Operation support device, operation support method, and program

Abstract

To provide a prediction technique combining machine learning and a control logic.SOLUTION: A operation support device comprises: a first prediction section that predicts a state variable representing a state of a control target using a machine learning model generated in advance; a second prediction section that predicts a logic variable representing a control value for controlling the control target using a control logic for the control target and the state variable predicted by the first prediction section; and an output section that outputs the state variable predicted by the first prediction section and the logic variable predicted by the second prediction section to a predetermined output destination determined in advance.SELECTED DRAWING: Figure 10

Description

[Technical Field]

[0001] The present disclosure relates to a driving assistance device, a driving assistance method, and a program. [Background technology]

[0002] In plants such as combustion furnaces, stable operation is achieved by operators changing set values ​​while taking into account past conditions, based on measured values ​​of the furnace's combustion state, etc. For this reason, various technologies have been proposed that use machine learning to predict the future state of the plant, with the aim of supporting operator operations, etc. (e.g., Patent Documents 1 to 3, Non-Patent Document 1, etc.). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-173495 [Patent Document 2] Japanese Patent Publication No. 2022-10199 [Patent Document 3] Japanese Patent Publication No. 2022-134592 [Non-patent literature]

[0004] [Non-Patent Document 1] Hiroto Ichikawa, Kenichiro Shimada, Daichi Kimura, Tomonori Izumiya, Online steam volume prediction using time series data from waste incineration plants, Proceedings of the 2020 Japanese Society for Artificial Intelligence National Conference Summary of the Invention [Problem to be solved by the invention]

[0005] However, conventional technology that uses machine learning to predict the future state of a plant does not reflect the plant's control logic, and predictions can sometimes be inconsistent with the control logic. This means that operators cannot trust the predictions, and the system is unable to fully support plant operations.

[0006] The present disclosure has been made in consideration of the above points, and aims to provide a prediction technology that combines machine learning and control logic. [Means for solving the problem]

[0007] A driving assistance device according to one embodiment of the present disclosure includes a first prediction unit that predicts a state variable representing a state of a control object using a pre-created machine learning model, a second prediction unit that predicts a logic variable representing a control value for controlling the control object using a control logic for the control object and the state variable predicted by the first prediction unit, and an output unit that outputs the state variable predicted by the first prediction unit and the logic variable predicted by the second prediction unit to a predetermined output destination. [Effects of the Invention]

[0008] A predictive technology that combines machine learning and control logic is provided. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a diagram illustrating an example of an overall configuration of a plant control system according to a first embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of a driving assistance device according to the first embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of a functional configuration of a driving assistance device according to the first embodiment. [Figure 4] 4 is a flowchart illustrating an example of a driving assistance process according to the first embodiment. [Figure 5] FIG. 10 is a diagram illustrating an example of a functional configuration of a driving assistance device according to a second embodiment. [Figure 6] 10 is a flowchart illustrating an example of a driving assistance process according to a second embodiment. [Figure 7] FIG. 1 illustrates input and output of a prediction model in one embodiment. [Figure 8]FIG. 2 is a diagram illustrating inputs and outputs of control logic in one embodiment. [Figure 9] FIG. 1 illustrates prediction of state variables and logic variables in one embodiment. [Figure 10] FIG. 10 is a diagram illustrating an example of a usage method. DETAILED DESCRIPTION OF THE INVENTION

[0010] First and second embodiments of the present invention will be described in detail below with reference to the drawings. In each of the following embodiments, a plant control system 1 will be described that includes an operation assistance device 10 that can assist an operator in operation by performing predictions that combine machine learning and the control logic of a plant such as a waste incineration plant.

[0011] Hereafter, the physical quantities (e.g., temperature, pressure, flow rate, gas concentration, etc.) that represent the plant state measured by various sensors are called "state variables," and x i (i=1, ,I). Similarly, the control value determined by the control logic is called a "logic variable," and y j (j=1, ,J). Similarly, the setting values ​​set as target values ​​or threshold values ​​of certain physical quantities for the control device that controls the plant are called "setting variables", and r k (k=1, ,K). Also, the state variable x at discrete time s (hereafter simply referred to as "time s") i , logic variable y j and the setting variable r k x i (s), y j (s) and r kLet us denote it as (s), where I is the total number of state variables, J is the total number of logic variables, and K is the total number of setting variables. A control value is a value that represents a control command for controlling the operation of equipment included in a plant. The control value may be a discrete value or a continuous value, but a specific example of a control value that takes a discrete value is a value that represents the start or stop of a feeder or stoker in a waste incineration plant.

