Prediction device, prediction method, and program
The prediction device accurately predicts time series data by calculating similarity matrices and searching for similar data, addressing the limitations of conventional methods in waveform consideration and model obsolescence.
Patent Information
- Application Number
- JP2024080054
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2024-05-16
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-05-16
AI Technical Summary
Conventional technologies face challenges in accurately predicting time series data, particularly due to the obsolescence of prediction models and the inability to consider waveform characteristics.
A prediction device that calculates similarity matrices for time series data, searches for similar data based on feature data, and predicts time series data of a predetermined objective variable using these similarities.
Enables high-accuracy prediction of time series data by considering waveform characteristics and avoiding model obsolescence.
Smart Images

Figure 0007586362000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to a prediction device, a prediction method, and a program. [Background technology]
[0002] A technique for constructing a prediction model using machine learning is known. For example, Patent Document 1 discloses an apparatus that divides points indicating actual measured values into multiple clusters, optimizes parameters of a prediction model for each of the multiple clusters, evaluates the difference between the predicted value output by the prediction model with optimized parameters and the actual measured value for each number of clusters, and determines the number of clusters based on the evaluation results. Patent Document 2 discloses a method for searching for similar past cases of the state of a process from a process time series database when predicting the future state of a process. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent No. 6637206 [Patent Document 2] Patent No. 5176206 Summary of the Invention [Problem to be solved by the invention]
[0004] However, with conventional techniques, it is difficult to predict time series data with high accuracy.
[0005] The present disclosure provides a technique capable of predicting time-series data with high accuracy. [Means for solving the problem]
[0006] A prediction device according to one embodiment of the present disclosure includes an acquisition unit that acquires multiple time series data including one or more explanatory variables, a matrix calculation unit that calculates, for each explanatory variable, a matrix whose elements represent the similarity between two of the multiple time series data, a feature calculation unit that calculates feature data for each time series data based on the matrix for each explanatory variable, a search unit that searches for similar data similar to the time series data to be predicted among the multiple time series data based on the feature data for each time series data, and a prediction unit that predicts time series data of a predetermined response variable based on the similar data. Effect of the Invention
[0007] According to one aspect of the present disclosure, time series data can be predicted with high accuracy. [Brief description of the drawings]
[0008] [Figure 1] 1 is a block diagram showing an example of an overall configuration of a prediction system. [Diagram 2] FIG. 2 is a block diagram showing an example of a hardware configuration of a computer. [Diagram 3] 2 is a block diagram showing an example of a functional configuration of the prediction system. FIG. [Figure 4] 1 is a flowchart illustrating an example of a prediction method. [Diagram 5] FIG. 11 is a diagram illustrating an example of batch process data. [Figure 6] FIG. 2 illustrates an example of a database. [Figure 7] FIG. 13 is a diagram illustrating an example of variable importance. [Figure 8] FIG. 13 is a diagram illustrating an example of a distance matrix. [Figure 9] FIG. 13 is a diagram illustrating an example of an eigenvector. [Figure 10] FIG. 13 is a diagram illustrating an example of a neighborhood search. [Figure 11] FIG. 11 is a diagram illustrating an example of prediction data. [Figure 12] FIG. 4 is a diagram showing an example of a display screen. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configurations are denoted by the same reference numerals, and redundant description will be omitted.
[0010] [Embodiment] One embodiment of the present disclosure is a prediction system that predicts a predetermined objective variable related to a prediction target. The prediction target according to this embodiment executes a predetermined batch process. As an example, if the prediction target is a steel or chemical plant, the batch process corresponds to one operation. The prediction target may execute a continuous process that can be interpreted as a batch process. For example, the demand for electricity or heat, the price of stock prices or electricity prices, or the inflow of a river or dam are continuous process data, but if they are divided into a predetermined time interval (for example, one day, one week, or one month), they can be interpreted as a batch process.
[0011] The objective variable may be any variable that can be obtained from the prediction target. In the present embodiment, as an example, the objective variable is the temperature measured by a temperature sensor installed in the plant. However, the objective variable is not limited to the temperature of the plant, and various variables such as the amount of material, pressure, power, or heat may be used as the objective variable depending on the application. In addition, the objective variable does not have to be a measurable variable, and may be a variable that can be obtained, calculated, or predicted by a predetermined information processing, for example.
