JP2006259937A - Data collection device and data recovery device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce a writing amount and thus reduce the entire period of time for writing by dynamically compressing input data. <P>SOLUTION: The device is equipped with an input processing part 11 for reading data from an input source such as an external device 20 and storing the data in an input data arrangement memory part 11a, a compression processing part 12 for reading the data in the input data arrangement memory part 11a stored by the input processing part 11 and performing compression processing, a storing processing part 13 for storing the compressed data compressed by the compression processing part 12 in a storage device 21, and a setting processing part 14 for setting operation/function of the input processing part 11 and the compression processing part 12. The input processing part 11 collects/stores a data to be classified into a bit data and a numerical value data and the data are compressed by the compression processing part 12. The compression processing part 12 classifies the input information into the bit data and the numerical value data, performs input value estimation from the characteristic of each data in a time sequence, obtains the difference between the data and an actual input value, represents a difference value with high appearing frequency by a short sign and deletes the data amount. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a data collection device and a data restoration device, and relates to a technique for compressing and storing acquired data and restoring the compressed data.

  A PLC (programmable logic controller) that controls the FA (factory automation) system installed in a production factory (manufacturing site) connects input devices such as CPU units, sensors, and switches that execute calculations based on control programs. An input unit that captures these on / off signals as input signals, an output unit that sends output signals to actuators, relays, and other output devices, and a communication that connects and exchanges information with higher-level terminal devices It is configured by combining a plurality of units such as a unit and a power supply unit that supplies power to each unit. Then, the PLC fetches the signal input from the input unit into the I / O memory of the CPU unit (IN refresh process), and performs a logical operation based on a pre-registered ladder language (program operation execution process). The operation execution result is written in the I / O memory, sent to the output unit (OUT refresh process), and then communicated with a higher-level terminal device or display connected to the network (peripheral service process). Repeat on click.

  An example of the processing of the CPU unit will be briefly described. First, after initial processing, I / O refresh processing for exchanging input / output data (I / O data) with an I / O unit or a high-function unit. Cyclic execution of program execution processing that performs logical operations on input data in accordance with the user program, and peripheral processing that mainly exchanges data with the outside through the network by the communication unit, and performs message request processing from the outside. To do. In general, the time required for one cycle in which each process is performed is called a cycle time.

  The CPU unit has an IO memory, and the input data from the IO unit and the high function unit is stored in the IO memory, and the result of executing the user program is stored as output data. The CPU unit sends output data to the IO unit and the high-function unit, and these units operate the output device based on the received output data.

  Since the IO memory stores control values and current values (ON / OFF, numerical values) of the input device and output device at each time, the sensors and actuators stored in the IO memory are stored in the IO memory. PLC history (how IO data has changed) can be recorded by periodically storing the current value, etc. as input information in a storage device such as a flash memory or hard disk at a predetermined timing. There are also things. By confirming this history, it is possible to know the history of the operation state of the PLC and the control history of the controlled device, and the operation history at the FA site can be analyzed in a subsequent process.

  As a configuration for collecting data in this way, for example, a data collection unit connected to the PLC's PLC bus as disclosed in Patent Document 1 is used, or connected to the PLC via the PLC's communication unit. There are various forms such as using a data collection device that accesses and communicates with the CPU unit from the outside and collects data stored in the IO memory.

At this time, it is generally assumed that it will be used in later analysis processes, and saved in a comma-delimited text file format so that it can be used in spreadsheet software. Thereby, for example, as shown in FIG. 1, each input data constitutes a column of the table, and a row of the table is constituted by a collection time unit.
Japanese Patent Laid-Open No. 2004-199670

  By the way, not only the number of input points to be collected / accumulated increases due to diversification of work processes and complicated control objects, but the collection cycle is shortened to perform more advanced analysis. For this reason, the amount of data to be collected / stored increases, but there is a limit to the size of storage devices that can be used at FA sites, etc., and there is a situation in which input information necessary for appropriate analysis cannot be stored. It has occurred.

  In addition, these storage devices have a lower writing speed than a hard disk used in a personal computer or the like. For this reason, in the current storage device performance, there are cases where writing cannot be performed in the collection cycle required for performing the above-described advanced analysis. In such a case, data is accumulated in the storage device device in the thinned state, and the accuracy of analysis is reduced.

  The present invention dynamically compresses data collected in units of collection or time series of input data, thereby reducing the amount of writing and shortening the entire writing time. The data can be stored beyond the writing capability, and the user of the stored compressed data can perform reversible compression that can completely reproduce the input data at the time of decompression, and provides a data collection device and a data restoration device. The purpose is to do.

