Coordinated learning system, monitoring system, and storage medium

Abstract

The present invention relates to a coordinated learning system and a monitoring system to improve the bottleneck of a processing process and improve the efficiency of the entire process. The coordinated learning system of the embodiment can be used to set monitoring models for each of a plurality of processes that are continuous in a time series at a predetermined transition time interval. The system stores, in a time series, the first monitoring data of the first process, the second monitoring data of the second process upstream or downstream of the first process, and the monitoring result of the first process output by the first monitoring model using the first monitoring data as an input parameter. In addition, the first monitoring model is subjected to a parent model learning process using the first monitoring data and the monitoring result of the first monitoring model, the monitoring result of the first monitoring model at the first moment is used as teacher data, and the second monitoring model is subjected to a child model learning process using the second monitoring data at the second moment that differs from the first moment by a transition time amount as an input parameter.

Description

Technical Field

[0001] Embodiments of the present invention relate to a process improvement method and a monitoring technology for a processing step or a manufacturing step that is continuous in time series. Background Art

[0002] For example, water treatment improves water quality through multiple treatment processes that are continuous in time series. In each treatment process where water flows continuously, monitoring data measured and detected by sensor equipment is collected. The collected monitoring data is input into the monitoring system, and the monitoring system determines normal operation / abnormal operation according to each treatment process.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Publication No. 2016-195974

[0006] Patent Document 2: Japanese Patent Application Publication No. 2017-157072

[0007] Patent Document 3: Japanese Patent Application Publication No. 2018-63656

[0008] Patent Document 4: Japanese Patent No. 5022610 Summary of the invention

[0009] Problems to be solved by the invention

[0010] Provided is a coordinated learning system for realizing overall process efficiency improvement in process monitoring of a plurality of processes that are continuous in time series.

[0011] Means for solving problems

[0012] The coordinated learning system of the embodiment is a coordinated learning system for process monitoring in which monitoring models are respectively provided for a plurality of processes that are continuous in time series at a predetermined transition time interval. The coordinated learning system comprises: a storage unit that stores, in time series, the first monitoring data of the first process, the second monitoring data of the second process that is upstream or downstream of the first process, and the monitoring result of the first process that is output by the first monitoring model using the first monitoring data as an input parameter; and a model learning unit that performs a parent model learning process on the first monitoring model using the first monitoring data and the monitoring result of the first monitoring model, uses the monitoring result of the first monitoring model at the first moment as teacher data, and performs a child model learning process on the second monitoring model that uses the second monitoring data at the second moment that is different from the first moment by the transition time as an input parameter. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a diagram for explaining the functional blocks of the monitoring system according to the first embodiment.

[0014] Figure 2 It is an explanatory diagram of the cooperative learning model according to the first embodiment.

[0015] Figure 3 This is a diagram showing an example of constructing a coordinated learning model for monitoring each process in the first embodiment (a monitoring model upstream of a starting point process).

[0016] Figure 4 This is a diagram showing an example of constructing a cooperative learning model for monitoring each process of the first embodiment (a monitoring model downstream of a starting point process).

[0017] Figure 5 This is a diagram for explaining the relationship between the cooperative learning models with respect to the starting point process in a plurality of steps that are continuous in time series in the first embodiment.

[0018] Figure 6 It is a diagram showing an example of process improvement in water treatment to which the cooperative learning model according to the first embodiment is applied.

[0019] Figure 7 This is a diagram for explaining an example of improvement of the overall process of the bottleneck step by the coordinated learning model according to the first embodiment.

[0020] Figure 8 This is a flowchart showing the construction process and update process of the cooperative learning model according to the first embodiment. DETAILED DESCRIPTION

[0021] Hereinafter, embodiments will be described with reference to the drawings.

[0022] (First embodiment)

[0023] Figures 1 to 8 is a diagram for explaining a process improvement method using a coordinated learning model according to a first embodiment. Figure 1 FIG. 1 is a diagram showing the functional blocks of a monitoring system to which a cooperative learning system is applied. Figure 1 As shown, monitoring data collected from a plurality of processes that are continuous in time series is input to the monitoring device 100, and the entire process is managed using monitoring results in monitoring models that are respectively set for the plurality of processes.

