DATA INTERPOLATION PROCESSING METHOD, DATA INTERPOLATION PROCESSING DEVICE, PROGRAM, LEARNING PROCESSING DEVICE, AND PREDICTION PROCESSING DEVICE

The data interpolation method using Hankel matrix decomposition and optimization processing addresses the issue of intermittent data in wireless communication, enabling accurate and efficient prediction of communication quality.

JP7593615B2Active Publication Date: 2024-12-03ATR ADVANCED TELECOMM RES INST INT
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
JP2020190715
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-11-17
Publication Date
2024-12-03
Estimated Expiration
2040-11-17

AI Technical Summary

Technical Problem

In wireless communication environments, especially in narrow spaces like factories, intermittent radio wave reception leads to poor learning efficiency when using neural networks for predicting communication quality due to intermittent observation data, resulting in suboptimal models and potential communication failures.

Method used

A data interpolation method involving Hankel matrix generation, matrix decomposition, and optimization processing is employed to interpolate missing data, allowing for high-precision learning and prediction of communication quality conditions.

Benefits of technology

The method enables accurate and efficient learning and prediction of future communication quality by ensuring consistent waveform patterns, even with intermittent data, thereby improving the accuracy and efficiency of communication quality prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007593615000021
    Figure 0007593615000021
  • Figure 0007593615000022
    Figure 0007593615000022
  • Figure 0007593615000023
    Figure 0007593615000023
Patent Text Reader

Abstract

To provide a wireless state prediction device capable of acquiring data obtained by appropriately interpolating observation data even under a situation where the observation data is acquired as intermittent data and by using the data, highly accurately and appropriately performing learning processing of a model predicting a future communication quality state.SOLUTION: In a case where observation data is intermittently acquired, a wireless state prediction device 100 is configured to, no matter where a missing part of the observation data is, when data interpolation processing is performed for the same waveform pattern, perform the data interpolation processing so as to acquire substantially the same waveform pattern, In other words, the wireless state prediction device 100 generates Hankel matrix R from the observation data, using the Hankel matrix R to acquire an optimization matrix R_opt acquired by performing optimization processing and performs the data interpolation processing based on the optimization matrix R_opt.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a technique for estimating a communication situation or a communication state of wireless communication or the like. [Background technology]

[0002] In a narrow space such as a factory, the wireless environment changes in a complex and dynamic manner due to changes in the propagation environment caused by the movement of machine tools, products, workers, etc., and interference between multiple wireless applications. In a narrow space such as a factory, it is difficult to grasp the position information of moving objects such as machine tools, and since the radio wave propagation is a multipath-rich environment, unexpected effects may occur in unexpected places. In this way, if unexpected effects occur in unexpected places in a narrow space, in the worst case, stable communication cannot be maintained, and at the manufacturing site, measures such as stopping the production line may be taken, resulting in a decrease in productivity. In other words, in a narrow space such as a factory, an unexpected decrease in communication quality (QoS: Quality of Service) may occur, resulting in a decrease in productivity.

[0003] Therefore, it is necessary to take preventive action before extreme degradation of QoS (communication failure) occurs, and in order to realize this, a technology that can predict the occurrence of communication failure with high accuracy is required. In order to realize such a technology, for example, a technology that obtains a distribution of values ​​indicating communication quality in a time series and predicts the future communication quality state based on the distribution (for example, the technology disclosed in Patent Document 1) can be used. In addition, a technology that realizes such a technology for predicting the future communication quality state by using a neural network model or the like has also been developed. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2005-18304 A Summary of the Invention [Problem to be solved by the invention]

[0005] For example, when predicting a future communication quality situation using a technology using a neural network model or the like, it is necessary to perform a learning process and obtain a learned model. For this purpose, in a wireless communication environment, a sensor receives radio waves to obtain observation data, and the observation data is used to perform a learning process of a model that predicts a future communication quality situation. At this time, there is a problem that the sensor cannot always receive radio waves, and radio wave reception (frame reception) becomes intermittent, and as a result, the obtained observation data is also intermittent. If a learning process of a model that predicts a communication quality situation is performed using such intermittent observation data (observation data with missing parts), the learning efficiency is poor, and an optimal model cannot be obtained.

[0006] In view of the above problems, the present invention aims to provide a wireless situation prediction device that can obtain data by appropriately interpolating observation data even in a situation where the observation data is obtained intermittently, and can use the obtained data to accurately and appropriately perform learning processing for a model that predicts future communication quality conditions, as well as a data interpolation processing method and program used therein. [Means for solving the problem]

[0007] In order to solve the above problems, a first invention is a data interpolation processing method for performing interpolation processing on intermittently acquired observation data, comprising an observation data acquisition step, a data holding step, a matrix generation step, a matrix decomposition processing step, an optimization processing step, and an interpolated data processing step.

[0008] The observation data acquiring step acquires observation data relating to communication quality from the wireless signal.

[0009] The data holding step stores and holds the observation data acquired in the observation data acquiring step for a predetermined period of time.

[0010] The matrix generating step generates a Hankel matrix R such that when a window corresponding to a predetermined unit period is shifted in the time direction by a first hour, which is a predetermined period, a group of data included in the period corresponding to the window becomes elements of one row. The matrix generating step then generates the Hankel matrix R by (1) setting the data corresponding to the elements of the Hankel matrix R to the observation data if the data corresponding to the elements is observation data stored and held in the data holding step, and (2) setting the data corresponding to the elements to a value indicating that the data is missing data if the data corresponding to the elements of the Hankel matrix R is not stored and held in the data holding step.

[0011] The matrix decomposition processing step performs matrix decomposition on the Hankel matrix R generated in the matrix generation step to obtain a decomposed matrix.

[0012] The optimization process step performs optimization processing based on element values ​​obtained by multiplying the matrices decomposed in the matrix decomposition process step and element values ​​of a Hankel matrix, thereby obtaining an optimized matrix.

[0013] The interpolated data processing step interpolates data for a portion where observation data is missing based on an optimization matrix.

[0014] In this data interpolation processing method, the Hankel matrix R is acquired by stacking data acquired while shifting a window corresponding to a predetermined unit period in the time direction by a first time (e.g., a sample period). In wireless applications such as AGVs (Automated Guided Vehicles), a wireless terminal (e.g., a wireless terminal mounted on an AGV) repeatedly moves along a fixed route, and therefore the communication quality information to be interpolated also varies periodically. For this reason, the Hankel matrix R acquired as described above has a low rank due to high correlation between rows. Therefore, the Hankel matrix R can be decomposed into a matrix of a low rank by matrix decomposition. In this data interpolation processing method, an optimization process is performed based on the element values ​​acquired by the product of the matrices decomposed in the matrix decomposition processing step and the element values ​​of the Hankel matrix, so that the optimization process can be performed while reducing the amount of calculation.

[0015] In this data interpolation method, the optimized matrix obtained by the optimization process is used to interpolate data of missing portions with high accuracy.

[0016] As a result, this data interpolation method can acquire an appropriate waveform pattern (waveform pattern after the interpolation process), and by using the waveform pattern to perform a learning process such as a PNN (Probabilistic neural network) model for future value prediction, highly accurate and efficient learning process can be performed. Furthermore, by performing a prediction process using a trained model acquired by the learning process on the waveform pattern acquired by this data interpolation method, highly accurate prediction process can be realized.

[0017] The "observation data" is, for example, data indicating communication quality (QoS data), such as an RSSI value.

[0018] A second aspect of the present invention is the first aspect of the present invention, in which the matrix decomposition processing step performs singular value decomposition on the Hankel matrix R to obtain a decomposition matrix.

[0019] As a result, in this data interpolation processing method, a decomposition matrix can be obtained by singular value decomposition.