[0012] Control logic refers to the process executed by the control device of a plant to control the plant. Specific examples of control logic include threshold processing for a certain state variable, control processing using a control method such as PID control, etc.

[0013] [First embodiment] The first embodiment will be described below.

[0014] <Example of Overall Configuration of Plant Control System 1 According to First Embodiment> An example of the overall configuration of a plant control system 1 according to the first embodiment is shown in Fig. 1. As shown in Fig. 1, the plant control system 1 according to the first embodiment includes an operation support device 10, a control device 20, and a target plant 30. Here, the operation support device 10 and the control device 20 are communicably connected via an arbitrary communication network. Similarly, the control device 20 and the target plant 30 are communicably connected via an arbitrary communication network.

[0015] The driving assistance device 10 calculates the state variable x obtained from the control device 20. i and logic variable y j and the setting variable r set in the control device 20. k Based on this, a prediction method that combines machine learning and the control logic of the target plant 30 is used to predict the state variable x up to a certain time in the future. i and logic variable y j The driving support device 10 also predicts the state variable x i and logic variable y jThis allows the operator to calculate the state variable x i and logic variable y j The setting variable r set in the control device 20 is determined based on the predicted value of k It will be possible to make changes to the above.

[0016] The control device 20 is configured with a predetermined control logic and a setting variable r set by an operator. k That is, the control device 20 is a controller that controls the target plant 30 based on the state variable x i and its state variable x i and the setting variable r set to itself k and logic variable y j and the logic variable y at the next time from the control logic. j Determine the logic variable y j to the target plant 30. As a result, the target plant 30 (more precisely, the operation of the equipment included in the target plant 30) is controlled in accordance with the control logic.

[0017] The target plant 30 is any plant that is the target of operation assistance. The target plant 30 is not limited to a specific plant, but a waste incineration plant is a specific example of the target plant 30. However, the target plant 30 is not necessarily limited to a plant, and may be, for example, a supply and demand system such as an energy management system.

[0018] 1 is merely an example, and other configurations may be used. For example, an operator terminal operated by an operator may be provided in addition to the operation support device 10.

[0019] <Example of Hardware Configuration of Driving Assistance Device 10 According to First Embodiment> An example of a hardware configuration of the driving assistance device 10 according to the first embodiment is shown in Fig. 2. As shown in Fig. 2, the driving assistance device 10 according to the first embodiment includes an input device 101, a display device 102, an external I / F 103, a communication I / F 104, a RAM (Random Access Memory) 105, a ROM (Read Only Memory) 106, an auxiliary storage device 107, and a processor 108. Each of these pieces of hardware is connected to each other via a bus 109 so as to be able to communicate with each other.

[0020] The input device 101 is, for example, a keyboard, a mouse, a touch panel, a physical button, etc. The display device 102 is, for example, a display, a display panel, etc. Note that the driving assistance device 10 does not necessarily have to have at least one of the input device 101 and the display device 102.

[0021] The external I / F 103 is an interface with an external device such as a recording medium 103a. Examples of the recording medium 103a include a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), and a USB (Universal Serial Bus) memory card.

[0022] The communication I / F 104 is an interface for communicating with other devices and equipment such as the control device 20. The RAM 105 is a volatile semiconductor memory (storage device) that temporarily stores programs and data. The ROM 106 is a non-volatile semiconductor memory (storage device) that can store programs and data even when the power is turned off. The auxiliary storage device 107 is a non-volatile storage device (storage device) such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory. The processor 108 is various types of arithmetic device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

[0023] 2 is an example, and the driving assistance device 10 may have other hardware configurations. For example, the driving assistance device 10 may have multiple auxiliary storage devices 107 or multiple processors 108, or may have various types of hardware other than the hardware shown in the figure.

[0024] <Example of functional configuration of the driving assistance device 10 according to the first embodiment> FIG. 3 shows an example of a functional configuration of the driving assistance device 10 according to the first embodiment. As shown in FIG. 3, the driving assistance device 10 according to the first embodiment includes a prediction target time update unit 201, a state variable prediction unit 202, a logic variable prediction unit 203, and an input / output unit 204. These units are realized, for example, by processing in which one or more programs installed in the driving assistance device 10 are executed by the processor 108 or the like. The driving assistance device 10 according to the first embodiment also includes a database 205. The database 205 is realized, for example, by a storage area of ​​a storage device such as the auxiliary storage device 107. Note that the database 205 may also be realized, for example, by a storage area of ​​a storage device or the like connected to the driving assistance device 10 via a communication network.