[0024] As an example of a process including multiple steps that are continuous in time series, there is water treatment. In water treatment, water is purified by flowing through each process such as an adjustment tank, an aeration tank, a sedimentation tank, and a coagulation reaction tank. These tanks are continuous in time series at a predetermined transition time interval, and the overall water treatment process is managed by monitoring each tank.

[0025] In addition, the coordinated learning structure of this embodiment can be applied to other process management besides the improvement and management of water treatment processes. For example, the coordinated learning model can be applied to process monitoring in which two or more processes are continuous in a time series at a predetermined transition time interval, such as a manufacturing line of products or materials, a processing line such as heat treatment or chemical treatment, or an incineration line of garbage or waste.

[0026] exist Figure 1 In the example, the processing process sets process stage n as the starting process, takes process stage n as the starting point, expresses the upstream process as process stages n-1 and n-2, and expresses the downstream stage as n+1 and n+2. These process stages are multiple processes that are continuous in time series, so the object processed at time t of process stage n is processed at the other processes at the time that differs from each other by the transition time. In other words, at the same time, the object being processed in process stage n is processed in the previous process stage n-1 and is waiting to be processed in the next process stage n+1.

[0027] In this embodiment, monitoring models are set according to each process stage, and the monitoring results of each monitoring model are output in each process stage. The so-called monitoring model refers to the judgment logic that uses monitoring data as input parameters to determine whether the process stage is in a normal state or an abnormal state. The monitoring model is customized according to the particularity of the process and the environment through machine learning.

[0028] Monitoring data is sensor information output from sensor equipment installed in each process. In addition to values ​​such as temperature information, equipment operation information, and water quality information, monitoring data also includes image data photographed by a camera or other photographic device. The monitoring model is set with, for example, thresholds for determining normal / abnormal conditions, reference image data, etc., and determines whether the sensor value exceeds the threshold, whether the matching rate with the reference image data is lower than the threshold, etc., and outputs the determination result as the monitoring result. By using these monitoring results and their monitoring data (measured values) to perform machine learning, for example, the determination criteria are updated and an optimized monitoring model is constructed.

[0029] The monitoring device 100 of this embodiment is input with monitoring data acquired at each process stage included in the processing process. The data collection control device 110 performs collection control of each monitoring data outputted from a plurality of processes, and stores it in a storage device 130 in a time series. The monitoring control device 120 sets each monitoring model according to a plurality of processes, and outputs monitoring results based on each monitoring data according to each process using each set monitoring model. The monitoring results are stored in the storage device 130 in a time series. These monitoring controls are executed by the monitoring control unit 121.

[0030] On the other hand, the monitoring control device 120 of the present embodiment includes a cooperative learning device that performs learning control of monitoring models that differ according to the process. Figure 1 In the example, a coordinated learning system is constructed by monitoring the model management unit 122 and the teacher data management unit 123. In addition, in the present embodiment, the coordinated learning system is described as an example in which the coordinated learning system is built into the monitoring device 100, but it is not limited to this. For example, it is also possible to construct a separately independent coordinated learning system externally connected to the monitoring device 100. In addition, it is also possible to apply it as a learning system connected to the monitoring device 100 via a network or the like.

[0031] Figure 2 1 is an explanatory diagram of the cooperative learning model of this embodiment. The learning process and the teacher data generation process of the cooperative learning system can be performed at any timing.

[0032] First, the monitoring model management unit 122 sets the parent model as the starting process. For example, the monitoring model Yn of the process stage n is set as the parent model, and for the monitoring model Yn, the monitoring data A (first monitoring data) and the monitoring results of the monitoring model Yn (first monitoring model) are used, as well as the teacher data pre-made for the monitoring model Yn, and the parent model learning process for the monitoring model Yn is performed based on machine learning. The parent model learning process can apply known techniques, and in the child model learning process described later, the machine learning itself using the teacher data can also apply known methods.