  A data collection apparatus according to the present invention predicts next input data from input data acquisition means for periodically acquiring data that changes in time series, and the most recent predetermined number of input data acquired by the input data acquisition unit. A prediction means, a comparison means for comparing the prediction value predicted by the prediction means and the input data acquired by the input data acquisition means (corresponding to “difference value calculation means” in the embodiment), and the comparison means Compression processing means for performing compression processing on the obtained comparison value and means for outputting the compressed data compressed by the compression processing means to the storage means are provided. In the embodiment, the comparison value is simply obtained as a difference value. However, the comparison value can be obtained through various calculations. The difference value is preferable because the calculation process is simple. Here, “most recent” includes, as shown in the embodiment, a predetermined number (including one) immediately before including the previous data (previous data in time series). For example, the previous data is not included in the previous time series, and the previous data is included immediately. However, it does not include past data that is so low in relevance (connection) as data to be acquired next time that the predicted value cannot be obtained. In other words, for example, data in a range in which data to be acquired next can be predicted based on previous data or previous data is included in the latest data.

  In general, data compression is performed by finding the redundancy of data and expressing the redundant portion in a straightforward manner to reduce the total data amount. However, for example, bit data and numerical data input in time series at the FA site vary in data bias depending on the input source, and regular redundancy cannot be defined. Therefore, it is difficult to apply a compression technique that is generally performed as it is.

  On the other hand, for these data, the next input value can be predicted to some extent from the already input data of several points close to the present time. As will be described in detail in the embodiment, when the predicted value matches or approximates the input value, the difference value between the input value and the predicted value is constant around the value 0 as shown in FIG. Concentrate on the range. Therefore, in the present invention, the bias in the appearance range of the data is regarded as redundancy, and compression is performed. Of course, the reason why the present invention has been reached is that the predicted value in the FA field and the actual input value are likely to match / approximate, and the bias of the appearance range of the data is compressed as redundancy. However, the application field of the present invention is not limited to the FA site in this way, and there is a bias in the same predicted value, actual input value, and comparison result, and this is regarded as redundancy and compression processing is applied. Anything that can be done.

  More specifically, for example, the target of the compression processing performed by the compression processing unit is a comparison value for a plurality of input data acquired in the same cycle, or the target of the compression processing performed by the compression processing unit is continuous. Or a comparison value of single data of a plurality of periods.

  Further, there may be a plurality of types of input data, and the compression processing means may perform different compression processing depending on the type of the input data. In the embodiments, suitable prediction algorithms are different such as bit data and numerical data in the embodiment. As a result, the accuracy of the predicted value increases for each type of input data, and the compression efficiency also increases. As an example, input data is classified into bit data and numerical data, and input value prediction is performed based on the characteristics of each data in time series.

  Further, the comparison value is a difference value between the input value and the predicted value, and the compression processing can be expressed by a smaller number of bits as the frequency of the difference value is higher. When the difference value is stored, a difference value having a high appearance frequency is represented by a short code to reduce the entire data amount.

  The data collection device of the present invention can be applied to various fields. For example, the input data can be input data in an FA system. Specifically, the input data is ON / OFF data, and the predicting means calculates the input data input next from the most recently obtained ON / OFF state (in the case of a plurality of patterns). The compression means represents 0 or 1 when the predicted value is hit, and expresses 1 or 0 (the opposite value to the case when the predicted value is hit) when the predicted value is lost. The function of compressing can be provided by expressing using consecutive numbers. The input data is analog data, and has storage means for storing at least the latest N data, and the prediction means presets the latest N data stored in the storage means. The prediction value to be input next is obtained by substituting into the predicted calculation formula, and the compression processing means expresses the case where the input value and the prediction value match with “0”, and the input value and the prediction The smaller the difference from the value, the smaller the number of bits.

  In the case of FA system, the same value is often taken continuously in the case of the ON / OFF value of IO information, and the value of analog data suddenly increases even if the same value is taken or fluctuates. It rarely changes and often fluctuates slowly. Therefore, there is a characteristic that it is easy to predict the next input data based on data acquired before that time. Therefore, the degree (probability) that the predicted value matches the actual input value is high, and even if the predicted value deviates, the difference from the predicted value when the predicted value deviates is small. Therefore, in the case of ON / OFF data, the number of consecutive 0s or 1s increases by setting 0 or 1 when the predictions match. As a result, compression can be achieved by expressing the numbers as consecutive numbers. Also, in the case of analog data, the case where it matches the predicted value is indicated by 0 (1 bit), and it is necessary for expressing the whole by expressing with a shorter number of bits as the difference between the predicted value and the actual input value is smaller. The number of bits can be reduced.

  The data decompression device of the present invention is a data decompression device that restores the data compressed by the above-described data collection device, the compressed data, at least an input value that does not have a comparison value, and the prediction means predicted Means for obtaining a prediction value or a prediction rule (Huffman tree, prediction table, etc.) of the prediction means, compressed data acquired by the acquisition means, an input value for which no comparison value exists, and prediction value / prediction Based on the rules, the data expansion means for obtaining the input value is provided.