[0012] For example, a technique is known in which data measured by a sensor installed in a factory or a plant is collected and a predictive model is constructed using, for example, machine learning. Predictions using machine learning are attracting attention in various situations in which the targets of prediction are, for example, the demand for heat or electricity, prices such as stock prices or electricity prices, or the amount of inflow into a river or a dam. In a factory or a plant, future values are predicted in order to determine in advance what operations should be performed.
[0013] However, it is difficult to predict time series data with high accuracy using conventional techniques. For example, in Patent Document 1, prediction is made using a pre-constructed prediction model, so the prediction model becomes outdated as the system is used, and the prediction accuracy decreases. In Patent Document 2, in order to predict variables at a single point in time, similar process data is searched for and a prediction model is constructed for each point in time included in the time period to be predicted, so prediction cannot be made taking into account the characteristics of the waveform.
[0014] The present embodiment aims to accurately predict time series data of a predetermined objective variable. To this end, the present embodiment calculates a matrix with elements representing the similarity between two time series data including one or more explanatory variables for each explanatory variable, searches for similar data similar to the time series data to be predicted based on feature data calculated based on the matrix for each explanatory variable, and predicts the time series data of the predetermined objective variable based on the similar data.
[0015] In one aspect, according to the present embodiment, time series data similar to the data to be predicted is searched for sequentially, so that a predetermined objective variable can be predicted with high accuracy. In another aspect, according to the present embodiment, time series data similar to the data to be predicted is searched for based on the similarity between the time series data, so that the time series data can be predicted while taking into account the characteristics of the waveform. Therefore, according to the present embodiment, time series data of a predetermined objective variable can be predicted with high accuracy.
[0016] <Overall composition> The overall configuration of the prediction system in this embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the overall configuration of the prediction system.
[0017] 1, the prediction system 1000 includes a prediction target 10, a prediction device 20, and one or more terminal devices 30. The prediction target 10, the prediction device 20, and the terminal devices 30 are connected to each other so as to be able to communicate data with each other via a communication network N such as a LAN (Local Area Network) or the Internet.
[0018] The prediction target 10 is an example of an object that executes a batch process. The prediction target 10 may be, for example, a factory or a plant, or equipment, devices, or the like installed therein. The prediction target 10 may include a plurality of factories, plants, equipment, devices, or devices.
[0019] The prediction target 10 generates batch process data during execution of a batch process. The batch process data is time series data indicating a time change of a variable indicating a state of the prediction target 10. The batch process data may include a plurality of variables acquired in one prediction target 10. The batch process data may include a plurality of variables acquired in each of a plurality of prediction targets 10.
[0020] The prediction device 20 is an example of an information processing device such as a personal computer, a workstation, or a server that predicts time-series data of a predetermined objective variable. The prediction device 20 acquires batch process data from a prediction target 10. The prediction device 20 predicts the time-series data of the objective variable based on the batch process data, and transmits the prediction result to a terminal device 30.
[0021] The terminal device 30 is an example of an information processing terminal such as a personal computer, a tablet terminal, or a smartphone operated by a user of the prediction system 1000. The terminal device 30 requests the prediction device 20 to predict a response variable in response to a user's operation. The terminal device 30 receives a prediction result from the prediction device 20 and outputs it to the user.
[0022] The overall configuration of the prediction system 1000 shown in FIG. 1 is an example, and various system configuration examples are possible depending on the application and purpose. For example, the prediction system 1000 may include a plurality of prediction target 10, prediction device 20, and terminal device 30. For example, the prediction device 20 may be realized by a plurality of computers, or may be realized as a cloud computing service. The classification of devices such as the prediction target 10, prediction device 20, and terminal device 30 shown in FIG. 1 is an example.
[0023] <Hardware configuration> The hardware configuration of each device included in the prediction system 1000 will be described with reference to Fig. 2. The prediction device 20 and the terminal device 30 included in the prediction system 1000 are realized by, for example, a computer. Fig. 2 is a block diagram showing an example of the hardware configuration of a computer.
[0024] 2, the computer 500 includes a CPU (Central Processing Unit) 501, a ROM (Read Only Memory) 502, a RAM (Random Access Memory) 503, a HDD (Hard Disk Drive) 504, an input device 505, a display device 506, a communication I / F (Interface) 507, and an external I / F 508. The CPU 501, the ROM 502, and the RAM 503 form a so-called computer. Each piece of hardware of the computer 500 is connected to each other via a bus line 509. The input device 505 and the display device 506 may be connected to the external I / F 508 for use.