[0020] A third aspect of the present invention is the first or second aspect of the present invention, wherein the optimization step is carried out by selecting an element R ij (i,j: natural number, R ij : A first extraction process is executed to extract only the element in the i-th row and j-th column of the Hankel matrix R.

[0021] Then, the optimization process step selects an element A, which is an element of the matrix product decomposed by the matrix decomposition process step corresponding to the element. ij (A ij : A second extraction process is executed to extract only the element in the i-th row and j-th column of the product of matrices decomposed in the matrix decomposition process step.

[0022] Then, the optimization process step is to extract the element R ij and element A ij Optimization processing is performed to reduce the error between the

[0023] As a result, in this data interpolation processing method, for example, the extracted element R ij and element A ij Optimization processing can be performed by evaluating the loss function for the error with respect to

[0024] A fourth invention is the third invention, in which the optimization processing step executes the first extraction processing and / or the second extraction processing by performing a matrix operation on the product of the Hankel matrix R and / or the matrix decomposed in the matrix decomposition processing step using a selection matrix that selects predetermined elements.

[0025] As a result, in this data interpolation processing method, the selection matrix can be used to easily extract only the elements necessary for the optimization processing.

[0026] A fifth invention is any of the first to fourth inventions, in which the interpolated data processing step obtains an average value or a weighted average value of elements corresponding to a portion where observation data is missing and arranged in an anti-diagonal direction of the optimization matrix, and interpolates data for the portion where observation data is missing based on the obtained average value or weighted average value.

[0027] As a result, this data interpolation method can obtain data for parts where observation data is missing by the simple method of obtaining the average or weighted average of the elements arranged in the opposite diagonal direction of the optimization matrix.

[0028] A sixth invention is a data interpolation processing device for performing interpolation processing on intermittently acquired observation data, comprising an observation data acquisition unit, a data storage unit, a matrix generation unit, a matrix decomposition processing unit, an optimization processing unit, and an interpolation data processing unit.

[0029] The observation data acquisition unit acquires observation data relating to communication quality from the wireless signal.

[0030] The data storage unit stores and holds the observation data acquired by the observation data acquisition unit for a predetermined period of time.

[0031] The matrix generation unit generates a Hankel matrix R so that when a window corresponding to a predetermined unit period is shifted in the time direction by a first time, which is a predetermined period, the group of data included within the period corresponding to the window becomes one row of elements.

[0032] Then, the matrix generation unit generates the Hankel matrix R by (1) setting the data corresponding to the element of the Hankel matrix R to the observation data if the data corresponding to the element is stored in the data storage unit, and (2) setting the data corresponding to the element of the Hankel matrix R to a value indicating that the data is missing data if the data corresponding to the element is not stored in the data storage unit.

[0033] The matrix decomposition processing unit performs matrix decomposition on the Hankel matrix R generated by the matrix generation unit to obtain a decomposed matrix.

[0034] The optimization processing unit performs optimization processing based on element values ​​obtained by multiplying the matrices decomposed by the matrix decomposition processing unit and element values ​​of the Hankel matrix, thereby obtaining an optimized matrix.

[0035] The interpolation data processing unit interpolates data for portions where observation data is missing, based on the optimization matrix.

[0036] This makes it possible to realize a data interpolation processing device that achieves the same effects as the first aspect of the invention.

[0037] A seventh aspect of the present invention is a program for causing a computer to execute the data interpolation processing method according to any one of the first to fourth aspects of the present invention.

[0038] This makes it possible to realize a program for causing a computer to execute a data interpolation processing method that achieves the same effects as the first aspect of the invention.

[0039] An eighth invention is a learning processing device that performs a learning process of a model that predicts future values ​​of observed data using interpolated data obtained by a data interpolation processing method that is any one of the first to fifth inventions.

[0040] This makes it possible for the learning processing device to perform learning processing using the interpolation data acquired by the data interpolation processing method according to any one of the first to fifth aspects of the present invention.

[0041] The ninth invention is a trained model obtained by performing a learning process of a model that predicts future values ​​of observation data using the interpolation data obtained by the data interpolation processing method according to any one of the first to fifth inventions, This is a prediction processing device that predicts future values ​​of observed data by inputting interpolated data obtained by the data interpolation processing method of any of the first to fifth inventions into a trained model.

[0042] As a result, in this learning processing device, by inputting the interpolated data obtained by the data interpolation processing method of any of the first to fifth inventions into the learned model, it is possible to perform highly accurate prediction of future values ​​of observed data. Effect of the Invention

[0043] According to the present invention, it is possible to realize a wireless situation prediction device that can obtain data that is appropriately interpolated from observation data even in a situation where the observation data is obtained intermittently, and can use the data to accurately and appropriately perform learning processing for a model that predicts future communication quality conditions, as well as a data interpolation processing method and program used therein. [Brief description of the drawings]

[0044] [Figure 1] 1 is a schematic configuration diagram of a wireless situation predicting device 100 according to a first embodiment. [Diagram 2] 1 is a schematic configuration diagram of a data interpolation processing unit 1 of a wireless situation predicting device 100 according to a first embodiment. [Diagram 3] 11A and 11B are diagrams for explaining linear interpolation processing for the same waveform pattern when observation data is acquired intermittently. [Figure 4] 11A and 11B are diagrams for explaining data interpolation processing for the same waveform pattern when observation data is acquired intermittently. [Diagram 5] 11A and 11B are diagrams for explaining a process of acquiring a Hankel matrix from intermittently acquired observation data in a waveform pattern of RSSI values. [Figure 6] FIG. 13 is a diagram for explaining a matrix decomposition process. [Figure 7] 4 is a flowchart of a data interpolation process executed by the wireless situation predicting device 100. [Figure 8]11 is a diagram for explaining a process of extracting elements for which observation data exists from a matrix R using a selection matrix Sk. [Figure 9] 13 is a diagram for explaining a process of extracting data in the same row as an element for which observed data exists in matrix R from matrix P using selection matrix Sk. [Figure 10] 11 is a diagram for explaining a process of extracting elements for which observation data exists from a matrix R using a selection matrix Si; [Figure 11] 13 is a diagram for explaining a process of extracting data in the same column as an element in matrix R for which observed data exists, from matrix P using selection matrix Si; [Figure 12] FIG. 13 is a diagram for explaining data interpolation processing, showing intermittently acquired observation data (RSSI values) and an optimization matrix R_opt (a matrix acquired by optimization processing for the Hankel matrix R). [Figure 13] FIG. 13 is a diagram for explaining data interpolation processing, showing intermittently acquired observation data (RSSI values) and interpolated data acquired using an optimization matrix R_opt. [Figure 14] This figure shows data acquired by interpolating the RSSI values ​​(data acquired intermittently) of wireless signals acquired at access points installed at predetermined positions in an environment where an AGV controlled by wireless communication travels a predetermined route within a warehouse. [Figure 15] This figure shows the predicted result data when a prediction process (prediction of RSSI value 10 seconds ahead) is performed using a PNN model using data acquired by interpolating RSSI values ​​(intermittently acquired data) of wireless signals acquired at access points installed at predetermined positions in an environment where AGVs controlled by wireless communication travel a predetermined route within a warehouse. [Figure 16] A diagram showing the CPU bus configuration. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0045] [First embodiment] The first embodiment will be described below with reference to the drawings.

[0046] <1.1: Configuration of wireless communication system> FIG. 1 is a schematic configuration diagram of a radio situation predicting device 100 according to the first embodiment.

[0047] FIG. 2 is a schematic configuration diagram of the data interpolation processing unit 1 of the radio situation predicting device 100 according to the first embodiment.