  In the present invention, for example, single or plural pieces of input information at the FA site can be saved in storage means such as a storage device by reducing the data amount without losing the original data. Further, the stored data can be restored to the original input information by a data restoration device (which can be realized by a personal computer or the like).

  In the data collection device of the present invention, the amount of data that can be expected can be reduced by dynamically compressing the input data collected in the collection unit or time series, thereby reducing the amount of writing and the overall writing time. Can be stored beyond the writing capability of the storage device. Further, the user of the stored compressed data can perform lossless compression that can completely reproduce the input data at the time of decompression by using the data decompression device.

  FIG. 2 shows a preferred embodiment of the present invention. As shown in FIG. 2, the data collection device 10 of the present embodiment is realized as one of the units (data collection unit) constituting the PLC 1. That is, the data collection device 10 (data collection unit) is connected to a unit for realizing various functions such as a power supply unit 2, a CPU unit 3, an IO unit 4, a master unit 5 that performs master-slave communication, and the PLC 1 Configure. Therefore, the data collection device 10 can collect data by accessing the IO memory in the CPU unit 3 and other unit memories via the PLC bus.

  Of course, the CPU unit 3 is connected to the IO unit 4 or the like via the PLC bus, and an IO device is connected to the IO unit. The IO device is an input device such as a sensor or a switch that performs an on / off operation, or an output device such as an actuator or a power relay. The IO memory of the CPU unit 3 stores data relating to the on / off state of the input device of the IO device, data relating to the operating state of the output device, and the like. Therefore, the data collection device 10 accesses the IO memory of the CPU unit 3 to acquire information on the IO devices connected to the IO unit 3 and information acquired by the master unit 5 through master-slave communication via the network. can do.

FIG. 3 shows the data collection device 10 of the present embodiment. That is, the data collection device 10 reads (collects) data from an input source such as the external device 20 and stores the data in the input data array storage unit 11a, and the input data stored in the input processing unit 11. The compression processing unit 12 that reads the array storage unit 11a and performs compression processing, the storage processing unit 13 that stores the compressed data compressed by the compression processing unit 12 in the storage device 21 that is a storage device, the input processing unit 11 and And a setting processing unit 14 for setting the operation / function of the compression processing unit 12.
The external devices 20 include various devices such as a CPU unit 3 (IO memory), other units, and devices connected to a network other than the PLC 1.

  The setting processing unit 14 specifies the data to be collected, and the input data such as the input cycle (time), input source (memory address, IO, network, etc.) for each piece of data to be collected, that is, input information. Define the existing location, etc.) and the type of input data (bit / numerical value (8 bits, 16 bits, 32 bits, integer, floating point (single precision, double precision)), and store them in the setting memory 14a An example of the data structure of the setting memory 14a is, for example, as shown in Fig. 4. As described above, there are usually a plurality of types of input data to be collected, such as BIT information and numerical data. However, the input periods are different, and of course, all the input periods may be equal.

  The input processing unit 11 reads the setting of each input to be collected from the setting memory 14a, and reads data from an input source such as various devices 20 according to the input cycle of each input. The read data is divided into BIT information and numerical information and stored in the input data array storage unit 11a due to the difference in the subsequent compression processing technique. In other words, the input data array storage unit 11a has an input data array for BIT information and input data for numerical information. Specifically, the processing function of the input processing unit 11 executes the flowchart shown in FIG.

  First, X is initialized to 1 (S1), and it is determined whether or not the input cycle is the input X (S2). Here, the input X is an input No. This input No. Is a kind of record No. It is the same. Processing may be executed based on the input name (input A, input B, etc.). In the example shown in FIG. 4, X takes a value from 1 to 10. Whether or not it is the cycle of the input X is determined by whether or not it corresponds to the input cycle corresponding to the input X stored in the setting memory 14a. When the condition of the input data to be collected is as in the setting memory 14a shown in FIG. 4, the input cycle of BIT data is 1 msec and the input cycle of numerical data is 2 msec. It can be easily judged by whether or not each period is reached.

  If the input cycle has not been reached, the process jumps to processing step S6, X is incremented by 1 (S6), and the value of X after the increment is larger than the maximum value (10 in the example of FIG. 4). It is determined whether or not (S7). If the maximum value has not been reached, the process returns to step S2 to determine whether or not the input cycle of the input X after the increment has been reached. If the value of X exceeds the maximum value (in the case of the setting memory 14a shown in FIG. 4, X after the increment is 11), the branch determination in processing step S7 is Yes, Returning to process step 1, X = 1 is initialized, and the process returns to the process for the data whose input No is 1.

  On the other hand, when the input cycle of the input X is reached, the branch determination in the processing step S2 is Yes, so that the process jumps to the processing step S3 and determines whether the input type of the current input X is BIT or a numerical value (S3). ). If the input type is BIT, data is read from the input source of the input X, and the read data is stored in the BIT input data array (S4). If the input type is a numerical value, data is read from the input source of the input X, and the read data is stored in the numerical input data array (S5). If any of the processes of S4 and S5 is executed and data is stored in a predetermined input data array, the process proceeds to process step S6, and X is incremented. By repeatedly executing the above-described processing, all input data stored in the setting memory 14a are sequentially executed from the beginning to the end, and when the acquisition timing comes, the data is read and the desired input data array storage It can be stored in the part 11a.