[0025] The prediction target time update unit 201 updates the state variable x up to a certain time in the future by a prediction method that combines machine learning and the control logic of the target plant 30. i and logic variable y j When predicting the state variable x i and logic variable y j The predicted target time update unit 201 also determines whether the predicted target time has reached a predetermined future time.

[0026] The state variable prediction unit 202 uses a prediction model created in advance by a machine learning technique to predict the state variable x at each prediction target time. i Furthermore, the state variable prediction unit 202 predicts the state variable x i is stored in the database 205.

[0027] The logic variable prediction unit 203 uses the control logic of the target plant 30 to predict the logic variable y j Furthermore, the logic variable prediction unit 203 predicts the logic variable y j The logic variable prediction unit 203 stores the logic variable y at each prediction target time by using a program or the like that simulates the control logic of the target plant 30. j can be predicted.

[0028] The input / output unit 204 receives the state variable x i and logic variable y j The state variable x contains the predicted results of i and logic variable y j to a predetermined output destination (for example, the display device 102 such as a display). k When setting the variable r k and the setting variable r k is output to the control device 20.

[0029] The database 205 stores the past state variables x acquired from the control device 20. i and logic variable y j and a setting variable r set in the control device 20 from the past to a certain time in the future. k The database 205 also stores the state variable x i The control unit 20 also stores a prediction model for predicting the state variable x obtained from the control device 20, a program for simulating the control logic of the target plant 30, etc. In the following, the current time is represented by t, an integer equal to or greater than 1 is represented by S, and the past state variable x obtained from the control device 20 is represented by i Let X be the set of i ={x i (s)|s=tS, ,t-1}. Similarly, the past logic variable y j Let Y be the set of j ={y j(s)|s=tS, , t-1}. Also, let T be a certain time in the future. Let r be a setting variable from the past to the future time T set in the control device 20. k Let R be the set of k ={r k (s)|s=tS, ,t-1,t,t+1, ,T}. Furthermore, in the following, a function representing a prediction model is defined as f, and a function representing an algorithm such as a program that simulates the control logic of the target plant 30 is defined as g.

[0030] In the following, each X i (i=1, ,I) is the "state variable set", and each Y j (j=1, ,J) is the "logic variable set", and each R k (k=1,···,K) is called the "configuration variable set."

[0031] <Driving assistance process according to the first embodiment> An example of the driving assistance process according to the first embodiment will be described with reference to Fig. 4. In the following, it is assumed that a future time T is set by an operator or the like, where t is the current time.

[0032] The prediction target time update unit 201 initializes a variable τ for expressing the prediction target time to 0 (step S101).

[0033] The state variable prediction unit 202 predicts the state variable x i (s)∈X i (i=1, ,I) and logic variable y j (s)∈Y j (j=1, ,J) and the setting variable r k (s)∈R k (k=1, ,K) and the state variable x at the time t+τ to be predicted by the prediction model f. i (t+τ) is predicted (step S102).

[0034] The prediction model f is the state variable x i (s)∈X i (i=1, ,I) and logic variable yj (s)∈Y j (j=1, ,J) and the setting variable r k (s)∈R k (k=1, ,K) as inputs, and the state variable x i Any machine learning model that predicts and outputs (t+τ)(i=1, ,I) can be used. Such a machine learning model can be a linear model (e.g., linear regression, partial least squares, Lasso regression, etc.) or a nonlinear model (e.g., neural network, random forest, Transformer, etc.).

[0035] The state variable prediction unit 202 predicts the state variable x predicted in step S102. i (t+τ) is stored in the database 205 (step S103). That is, the state variable prediction unit 202 stores X i ←X i ∪{x i (t+τ)}, the state variable set X i Update.

[0036] The logic variable prediction unit 203 predicts the state variable x i (s)∈X i (i=1, ,I) and logic variable y j (s)∈Y j (j=1, ,J) and the setting variable r k (s)∈R k (k=1, ,K) and the logic variable y at the target time t+τ is predicted by the control logic g. j (t+τ) is predicted (step S104). At this time, the state variable x i (t+τ)(i=1, ,I) is also used to predict the logic variable y j Note that (t+τ) is predicted.