[0048] As shown in FIG. 1, the radio situation predicting device 100 includes an antenna Ant1, a demodulation processing unit Dm1, an observation data acquisition unit FE1, a data interpolation processing unit 1, and a prediction processing unit 2.

[0049] The antenna Ant1 is an antenna for receiving an RF signal (e.g., a carrier band OFDM signal) from the outside. The antenna Ant1 outputs the received RF signal (e.g., a carrier band OFDM signal) to the demodulation processing unit Dm1 as a signal RFin. The antenna Ant1 may be an antenna that also functions as a transmitting antenna (a transmitting and receiving antenna).

[0050] The demodulation processing unit Dm1 includes an RF processing unit Dm11 and a BB processing unit Dm12.

[0051] The RF processing unit Dm11 performs RF demodulation processing on the signal RFin to obtain a baseband signal.

[0052] The BB processing unit Dm12 performs baseband demodulation processing on the baseband signal demodulated by the RF processing unit Dm11, acquires the signal (data) after the baseband demodulation processing as data D0, and outputs the acquired data D0 to the observation data acquisition unit FE1.

[0053] The observation data acquisition unit FE1 inputs data D0 output from the demodulation processing unit Dm1, and acquires from the data D0 a QoS value that is data indicating the Quality of Service (QoS) of the wireless communication environment in which the wireless situation predicting device 100 is placed (data representing communication quality). Then, the observation data acquisition unit FE1 outputs the data including the acquired QoS value to the data interpolation processing unit 1 as data D1.

[0054] The QoS value is an index value indicating communication quality, such as received signal strength indicator (RSSI), throughput, the number of (continuous) packet losses, and the amount of delay.

[0055] As shown in FIG. 2, the data interpolation processing unit 1 includes a data holding unit 11, a matrix generation unit 12, a matrix decomposition processing unit 13, an optimization processing unit 14, and an interpolation data acquisition unit 15.

[0056] The data holding unit 11 inputs data D1 from the observation data acquisition unit FE1. The data holding unit 11 stores and holds the input data D1 for a predetermined period of time. The data holding unit 11 is realized, for example, by using a FIFO (First in First out) memory. In response to a command from the matrix generation unit 12, the data holding unit 11 reads out data stored in a predetermined address and outputs the read out data to the matrix generation unit 12 as data D_ptn.

[0057] The matrix generation unit 12 reads data stored at a predetermined address from the data storage unit 11, and inputs the read data as data D_ptn from the data storage unit 11. The matrix generation unit 12 acquires a matrix R from the data D_ptn, and outputs the acquired matrix R (data including the matrix R) to the matrix decomposition processing unit 13 and the optimization processing unit 14.

[0058] The matrix decomposition processing unit 13 receives data including the matrix R output from the matrix generation unit 12, and performs matrix decomposition processing on the matrix R. The matrix decomposition processing unit 13 outputs data including the matrices R and Q acquired by the matrix decomposition processing to the optimization processing unit 14.

[0059] The optimization processing unit 14 receives data including matrix R output from the matrix generation unit 12 and data including matrices R and Q output from the matrix decomposition processing unit 13. The optimization processing unit 14 performs optimization processing using matrix R, matrices P, and Q (matrices P and Q obtained by decomposing matrix R) to obtain an optimized matrix R_opt. The optimization processing unit 14 then outputs the data including the obtained optimized matrix R_opt to the interpolation data acquisition unit 15.

[0060] The interpolation data acquisition unit 15 inputs the optimization matrix R_opt output from the optimization processing unit 14, performs an interpolation process based on the optimization matrix R_opt, and acquires the interpolation data. Then, the interpolation data acquisition unit 15 outputs data including the acquired interpolation data to the prediction processing unit 2 as data D1_intpl.

[0061] The prediction processing unit 2 inputs data D1_intpl output from the data interpolation processing unit 1 and a mode signal Mode output from a control unit (not shown) specifying a processing mode (e.g., (1) learning processing mode and (2) prediction processing execution mode).

[0062] When the mode signal Mode is a signal specifying the learning processing mode, the prediction processing unit 2 uses the data D1_intpl to execute a learning process (for example, a learning process (training process) of a learning model using a PNN model or a neural network model).

[0063] When the mode signal Mode is a signal specifying the prediction process execution mode, the prediction processor 2 executes prediction processing on the data D1_intpl, and outputs data including the result data of the prediction processing as data Dout.

[0064] The "data interpolation processing device" is realized by the data interpolation processing unit.

[0065] <1.2: Operation of the wireless situation prediction device> The operation of the wireless situation predicting device 100 configured as above will be described below.

[0066] Fig. 3 is a diagram for explaining linear interpolation processing for the same waveform pattern when observation data is intermittently acquired. The upper diagram of Fig. 3 is a diagram showing a waveform pattern (waveform pattern of RSSI value) when observation data is intermittently acquired, with the horizontal axis representing time and the vertical axis representing RSSI value. The lower diagram of Fig. 3 is a diagram showing a waveform pattern (waveform pattern of RSSI value) of observation data acquired by linear interpolation for the waveform pattern in the upper diagram of Fig. 3, with the horizontal axis representing time and the vertical axis representing RSSI value.

[0067] Fig. 4 is a diagram for explaining data interpolation processing for the same waveform pattern when observation data is intermittently acquired. The upper diagram of Fig. 4 is a diagram showing a waveform pattern (waveform pattern of RSSI value) when observation data is intermittently acquired, with the horizontal axis representing time and the vertical axis representing RSSI value. The lower diagram of Fig. 4 is a diagram showing a waveform pattern (waveform pattern of RSSI value) of observation data acquired by data interpolation processing (data interpolation processing executed by wireless situation predicting device 100) for the waveform pattern in the upper diagram of Fig. 4, with the horizontal axis representing time and the vertical axis representing RSSI value.

[0068] FIG. 5 is a diagram for explaining a process of acquiring a Hankel matrix from intermittently acquired observation data in a waveform pattern of an RSSI value.

[0069] FIG. 6 is a diagram for explaining the matrix decomposition process.

[0070] FIG. 7 is a flowchart of the data interpolation process executed by the wireless situation predicting device 100. As shown in FIG.

[0071] Figure 8 shows the selection matrix S k13 is a diagram for explaining a process of extracting elements for which observation data exists from matrix R by

[0072] Figure 9 shows the selection matrix S k 13 is a diagram for explaining a process of extracting data in the same row as an element for which observed data exists in matrix R from matrix P by

[0073] Figure 10 shows the selection matrix S i 13 is a diagram for explaining a process of extracting elements for which observation data exists from matrix R by

[0074] Figure 11 shows the selection matrix S i 13 is a diagram for explaining a process of extracting data in the same column as an element for which observed data exists in matrix R from matrix P by

[0075] First, an interpolation process for a waveform pattern in which observed data (for example, RSSI values) are intermittently acquired will be described.

[0076] For example, a case will be described where linear interpolation is performed using observed data when the waveform pattern of the RSSI value from time tt1 to time tt2 is the same as the waveform pattern of the RSSI value from time tt3 to time tt4 as shown in Fig. 3. As shown in the left diagram of Fig. 3, it is assumed that the data shown in the portions indicated by D1_observed and D2_observed is observed (data was acquired) from time tt1 to time tt2, and as shown in the right diagram of Fig. 3, it is assumed that the data shown in the portions indicated by D3_observed and D4_observed is observed (data was acquired) from time tt3 to time tt4.

[0077] In this case, when linear interpolation is performed using the observation data in the left diagram of Figure 3 (the data indicated by D1_observed and D2_observed), a waveform pattern like the one shown in the lower left diagram of Figure 3 (waveform pattern after linear interpolation) is obtained.