  The compression processing unit 12 reads the input cycle and input type for each input data from the setting memory 14a. Then, input data is read from the input data array for each input cycle and input type. In the case of the present embodiment, the input processing unit actually reads out data in accordance with the input cycle of the input data and stores it in the input data array storage unit 11a, and the input cycle is in units of input types. Since they are equal (BIT: 1 msec, numerical value: 2 msec), the data stored in each input data array storage unit 11a may be read in batches in units of input types.

  The compression processing in the compression processing unit 12 basically obtains an input value, predicts the next input value for each of the input data based on the input value so far, Next, the difference from the acquired input value (actual value) is taken. When the predicted value approximates the input value, the difference value has a distribution as shown in FIG. At this time, by encoding a value with a large number of appearances with a small number of bits, the total number of bits used can be reduced as compared with a case where all values are expressed with 16 bits as shown in FIG.

  For example, if the difference value is 0, 50% of the total value is represented by 1 bit. Furthermore, if the value that can be represented by 4 bits is 40% of the whole, the value that is represented by 12 bits is 9% of the whole, and the value that is represented by 18 bits is 1% of the whole, the required bits per value The number is "0.5 + 1.6 + 1.08 + 0.18 = 3.36 bits". Compared to the case where all values are expressed in 16 bits, the whole can be expressed with a bit number of 1/4 or less and can be compressed. .

  The compression effect is more preferable as the difference between the input value and the predicted value concentrates on a certain value. Therefore, the higher the prediction accuracy, the higher the peak value of the distribution shown in FIG. 6, the higher the frequency at which the difference value becomes 0 (or a value approximating 0), and the higher the compression effect. Therefore, in the present embodiment, there are a plurality of types of input such as BIT and numerical values, and the prediction value calculation algorithm suitable for each of them is different. Therefore, the compression processing unit 12 uses the numerical value compression processing unit 15. And a BIT compression processing unit 16.

  FIG. 7 shows the numerical value compression processing unit 15. The numerical value compression processing unit 15 reads numerical data from the numerical value input data array storage unit 11a in the input processing unit 11 and temporarily stores the numerical value data, and the input value buffer 15a stores the numerical value data. Each input value is sequentially read out, and based on the past input value stored in the prediction table 15b and the prediction table 15b that stores a predetermined number of past input values (in the present embodiment, the past four times) A predicted value calculation unit 15c that obtains a predicted value of the input value, a difference value calculation unit 15d that compares the input value stored in the input value buffer 15a with the predicted value obtained by the predicted value calculation unit 15c, and obtains the difference between them. And the difference value obtained by the difference value calculation unit 15d and the difference value are acquired, and the acquired difference value is encoded based on the Huffman tree stored in the Huffman tree storage unit 15e. A coding unit 15f, and an output buffer 15g for temporarily storing the encoded data (compressed data) by the coding unit 15f.

Therefore, in the input value buffer 15a, if the contents of the setting memory 14a are in the state shown in FIG. 4, the input names in the input data array storage unit 11a are input C, input E, input F, input G, input H, Data corresponding to input J is stored. These six input data are transferred from the input value buffer 15a to the prediction table 15b and stored in time series for each input data.
The predicted value calculation unit 15c reads four data before the input from the prediction table 15b, and calculates the predicted value Y (n) using the following calculation formula.

Y (n) = − Y (n−4) + 4 × Y (n−3) −6 × Y (n−2) + 4 × Y (n−1)
That is, analog data values used in FA are expected to draw patterns such as those shown in FIGS. 8 and 9, for example, where the horizontal axis represents time and the vertical axis represents values. That is, it changes in a pattern similar to a sine waveform or cosine waveform as shown in FIG. 8A, or a pattern similar to FIG. 8A as shown in FIG. As shown in the enlarged view in the figure, when the data does not change smoothly, or as shown in FIG. 9A, a pattern that changes to a certain value over time and maintains that value, or FIG. As shown in b), the pattern changes in the same pattern as in FIG. 9A, but there are cases where pulses due to noise sometimes occur.

When data that transitions in such a pattern is read as an input value at a period of 1 msec or the like, the next input value can be predicted from the data for several past periods by N-order polynomial approximation. For example, cubic curve formula Y = aX ³ + bX ² + cX + d
Simplified formula of the large range interpolation formula of Y (n) = − Y (n−4) + 4 × Y (n−3) −6 × Y (n−2) + 4 × Y (n−1)
Can be used to predict the next input value X (n) from the values (Y (n−4) to Y (n−1)) four times immediately before the input value. Therefore, as described above, the predicted value calculation unit 15c calculates the predicted value Y (n) of the next input value X (n).