[0025] The CPU 501 is a calculation device that reads out programs and data from a storage device such as the ROM 502 or the HDD 504 onto the RAM 503 and executes processing to realize the overall control and functions of the computer 500. The computer 500 may have a GPU (Graphics Processing Unit) in addition to or instead of the CPU 501.
[0026] The ROM 502 is an example of a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The ROM 502 functions as a main storage device that stores various programs and data required for the CPU 501 to execute various programs installed in the HDD 504. Specifically, the ROM 502 stores boot programs such as a Basic Input / Output System (BIOS) and an Extensible Firmware Interface (EFI) that are executed when the computer 500 is started, and data such as an Operating System (OS) setting and a network setting.
[0027] The RAM 503 is an example of a volatile semiconductor memory (storage device) in which programs and data are erased when the power is turned off. The RAM 503 is, for example, a dynamic random access memory (DRAM) or a static random access memory (SRAM). The RAM 503 provides a working area in which various programs installed in the HDD 504 are expanded when the CPU 501 executes them.
[0028] The HDD 504 is an example of a non-volatile storage device that stores programs and data. The programs and data stored in the HDD 504 include an OS, which is basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that the computer 500 may use a storage device that uses a flash memory as a storage medium (e.g., an SSD (Solid State Drive) or the like) instead of the HDD 504.
[0029] The input device 505 includes a touch panel, operation keys or buttons, a keyboard or mouse, a microphone for inputting sound data such as voice, etc., which are used by the user to input various signals.
[0030] The display device 506 is composed of a display such as a liquid crystal display or an organic EL (Electro-Luminescence) display for displaying a screen, a speaker for outputting sound data such as voice, and the like.
[0031] The communication I / F 507 is an interface that connects to a communication network and enables the computer 500 to perform data communication.
[0032] The external I / F 508 is an interface with external devices, such as a drive device 510.
[0033] The drive device 510 is a device for setting the recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, or a magneto-optical disk. The recording medium 511 may also include semiconductor memories that record information electrically, such as a ROM or a flash memory. This allows the computer 500 to read and / or write data from and to the recording medium 511 via the external I / F 508.
[0034] The various programs to be installed in the HDD 504 are installed, for example, by setting the distributed recording medium 511 in a drive device 510 connected to the external I / F 508 and reading out the various programs recorded in the recording medium 511 by the drive device 510. Alternatively, the various programs to be installed in the HDD 504 may be installed by being downloaded from a network different from the communication network via the communication I / F 507.
[0035] <Functional configuration> The functional configuration of the prediction system 1000 will be described with reference to Fig. 3. Fig. 3 is a block diagram showing an example of the functional configuration of the prediction system.
[0036] As shown in FIG. 3, the prediction device 20 includes an acquisition unit 110, a data storage unit 120, a variable selection unit 130, a matrix calculation unit 140, a feature calculation unit 150, a search unit 160, a prediction unit 170, and a visualization unit 180.
[0037] The acquisition unit 110, the variable selection unit 130, the matrix calculation unit 140, the feature calculation unit 150, the search unit 160, the prediction unit 170 and the visualization unit 180 are realized by processing executed by the CPU 501 of a program expanded from the HDD 504 shown in FIG. 2 onto the RAM 503.
[0038] The data storage unit 120 is realized by the HDD 504 shown in FIG.
[0039] The acquiring unit 110 acquires batch process data from the prediction target 10. The acquiring unit 110 may receive batch process data output by the prediction target 10 from the prediction target 10. The acquiring unit 110 may receive batch process data input to the terminal device 30 from the terminal device 30. The acquiring unit 110 may accept batch process data input to the input device 505 of the prediction device 20.
[0040] The data storage unit 120 stores the batch process data acquired by the acquisition unit 110. The data storage unit 120 may also store other data used in processing by the prediction device 20, data in the middle of processing, and the like.
[0041] Previously acquired batch process data may be accumulated in the data storage unit 120. The batch process data stored in the data storage unit 120 may be deleted when a predetermined time has elapsed. Here, it is assumed that m pieces of batch process data are stored in the data storage unit 120, where m is an integer of 2 or more.