[0037] The control logic g is the state variable x i (s)∈X i (i=1, ,I) and logic variable y j (s)∈Y j(j=1, ,J) and the setting variable r k (s)∈R k (k=1, ,K) as inputs, and the logic variable y j (t+τ)(j=1,...,J). Specific examples of control logic g include threshold processing (on-off control processing) that sets a logic variable to 1 or 0 when a certain state variable exceeds (or falls below) a threshold, processing that sets a logic variable to 1 or 0 when a certain state variable exceeds (or falls below) a set variable, and control processing that calculates a logic variable of continuous values ​​(which may have upper and lower limits) using a control method such as PID control. Other specific examples of control logic g include processing that sets a logic variable to 1 or 0 when a certain state variable deviates from a certain range set for a set variable.

[0038] The logic variable prediction unit 203 predicts the logic variable y j (t+τ) is stored in the database 205 (step S105). That is, the logic variable prediction unit 203 stores Y j ←Y j ∪{y j (t+τ)} is used to define the logic variable set Y j Update.

[0039] The predicted target time update unit 201 determines whether or not T>t+τ is satisfied (step S106).

[0040] If it is determined in step S106 that T>t+τ is satisfied, the prediction target time update unit 201 updates the value of the variable τ by τ←τ+1 (step S107). As a result, the processing from step S102 onwards is executed using the updated τ.

[0041] On the other hand, if it is not determined in step S106 that T>t+τ is satisfied (that is, if T=t+τ), the input / output unit 204 calculates the state variable x i and logic variable yj to a predetermined output destination (step S108). i ={x i (s)|s=tS,···,t-1,t,t+1,···,T} and Y j ={y j (s)|s=tS, , t-1, t, t+1, , T} is visualized on a display device 102 such as a display. This allows the operator to visualize these state variables x i and logic variable y j The setting variable r set in the control device 20 is k It becomes possible to input (s)(s≧t), etc.

[0042] [Second embodiment] The second embodiment will be described below. Note that in the second embodiment, differences from the first embodiment will be mainly described, and descriptions of components similar to those in the first embodiment will be omitted.

[0043] <Example of functional configuration of driving assistance device 10 according to second embodiment> An example of the functional configuration of the driving assistance device 10 according to the second embodiment is shown in Fig. 5. As shown in Fig. 5, the driving assistance device 10 according to the second embodiment further includes a relearning unit 206. The relearning unit 206 is realized, for example, by processing in which one or more programs installed in the driving assistance device 10 are executed by the processor 108 or the like.

[0044] The re-learning unit 206 retrains the state variable set X i (i=1, ,I) and a logic variable set Y j (j=1, ,J) and the set of variables R k (k=1, ,K) to retrain the prediction model f.

[0045] <Driving assistance process according to the second embodiment> An example of the driving assistance process according to the second embodiment will be described with reference to Fig. 6. Note that steps S202 to S208 in Fig. 6 are similar to steps S101 to S107 in Fig. 4, respectively, and therefore description thereof will be omitted.

[0046] The re-learning unit 206 retrains the state variable set X i (i=1, ,I) and a logic variable set Y j (j=1, ,J) and the set of variables R k (k=1, , K) to retrain the prediction model f (step S201). That is, the retraining unit 206 retrains the state variable x at s=t−(S+1), , t−2, for example. i (s)(i=1, ,I) and logic variable y j (s)(j=1, ,J) and the setting variable r k (s)(k=1, ,K) and the state variable x when input to the prediction model f i Update the learnable parameters of the prediction model f so that the prediction accuracy of (s+1) is improved. This allows for more accurate prediction of the state variable x i A prediction model f that can predict

[0047] [Example] As an example of the first embodiment, a case where the target plant 30 is a stoker-type waste incineration plant will be described below.

[0048] In a stoker-type waste incineration plant, waste is sent into the furnace by moving the feeder or stoker, and air is blown into the waste inside the furnace to combust it.The heat generated by combustion is used to generate steam to generate electricity, and exhaust gases (NOx, CO, HCl, SO2, mercury, etc.) are generated during combustion.