  As described above, in this embodiment, since the predicted value of the next input value is calculated from the previous four input values, the prediction table 15b stores at least the previous four input values. Table. The previous data may be stored and held, or may be overwritten and deleted sequentially like a ring buffer. However, at least the first four input data are stored and held by using the prediction table 15b, the output buffer 15g, or another storage means.

Taking the input C as an example, a specific example of input value prediction processing in the prediction value calculation unit 15c and difference value calculation processing in the difference value calculation unit 15d will be described. Assume that the input values from the first time are as shown in FIG. Then, in order to calculate the predicted value, data for the past four times are required, and therefore the predicted value cannot be obtained from the first time to the fourth time. When the first to fourth input values are substituted into the above-described prediction calculation formula, the fifth predicted value Y (5) is
− (First input value) + 4 × (second input value) −6 × (third input value) + 4 × (fourth input value)
= −11 + 4 × 12−6 × 13 + 4 × 14
= 15
It becomes. Thereafter, when the same processing is executed in order, and the next input value is predicted from the previous four input values, it is as shown in the predicted value column of FIG. In this embodiment, the prediction is performed using the large-range interpolation formula, but it is needless to say that the prediction value may be obtained using another prediction method.

  Then, the difference value calculation unit 15b calculates “X (n) −Y (n)”. Therefore, the difference value for the input C can be calculated for the fifth and subsequent times when the predicted value is obtained, and specifically, the difference value is calculated for each time as shown in FIG. That is, the difference values after the fifth time of the input C are 0, 1, -2, 0, 0, and in this example, the difference value is 0, that is, the number of times the prediction has been made is the largest. Similarly, for the other inputs E, F, G, H, and J, the difference value of each input time is obtained for the fifth time and thereafter.

  The encoding unit 15f performs Huffman encoding on the difference value for each input data obtained by the difference calculation unit 15b. Here, the difference value is set to a fixed range such as −256 to +255, −128 to +127. The code corresponding to the difference value is searched from the Huffman tree stored and held in the Huffman tree storage unit 15e. If a difference value corresponding to the Huffman tree exists (hits), a code corresponding to the hit difference value is output. If there is no hit, there is no corresponding code, so the difference value is output as it is without encoding.

  FIG. 12 shows an example of a Huffman tree (value: a table indicating the correspondence between difference values and codes) stored in the Huffman tree storage unit 15e. In this example, the range of the difference value is −128 to +127.

  For example, if the difference values for the sixth input data (inputs C, E, F, G, H, and J) for the fifth time are as shown in FIG. 13, the difference values of the input data are shown in FIG. Encoding is performed as shown in FIG. 14 by performing Huffman encoding using the Huffman tree shown. Therefore, the 6th input data (inputs C, E, F, G, H, J) for the fifth time is “010100101100101” (binary number), and the data that originally required 48 bits (8 × 6) is required. Can be expressed in 15 bits.

  The encoded data is sequentially stored in the output buffer 15g each time, and when the compression process (encoding) is completed for all the numerical values, the compressed data stored in the output buffer 15g is stored. The data is passed to the storage processing unit 13.

  It should be noted that no compression is performed on the noise (abnormal value) that occurs suddenly (instantaneously) as shown in FIG. However, such a value has a small ratio to the whole data, and its influence can be ignored.

  In this embodiment, the case where the Huffman tree is statically constant data has been described. However, as a technique for further increasing the compression ratio, a technique for dynamically optimizing the Huffman tree based on the result of each input is used. May be. Specifically, it is realized to recalculate the appearance frequency of the difference value for each input and dynamically add a function for reconfiguring the Huffman tree so that a high appearance frequency can be expressed with a small number of bits.

  FIG. 15 shows the BIT compression processing unit 16. The BIT compression processing unit 16 reads an input type of BIT data from the input data array storage unit 11a in the input processing unit 11, stores it in the input value buffer 16a for temporary storage, and stores it in the input value buffer 16a. Each input value is sequentially read out, based on a past predetermined value stored in the prediction table 16b, and a prediction table 16b that stores a predetermined number of past input values (in the present embodiment, past two times), A prediction value calculation unit 16c for obtaining a prediction value of the next input value, a difference value calculation for comparing the input value stored in the input value buffer 16a with the prediction value obtained by the prediction value calculation unit 16c, and obtaining the difference Section 16d, RLE compression spare buffer 16f for storing spare data for processing based on the difference value obtained by difference value calculation section 16d, and its RLE pressure It includes a RLE compression section 16h for compressed by RLE (Run Length Encoding) based on the stored data to use the spare buffer 16F16f, and an output buffer 16g for storing the data compressed by the RLE compression unit 16H. Furthermore, in order to increase the compression rate of final RLE compression, an input extraction unit 16e having a large coefficient difference is also provided.