[0078] In addition, if linear interpolation is performed using the observation data in the right diagram of Figure 3 (the data indicated by D3_observed and D4_observed), a waveform pattern like the one shown in the lower right diagram of Figure 3 (waveform pattern after linear interpolation) is obtained.

[0079] In this way, when data of a data missing portion is interpolated by linear interpolation for the same waveform pattern, the waveform pattern acquired by linear interpolation will be different depending on which portion of the waveform pattern the observed data is (depending on the position of the missing portion of the observed data). When the missing portion of the observed data is different, if the same waveform pattern is linearly interpolated to obtain a different waveform pattern (different waveform pattern), the accuracy and efficiency of the learning process will be deteriorated when a learning process of a PNN model for future value prediction is performed using the waveform pattern acquired by linear interpolation. In other words, even if the waveform pattern is originally the same, if the learning process is performed using a waveform pattern that is different from the waveform pattern acquired by linear interpolation, the learning process will be performed as a different waveform pattern, and a learned model that performs appropriate future value prediction will not be acquired. In addition, even if the waveform pattern is originally the same, if the waveform pattern acquired by linear interpolation becomes a different waveform pattern, the number of waveform patterns will be inappropriately increased, and the efficiency of the learning process will be deteriorated.

[0080] In order to solve such problems with linear interpolation, the radio situation predicting device 100 performs data interpolation processing so that when observation data is intermittently acquired, no matter where the acquired portion of the observation data (the portion of missing data) is, when data interpolation processing is performed on the same waveform pattern, a substantially identical waveform pattern (waveform pattern after interpolation processing) is acquired, as shown in Fig. 4. This allows the radio situation predicting device 100 to acquire an appropriate waveform pattern (waveform pattern after interpolation processing) and perform highly accurate and efficient learning processing (learning processing of a PNN model for predicting future values, etc.).

[0081] The following describes the operation of the radio situation predicting device 100, dividing it into (1) the learning process operation (the learning process mode operation) and (2) the prediction process operation (the prediction process execution mode operation). The description will be made with reference to the flowchart in FIG. 7.

[0082] (1.2.1: Learning process (operation in learning process mode)) First, the learning process (operation in the learning process mode) will be described.

[0083] The demodulation processing unit Dm1 of the radio situation predicting device 100 executes demodulation processing on the radio signal received via the antenna Ant1, and acquires an RF demodulated signal and a BB demodulated signal. The observation data acquisition unit FE1 acquires a QoS value (an index value indicating communication quality such as throughput, the number of (continuous) packet losses, and the amount of delay) from the RF demodulated signal and / or the BB demodulated signal acquired by the demodulation processing unit Dm1. For ease of explanation, the following description will be given assuming that the QoS value is an RSSI value (signal reception strength).

[0084] Then, the observation data acquisition unit FE1 outputs data including the acquired RSSI value to the data interpolation processing unit 1 as data D1.

[0085] The data storage unit 11 of the data interpolation processing unit 1 stores and holds the input data D1 for a predetermined period. Here, the data storage unit 11 is assumed to be realized using a FIFO (First in First out) memory. The FIFO memory may be realized using, for example, a dual-port RAM.

[0086] If the period for acquiring one sample from the FIFO memory (the period for inputting one piece of data to the FIFO memory) is a sample period Ts, the data holding unit 11 stores and holds data for a period equivalent to Ts×(Nw+Ns-1). In other words, the data holding unit 11 stores and holds (Nw+Ns-1) pieces of sample data.

[0087] In addition, during a period when observation data cannot be acquired from the observation data acquisition unit FE1 (a data missing period), the data storage unit 11 inputs data with a value of "0" (RSSI value) to the FIFO memory, and during a period when observation data can be acquired, inputs the RSSI value of the observation data to the FIFO memory.

[0088] The matrix generation unit 12 reads out data for a predetermined period stored in the data storage unit 11 as data D_ptn for the data storage unit 11. Specifically, the matrix generation unit 12 reads out (Nw+Ns-1) pieces of sample data stored in the data storage unit 11 and acquires Ns pieces of row data as follows, thereby acquiring a matrix R with Ns rows and Nw columns (Hankel matrix R) (step S1 in FIG. 7). (1) Time t 11 ~t 12 Data (samples) r1 to r2 stored (input) in the FIFO memory of the data holding unit 11 during the period Nw be the elements of the first row of matrix R (Nw elements). (2) Time t 11 +Ts~t 12 Data (samples) r2 to r4 stored (input) in the FIFO memory of the data storage unit 11 during the period +Ts Nw+1 are the elements (Nw elements) of the second row of the matrix R. Note that Ts is the sampling period when acquiring data in the FIFO memory of the data holding unit 11. (3) Time t 11 +Ts×2~t 12 Data (samples) r3 to r4 stored (input) in the FIFO memory of the data holding unit 11 during the period +Ts×2 Nw+2 be the elements in the third row of matrix R (Nw elements). ... (k) Time t 11 +Ts×(k-1)~t 12 Data (samples) r stored (input) in the FIFO memory of the data holding unit 11 during the period +Ts×(k−1) k ~r Nw+k-1 is the element (Nw elements) of the kth row of matrix R (k: natural number, 1≦k≦Ns). ... (Ns) Time t 11 +Ts×(Ns-1)~t 12 Data (samples) r stored (input) in the FIFO memory of the data holding unit 11 during the period +Ts×(Ns-1) Ns ~r Nw+Ns-1 is the element (Nw elements) of the Ns-th row of matrix R (k: natural number, 1≦k≦Ns).

[0089] The matrix R (Hankel matrix R) of Ns rows and Nw columns acquired by the matrix generating unit 12 as described above corresponds to a matrix acquired by shifting the sliding window Win1 shown in FIG. 5 by one sample period in the direction of the arrow Dir1 in FIG. 5, and taking the data in the sliding window as one row of elements. As can be seen from FIG. 5, the matrix R is a matrix in which the observation data moves diagonally downward to the left, and elements of the part without observation data (elements of the data missing part) have a value of "0". In the Hankel matrix R shown in the lower left diagram of FIG. 5, the data in the part indicated by the solid line is the RSSI value of the observation data as an element, and the white part is the element with a value of "0". In the Hankel matrix R shown in the lower right diagram of FIG. 5, the data in the gray part is the RSSI value of the observation data as an element, and the white part is the element with a value of "0".

[0090] That is, the matrix R with Ns rows and Nw columns obtained as described above is a Hankel matrix in which only the anti-diagonal components have a value other than "0." Note that, although a Hankel matrix is ​​defined as a square matrix, even in the case of a matrix other than a square matrix, if the anti-diagonal components are the same (if the element (observation data) in row i and column j is the same as the element in row i+1 and column j-1), it is considered to be a Hankel matrix in the broad sense, and such a matrix will be referred to as a Hankel matrix hereinafter.

[0091] The matrix generation unit 12 outputs data including the Hankel matrix R acquired as described above to the matrix decomposition processing unit 13 and the optimization processing unit 14.

[0092] The matrix decomposition processing unit 13 performs a matrix decomposition process on the Hankel matrix R acquired by the matrix generation unit 12. Specifically, the matrix decomposition processing unit 13 performs a process of acquiring (calculating) an effective rank number (order) L of the Hankel matrix R by a known method, and acquires the effective rank number L of the Hankel matrix R.