[0049] In this embodiment, the state variables, logic variables, and setting variables are as follows:

[0050] x1:Furnace temperature x2: Steam flow rate x3: Exhaust gas concentration y1: Feeder control command y2: Stoker control command r1: Steam flow rate setting value r2: Air flow rate setting value

[0051] Also, the state variable x i Figure 7 shows the forecast model f that predicts the logic variable y j The control logic g for calculating the state variable x predicted by the prediction model f is shown in Fig. 8. i Note that (t+τ) is also input. Note that the logic variable y1 represents the stoppage or operation of the feeder as 0 or 1, and similarly the logic variable y2 represents the stoppage or operation of the stoker as 0 or 1. However, the logic variable y1 may represent the feeder speed and the logic variable y2 may represent the stoker speed as continuous values.

[0052] At this time, as an example, predictions of the state variable x2, logic variable y1, and setting variable r1 when the driving assistance process shown in FIG. 4 is executed are shown in FIG. 9. As shown in FIG. 9, when τ=0, first, the state variable x2(t) at time t is predicted, and then the logic variable y1(t) at time t is calculated. Next, when τ=1, the state variable x2(t+1) at time t+1 is predicted, and then the logic variable y1(t+1) at time t is calculated. Note that the example shown in FIG. 9 shows a case where there is a control logic g that operates the feeder when the state variable x2 falls below the setting variable r1, and stops the feeder otherwise. Also, PV represents the state variable, LV represents the logic variable, and SV represents the setting variable.

[0053] In this way, at each prediction target time t+τ, the state variable x i After predicting (t+τ), the logic variable y j (t+τ) is calculated. This calculates the state variable x that reflects the control logic g. i It becomes possible to predict (t+τ).

[0054] ·How to use A method for using this embodiment will be described with reference to FIG. 10. For example, as shown in the left diagram of FIG. 10, if it is predicted that the steam flow rate will fall significantly below the steam flow rate set value due to a feeder stop, the operator can quickly start up the feeder by changing the set value for a short period of time, as shown in the right diagram of FIG. 10. This makes it possible to avoid a situation in which the steam flow rate drops significantly. In this way, by appropriately changing the set variable with reference to the future predicted value, the operator can change the logic variable, which cannot be directly manipulated, and achieve the desired operation.

[0055] [Variations] Modifications of the first and second embodiments will be described below. Note that the following modifications can be combined with other modifications as appropriate, as long as they do not contradict each other.

[0056] Variation 1 In the first and second embodiments described above, a common prediction model f is used for all prediction target times t+τ, but a different prediction model f may be used depending on the prediction target time t+τ. For example, f=f (τ) A prediction model f may be used, which is expressed as follows:

[0057] Variation 2 The state variable x at the time t+τ to be predicted by the prediction model f i When predicting (t+τ), predetermined preprocessing may be performed on some of the variables (at least some of the state variables, logic variables, and setting variables) that are input to the prediction model f. Specific examples of preprocessing include, for example, a process of calculating a moving average of the variables that are input to the prediction model f, a process of calculating the rate of change (e.g., differential value) of the variables that are input to the prediction model f, a process of calculating the integral value of the variables that are input to the prediction model f, and a process of compressing the dimensions of the variables that are input to the prediction model f using a predetermined dimension reduction technique.

[0058] The state variable x at the time t+τ is iIf preprocessing is performed on the input of prediction model f when predicting (t+τ), preprocessing is also performed on the input of prediction model f when creating or retraining prediction model f.

[0059] Variation 3 A graph neural network that can input a graph structure may be used as the prediction model f. In this case, the prediction model f is a function of the state variable x i (s)∈X i (i=1, ,I) and logic variable y j (s)∈Y j (j=1, ,J) and the setting variable r k (s)∈R k (k=1, ,K) and the state variable x i and logic variable y j and the configuration variable r k and a graph showing at least a part of the relationship between them, and the state variable x at the time t+τ to be predicted is input. i (t+τ) is predicted and output. By using a graph neural network as the prediction model f, it is possible to filter out noise, which is expected to improve prediction accuracy.

[0060] The graph that is the input of the graph neural network may be constructed in advance using domain knowledge, or may be constructed by using LinGAM or the magnitude of the correlation coefficient, etc., to estimate the past state variables x. i and past logic variable y j , configuration variable r k The graph that is input to the graph neural network may be an undirected graph or a directed graph.