[0042] The variable selection unit 130 selects one or more explanatory variables capable of appropriately predicting a predetermined objective variable from a plurality of variables included in the batch process data read from the data storage unit 120. The variable selection unit 130 may select a predetermined variable as the explanatory variable. The variable selection unit 130 may select the explanatory variable based on the importance of each of the plurality of variables. As an example, the importance may be a variable importance that can be calculated from a decision tree. The variable importance may be weighted based on the correlation with the objective variable.
[0043] The matrix calculation unit 140 calculates a matrix whose elements are the similarity between two batch process data among the batch process data read from the data storage unit 120, for each explanatory variable selected by the variable selection unit 130. Hereinafter, the matrix calculated by the matrix calculation unit 140 will be referred to as a "distance matrix."
[0044] The matrix calculation unit 140 may calculate the similarity based on the waveform of the batch process data. For example, the matrix calculation unit 140 may calculate the similarity using Euclidean distance, cross-correlation function, shapelet, or dynamic time warping. For another example, the matrix calculation unit 140 may identify the waveform diagram of each batch process data by image recognition technology and calculate the similarity. The similarity may use any index as long as it is a calculation that can be defined as a distance. The similarity may be a conversion of a predetermined distance as long as it follows the axiom of distance. The similarity may be, for example, a correlation coefficient or the like.
[0045] The feature calculation unit 150 calculates feature data for each batch process data based on the distance matrix calculated by the matrix calculation unit 140. The feature calculation unit 150 may calculate feature data by compressing the dimensions of the distance matrix using eigenvalue decomposition. Eigenvalue decomposition is a method of finding eigenvalues from an eigenvalue equation for a matrix to be analyzed and finding eigenvectors corresponding to the eigenvalues. An eigenvector is an example of feature data. The number of dimensions of the eigenvector is 1 to m.
[0046] The search unit 160 searches for batch process data similar to the prediction target data (hereinafter also referred to as "similar data") among the batch process data read from the data storage unit 120 based on the feature data calculated by the feature calculation unit 150. The similar data includes one or more batch process data corresponding to feature data similar to the feature data of the prediction target data.
[0047] The prediction unit 170 predicts time series data of the objective variable (hereinafter, also referred to as "predicted data") based on the similar data searched for by the search unit 160. The prediction unit 170 may calculate the predicted data based on the time series data of the objective variable included in the similar data.
[0048] The prediction unit 170 may calculate the predicted data by extracting the time series data of the objective variable from the similar data and calculating the average value of the time series data of the objective variable at each time point. The prediction unit 170 may construct a statistical model or a machine learning model based on the similar data and infer the predicted data based on the model. If there is one similar data, the prediction unit 170 may output the time series data of the objective variable included in the similar data as the predicted data.
[0049] The visualization unit 180 outputs the prediction result by the prediction unit 170. The visualization unit 180 may cause a display screen for visualizing the prediction result to be displayed on the display device 506 of the terminal device 30. The visualization unit 180 may also display a display screen for visualizing the prediction result on the display device 506 of the prediction device 20.
[0050] The prediction result includes the prediction data calculated by the prediction unit 170. The prediction result may include information used in predicting the prediction data. The information used in the prediction may include, for example, the importance and weight calculated by the variable selection unit 130, the distance matrix calculated by the matrix calculation unit 140, the feature data calculated by the feature calculation unit 150, and similar data searched for by the search unit 160.
[0051] <Processing Procedure> The processing procedure of the prediction method executed by the prediction system 1000 will be described with reference to Fig. 4. Fig. 4 is a flowchart showing an example of the prediction method.
[0052] In step S1, the prediction target 10 executes a batch process. The prediction target 10 generates batch process data during the execution of the batch process. The prediction target 10 transmits the generated batch process data to the prediction device 20.
[0053] The batch process data may include operation data or sensor data of the prediction target 10. The operation data may include normal or abnormal ranges related to a factory, plant, equipment, device, or apparatus, or event information such as operation or equipment inspection. The sensor data may include sensor values measured by various sensors provided in the prediction target 10.