[0093] Then, the matrix decomposition processing unit 13 performs singular value decomposition on the Hankel matrix R. The effective rank number refers to the number of non-zero singular values ​​after considering singular values ​​that are very small compared to the maximum singular value (the singular value with the largest absolute value) as 0. For example, the singular value considered as 0 is determined as a singular value having an absolute value of 1 / 10 or less of the absolute value of the maximum singular value. By using the effective rank number, it is possible to reduce the influence of noise caused by error factors such as a shift in observation timing and a shift in the route traveled by a wireless terminal. In general, when singular value decomposition is performed on a matrix R with an effective rank number L, the matrix R can be expressed as follows. R=UΣV T U: unitary matrix composed of left singular vectors Σ: A matrix whose diagonal elements are singular values ​​(hereafter referred to as a "singular value matrix") V: unitary matrix composed of right singular vectors Here, when the effective rank number of the singular value matrix Σ is L, the elements in the first row to the Lth row of the diagonal components are non-zero, and the values ​​of the other elements are "0". In other words, the diagonal components σ 11 , σ 22 , , σ LL The value of σ is non-zero. ij represents the element in row i, column j. Also, A T represents the transpose of matrix A (A T is the transpose of matrix A).

[0094] The matrix decomposition processing unit 13 performs singular value decomposition to obtain the unitary matrix U, the singular value matrix Σ, and the unitary matrix V T Then, the matrix decomposition processing unit 13 obtains P=U :,1:L Σ 0.51:L,1:L Q=Σ 0.5 1:L,1:L V T 1:L,: L: Effective rank of matrix R Then, we obtain matrices P and Q. Let us assume that matrix A is an m×n (m rows, n columns) matrix with real numbers as elements, and the element of row i, column j of matrix A is x ij Then, (1)A 0.5 , (2)A a:b,c:d , (3)A a,: , and (4)A :,a shall mean the following: (1)A 0.5 :The component in row i, column j is sqrt(x ij ) matrix (m × n matrix) (sqrt(x) represents the square root of x) (2)A a:b,c:d : A matrix obtained by extracting elements from rows a to b of matrix A and elements from columns c to d of matrix A (3)A a,: : A matrix obtained by extracting the elements of row a of matrix A (4)A :,a : A matrix obtained by extracting the elements of column a of matrix A As shown in Figure 6, the m×n matrix R is a combination of the n×n unitary matrix U, the n×m singular value matrix Σ, and the transpose matrix V of the m×m unitary matrix V. T When the effective rank number of the matrix R is L, the diagonal elements σ of the singular value matrix Σ are 11 , σ 22 , , σ LL Since only the value of is non-zero, as shown in Figure 6, R=UΣV T =U :,1:L Σ 1:L,1:L V T 1:L,: =U :,1:L Σ 0.5 1:L,1:L Σ 0.5 1:L,1:L V T 1:L,: Therefore, P=U :,1:L Σ 0.5 1:L,1:L Q=Σ 0.5 1:L,1:L V T 1:L,: By doing so, R ≒ PQ That is, the matrix R can be expressed as the product of an n×L matrix P and an L×m matrix Q. In singular value decomposition, small singular values ​​caused by the influence of noise, etc. are ignored, so R and PQ are not strictly equivalent, and therefore, as described above, they are expressed as an approximation (R and PQ are expressed as being approximately the same). In addition, in FIG. 6, singular values ​​smaller than a predetermined value (for example, singular values ​​within an error range) are regarded as "0", and the singular value matrix Σ, etc. are shown.

[0095] The matrix decomposition processing unit 13 performs processing corresponding to the above formula on the Hankel matrix R (Ns×Nw matrix) with effective rank number L to obtain the Ns×L matrix P and the L×Nw matrix Q (step S2 in FIG. 7). Then, the matrix decomposition processing unit 13 outputs data including the obtained matrices P and Q to the optimization processing unit 14.

[0096] The optimization processing unit 14 performs optimization processing using the Hankel matrix R, matrices P, and Q (matrices P and Q obtained by decomposing the Hankel matrix R) to obtain an optimized matrix R_opt. Specifically, the optimization processing unit 14 sets the loss function Loss(P,Q) as shown in the following formula.

number

number

number

[0097] The optimization processor 14 performs processing corresponding to the following formula to obtain the optimal solution P opt , Q opt Get the.

number

number

number

number

number

[0098] ≪Optimization process for 1:Q (step S4)≫ The optimization processor 14 extracts only elements for which observed data exists in the kth column (column for which observed data exists) of the Hankel matrix. Specifically, the optimization processor 14 performs processing equivalent to the following formula, and extracts only elements for which observed data exists in the kth column (column for which observed data exists) of the Hankel matrix R.

number

[0099] When elements having observation data in the kth column of the Hankel matrix R are detected in ascending order of row number, the elements are taken as the first element to the αth element, and the row number of the qth element (q: natural number, 1≦q≦α) is taken as row(q). The optimization processing unit 14 then extracts the row number of the selection matrix S k The element in the qth row of is set so that only the row(q) column has the value "1" and the values ​​of the other columns are "0".

[0100] The selection matrix S thus generated k An example of the above is shown in Fig. 8. Fig. 8 shows a case where the elements (elements of the known part) having the observation data of the k-th column of the Hankel matrix R are the elements of the 1st row, 6th row, 7th row, and n-th row (four elements). As shown in the lower diagram of Fig. 8, the optimization processing unit 14 calculates S k R :,k By performing processing equivalent to the above, it is possible to obtain a matrix x that extracts only elements that have observation data in the kth column of the Hankel matrix R.

[0101] In addition, the optimization processor 14 selects the selection matrix S kIn the matrix P, data (elements) in the same row as the elements in the kth column (column in which observed data exists) of the Hankel matrix R are extracted using the selection matrix S k By multiplying matrix P by matrix P, data (elements) in the same row as elements in the kth column (column in which observed data exists) of Hankel matrix R are extracted. That is, by executing a process corresponding to the following formula, optimization processing unit 14 extracts matrix P part can be obtained.

number

[0102] The optimization processing unit 14 calculates the matrix P part and matrix Q obtained by extracting the k-th column of matrix Q :,k By performing processing corresponding to the following formula using the matrix x, the optimization processor 14 can efficiently (by reducing the amount of calculation) eliminate the error between the RSSI value of the observation data in the kth column of the Hankel matrix R and the value obtained by multiplying the matrices P and Q obtained by matrix decomposition. k The matrix P obtained by part , and matrix Q obtained by extracting the k-th column of matrix Q :,kThis can be efficiently achieved by carrying out processing equivalent to the following formula to remove the error from the matrix obtained by multiplying the above.

number

number

number

number

[0103] Then, the optimization processing unit 14 repeatedly executes the above process for all k. That is, the optimization processing unit 14 repeatedly executes the above process (the optimization process for the matrix Q with the matrix P fixed) for all columns in the Hankel matrix R in which observed data exists.

[0104] <2: Optimization process for P (step S5)> Next, the optimization processing unit 14 fixes the matrix Q and performs optimization processing on the matrix P. Specifically, the optimization processing unit 14 performs processing corresponding to the following formula, and extracts only elements for which observed data exists in the i-th row (row for which observed data exists) of the Hankel matrix R.

number

[0105] When elements having observation data in the i-th row of the Hankel matrix R are detected in ascending order of column number, the elements are taken as the first element to the α-th element, and the column number of the q-th element (q: natural number, 1≦q≦α) is taken as col(q). The optimization processor 14 then finds the selection matrix S i The element in the qth row of is set so that only the col(q)th column has the value "1" and the values ​​of the other columns are "0".

[0106] The selection matrix S thus generated i An example of the above is shown in Fig. 10. Fig. 10 shows a case where the elements (the elements of the known part) having the observation data in the i-th row of the Hankel matrix R are the elements in the 1st, 6th, 7th, and m-th columns (four elements). As shown in the lower diagram of Fig. 10, the optimization processing unit 14 calculates S i R T i,: By performing a process equivalent to the above, it is possible to obtain a matrix y that extracts only elements that have observation data in the i-th row of the Hankel matrix R.

