate2 ,Lrate m ) (4)

[0017] Set the minimum arrival rate required for data joint analysis to Threshold. min If the value is ≥Threshold, the subsequent analysis will be performed. Otherwise, the minimum requirement of the analysis is not met and Threshold is set to 80%;

[0018] Step 4: Calculate the benchmark data frequency step base ,

[0019] step base = func(step i ),i=1,2L,m (5)

[0020] Among them, func is the least common multiple or greatest common divisor according to the application scenario. The least common multiple indicates downsampling, and the greatest common divisor indicates upsampling. Downsampling refers to reducing the sampling rate of the signal, which is characterized by reducing the amount of data; upsampling refers to increasing the sampling rate of the signal, which is characterized by increasing the amount of data.

[0021] Step 5: Calculate the base timestamp sequence t of the data item base , each timestamp is an integer multiple of the benchmark data frequency, and the timestamp start is divided into two parts: the date and time part start other and the millisecond part start SSS ,

[0022]

[0023] Among them, round means rounding to the nearest integer, and N is the length of the reference timestamp sequence;

[0024] Step 6: Calculate the numerical sequence corresponding to the base timestamp sequence of each data item Each base timestamp With time series The most recent timestamp The corresponding value

[0025]

[0026] Step 7. The final time series is:

[0027]

[0028] Since the timestamp alignment method for joint analysis of high-frequency data in the industrial field of the present invention adopts the above technical solution, that is, the time series of the method is composed of the time window length, timestamp, value and data acquisition frequency; the arrival rate of each data item in the time window is calculated; the minimum arrival rate of all data items is calculated, and compared with the minimum arrival rate threshold to confirm whether to perform subsequent analysis; the reference data frequency is calculated, and the least common multiple or the greatest common divisor can be selected according to the application scenario; the reference timestamp sequence is calculated, and each timestamp is an integer multiple of the reference data frequency; the nearest neighbor principle is used to calculate the numerical sequence corresponding to the reference timestamp sequence of each data item to obtain the final time series. This method overcomes the defects of the traditional high-frequency data timestamp alignment method, finds a suitable reference time series based on the application scenario, keeps the curve shape and distribution of the aligned data unchanged, retains the characteristics of the original data, reduces the error rate of the data, and improves the accuracy and credibility of the joint analysis of multiple data. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments:

[0030] Figure 1 It is a flowchart of the timestamp alignment method for joint analysis of high-frequency data in the industrial field according to the present invention;

[0031] Figure 2a and Figure 2b The curve shape and distribution diagram of the motor bearing seat lifting pressure before and after alignment in the application of this method;

[0032] Figure 3a and Figure 3b The curve shape and distribution diagram of the rolling mill steel signal before and after alignment in the application of this method;

[0033] Figure 4a and Figure 4b The curve shape and distribution diagram of the actual motor speed before and after alignment in the application of this method. DETAILED DESCRIPTION

[0034] Implementation example Figure 1 As shown, the timestamp alignment method for joint analysis of high-frequency data in the industrial field of the present invention includes the following steps:

[0035] Step 1: For the time series data item X within the time window length window i , which is timestamped by The corresponding value And data frequency step i composition,

[0036]

[0037] Where i = 1, 2, L, m, where m represents the time series data item X i The number of i Represents a time series data item X i The length of the sequence, window is the length of the time window to be analyzed, and the time series t i The format of each timestamp is yyyy-MM-ddHH:mm:ss.SSS, which means year-month-day hour:minute:second.millisecond. Step means the data frequency, that is, how long it takes to collect a data point, in milliseconds.

[0038] Step 2: Calculate the data item X within the time window length window i The arrival rate i , get the data points in the time interval [start, end], the arrival rate rate i It is equal to the ratio of the actual points received to the expected points received.

[0039]

[0040] end=start+window (3)

[0041] Among them, start is the start timestamp of this data analysis, and end is the end timestamp of this data analysis;

[0042] Step 3: Calculate the minimum arrival rate among m data items min ,

[0043] rate min =min(rate 1 ,rate 2 ,Lrate m ) (4)

[0044] Set the minimum arrival rate required for data joint analysis to Threshold. min If the value is ≥Threshold, the subsequent analysis will be performed. Otherwise, the minimum requirement of the analysis is not met and Threshold is set to 80%;

[0045] Step 4: Calculate the benchmark data frequency step base ,

[0046] step base = func(step i ),i=1,2L,m (5)

[0047] Among them, func is the least common multiple or greatest common divisor according to the application scenario. The least common multiple indicates downsampling, and the greatest common divisor indicates upsampling. Downsampling refers to reducing the sampling rate of the signal, and its main feature is that the amount of data is reduced; upsampling refers to increasing the sampling rate of the signal, and its main feature is that the amount of data is increased.