  Accordingly, in the input value buffer 16a, when the contents of the setting memory 14a are in the state shown in FIG. 4, the input names in the input data array storage unit 11a for BIT are data corresponding to the inputs A, B and D. Is stored. Then, these three input data are transferred from the input value buffer 16a to the prediction table 16b and stored in time series for each input data.

  In the prediction table storage unit 16b, as shown in FIG. 16, there is a prediction table having a table structure in which the frequency of the next occurrence of 0 and the frequency of occurrence of 1 for the combination of the previous m values are prediction coefficients. Stored. Such a prediction table is created and stored for each input data. The value of m is 1 or more, but is determined by the type of input value (m = 2 in the figure).

  In this embodiment, in order to increase the compression efficiency of the final RLE compression, processing is performed in order from the input with the highest predicted matching probability. That is, among the bit inputs, processing is performed in order from the input with the largest difference in the prediction coefficients in the table that matches the previous two input values from the prediction table 16b. This is to use the fact that “there is a high probability that the prediction matches when the difference in prediction coefficients is large”. The calculation for establishing the prediction match can be similarly calculated by dividing the number of occurrences of the predicted value (0 or 1) by the sum of the prediction coefficients. However, in the present embodiment, the difference is calculated in consideration of the calculation efficiency. As a way to compare. Such processing is performed by the input extraction unit 16e having a large coefficient difference.

  The BIT compression processing unit 16 also predicts the next input value from the value m times (for example, twice) immediately before the input value, and compresses the compressed data based on the match / mismatch between the predicted value and the actual input value. Create That is, the predicted value calculation unit 16c refers to the prediction table as illustrated in FIG. 16 stored in the prediction table storage unit 16b, and based on the combination of the previous m values, the value of the next appearing coefficient (prediction coefficient) is high. Is a predicted value. As the prediction coefficient has a larger number of samples of the prediction table coefficient and the difference is larger, the prediction probability is higher. In the case of input data having the prediction table shown in FIG. 16, when the last two bit data are “0, 0”, the next input has a prediction coefficient of 25 when 0 and a prediction coefficient of 1 Since the value is 1, the value “0” having the larger coefficient is set as the predicted value. In addition, based on each input value stored in the input value buffer 16a, a process of adding one prediction coefficient that matches the input value of the prediction table is also performed. As a result, the prediction coefficient is sequentially updated, and a prediction table capable of performing prediction with higher accuracy is created.

The difference value calculation unit 16 compares the predicted value with the input value, and sets the difference value to 0 if the difference value matches the input value, and sets the difference value to 1 if the prediction does not match. Therefore, when the prediction coincidence is high, the difference value is mostly 0. In addition, since the input extraction unit 16e having a large coefficient difference is used to extract an input having a large coefficient difference and then processed in order, the probability that the predicted values match is higher toward the beginning, and the output of the difference value calculation unit 16d is the same. Probability of consecutive numbers increases. Therefore, the compression rate by RLE compression also becomes high. Since the input data is 1/0 bit data, the actual calculation process in the difference value calculation unit 16d is, for example,
Difference value = predicted value XOR input value can be obtained.

  The difference value for each input data obtained by the difference value calculation unit 16d is stored in the RLE compression spare buffer 16f every same input time. For example, it is assumed that the difference value for each input with respect to a plurality of inputs is as shown in FIG. 17 (0 means prediction match, 1 means mismatch). In the present embodiment, since the predicted value is obtained based on the past two input values, the first and second difference value data are left as input values. The difference value data stored in the RLE compression spare buffer 16f is collected and held in a line for each time. Then, the RLE compression processing unit 16h compresses the stored data and stores the obtained compressed data in the output buffer 16g.

As shown in FIG. 17, if there are 48 input types of BIT data of the input type, normally, 48 bits (6 bytes) are necessary to express this. If this is expressed by RLE by the continuous number of 0, 1, each data shown in FIG.
1st time 35 0, 1 1, 12 0
Second time 48 0
3rd time 10 0, 1 1,37 0
4th time 48 0
Therefore, when expressed in RLE, 35,1,12
48
10,1,37
48
In this example, it can be expressed by 1 byte (8 bits) at the minimum and 3 bytes (24 bits) at the maximum. In this case, at all times, the amount of data can be compressed by being less than 48 bits required by the original data.

  Further, in the case of normal RLE, “value” and “number of consecutive values” are recorded. However, in the case of this embodiment, only two types of values 0 and 1 appear in a toggle. Only the “continuous number” is required, and the recording of “value” is not necessary.

  The function (action) of the BIT compression processing unit 16 will be described with a specific example. For example, when there are three bit inputs A, B, and D having a prediction table as shown in FIGS. 18A to 18C, the input values of the previous two times at a certain time are “input A is Assume that “0, 1”, “input B is 1, 0”, and “input D is 0, 0”.