[0107] In addition, the optimization processor 14 selects the selection matrix S i In the matrix Q, data (elements) in the same column as the elements in the i-th row (row in which observed data exists) of the Hankel matrix R are extracted using the selection matrix S i and matrix Q TBy taking the product of and, in matrix Q, data (elements) in the same column as the elements in which observed data exists in row i (rows in which observed data exists) of Hankel matrix R are extracted. That is, the optimization processing unit 14 executes processing corresponding to the following formula to extract data (elements) in the same column as the elements in which observed data exists in row i (rows in which observed data exists) of Hankel matrix R in matrix Q. part can be obtained.

number

[0108] The optimization processing unit 14 extracts the i-th row of the matrix P obtained as described above, T i,: and matrix Q part By performing processing corresponding to the following formula using the matrix y, the error between the RSSI value of the observation data in the i-th row of the Hankel matrix R and the value obtained by multiplying the matrices P and Q obtained by matrix decomposition can be efficiently (by reducing the amount of calculation). T i,: , and the selection matrix S i The matrix Q obtained by part This can be efficiently achieved by carrying out processing equivalent to the following formula to remove the error from the matrix obtained by multiplying the above.

number

number

number

number

[0109] Then, the optimization processing unit 14 repeatedly executes the above process for all i. That is, the optimization processing unit 14 repeatedly executes the above process (the optimization process for the matrix P with the matrix Q fixed) for all rows in the Hankel matrix R in which observed data exists.

[0110] Then, after performing the optimization process for the matrix Q (step S4) and the optimization process for the matrix P (step S5), the optimization processing unit 14 updates the Hankel matrix R (R=PQ) (step S6) using the matrix Q obtained in the optimization process for the matrix Q and the matrix P obtained in the optimization process for the matrix P.

[0111] Then, as shown in the flowchart of FIG. 7, the above process (steps S4 to S6) is repeated a predetermined number of times (in the flowchart of FIG. 7, the variable t is incremented by +1, and the process of steps S4 to S6 is repeated (steps S3 to S8) while t is smaller than a predetermined number of times tmax).

[0112] 7 shows a case where the process of steps S4 to S6 is repeated a predetermined number of times, but is not limited thereto. For example, when the value of the loss function Loss(P,Q) does not vary by a value within a predetermined range even after the process of steps S4 to S6 is repeatedly executed, it may be determined that the optimization process of matrix R has converged, and the optimization processor 14 may end the process of steps S4 to S6. Also, when the value of the loss function Loss(P,Q) becomes smaller than a predetermined value, it may be determined that the optimization process of matrix R has converged, and the optimization processor 14 may end the process of steps S4 to S6.

[0113] The optimization processing unit 14 outputs the Hankel matrix R acquired as described above (the Hankel matrix R after the optimization processing) to the data interpolation processing unit 15 as an optimization matrix R_opt.

[0114] <<Data Interpolation Processing (Step S9)>> The interpolation data acquisition unit 15 performs an interpolation process and acquires the interpolation data based on the optimization matrix R_opt output from the optimization processing unit 14. A specific process will be described with reference to Figs.

[0115] Fig. 12 is a diagram for explaining the data interpolation process, showing intermittently acquired observation data (RSSI values) and an optimization matrix R_opt (a matrix acquired by an optimization process for the Hankel matrix R). In the upper diagram of Fig. 12, the horizontal axis represents time, and the vertical axis represents the RSSI value.

[0116] Fig. 13 is a diagram for explaining the data interpolation process, showing the intermittently acquired observation data (RSSI value) and the interpolated data acquired by the optimization matrix R_opt. In the upper diagram of Fig. 13, the horizontal axis is time and the vertical axis is RSSI value.

[0117] 12 and 13 show the case where the optimization process is performed using the Hankel matrix R (Ns=8, Nw=13) of Ns×Nw to obtain the optimized matrix R_opt of Ns×Nw. In addition, in FIG. 12 and FIG. 13, the black circles indicate the observed data (known part).

[0118] In the case of Figures 12 and 13, r1 to r3, r8 to r 10 , r 13 , r 16 , r 21 is the data (sample) with observed data, and the other r4 to r7, r 11 ~r 12 , r 14 ~r 15 , r 17 ~r 20 are parts (samples) with no observed data (parts with missing data). In the optimization matrix R_opt shown in Figures 12 and 13, elements with black circles connected by a solid line indicate elements with the same value (same RSSI value).

[0119] In this case, as described above, in the wireless situation predicting device 100, an Ns×Nw Hankel matrix R (Ns=8, Nw=13) is generated (the value “0” is inserted as the initial value into the elements of the data missing portion), and the optimization matrix R_opt is obtained by performing the above processing (the processing of steps S2 to S8 in FIG. 7).

[0120] Then, the interpolation data acquisition unit 15 of the wireless situation predicting device 100 acquires the average value of the components (elements) in the opposite diagonal direction, which are elements corresponding to the data (samples) of the data missing portion, as shown in Fig. 13, and sets the acquired average value as the data (interpolation data) of the data missing portion. For example, in the case of Fig. 13, the element of the i-th row and j-th column of the optimization matrix R_opt is x i,j Then, r4 to r7, r 11 ~r 12 , r 14 ~r 15 , r 17 ~r 20 Interpolated data rh4~rh7, rh 11 ~rh 12 , rh 14 ~rh 15 , rh 17 ~rh 20 becomes rh4 = Ave(x 1,4 ,x 2,3 ,x 3,2 ,x 4,1 ) rh5=Ave(x 1,5 ,x 2,4 ,x 3,3 ,x 4,2 ,x 5,1 ) rh6 = Ave(x 1,6 ,x 2,5 ,x 3,4 ,x 4,3 ,x 5,2 ,x 6,1 ) rh7=Ave(x 1,7 ,x 2,6 ,x 3,5 ,x 4,4 ,x 5,3 ,x 6,2 ,x 7,1 ) rh 11 =Ave(x 1,11 ,x 2,10 ,x 3,9 ,x 4,8 ,x 5,7 ,x 6,6 ,x 7,5 ,x 8,4 ,x 9,3 ) rh 12 =Ave(x1,12 ,x 2,11 ,x 3,10 ,x 4,9 ,x 5,8 ,x 6,7 ,x 7,6 ,x 8,5 ,x 9,4 ) rh 14 =Ave(x 2,13 ,x 3,12 ,x 4,11 ,x 5,10 ,x 6,9 ,x 7,8 ,x 8,7 ,x 9,6 ) rh 15 =Ave(x 3,13 ,x 4,12 ,x 5,11 ,x 6,10 ,x 7,9 ,x 8,8 ,x 9,7 ) rh 17 =Ave(x 5,13 ,x 6,12 ,x 7,11 ,x 8,10 ,x 9,9 ) rh 18 =Ave(x 6,13 ,x 7,12 ,x 8,11 ,x 9,10 ) rh 19 =Ave(x 7,13 ,x 8,12 ,x 9,11 ) rh 20 =Ave(x 8,13 ,x 9,12 ) Ave(): A function to get the average value of elements In FIG. 13, the interpolation data rh4 to rh7 and rh 11 ~rh 12 , rh 14 ~rh 15 , rh 17 ~rh 20are indicated by white circles in the upper diagram of Fig. 13. By performing the above processing in the data interpolation processing unit 1 of the wireless situation predicting device 100, appropriate interpolated data is acquired as shown in the upper diagram of Fig. 13.

[0121] In the above, the function Ave() is used to obtain the average value of the components (elements) in the opposite diagonal direction of the optimization matrix R_opt. However, this is not limited to this. For example, a weighted average value of the components (elements) in the opposite diagonal direction of the optimization matrix R_opt may be obtained.