[0048] Step 5: Calculate the base timestamp sequence t of the data item base , each timestamp is an integer multiple of the benchmark data frequency, and the timestamp start is divided into two parts: the date and time part start other and the millisecond part start SSS ,

[0049]

[0050] Among them, round means rounding to the nearest integer, and N is the length of the reference timestamp sequence;

[0051] Step 6: Calculate the numerical sequence corresponding to the base timestamp sequence of each data item Each base timestamp With time series The most recent timestamp The corresponding value

[0052]

[0053] Step 7. The final time series is:

[0054]

[0055] This method is applied to the finishing mill of a thick plate plant. The operating condition of the finishing mill stand roller is bad. How to use and maintain the stand roller is very important for the normal and continuous production of the mill. Through the joint analysis of the three data items of the actual speed of the stand roller motor, the steel signal of the rolling mill, and the lifting pressure of the motor bearing seat, the bearing seat pressure under the condition of no steel and uniform speed is calculated, and compared with the maximum and minimum thresholds of the bearing seat pressure, an early warning of the bearing seat pressure exceeding the limit can be given.

[0056] As shown in Table 1, there are three data items, namely, the steel signal of the rolling mill, the actual speed of the motor, and the lifting pressure of the motor bearing seat, for the collection time stamps. Although the collection frequency of the three data items is 100 milliseconds, the timestamps at the same moment have slight differences in milliseconds.

[0057] Table 1:

[0058]

[0059] As shown in Table 2, the arrival rates of the data collected by the three data items in the same time window [2024-04-26 05:06:12.541, 2024-04-26 05:09:12.541] are compared with the minimum arrival rate threshold of 80%, which can be used for subsequent joint analysis;

[0060] Table 2:

[0061] Data Item Collection frequency Time Window Points due Actual points Reach ratio Steel signal from rolling mill 100 ms 180 seconds 1800 1582 87.88％ Actual motor speed 100 ms 180 seconds 1800 1618 89.88％ Motor bearing seat lifting pressure 100 ms 180 seconds 1800 1781 98.94％

[0062] Confirm that the benchmark data frequency is 100 milliseconds, then calculate the benchmark timestamp sequence, and use the nearest neighbor principle to match each data item to obtain the aligned numerical sequence.

[0063] Table 3 shows the timestamps and values ​​of the actual motor speed before and after alignment.

[0064] Table 3:

[0065]

[0066] As shown in Figures 2, 3 and 4, whether it is a control signal with data values ​​of 0 or 1, or a process signal with data values ​​fluctuating up and down, the curve shape and value distribution before and after alignment have not changed. There is only a slight difference in the frequency of different data value ranges, and the main characteristics of the data have not changed. It can be seen that the data alignment method of this method has a good effect, and the aligned data tends to the true value. This method has a small impact on the amount of data and can better retain the authenticity of the data.

Claims

1. A timestamp alignment method for joint analysis of high-frequency data in the industrial field, characterized by The steps include: Step 1: For the time series data item X within the time window length window i , which is timestamped by The corresponding value And data frequency step i composition, Where i = 1, 2, L, m, where m represents the time series data item X i The number of i Represents a time series data item X i The sequence length, time series t i The format of each timestamp is yyyy-MM-dd HH:mm:ss.SSS, which means year-month-day hour:minute:second.millisecond. Step means the data frequency, that is, how long it takes to collect a data point, in milliseconds. Step 2: Calculate the data item X within the time window length window i The arrival rate i , get the data points in the time interval [start, end], the arrival rate rate i It is equal to the ratio of the actual points received to the expected points received. end=start+window (3) Among them, start is the start timestamp of this data analysis, and end is the end timestamp of this data analysis; Step 3: Calculate the minimum arrival rate among m data items min , rate min =min(rate 1 ,rate 2 ,L rate m ) (4) Set the minimum arrival rate required for data joint analysis to Threshold. min If the value is ≥Threshold, the subsequent analysis will be performed. Otherwise, the minimum requirement of the analysis is not met and Threshold is set to 80%; Step 4: Calculate the benchmark data frequency step base , step base =func(step i ),i=1,2L,m (5) Among them, func is the least common multiple or greatest common divisor according to the application scenario. The least common multiple indicates downsampling, and the greatest common divisor indicates upsampling. Downsampling refers to reducing the sampling rate of the signal, which is characterized by reducing the amount of data; upsampling refers to increasing the sampling rate of the signal, which is characterized by increasing the amount of data. Step 5: Calculate the base timestamp sequence t of the data item base , each timestamp is an integer multiple of the benchmark data frequency, and the timestamp start is divided into two parts: the date and time part start other and the millisecond part start SSS , t base = [(t1) base , L, (t N ) base , and (t k ) base - (t k-1 ) base = step base , k = 2, L N Among them, round means rounding to the nearest integer, and N is the length of the reference timestamp sequence; Step 6: Calculate the numerical sequence corresponding to the base timestamp sequence of each data item Each base timestamp With time series The most recent timestamp The corresponding value Step 7. The final time series is:

Images