[0033] Each monitoring model of each process stage upstream or downstream of the starting process is constructed by using the monitoring results of the parent model, i.e., the monitoring model Yn, as the learning process of the teacher data. The monitoring model management unit 122 uses the monitoring results of the monitoring model Yn constructed by the parent model learning process as the teacher data, and performs each child model learning process on each monitoring model Yn-1 and Yn+1 of the process stage n-1 and n+1.

[0034] Furthermore, in the process stage n-2 upstream of the process stage n-1, the monitoring results of the monitoring model Yn-1 constructed by the sub-model learning process using the teacher data of the parent model are used as the teacher data, and the sub-model learning process is performed on the monitoring model Yn-2. Similarly, in the process stage n+2 downstream of the process stage n+1, the monitoring results of the monitoring model Yn+1 constructed by the sub-model learning process using the teacher data of the parent model are used as the teacher data, and the sub-model learning process is performed on the monitoring model Yn+2.

[0035] In addition, in each process of process stage n-2 and process stage n+2, as shown by the double dotted lines, it can also be constructed so that the monitoring results of the monitoring model Yn constructed through the parent model learning process are used as teacher data, and each sub-model learning process is performed on each monitoring model Yn-2 and Yn+2 of the process stages n-2 and n+2.

[0036] Figure 3 This is a diagram showing an example of constructing a coordinated learning model for monitoring each process according to the present embodiment, and is a diagram for explaining a coordinated learning process of a monitoring model upstream of a starting point process.

[0037] The monitoring data A of process stage n (first process), the monitoring data B of process stage n-1 (second process) upstream of process stage n, the monitoring result of process stage n output by the monitoring model Yn using the monitoring data A as an input parameter, and the monitoring result of process stage n-1 output by the monitoring model Yn-1 using the monitoring data B as an input parameter are stored in time series.

[0038] At this time, when an abnormality is detected in the process stage n as the starting point process at time t, the monitoring result in the upstream process stage n-1 is normal. That is, even if the monitoring result of the process stage n-1 is "normal", the monitoring result in the downstream process stage n after the transition time s1 has passed becomes "abnormal". Therefore, even if the monitoring result of the process stage n-1 is normal at time t-s1, if s1 time has passed, the monitoring result of the downstream process stage n also becomes abnormal. Therefore, for the monitoring model Yn-1 of the process stage n-1 that is judged as "normal" only by the monitoring data B, the monitoring result of the downstream monitoring model Yn is used as the teacher data (solution) for learning processing.

[0039] The monitoring model management unit 122 uses the monitoring result of the monitoring model Yn (first monitoring model) at time t (first moment) as the teacher data (solution), and performs sub-model learning processing on the monitoring model Yn-1 (second monitoring model) using the monitoring data B (second monitoring data) at time t-s1 (second moment) which differs from time t (first moment) by the transition time s1 as an input parameter. At this time, the teacher data management unit 123 takes the transition time between processes into consideration, manages the monitoring results of the process used as the teacher data and each monitoring data for the monitoring result in a time series, and provides the monitoring model management unit 122 with the data required for the learning process.

[0040] By configuring in this way, the monitoring model Yn-1 can determine the "abnormality" of the process stage n on the downstream side through the monitoring data B before the time s1 has passed. Therefore, (1) it is possible to respond before the downstream process stage n becomes "abnormal", (2) since it is possible to respond before the downstream process stage n becomes "abnormal", even if it becomes "abnormal" later, the recovery time to become "normal" can be shortened, and (3) it is also possible to detect the omission of the "abnormality" determination of the downstream process stage n. The above (3) can be configured so that when it is necessary to determine that it is abnormal based on the monitoring result after the time s1 has passed, but it is not determined as abnormal for some reason, the previous "abnormality" detection in the monitoring model Yn-1 before the time s1 is recorded, and if the monitoring result of the monitoring model Yn at the time t after the time s1 has passed is "normal", a warning of abnormality detection is output in the process stage n.