[0054] FIG. 5 is a diagram showing an example of batch process data. FIG. 5 shows batch process data generated by two batch processes A and B that were executed at different times. The batch process data shown in FIG. 5 includes n+1 variables (Y, X1,...,Xn), where n is a natural number. Each of the variables Y, X1,...,Xn is a variable that changes over time, and is time-series data that can be expressed in a graph with the horizontal axis representing time and the vertical axis representing the variable value. Note that the variable Y is a response variable, and the variables X1,...,Xn are variables other than the response variable.
[0055] In this embodiment, the objective variable Y may be temperature. The variables X1,...,Xn may include humidity, carbon dioxide concentration 1, carbon dioxide concentration 2, oxygen concentration 1, oxygen concentration 2, the amount of material A, the amount of material B, pressure 1, pressure 2, and outside air temperature. The carbon dioxide concentration, oxygen concentration, amount of material, or pressure may include multiple variables depending on the number of measurement points, etc.
[0056] 4, in step S2, the acquisition unit 110 of the prediction device 20 receives batch process data from the prediction target 10. The acquisition unit 110 stores the received batch process data in the data storage unit 120.
[0057] A database for storing the batch process data may be constructed in the data storage unit 120. The acquisition unit 110 may store the batch process data in the database in association with identification information for identifying the batch process. The identification information may be time information indicating the time point when the batch process is executed.
[0058] Fig. 6 is a diagram showing an example of a database. As shown in Fig. 6, the database includes, as data items, a label DATE, a response variable Y, and n explanatory variables X1,...,Xn. The label DATE is an example of identification information. The response variable Y and the explanatory variables X1,...,Xn are batch process data.
[0059] In this embodiment, one day's operation is treated as a batch process, and sensor values measured at multiple points in time on the same day are treated as one batch process data. In the example shown in Fig. 6, 10 records with labels 20240101-1 to 20240101-10 correspond to one batch process data. The batch process data corresponding to the batch process with the latest date is the batch process data to be predicted (hereinafter, also referred to as "prediction target data").
[0060] The prediction target data may be batch process data related to a currently running batch process or may be batch process data related to a batch process to be executed in the future. The explanatory variables included in the batch process data related to a batch process to be executed in the future may be values predicted based on a trained model.
[0061] Returning to Fig. 4, in step S3, the user of the terminal device 30 performs an operation for requesting a prediction of the dependent variable. In response to the user's operation, the terminal device 30 transmits a prediction request for the dependent variable to the prediction device 20. The prediction device 20 receives the prediction request from the terminal device 30.
[0062] The variable selection unit 130 of the prediction device 20 reads out batch process data from the data storage unit 120 in response to a prediction request from the terminal device 30. The variable selection unit 130 selects one or more explanatory variables from a plurality of variables included in the batch process data. In this embodiment, the variable selection unit 130 selects one or more explanatory variables based on the variable importance calculated by the decision tree. The variable selection unit 130 sends the selected explanatory variables to the matrix calculation unit 140.
[0063] Fig. 7 is a diagram showing an example of variable importance. As shown in Fig. 7, the variable importance calculated by the decision tree differs depending on the explanatory variables X1,...,Xn. The variable selection unit 130 may select variables whose variable importance is equal to or greater than a predetermined threshold as explanatory variables, or may select a predetermined number of variables in descending order of variable importance as explanatory variables.
[0064] 6, the batch process data is time-series data including variables acquired at multiple time points. The variable selection unit 130 may calculate the variable importance for each time point and integrate the variable importance for each time point. In this embodiment, it is assumed that, as a result of integrating the variable importance based on the decision tree, it is found that the variable importance of carbon dioxide concentration 1, oxygen concentration 2, the amount of material A, the amount of material B, and the outside air temperature is high with respect to the target variable temperature.
[0065] Returning to FIG. 4, in step S4, the matrix calculation unit 140 of the prediction device 20 receives one or more explanatory variables from the variable selection unit 130. The matrix calculation unit 140 reads out m pieces of batch process data from the data storage unit 120. The matrix calculation unit 140 calculates, for each explanatory variable, a matrix whose elements are the similarity between two pieces of batch process data among the m pieces of batch process data. The matrix calculation unit 140 sends the distance matrix for each explanatory variable to the feature calculation unit 150.
[0066] In this embodiment, the matrix calculation unit 140 calculates the similarity for all combinations of two batch process data out of the m batch process data by using the dynamic time warping method. The matrix calculation unit 140 generates a distance matrix whose elements are the similarities calculated for all combinations for each explanatory variable.