In this case, the prediction table to be used uses 01 for input A, 10 for input B, and 00 for input C, and the predicted values obtained by the predicted value calculation unit 16c are predicted as follows.
Input A is predicted to be 1 because the prediction coefficient is 0: 2 <1: 111
Input B has a prediction coefficient of 0: 173 times> 1:35 times, so next is predicted to be 0,
Since the prediction coefficient of the input C is 0:25 times> 1:10 times, the next is predicted to be 0.

  If the actual input value has a history as shown in the input value column shown in FIG. 19, the predicted value and the difference value take the values shown in the respective columns in FIG. Then, the difference value for a certain time with respect to each input data obtained by the difference value calculation unit 16d is left-justified and sequentially stored in the RLE compression spare buffer 16f. Thus, for example, data as shown in FIG. 20 is stored in the RLE compression spare buffer 16f. Here, the input values are used as they are for the first and second times, and the difference values are used for the third and subsequent times.

  After processing all the bits, the RLE compression unit 16h executes RLE compression and creates compressed data to be stored in 16g in the output buffer. RLE compression records the number of times the same numerical value appears consecutively, and the compression rate increases when the same numerical value appears unevenly.

In the present embodiment, compression is executed by the number of consecutive appearances of 0/1 difference values in the horizontal direction in the RLE compression spare buffer 16f shown in FIG. In FIG. 20, since the difference value after input D is unknown, for the sake of convenience, when implemented with three inputs of A / B / D,
3rd: 3, ...
4th: 1, 2, ...
5th: 3, ...
It becomes.

  The storage processing unit 13 stores, in the storage device 21, output buffers that are compression results of the numerical value compression processing and the bit compression processing obtained by the compression processing unit 12. At this time, considering the efficiency at the time of expansion, it is stored for each numerical value, bit, and collection cycle with different compression prescriptions. Instead of saving in the storage device 21, it is also possible to output it as stream data via a network. Further, the contents of the setting memory are saved as a file. This is used when performing deployment.

  FIG. 21 shows a deployment system which is a preferred embodiment of the data restoration apparatus according to the present invention. This deployment system can be realized, for example, by stamping a predetermined application program on a personal computer. The decompression executed here is decompressed from the data file obtained by dynamic compression to the data before compression. At this time, the data is easily rearranged, added with a time stamp, and converted into a text file to facilitate data analysis.

  The expansion system 40 includes a numerical data expansion unit 41, a bit data expansion unit 42, and a shaping unit 43. Further, a storage device 44 that stores the file shaped by the shaping unit 43 is also provided. The storage device 44 may be installed outside. That is, as described in the compression process, the numerical data and the BIT data are compressed by different methods, so that the decompression (restoration) processing is also performed using the numerical data decompression unit 41 and the BIT data decompression unit 42. This is done for each data. Then, the file divided according to the difference in the compression method of numerical values and bits and the difference in the collection cycle is made into one shaped file by the shaping unit 43. Specifically, the following processing is performed.

  First, an N-th input value is obtained for each numerical data and bit data. That is, in the numerical data development unit 41, the first, second, third, and fourth times are used without calculation because the input values are stored as they are. From the fifth time onward, the Nth input value of each input is calculated from the stored compressed data, Huffman tree, and prediction table. That is, the Nth compressed data is read, the difference value of each input is expanded from the Huffman table, and the input value can be calculated from the predicted value and the difference value.

  For example, if the Nth compressed data is “010100101100101”, the difference value is “0, 0, 1, 0, −1, 0” from the Huffman tree shown in FIG. Also, the Nth predicted value of each input is calculated from the prediction table. The predicted value is calculated based on the data of the past four input values. When the predicted value of each input is input C = 15, input E = 100, input F = 200, input G = 1000, input H = 1000, input J = 99, the input value is calculated from the predicted value and the difference value. The result (input value = predicted value−difference value) is as shown in FIG. While incrementing one by one from N = 5, the input value for each time can be calculated from the predicted value and the difference value. Note that the predicted values can be calculated by using the values obtained by the arithmetic processing (development processing) as described above as the input values for the past four times.

  The bit data expansion unit 42 performs expansion processing according to the following procedure. First, the first and second times are used without calculation because the input values are stored as they are. From the third time onward, the Nth input value of each input is calculated from the stored compressed data and the prediction table.

  That is, for example, as a result of RLE decoding from the Nth compressed data, a difference value of each input is obtained as shown in FIG. Then, since the Nth difference values of the inputs A, B, and D are 0, 0, and 0, respectively, “input value = difference value XOR predicted value” is calculated from the predicted values of the Nth input A, B, and D. The input value is obtained as shown in FIG.

  As described above, numerical data and bit data are individually developed by the development units 41 and 42, respectively. Therefore, the shaping unit 43 rearranges the data for each input value in the order of the setting table based on the expansion result (input value) acquired from each of the expansion units 41 and 42 and the information of the setting table. Thereby, as shown in FIG. 25, the Nth input data can be reversibly compressed and reproduced.