[0041] Figure 4 This is a diagram showing an example of constructing a coordinated learning model for monitoring each process according to the present embodiment, and is a diagram for explaining a coordinated learning process of a monitoring model downstream of a starting point process.

[0042] The monitoring data A of process stage n (first process), the monitoring data C of process stage n+1 (second process) downstream of process stage n, the monitoring result of process stage n output by the monitoring model Yn using the monitoring data A as an input parameter, and the monitoring result of process stage n+1 output by the monitoring model Yn+1 using the monitoring data C as an input parameter are stored in time series.

[0043] Then, even if an abnormality is detected in the process stage n as the starting point process at time t, the monitoring result in the downstream process stage n+1 at the same time t becomes normal. That is, even if the monitoring result of the process stage n+1 is "normal", the monitoring result in the upstream process stage n before the transition time s2 becomes "abnormal". Therefore, even if the monitoring result of the process stage n+1 is normal at time t, if the s2 time passes, the monitoring result becomes abnormal. Therefore, for the monitoring model Yn+1 of the process stage n+1 that is judged as "normal" only by the monitoring data C, the monitoring result of the upstream monitoring model Yn is used as the teacher data (solution) for learning processing.

[0044] The monitoring model management unit 122 uses the monitoring result of the monitoring model Yn (first monitoring model) at time t (first moment) as the teacher data (solution), and performs sub-model learning processing on the monitoring model Yn+1 (second monitoring model) that uses the monitoring data C (second monitoring data) at time t+s2 (second moment) that differs from time t (first moment) by the transition time s2 as an input parameter. At this time, the teacher data management unit 123 also takes the transition time between processes into consideration, manages the monitoring results of the process used as the teacher data and each monitoring data for the monitoring result in a time series, and provides the monitoring model management unit 122 with the data required for the learning process.

[0045] Therefore, the monitoring model Yn+1 is linked to the abnormality of the upstream process stage n before the time s2 has passed, and the "abnormality" is detected through the monitoring data C. Therefore, (4) in the downstream process stage n+1, the tendency of the upstream process stage n to become "abnormal" can be grasped in advance, (5) according to the "abnormality" of the upstream process stage n, the downstream process stage n+1 can be dealt with in advance, and (6) the omission of the "abnormality" judgment of the upstream process stage n can also be detected. The above (6) can be constructed so that when the monitoring result at time t in the process stage n should be judged as abnormal, but it is not judged as abnormal for some reason, the prior "abnormality" detection in the monitoring model Yn+1 before the time s2 is recorded, and if the monitoring result of the monitoring model Yn at time t before the time s2 has passed is "normal", a warning of abnormality detection is output in the process stage n+1.

[0046] In addition, as described above, for process stage n-2 upstream of process stage n-1 (second process), monitoring data D (third monitoring data) of process stage n-2 (third process) and monitoring results of process stage n-1 (second process) output by monitoring model Yn-1 (second monitoring model) constructed by sub-model learning processing with monitoring data B (second monitoring data) as input parameter are also stored in time series, so that the monitoring model management unit 122 can use the monitoring results of monitoring model Yn-1 (second monitoring model) at time t-s1 (third moment) as teacher data (solution) and perform sub-model learning processing on monitoring model Yn-2 (third monitoring model) with monitoring data D (third monitoring data) at time t-s1-s3 (fourth moment) which differs from time t-s1 (third moment) by transition time s3 as input parameter. When observed from the parent model, monitoring model Yn-2 becomes a grandchild model. The same is true for process stage Yn+2.