[0067] FIG. 8 is a diagram showing an example of a distance matrix. As shown in FIG. 8, the distance matrix includes a matrix for each explanatory variable, whose elements are distances between batch process data. The distance matrix is an m×m matrix whose vertical and horizontal axes correspond to m batch process data. Each element of the distance matrix indicates the distance between the batch process data on the horizontal axis and the batch process data on the vertical axis. Note that the diagonal elements of the distance matrix are combinations of the same batch process data, so the distance is 0.
[0068] Returning to FIG. 4, in step S5, the feature calculation unit 150 of the prediction device 20 receives a distance matrix for each explanatory variable from the matrix calculation unit 140. The feature calculation unit 150 calculates feature data for each batch process data based on the distance matrix for each explanatory variable. The feature calculation unit 150 sends the feature data for each batch process data to the search unit 160.
[0069] In this embodiment, the feature calculation unit 150 calculates the eigenvector based on the multidimensional scaling method. The multidimensional scaling method is a method that uses eigenvalue decomposition when the matrix based on the similarity is a distance matrix. The multidimensional scaling method is a method that, when the distance between any two pieces of data among a plurality of pieces of data is given, calculates the coordinates of each piece of data so that two pieces of data that are close to each other are placed close to each other and two pieces of data that are far from each other are placed far from each other. The distance between the coordinates calculated by the multidimensional scaling method corresponds to the length of the relative distance between the m pieces of batch process data.
[0070] 9 is a diagram showing an example of an eigenvector. As shown in FIG. 9, the eigenvector is a label DATE for identifying a batch process and two variables X b 1,X b Contains 2. Variable X b 1,X b 2 is an example of feature data. The label DATE corresponds to the label DATE of the batch process data. Therefore, the eigenvector of the latest date corresponds to the data to be predicted. The eigenvector is b 1,X b It can be interpreted as a two-dimensional space with X and Y coordinates of 2. The two-dimensional space is an example of a feature space. The number of dimensions of the feature space is the same as the number of dimensions of the eigenvector.
[0071] Returning to Fig. 4, in step S6, the search unit 160 of the prediction device 20 receives feature data for each batch process data from the feature calculation unit 150. The search unit 160 searches for batch process data similar to the data to be predicted based on the feature data. The search unit 160 sends the searched batch process data to the prediction unit 170 as similar data.
[0072] Specifically, the search unit 160 arranges feature data for each batch process data in a feature space. The search unit 160 searches for feature data that is close to the feature data of the prediction target data in the feature space. The search unit 160 may search for feature data whose distance to the feature data of the prediction target data is equal to or less than a predetermined threshold. The search unit 160 may search for a predetermined number of feature data in order of shortest distance to the feature data of the prediction target data.
[0073] The search unit 160 reads out batch process data corresponding to the searched feature data from the data storage unit 120. The search unit 160 searches for feature data that has been dimensionally compressed by eigenvalue decomposition, but since it is not possible to predict a target variable from the feature data, it is necessary to acquire batch process data corresponding to the feature data. As shown in FIG. 9, the feature data is associated with identification information (label) that is common to the batch process data. The search unit 160 can acquire batch process data similar to the prediction target data by reading out from the data storage unit 120 batch process data associated with the same identification information as the searched feature data.
[0074] The search unit 160 may search for similar data by a neighborhood search based on the distance between feature data. The distance used in the neighborhood search may be, for example, the Euclidean distance or the Mahalanobis distance. The search unit 160 may divide the feature data arranged in the feature space into clusters, and treat the cluster including the feature data of the prediction target data as similar data.
[0075] Fig. 10 is a diagram showing an example of a neighborhood search. As shown in Fig. 10, feature data corresponding to batch process data is arranged in a feature space FS, which is a two-dimensional space. Fig. 10 shows how four feature data D2 to D5 are searched for by performing a neighborhood search from prediction target data D1.
[0076] 4, in step S7, the prediction unit 170 of the prediction device 20 receives the similar data from the search unit 160. The prediction unit 170 calculates predicted data based on the similar data. The prediction unit 170 sends the predicted data to the visualization unit 180.