[0061] Variation 4 An image or video or its features may be used as input for the prediction model. That is, the prediction model f is a function of the state variable x i (s)∈X i (i=1, ,I) and logic variable y j (s)∈Y j (j=1, ,J) and the setting variable rk (s)∈R k (k=1, ,K) and an image or video or its features are input, and the state variable x at the prediction target time t+τ is calculated. i (t+τ) may be predicted and output. In this case, when converting an image or video into its feature, for example, the image or video is used as input, and each state variable x i A neural network that outputs features that represent the influence on (i=1, . . . , I) may be used.

[0062] As the image or video, for example, an image or video photographed of the state of the target plant 30 (e.g., if the target plant 30 is a waste incineration plant, an image or video photographed of the combustion state of the incinerator, etc.) may be used.

[0063] Variation 5 The logic variable y at the time t+τ to be predicted by the control logic g j When predicting (t+τ), predetermined pre-processing may be performed on some of the variables that are input to the control logic g, as in the above-described Modification 2. Specific examples of pre-processing include, for example, a process of calculating a moving average of the variables that are input to the control logic g, a process of calculating a rate of change (e.g., a differential value) of the variables that are input to the control logic g, a process of calculating an integral value of the variables that are input to the control logic g, and a process of compressing the dimensions of the variables that are input to the control logic g using a predetermined dimension reduction technique.

[0064] Variation 6 In the first and second embodiments, the prediction model f and the control logic g are i (s)∈X i (i=1, ,I) and logic variable y j (s)∈Y j (j=1, ,J) and the setting variable r k (s)∈R k Although all of the variables (k=1, , K) are input, only some of the variables may be input.

[0065] Variation 7 When re - learning the prediction model f in the above - mentioned second embodiment, for example, after determining whether a certain predetermined condition is satisfied, the re - learning of the prediction model f may be performed only when the condition is satisfied. Examples of such conditions include that the prediction error of the prediction model f exceeds a predetermined threshold, that re - learning is instructed by an operator, etc.

[0066] · Variant 8 When re - learning the prediction model f in the above - mentioned second embodiment, the most recent state variables x i (s)(i = 1, ···, I), the logic variables y j (s)(j = 1, ···, J), and the setting variables r k (s)(k = 1, ···, K) may be used as inputs to the prediction model f. Specifically, using an integer of 1 or more that satisfies S'<S, the state variables x i (s)(i = 1, ···, I), the logic variables y j (s)(j = 1, ···, J), and the setting variables r k (s)(k = 1, ···, K) at s=t-(S'+1), ···, t - 2 may be used as inputs to the prediction model f.

[0067] · Variant 9 When re - learning the prediction model f in the above - mentioned second embodiment, the most recent state variables x i (i = 1, ···, I) and the similar state variables x i (s), the most recent logic variables y j (j = 1, ···, J) and the similar logic variables y j (s), the most recent setting variables r k (k = 1, ···, K) and the similar setting variables r k (s) are extracted respectively, and then the extracted state variables x i (s), the logic variables y j (s), and the setting variables r k (s) may be used as inputs to the prediction model f. Note that any scale for measuring similarity can be used. For example, cosine similarity, etc. can be used.

[0068] · Variant 10 As the prediction model f, a large-scale base model for time series trained using data of a target other than the target plant 30 to be predicted may be used. In this way, for example, if the target plant 30 is a newly constructed facility or the like, the data of the target plant 30 (i.e., the state variable x i Set X of i , logic variable y j Set Y of j , configuration variable r k Set R of k ), it is possible to make highly accurate predictions. Furthermore, it is possible to detect, for example, anomalies that occur infrequently (or anomalies that have never occurred). Furthermore, for example, it is possible to improve the prediction accuracy by evolving (i.e., additional learning, etc.) the large-scale foundation model for time series. Note that, for example, TimeGPT (Reference 1) can be used as a large-scale foundation model for time series. Large-scale foundation models are sometimes simply called foundation models.

[0069] Variation 11 In the above-described modification 10, the data of the target plant 30 may be used as learning data, and the large-scale infrastructure model may be fine-tuned using the learning data. This makes it possible to improve the prediction accuracy immediately after the installation or renewal of equipment in the target plant 30, for example. Note that, for fine-tuning, for example, LoRA (Reference 2) or the like may be used.