[0047] In addition, as the teacher data for the sub-model learning process of the process stage n-2, the monitoring results of the parent model of the process stage Yn can be used instead of the process stage Yn-1. In this case, the monitoring data D (third monitoring data) of the process stage n-2 (third process) upstream of the process stage n-1 (second process) is stored in a time series together with the monitoring results of the monitoring model Yn, and the monitoring model management unit 122 can use the monitoring results of the monitoring model Yn (first monitoring model) at the time t (first time) as the teacher data, and perform the sub-model learning process on the monitoring model Yn-2 (third monitoring model) that uses the monitoring data D (third monitoring data) at the time t-s1-s3 (fifth time) that differs from the time t (first time) by the transition time s1+s3 as the input parameter. In this case, when viewed from the parent model, the monitoring model Yn-2 becomes a sub-model like the monitoring model Yn-1. The same is true for the process stage Yn+2.

[0048] Figure 5 1 is a diagram for explaining the relationship between the coordinated learning models with respect to the starting process in a plurality of processes that are continuous in time series in this embodiment. Figure 5 As shown, by constructing a coordinated learning model, the monitoring model in each process stage is constructed as a precursor detection model centered on the starting process. The monitoring model Yn of the starting process is the parent abnormality detection model, the monitoring model Yn-1 of the previous process is constructed as a precursor detection model for process stage n, and the monitoring model Yn-2 of the previous process is constructed as a precursor detection model for process stage n-1. In addition, the monitoring model Yn+1 of the subsequent process is constructed as a precursor detection model for process stage n, and the monitoring model Yn+2 of the subsequent process is constructed as a precursor detection model for process stage n+1.

[0049] Figure 6 It is a diagram showing an example of process improvement of water treatment to which the cooperative learning model according to the present embodiment is applied. Figure 7 This is a diagram for explaining an example of improvement of the overall process of a bottleneck step by the coordinated learning model of this embodiment.

[0050] The coordinated learning model of this embodiment is such that when the judgment accuracy is improved through the learning process of the parent model, the judgment accuracy of the child model is also improved in conjunction. Figure 7 As shown, in water treatment, there is a bottleneck process that becomes a bottleneck, and the processing capacity in this bottleneck process affects the processing capacity of the entire process. Therefore, it is important to improve the accuracy of the monitoring model of the bottleneck process to improve the production capacity. However, in order to improve the production capacity of the bottleneck process, it is not necessary to adjust the monitoring models of other processes independently, but to improve the accuracy of the monitoring models of other processes through coordinated learning, so as to improve the production capacity of the entire process. Figure 7In the example, the white area is the part where the production capacity is improved (increased), and the production capacity of other processes is improved in conjunction with the bottleneck process.

[0051] The following improvement cycle is formed: by improving the monitoring accuracy of the parent model, a precursor detection model for process stage n (bottleneck process) of the parent model is constructed in process stages n-1 and n+1, and the monitoring accuracy of process stage n is further improved by constructing the precursor detection model. Therefore, when the monitoring accuracy of process stage n, which becomes a bottleneck, is improved due to the precursor detection model, the production capacity of process stage n is improved, so the production capacity (processing capacity) of process stages n-1 and n+1 can be improved, and the bottom line of the overall production capacity of the process can be improved.

[0052] Figure 8 Flowchart showing the construction process and update process of the coordinated learning model of the present embodiment. First, a specific process among a plurality of processes that are continuous in time series at a predetermined transition time interval is set as the starting process (for example, a bottleneck process) of the parent model. Then, monitoring data of the process stage of the parent model and monitoring data of each process stage of a plurality of child models are collected in time series. The data collection control device 110 as an interface for each process stage performs a collection process of monitoring data and stores it in the storage device 130 (monitoring data storage unit) (S101, S102).

[0053] At this time, while collecting monitoring data, the monitoring control unit 121 performs monitoring control using each monitoring model set for each process, and outputs the monitoring result. The monitoring result is associated with the collected monitoring data and stored in the storage device 130 in a time series.

[0054] Next, the monitoring model management unit 122 performs a model update process, that is, a model learning process, at a predetermined arbitrary timing. The monitoring model management unit 122 first performs a parent model update process (S103: Yes).

[0055] The monitor model management unit 122 performs a parent model learning process ( S104 ) using the monitoring data of the process stage of the parent model and the monitoring result of the monitor model of the parent model, and updates the monitor model to a learned parent model ( S105 ).