[0077] In this embodiment, the prediction unit 170 extracts time series data of the objective variables from multiple batch process data included in the similar data, and calculates the predicted data by averaging the extracted time series data of the objective variables. Specifically, the prediction unit 170 calculates the average value of the objective variables included in the batch process data at each time point of the batch process data, and generates the predicted data by arranging the average values of the objective variables in chronological order.
[0078] Fig. 11 is a diagram showing an example of predicted data. Fig. 11 shows how, when four similar data are searched for, the temperatures contained in the four similar data are averaged at each time point, and the time series data of the average temperature is output as predicted data.
[0079] Returning to FIG. 4 , the explanation will be given. In step S8, the visualization unit 180 of the prediction device 20 receives the prediction data from the prediction unit 170. The visualization unit 180 generates a display screen for visualizing the prediction result including the prediction data. The visualization unit 180 may include a waveform diagram of the prediction data as the prediction result on the display screen. The visualization unit 180 transmits screen data for displaying the display screen to the terminal device 30.
[0080] The visualization unit 180 may include information used in the prediction on the display screen together with the prediction result. For example, the visualization unit 180 may acquire importance and weight from the variable selection unit 130 and include a graph showing the importance and weight on the display screen. For example, the visualization unit 180 may acquire a distance matrix from the matrix calculation unit 140 and include a table showing the distance matrix on the display screen. For example, the visualization unit 180 may acquire a feature space in which feature data is arranged from the feature calculation unit 150 and include a scatter diagram of the feature data on the display screen. For example, the visualization unit 180 may acquire similar data from the search unit 160 and include a waveform diagram of the similar data on the display screen. When a large number of batch process data are included in the similar data, only a predetermined number of batch process data may be acquired in order of proximity to the prediction target data.
[0081] The terminal device 30 receives screen data for a display screen from the prediction device 20. The terminal device 30 displays the display screen on the display device 506 based on the screen data.
[0082] Fig. 12 is a diagram showing an example of a display screen. As shown in Fig. 12, a display screen 600 displays a variable importance 610 of an explanatory variable, a scatter plot 620 of feature data, a waveform diagram 630 of similar data, and a waveform diagram 640 of predicted data. The display screen 600 may further display any information used by the prediction device 20 for prediction.
[0083] By visualizing the variable importance 610 of the explanatory variables, the reason for using the explanatory variables can be clearly shown to the user. By visualizing the scatter plot 620 of the feature data and the waveform diagram 630 of the similar data, the basis of the prediction results can be explained.
[0084] The user of the terminal device 30 may consider the operation of the prediction target 10 based on the prediction result displayed on the display screen 600. For example, the user of the terminal device 30 may perform an operation related to temperature control by referring to the temperature prediction result. In addition, the user of the terminal device 30 can easily determine the validity of the prediction result by referring to the information used for the prediction.
[0085] <Effects of the embodiment> The prediction device 20 of this embodiment calculates a matrix for each explanatory variable, whose elements represent the similarity between two batch process data containing one or more explanatory variables, searches for similar data that is similar to the time series data to be predicted based on the feature data calculated based on the matrix for each explanatory variable, and predicts the time series data of a specified dependent variable based on the similar data.
[0086] In one aspect, according to this embodiment, it is possible to accurately predict the time series data of the objective variable. For example, this embodiment can sequentially search for similar data similar to the data to be predicted, and build a prediction model based on the similar data. Therefore, it is possible to predict the objective variable with high accuracy without the prediction model becoming obsolete. Also, for example, this embodiment searches for similar data based on feature data based on the similarity between batch process data, so it is possible to predict the entire time series data without losing the waveform features. Therefore, it is possible to accurately predict the time series data of the objective variable while taking into account the waveform features of similar data.
[0087] The prediction device 20 may select an explanatory variable from a plurality of variables. The prediction device 20 may select an explanatory variable based on the importance of each of the plurality of variables. In one aspect, according to the present embodiment, an explanatory variable capable of predicting a dependent variable is appropriately selected, and therefore the dependent variable can be predicted with high accuracy.
[0088] The prediction device 20 may calculate the similarity based on the waveform of the time series data. The prediction device 20 may calculate the similarity by a dynamic time warping method. According to one aspect, the present embodiment can accurately search for similar data that is similar to the prediction target data while taking into account the characteristics of the waveform.