[0122] Then, the interpolation data acquisition unit 15 outputs data including the interpolation data acquired in the above process to the prediction processing unit 2 as data D1_intpl.

[0123] The prediction processing unit 2 uses the data D1_intpl output from the data interpolation processing unit 1 to execute a learning process (for example, a learning process (training process) of a learning model using a PNN model or a neural network model).

[0124] In the radio situation predicting device 100, when observation data is intermittently acquired, data interpolation processing is performed so that a substantially identical waveform pattern (waveform pattern after interpolation processing) is acquired when data interpolation processing is performed on the same waveform pattern, regardless of where the acquired portion of the observation data (portion of missing data) is located. That is, in the radio situation predicting device 100, as described above, a Hankel matrix R is generated from the observation data, and optimization processing is performed using the Hankel matrix R to acquire an acquired optimization matrix R_opt. Then, the radio situation predicting device 100 interpolates data of the data missing portion using the acquired optimization matrix R_opt.

[0125] Therefore, the radio situation prediction device 100 can acquire an appropriate waveform pattern (waveform pattern after interpolation processing), and by using the waveform pattern to perform learning processing of a PNN model for future value prediction, etc., which is mounted in the prediction processing unit 2, it is possible to perform highly accurate and efficient learning processing (learning processing of a PNN model for future value prediction, etc.).

[0126] (1.2.2: Prediction process (operation in prediction process execution mode)) Next, the prediction process (operation in the prediction process execution mode) will be described.

[0127] In the radio situation predicting device 100, in the prediction process execution mode, data interpolation process is performed in the same manner as in the learning process mode, and data D1_intpl after the interpolation process is input to the prediction processing unit 2.

[0128] In the prediction processing unit 2, a trained model (e.g., a PNN model for future value prediction) is constructed by executing a learning process using the data D1_intpl (waveform pattern) after the interpolation process, so the data D1_intpl (waveform pattern) after the interpolation process is input to the trained model, and data of the future value prediction result, for example, is acquired based on the output from the trained model.Then, the prediction processing unit 2 outputs the data of the future value prediction result acquired as described above, for example, as data Dout.

[0129] In addition, when performing processing using a PNN model in the prediction processing in the prediction processing unit 2, for example, the techniques disclosed in Japanese Patent Application No. 2017-232442 and Japanese Patent Application No. 2018-042747 may be used.

[0130] In the radio situation predicting device 100, when observation data is acquired intermittently, prediction processing is performed using a learned model acquired by performing learning processing using data D1_intpl after data interpolation processing in which substantially the same waveform pattern (waveform pattern after interpolation processing) is acquired regardless of where the acquired portion of the observation data (portion of missing data) is located. Therefore, even when observation data can only be acquired intermittently, the radio situation predicting device 100 performs data interpolation processing using the observation data and performs prediction processing using the data after data interpolation processing, thereby enabling highly accurate prediction processing (for example, prediction processing of future values ​​of RSSI values).

[0131] <Summary> As described above, in the wireless situation prediction device 100, when observation data is intermittently acquired, data interpolation processing is performed so that a substantially identical waveform pattern (waveform pattern after interpolation processing) is acquired when data interpolation processing is performed on the same waveform pattern, regardless of where the acquired portion of the observation data (part of missing data) is located. That is, in the wireless situation prediction device 100, as described above, a Hankel matrix R is generated from the observation data, and optimization processing is performed using the Hankel matrix R to acquire an optimization matrix R_opt acquired. The Hankel matrix R generated by the wireless situation prediction device 100 is a matrix acquired by stacking while shifting samples one by one in the data holding unit 11 of the data interpolation processing unit 1. In wireless applications such as AGV, since a wireless terminal (for example, a wireless terminal mounted on an AGV) repeatedly moves along a fixed route, the communication quality information to be interpolated also fluctuates periodically. For this reason, the Hankel matrix generated by the wireless situation prediction device 100 under such circumstances is ranked low. Therefore, by subjecting the Hankel matrix to singular value decomposition, the number of non-zero singular values ​​is reduced (the number of non-zero singular values ​​is approximately the same as the effective rank number of the Hankel matrix R), and the matrix can be decomposed into low-rank matrices P and Q. Then, by performing optimization processing using the matrices P and Q obtained by decomposing the Hankel matrix R, it is possible to perform high-precision optimization processing while suppressing the amount of calculation.

[0132] Then, in the radio situation predicting device 100, the optimized matrix R_opt (a matrix obtained by performing an optimization process on a Hankel matrix) obtained as described above can be used to highly accurately interpolate data of missing parts.

[0133] As a result, the radio situation prediction device 100 can acquire an appropriate waveform pattern (waveform pattern after interpolation processing), and by using the waveform pattern to perform learning processing of a PNN model for future value prediction, etc., which is mounted in the prediction processing unit 2, it is possible to perform highly accurate and efficient learning processing (learning processing of a PNN model for future value prediction, etc.).

[0134] Furthermore, the radio situation prediction device 100 can perform highly accurate prediction processing by performing prediction processing using a trained model (e.g., a PNN model for predicting future values) acquired by the learning processing on the appropriate waveform pattern acquired by the data interpolation processing.

[0135] As an example, Fig. 14 shows data acquired by interpolating RSSI values ​​(intermittently acquired data) of wireless signals acquired at access points installed at predetermined positions in an environment where an AGV controlled by wireless communication travels a predetermined route in a warehouse. In Fig. 14, the horizontal axis is the sample number (time-series data number (corresponding to time)), and the vertical axis is the RSSI value.

[0136] As can be seen from FIG. 14, even when observation data is acquired intermittently, the accuracy (interpolation accuracy) of the data acquired by interpolation using data interpolation processing (interpolation processing executed by the wireless situation prediction device 100) is higher than data acquired by interpolation using the average value of the observation data or linear interpolation.

[0137] FIG. 15 shows, as an example, prediction result data when a prediction process (prediction of RSSI value 10 seconds ahead) is performed using a PNN model using data acquired by interpolating RSSI values ​​(intermittently acquired data) of wireless signals acquired at an access point installed at a predetermined position in an environment where an AGV controlled by wireless communication travels along a predetermined route within a warehouse.

[0138] As can be seen from FIG. 15, even when observation data is acquired intermittently, the accuracy (interpolation accuracy) of prediction processing using data acquired by interpolation through data interpolation processing (interpolation processing executed by the wireless situation prediction device 100) is higher than the accuracy of prediction processing using data acquired by interpolation based on the average value of the observation data or by linear interpolation.

[0139] In this way, by performing processing using the interpolated data obtained by executing data interpolation processing in the wireless situation predicting device 100, the prediction processing can be performed with high accuracy even when observation data is only obtained intermittently.

[0140] [Other embodiments] In the above embodiment, the case where the Hankel matrix R is subjected to singular value decomposition when performing the data interpolation process in the data interpolation processing unit 1 has been described, but the present invention is not limited to this, and matrix decomposition may be performed using other methods as long as they have the characteristics of compressing the redundancy of the Hankel matrix R (methods that can effectively utilize the low rank of the Hankel matrix R). For example, the Hankel matrix R may be decomposed using a method such as eigenvalue decomposition.

[0141] In the above embodiment, the second term of the loss function Loss(P,Q) is an L2 regularization term, but this is not limited thereto. The second term of the loss function Loss(P,Q) may be an L1 regularization term, or may be another term that can appropriately prevent overlearning.

[0142] In addition, in the above embodiment, the case where the values ​​of the elements of the Hankel matrix R in the portion where there is no observed data are set to the value “0” and processing is performed has been described. However, this is not limited to this, and the values ​​of the elements of the Hankel matrix R in the portion where there is no observed data may be set to other values ​​(for example, values ​​using random numbers) and used as the initial values ​​of the Hankel matrix R to perform optimization processing.