[0056] The teacher data management unit 122 generates teacher data used in the child model learning process as the parent model is learned and updated, and stores it in the storage device 130 (S106, S107). Specifically, based on the updated parent model and the monitoring data accumulated in time series, the updated parent model is verified, and the verification result is output as the inference result. For example, at time t+s1, the inference of the judgment result can be verified as "abnormal". The inference result at time t+s1 is generated as teacher data and saved.

[0057] When teacher data is generated, the monitoring model management unit 122 performs sub-model learning processing. At this time, it can be configured so that, in the process stage of the sub-model learning object, it is checked whether the accumulation of monitoring data is sufficient, and if the accumulation amount of monitoring data is lower than a specified value (number or amount), the sub-model learning processing is not performed, and the sub-model learning processing is performed after sufficient information is accumulated.

[0058] When the monitoring model management unit 122 determines that the monitoring data is accumulated sufficiently in the process stage of the sub-model learning object ("Yes" in S108), it uses the generated teacher data to perform sub-model learning processing (S109). Specifically, the inference result at time t+s1 is taken as the solution, the monitoring data of the process stage of the sub-model at time t (the transition time relative to the parent model is s1 time ago) is obtained from the storage device 130, and the monitoring data of the process stage of the sub-model at time t is learned, and the learning is completed. The sub-model update processing is performed (S110). The monitoring model management unit 122 performs sub-model learning processing on all or selected sub-models in sequence, and ends the sub-model learning processing ("Yes" in S111).

[0059] In the above description, the monitoring model is configured to output monitoring results only using monitoring data of each process, but the present invention is not limited thereto. For example, the monitoring data of the parent model may also be used to construct a child model.

[0060] For example, it is possible to use the monitoring result of monitoring model Yn at time t+s1 as teacher data to perform sub-model learning processing on monitoring model Yn-1 that uses monitoring data B at time t and monitoring data A at time t as input parameters. In this case, the input parameters of monitoring model Yn-1 in monitoring control become both monitoring data B and monitoring data A at the same specified time.

[0061] Therefore, in Figure 2In the example, as shown by the single-dotted line, the monitoring model Yn-1 of the process stage n-1 is constructed as a monitoring model (Yn-1=f(B, A)) that uses the monitoring data B and the monitoring data A at the same time as explanatory variables to obtain the monitoring result (target variable) Yn-1 of the process stage n-1. At this time, the learning structure of the monitoring model Yn-1 is to use the monitoring result of the monitoring model Yn at time t as teacher data, and to perform learning processing on the two input parameters of the monitoring data B at time t-s1 that differs from time t by the transition time s1 and the monitoring data A at the same time t-s1. In addition, it can also be constructed so that the monitoring data at a specified time and the monitoring result of the monitoring model are grouped as teacher data for learning processing.

[0062] According to the present embodiment, in the process monitoring of a plurality of steps that are continuous in time series, it is possible to improve the efficiency of the entire process.

[0063] In particular, with respect to the time t at which the monitoring result of the starting process that becomes the bottleneck is output, the status of the upstream process or the downstream process of the starting process at the time t′ which is the transition time interval between the different processes is understood, and the monitoring result of the starting process (which may also include the corresponding monitoring data) is reflected to the time t′ which is backtracked by the transition time interval or advanced by the transition time interval (refer to Figure 3 , Figure 4 ) is performed in each monitoring model of the upstream process or downstream process. Therefore, the monitoring model for monitoring each process is constructed as a precursor detection model coordinated with the monitoring model of the starting process, which can improve the efficiency of the entire process.

[0064] The above describes the implementation method, but the monitoring device 100 can also have a memory (main storage device), a mouse, a keyboard, a touch panel, an operation input unit such as a scanner, an output unit such as a printer, an auxiliary storage device (hard disk, etc.), etc. as a hardware structure in addition to the above.