[0089] The prediction device 20 may calculate feature data by compressing the dimensions of a matrix using eigenvalue decomposition. The prediction device 20 may search for similar data in a feature space in which the feature data is arranged. In one aspect, according to the present embodiment, a neighborhood search for feature data can be performed in the feature space, so that similar data can be searched for with a small amount of calculation.
[0090] The prediction device 20 may visualize a prediction result including time series data of the objective variable. The prediction result may include information used in predicting the objective variable. The information used in the prediction may include at least one of the variable importance of the explanatory variable, a scatter plot of feature data, or a waveform of similar data. In one aspect, according to the present embodiment, the basis of the prediction is shown together with the prediction data, so that the user can easily determine the validity of the prediction.
[0091] [supplement] Each function of the above-described embodiments can be realized by one or more processing circuits. Here, the term "processing circuit" in this specification includes a processor programmed to execute each function by software, such as a processor implemented by an electronic circuit, and an ASIC (Application Specific Integrated Circuit), DSP (Digital Signal Processor), FPGA (Field Programmable Gate Array), conventional circuit module, and other devices designed to execute each function described above.
[0092] Although the embodiments of the present invention have been described in detail above, the present invention is not limited to these embodiments, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims. [Explanation of symbols]
[0093] 1000: Prediction System 10: Prediction target 20: Prediction device 30: Terminal device 110: Acquisition Department 120: Data storage unit 130: Variable selection section 140: Matrix calculation section 150: Feature calculation unit 160: Search Department 170: Prediction Department 180: Visualization section
Claims
1. An acquisition unit configured to acquire a plurality of time series data including a plurality of variables; a variable selection unit configured to select one or more explanatory variables based on the importance of each of the plurality of variables; a matrix calculation unit configured to calculate, for each of the explanatory variables, a matrix having elements each representing a similarity between two of the plurality of time series data; a feature calculation unit configured to calculate feature data for each of the time series data based on a matrix for each of the explanatory variables; a search unit configured to search for similar data similar to the time series data to be predicted among the plurality of time series data based on feature data for each of the time series data; a prediction unit configured to predict time series data of a predetermined objective variable based on the similarity data; A prediction device comprising:
2. The matrix calculation unit is configured to calculate the similarity based on a waveform of the time series data. The prediction device according to claim 1 .
3. The matrix calculation unit is configured to calculate the similarity by a dynamic time warping method. The prediction device according to claim 2 .
4. The feature calculation unit is configured to calculate the feature data by compressing the dimension of the matrix by eigenvalue decomposition. The prediction device according to claim 1 .
5. the search unit is configured to search for the similar data in a feature space in which the feature data is arranged. The prediction device according to claim 1 .
6. A visualization unit configured to visualize a prediction result including time series data of the response variable. The prediction device according to claim 1 .
7. The prediction result includes information used to predict the dependent variable. The prediction device according to claim 6 .
8. The information used for the prediction includes at least one of the variable importance of the explanatory variables, a scatter plot of the feature data, or a waveform of the similar data. The prediction device according to claim 7 .
9. The computer A procedure for obtaining multiple time series data containing multiple variables; selecting one or more explanatory variables based on the importance of each of the plurality of variables; A step of calculating a matrix having elements each representing a similarity between two pieces of time series data among the plurality of time series data for each explanatory variable; calculating feature data for each of the time series data based on a matrix for each of the explanatory variables; searching for similar data similar to the time series data to be predicted among the plurality of time series data based on feature data for each of the time series data; A step of predicting time series data of a predetermined objective variable based on the similarity data; A forecasting method to perform.
10. On the computer, A procedure for obtaining multiple time series data containing multiple variables; selecting one or more explanatory variables based on the importance of each of the plurality of variables; A step of calculating a matrix having elements each representing a similarity between two pieces of time series data among the plurality of time series data for each explanatory variable; calculating feature data for each of the time series data based on a matrix for each of the explanatory variables; searching for similar data similar to the time series data to be predicted among the plurality of time series data based on feature data for each of the time series data; A step of predicting time series data of a predetermined objective variable based on the similarity data; A program for executing.
Citation Information
Patent Citations
2*66 toransu fuarunesooruno seizohoho
JP1976076206A
Information processing device, information processing method and information processing program
JP2023122124A
Cluster division evaluation device, cluster division evaluation method, and cluster division evaluation program
JP6637206B1
Time-series data analysis device, time-series data analysis method, and time-series data analysis program
WO2022113273A1