  Further, in the present embodiment, the shaping unit 43 has a function of creating a data file obtained by converting each data into text data separated by commas. As a result, one line of text data separated by commas is created each time, and one line is added each time the text data is expanded. In this way, numerical data and bit data are rearranged according to the setting table, converted into text, and one line of a comma-separated file is sequentially added to create a formatted data file as shown in FIG. 26, for example. This data file is stored in the storage device 44. Since this data file has a comma-delimited text data format that can be processed by a spreadsheet application software, subsequent analysis can be easily performed.

  In the embodiment described above, the input data input at the same time is compressed as a unit, but the present invention is not limited to this, for example, the same input (for example, input C) is acquired in time series, You may make it compress.

  The range to which the present invention is applied is not limited to the FA system described above, and compression processing can be appropriately performed based on values (difference values, etc.) obtained by comparing predicted values with input values. If so, it can be applied to various types.

It is a figure which shows an example of the data collection result when not compressing. It is a figure which shows an example of PLC. It is a figure which shows the internal structure of one Embodiment of the data collection device which concerns on this invention. It is a figure which shows an example of the data structure of a setting memory. It is a flowchart which shows the function of an input process part. It is a figure which shows an example of the distribution map of the difference value of an input value and a predicted value. It is a figure which shows an example of the internal structure of the compression process part for numerical values. It is a figure which shows an example of the transition example on the time axis of numerical data. It is a figure which shows an example of the transition example on the time axis of numerical data. It is a figure which shows the example of a prediction of numerical data. It is a figure which shows the example of a difference calculation of numerical data. It is a figure which shows an example of a Huffman tree. It is a figure which shows the example of a difference value of each input data (numerical data) of the 5th time. It is a figure which shows the example which encoded the difference value shown in FIG. 13 using the Huffman tree. It is a figure which shows an example of the internal structure of the compression process part 16 for BIT. It is a figure which shows the example of a prediction table of bit data. It is a figure which shows an example of the difference value for every input with respect to several input. It is a figure which shows the example of bit data prediction. It is a figure which shows the example of a bit data difference value. It is a figure which shows an example of the data structure of the reserve buffer for bit data RLE compression. It is a block diagram which shows one Embodiment of an expansion | deployment system. It is a figure which shows the input value calculation example of numerical data. It is a figure which shows an example of the RLE decoding result of bit data. It is a figure which shows the example of input value calculation of bit data. It is a figure which shows the example of input data rearrangement. It is a figure which shows an example of an expansion result example (text file).

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Data collection device 11 Input processing part 11a Input data arrangement | sequence storage part 12 Compression processing part 13 Storage processing part 14 Setting processing part 14a Setting memory 20 External devices

Claims (9)

Input data acquisition means for periodically acquiring data that changes in time series;
Predicting means for predicting next input data from the most recent predetermined number of input data acquired by the input data acquiring unit;
A comparison unit that compares the predicted value predicted by the prediction unit with the input data acquired by the input data acquisition unit;
Compression processing means for performing compression processing on the comparison value obtained by the comparison means;
A data collection apparatus comprising: means for outputting the compressed data compressed by the compression processing means to the storage means;   2. The data collection apparatus according to claim 1, wherein a target of compression processing performed by the compression processing means is a comparison value for a plurality of input data acquired at the same period.   2. The data collection apparatus according to claim 1, wherein the object of the compression processing performed by the compression processing means is a comparison value of single data having a plurality of consecutive periods. There are multiple types of the input data,
The data collection device according to claim 1 or 2, wherein the compression processing means performs different compression processing depending on the type of input data. The comparison value is a difference value between the input value and the predicted value,
5. The data collection device according to claim 1, wherein the compression processing is expressed by a smaller number of bits as the frequency of the difference value is higher. 6.   The data collection device according to claim 1, wherein the input data is input data in an FA system. The input data is ON / OFF data,
The prediction means predicts the next input data from the most recently obtained ON / OFF state,
The compression means has a function of compressing by expressing 0 or 1 when the predicted value is hit, expressing 1 or 0 when the predicted value is off, and expressing using consecutive numbers of the same number The data collection device according to claim 6, wherein The input data is analog data, and has storage means for storing at least the latest N data,
The prediction means substitutes the data for the latest N times stored in the storage means into a preset prediction arithmetic expression to obtain a predicted value to be input next,
The compression processing means expresses a case where the input value and the predicted value match with “0”, and expresses with a smaller number of bits as the difference between the input value and the predicted value is smaller. The data collection device according to claim 6. A data decompression device for decompressing data compressed by the data collection device according to any one of claims 1 to 8,
The compressed data;
At least an input value for which no comparison value exists,
Means for obtaining a prediction value predicted by the prediction means or a prediction rule of the prediction means;
A data restoration apparatus comprising: compressed data acquired by the acquiring means; an input value for which no comparison value exists; and data expansion means for obtaining an input value based on a prediction value / prediction rule.

Images