[0065] In addition, each function of the present invention can be realized by a program. A computer program prepared in advance to realize each function is stored in an auxiliary storage device. A control unit such as a CPU reads the program stored in the auxiliary storage device into a main storage device. The control unit executes the program read into the main storage device, and the computer can execute the functions of each part of the present invention. On the other hand, each function of the monitoring device 100 and the cooperative learning system can also be composed of individual devices, or a plurality of devices can be connected directly or via a network to form a computer system.

[0066] In addition, the above-mentioned program can also be provided to the computer in a state of being recorded on a computer-readable storage medium. As computer-readable storage media, there can be listed optical disks such as CD-ROM, phase change optical disks such as DVD-ROM, optical magnetic disks such as MO (Magnet Optical), MD (Mini Disk), floppy disks (registered trademark), magnetic disks such as removable hard disks, CF cards (registered trademark), SM cards, SD cards, memory sticks and other storage cards. In addition, as recording media, hardware devices such as integrated circuits (IC chips, etc.) specially designed and constructed for the purpose of the present invention are also included.

[0067] In addition, the embodiments of the present invention are described, but the embodiments are presented as examples and are not intended to limit the scope of the invention. The new embodiments can be implemented in various other ways, and various omissions, substitutions, and changes can be made without departing from the scope of the main purpose of the invention. These embodiments and their variations are included in the scope and main purpose of the invention, and are included in the invention described in the scope of the patent claim and the scope equivalent thereto.

[0068] Explanation of symbols

[0069] 100: monitoring device; 110: data collection control device; 120: monitoring control device; 121: monitoring control unit; 122: monitoring model management unit; 123: teacher data management unit; 130: storage device; n: process; Yn: monitoring model.

Claims

1. A coordinated learning system for process monitoring, wherein monitoring models are provided for each of a plurality of processes that are continuous in time series at a predetermined transition time interval, wherein: have: a storage unit storing, in time series, first monitoring data of a first process, second monitoring data of a second process that is upstream or downstream of the first process, and a monitoring result of the first process output by a first monitoring model using the first monitoring data as an input parameter; as well as The model learning unit uses the above-mentioned first monitoring data, the monitoring result of the above-mentioned first monitoring model and the teacher data pre-produced for the above-mentioned first monitoring model to perform a parent model learning process on the above-mentioned first monitoring model, sets any moment in the above-mentioned first process as the first moment, uses the monitoring result of the above-mentioned first monitoring model constructed by the above-mentioned parent model learning process at the above-mentioned first moment as the teacher data, and performs a child model learning process on the second monitoring model, and the second monitoring model uses the above-mentioned second monitoring data at the second moment that is different from the above-mentioned first moment by a transition time as an input parameter.

2. The coordinated learning system according to claim 1, characterized in that: The above-mentioned model learning unit uses the monitoring result of the above-mentioned first monitoring model at the first moment as teacher data, and performs sub-model learning processing on the above-mentioned second monitoring model. The above-mentioned second monitoring model uses the above-mentioned second monitoring data at the second moment that differs from the above-mentioned first moment by a transition time amount and the above-mentioned first monitoring data at the above-mentioned second moment as input parameters.

3. The coordinated learning system according to claim 1 or 2, characterized in that: The third monitoring data of the third process which is upstream or downstream of the second process, and the monitoring result of the second process output by the second monitoring model constructed by the sub-model learning process using the second monitoring data as an input parameter are stored in time series. The above-mentioned model learning unit uses the monitoring result of the above-mentioned second monitoring model at the third moment as teacher data, and performs the above-mentioned sub-model learning processing on the third monitoring model. The third monitoring model uses the above-mentioned third monitoring data at the fourth moment which is different from the above-mentioned third moment by a transition time as an input parameter.

4. The coordinated learning system according to claim 1 or 2, characterized in that: The third monitoring data of the third process which is upstream or downstream of the second process is stored in time series. The above-mentioned model learning unit uses the monitoring result of the above-mentioned first monitoring model at the first moment as teacher data, and performs the above-mentioned sub-model learning processing on the third monitoring model. The third monitoring model uses the above-mentioned third monitoring data at the fifth moment which is different from the above-mentioned first moment by the transition time as input parameters.