[0070] [summary] As described above, when predicting state variables and logic variables up to a certain time in the future, the operation assistance device 10 according to this embodiment predicts the state variables at each target prediction time using the state variables, logic variables, and setting variables at a time prior to the target prediction time, and then predicts the logic variables at the target prediction time using the prediction result and the state variables, logic variables, and setting variables at the time prior to the target prediction time. This enables predictions that are consistent with the control logic used by the control device 20 to control the target plant 30, resulting in highly reliable predictions. Therefore, the operator can make decisions, such as changing setting variables, based on the prediction results, thereby enabling operation assistance for the target plant 30.

[0071] The present invention is not limited to the above-described specifically disclosed embodiments, and various modifications, changes, and combinations with known technologies are possible without departing from the scope of the claims.

[0072] [References] Reference 1: Wenlong Liao, Fernando Porte-Agel, Jiannong Fang, Christian Rehtanz, Shouxiang Wang, Dechang Yang, Zhe Yang, "TimeGPT in Load Forecasting: A Large Time Series Model Perspective," arXiv:2404.04885 [cs.LG]. Reference 2: Divij Gupta, Anubhav Bhatti, Suraj Parmar, Chen Dan, Yuwei Liu, Bingjie Shen, San Lee, "Low-Rank Adaptation of Time Series Foundational Models for Out-of-Domain Modality Forecasting," arXiv:2405.10216 [cs.LG]. [Explanation of symbols]

[0073] 1 Plant control system 10 Driving assistance devices 20 Control device 30 Target plants 101 Input Device 102 Display device 103 External I / F 103a Recording media 104 Communication I / F 105 RAM 106 ROM 107 Auxiliary storage 108 processors 109 Bus 201 Forecast target time update unit 202 State variable prediction unit 203 Logic Variable Prediction Unit 204 Input / output section 205 databases 206 Re-learning Department

Claims

1. a first prediction unit that predicts a state variable representing a state of a controlled object using a machine learning model created in advance; a second prediction unit that predicts a logic variable representing a control value for controlling the controlled object using a control logic for the controlled object and the state variable predicted by the first prediction unit; an output unit that outputs the state variables predicted by the first prediction unit and the logic variables predicted by the second prediction unit to a predetermined output destination; A driving assistance device having the above configuration.

2. The first prediction unit For each prediction target time that is a time in the future than the current time, predicting a state variable at the prediction target time using the machine learning model; The second prediction unit The driving assistance device according to claim 1 , wherein, for each of the target prediction times, logic variables at the target prediction times are predicted using the control logic and state variables at the target prediction times.

3. The first prediction unit For each target time of prediction, predicting a state variable at the target time of prediction using the machine learning model, a state variable at a time earlier than the target time of prediction, a logic variable at the past time, and a setting variable representing a setting value for the control object at the past time; The second prediction unit 3. The driving assistance device according to claim 2, wherein, for each target prediction time, a logic variable at the target prediction time is predicted using the control logic, a state variable at the target prediction time, a state variable at the past time, a logic variable at the past time, and a setting variable at the past time.

4. 2. The driving assistance device according to claim 1, wherein the machine learning model is a graph neural network to which a graph structure expressing a relationship between the state variables and the logic variables can be input, or a neural network to which an image or video capturing a state of the control object can be input.

5. The driving assistance device according to claim 3 , wherein the machine learning model is a base model.

6. The driving assistance device according to claim 5 , wherein the base model is a model that is fine-tuned using learning data that includes state variables, logic variables, and setting variables of the controlled object.

7. 4. The driving assistance device according to claim 3, further comprising: a re-learning unit that re-learns the machine learning model for each prediction target time using state variables at the past time, logic variables at the past time, and setting variables at the past time.

8. The driving assistance device according to claim 1 , wherein the control logic is a threshold process for a predetermined state variable or a control process according to a predetermined control method including PID control.

9. a first prediction step of predicting a state variable representing a state of a controlled object using a machine learning model created in advance; a second prediction procedure for predicting a logic variable representing a control value for controlling the controlled object using a control logic for the controlled object and the state variable predicted in the first prediction procedure; an output step of outputting the state variables predicted in the first prediction step and the logic variables predicted in the second prediction step to a predetermined output destination; A driving assistance method performed by a computer.

10. a first prediction step of predicting a state variable representing a state of a controlled object using a machine learning model created in advance; a second prediction procedure for predicting a logic variable representing a control value for controlling the controlled object using a control logic for the controlled object and the state variable predicted in the first prediction procedure; an output step of outputting the state variables predicted in the first prediction step and the logic variables predicted in the second prediction step to a predetermined output destination; A program that causes a computer to execute the following.

Images