[0143] Furthermore, in the radio situation predicting device 100 described in the above embodiment, each block may be individually implemented as a single chip using a semiconductor device such as an LSI, or some or all of the blocks may be included in a single chip.

[0144] Although we refer to it as an LSI here, it may also be called an IC, system LSI, super LSI, or ultra LSI depending on the level of integration.

[0145] The method of integration is not limited to LSI, but may be realized by a dedicated circuit or a general-purpose processor. It is also possible to use a field programmable gate array (FPGA) that can be programmed after LSI manufacturing, or a reconfigurable processor that can reconfigure the connections and settings of circuit cells inside the LSI.

[0146] In addition, a part or all of the processing of each functional block in each of the above embodiments may be realized by a program. Then, a part or all of the processing of each functional block in each of the above embodiments is performed by a central processing unit (CPU) in a computer. Also, the programs for performing each processing are stored in a storage device such as a hard disk or a ROM, and are executed in the ROM or by being read into the RAM.

[0147] In addition, each process in the above embodiment may be realized by hardware, or may be realized by software (including the case where it is realized together with an OS (operating system), middleware, or a predetermined library). Furthermore, it may be realized by a mixed process of software and hardware.

[0148] Furthermore, for example, when each functional unit of the above embodiment is realized by software, each functional unit may be realized by software processing using the hardware configuration shown in FIG. 16 (for example, a hardware configuration in which a CPU (which may be a GPU), ROM, RAM, input units, output units, etc. are connected via a bus).

[0149] Furthermore, when each functional unit of the above embodiment is realized by software, the software may be realized by a single computer having the hardware configuration shown in FIG. 16, or may be realized by distributed processing using multiple computers.

[0150] Furthermore, the order of execution of the processing methods in the above embodiments is not necessarily limited to that described in the above embodiments, and the order of execution can be changed without departing from the gist of the invention.

[0151] The scope of the present invention includes a computer program for causing a computer to execute the above-mentioned method and a computer-readable recording medium having the program recorded thereon. Here, examples of the computer-readable recording medium include a flexible disk, a hard disk, a CD-ROM, an MO, a DVD, a DVD-ROM, a DVD-RAM, a large-capacity DVD, a next-generation DVD, and a semiconductor memory.

[0152] The computer program is not limited to one recorded on the recording medium, but may be one transmitted via a telecommunications line, a wireless or wired communication line, a network such as the Internet, or the like.

[0153] The specific configuration of the present invention is not limited to the above-described embodiment, and various changes and modifications are possible without departing from the gist of the invention. [Explanation of symbols]

[0154] 100 Radio situation prediction device FE1 Observation data acquisition section 1 Data Interpolation Processing Unit 11 Data storage unit 12 Matrix generator 13 Matrix decomposition processing section 14 Optimization Processing Section 15 Interpolation data acquisition section 2 Prediction processing unit (prediction processing device, learning processing device)

Claims

1. A data interpolation method for performing an interpolation process on intermittently acquired observation data, the method comprising: an observation data acquisition step of acquiring observation data on communication quality from a wireless signal using the computer; a data storage step of storing and holding the observation data acquired by the observation data acquisition step for a predetermined period of time using the computer; a matrix generation step of generating a Hankel matrix R using the computer such that a group of data included in a period corresponding to a predetermined unit period becomes elements of one row when the window corresponding to the window is shifted in the time direction by a first hour, which is a predetermined period, and the matrix generation step of generating the Hankel matrix R by (1) setting the data corresponding to the element of the Hankel matrix R to the observation data when the data corresponding to the element is the observation data stored and held in the data holding step, and (2) setting the data corresponding to the element of the Hankel matrix R to a value indicating that the data is missing data when the data corresponding to the element is not stored and held in the data holding step; a matrix decomposition processing step of performing matrix decomposition on the Hankel matrix R generated by the matrix generation step, using the computer, to obtain a decomposition matrix; an optimization processing step of performing optimization processing using the computer based on element values ​​obtained by multiplying the matrices decomposed in the matrix decomposition processing step and element values ​​of the Hankel matrix to obtain an optimization matrix; an interpolation data processing step of interpolating data for a portion where the observation data is missing based on the optimization matrix using the computer; A data interpolation processing method comprising:

2. The matrix decomposition processing step includes: The decomposition matrix is ​​obtained by performing singular value decomposition on the Hankel matrix R.

2. The data interpolation processing method according to claim 1.

3. The optimization process step includes: Among the elements of the Hankel matrix R, an element R in which the observation data exists is ij (i, j: natural number, R ij : a first extraction process for extracting only the element in the i-th row and j-th column of the Hankel matrix R; Element A, which is an element of the matrix product decomposed by the matrix decomposition processing step corresponding to the element ij (A ij : executing a second extraction process of extracting only the element in row i and column j of the matrix product decomposed by the matrix decomposition process step; The extracted element R ij and the element A ij The optimization process is performed so as to reduce the error between the 3. The data interpolation processing method according to claim 1.

4. The optimization process step includes: performing a matrix operation on the product of the Hankel matrix R and / or the matrix decomposed in the matrix decomposition processing step using a selection matrix that selects a predetermined element, thereby executing the first extraction processing and / or the second extraction processing; 4. The data interpolation processing method according to claim 3.

5. The interpolation data processing step includes: obtaining an average value or a weighted average value of elements corresponding to a portion where the observation data is missing and arranged in an anti-diagonal direction of the optimization matrix, and interpolating data for the portion where the observation data is missing based on the obtained average value or weighted average value; 5. A data interpolation processing method according to claim 1.

6. A data interpolation processing device for performing an interpolation process on intermittently acquired observation data, comprising: an observation data acquisition unit that acquires observation data on communication quality from a wireless signal; a data storage unit that stores and holds the observation data acquired by the observation data acquisition unit for a predetermined period of time; a matrix generation unit that generates a Hankel matrix R such that a group of data included in a period corresponding to a predetermined unit period becomes elements of one row when a window corresponding to the predetermined unit period is shifted in a time direction by a first hour, which is a predetermined period, the matrix generation unit generating the Hankel matrix R by: (1) setting data corresponding to an element of the Hankel matrix R to the observation data stored in the data storage unit when the data corresponding to the element is the observation data stored in the data storage unit; and (2) setting data corresponding to the element of the Hankel matrix R to a value indicating that the element is missing data when the data corresponding to the element is not stored in the data storage unit, a matrix decomposition processing unit that performs matrix decomposition on the Hankel matrix R generated by the matrix generation unit to obtain a decomposition matrix; an optimization processing unit that performs optimization processing based on element values ​​obtained by multiplying the matrices decomposed by the matrix decomposition processing unit and element values ​​of the Hankel matrix to obtain an optimization matrix; an interpolation data processing unit that interpolates data for a portion where the observation data is missing based on the optimization matrix; A data interpolation processing device comprising:

7. 6. A program for causing a computer to execute the data interpolation processing method according to claim 1.

8. A learning processing device that performs learning processing of a model for predicting future values ​​of observed data, using the interpolated data acquired by the data interpolation processing method according to any one of claims 1 to 5.

9. A trained model is installed, the trained model being acquired by performing a training process for a model that predicts future values ​​of observed data using the interpolated data acquired by the data interpolation processing method according to any one of claims 1 to 5; A prediction processing device that predicts future values ​​of the observation data by inputting interpolated data acquired by the data interpolation processing method according to any one of claims 1 to 5 into the trained model.

Citation Information

Patent Citations

  • Time series data prediction method and device, time series data prediction program, and recording medium with the same

    JP2005018304A

  • System and method for interpolating seismic data

    US20140288838A1