5. The coordinated learning system according to claim 1 or 2, characterized in that: The first step is a step that becomes a bottleneck of the entire process among a plurality of steps that are performed continuously in time series at a predetermined transition time interval.

6. A monitoring system for monitoring a plurality of processes that are continuous in time series at predetermined transition time intervals, characterized in that: have: A data collection control device collects each monitoring data outputted from a plurality of processes; A monitoring control device that uses monitoring models set for each of the plurality of processes to output monitoring results based on the monitoring data for each process; and The coordinated learning device performs learning control according to different monitoring models of the above process. The above-mentioned coordinated learning device is, storing, in time series, first monitoring data of a first process, second monitoring data of a second process that is upstream or downstream of the first process, and a monitoring result of the first process output by a first monitoring model using the first monitoring data as an input parameter, Using the above-mentioned first monitoring data, the monitoring results of the above-mentioned first monitoring model and the teacher data pre-produced for the above-mentioned first monitoring model, the above-mentioned first monitoring model is subjected to a parent model learning process, any moment in the above-mentioned first process is set as the first moment, and the monitoring results of the above-mentioned first monitoring model constructed through the above-mentioned parent model learning process at the above-mentioned first moment are used as teacher data to perform a child model learning process on the second monitoring model, and the second monitoring model uses the above-mentioned second monitoring data at the second moment that differs from the above-mentioned first moment by a transition time as an input parameter.

7. The monitoring system according to claim 6, characterized in that The first step is a bottleneck step for the entire process among a plurality of steps that are performed in a time series at a predetermined transition time interval. The above-mentioned multiple processes are water treatment processes.

8. A storage medium storing a program, the program being executed by a computer, the computer performing a coordinated learning process in process monitoring in which monitoring models are respectively provided for a plurality of processes that are continuous in time series at a predetermined transition time interval, the program being used to cause the computer to implement: a first function for storing, in time series, first monitoring data of a first process, second monitoring data of a second process that is upstream or downstream of the first process, and a monitoring result of the first process output by a first monitoring model using the first monitoring data as an input parameter; and The second function uses the above-mentioned first monitoring data, the monitoring results of the above-mentioned first monitoring model and the teacher data pre-produced for the above-mentioned first monitoring model to perform a parent model learning process on the above-mentioned first monitoring model, set any moment in the above-mentioned first process as the first moment, use the monitoring results of the above-mentioned first monitoring model constructed by the above-mentioned parent model learning process at the above-mentioned first moment as the teacher data, and perform a child model learning process on the second monitoring model, and the second monitoring model uses the above-mentioned second monitoring data at the second moment that is different from the above-mentioned first moment by a transition time as an input parameter.

9. A storage medium storing a program, wherein the program is executed by a computer that performs process monitoring of a plurality of processes that are continuous in time series at a predetermined transition time interval, the program causing the computer to implement: The first function is to collect monitoring data outputted from a plurality of processes; A second function of outputting monitoring results based on the monitoring data for each process using monitoring models set for each of the plurality of processes; and The third function is to perform learning control according to different monitoring models in accordance with the above process, and The third function mentioned above is, storing, in time series, first monitoring data of a first process, second monitoring data of a second process that is upstream or downstream of the first process, and a monitoring result of the first process output by a first monitoring model using the first monitoring data as an input parameter, Using the above-mentioned first monitoring data, the monitoring results of the above-mentioned first monitoring model and the teacher data pre-produced for the above-mentioned first monitoring model, the above-mentioned first monitoring model is subjected to a parent model learning process, any moment in the above-mentioned first process is set as the first moment, the monitoring results of the above-mentioned first monitoring model constructed through the above-mentioned parent model learning process at the above-mentioned first moment are used as teacher data, and the second monitoring model is subjected to a child model learning process, and the second monitoring model uses the above-mentioned second monitoring data at the second moment that is different from the above-mentioned first moment by a transition time as an input parameter.

Images