Condition diagnosis support system for water treatment base

Abstract

To shorten the time required for calculating the condition diagnosis of a water treatment base.SOLUTION: The situation diagnosis support system of a water treatment base, comprises a memory memorizing a plurality of unit models, i.e., Bayesian network models corresponding to each of a plurality of water treatment units in the base, a base model generation unit generating the Bayesian network models of the water treatment base, a conditional probability distribution set on the node present in the unit model and a calculation unit calculating a probability distribution based on a probability distribution inputted from another node. The unit model comprises an input port functioning as an assigning point for the probability distribution calculation result, and an output port functioning as an extraction point for the probability distribution calculation result. The base model generation unit connects the output port of the first unit model and the input port of the second unit model to generate the Bayesian network model of the water treatment base. The probability distribution calculation result calculated in the first unit model, is assigned to the second unit model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a status diagnosis support system for a water treatment plant. [Background technology]

[0002] Conventionally, diagnostic systems have been known that detect abnormalities or signs of abnormalities in equipment in power plants, water treatment plants, etc., and instruct maintenance work required for the detected abnormalities. For example, Patent Document 1 describes a plant operation system that monitors signs of abnormalities in a plant based on the plant's operation history, inputs measurement parameters that are determined to have abnormality signs into an abnormality diagnostic system as abnormality sign data, and the abnormality diagnostic system compares the abnormality sign data with abnormality model patterns and uses a Bayesian network as a statistical model to identify the cause of the abnormality.

[0003] When applying Bayesian network technology to the diagnosis of water treatment facilities, a unit model that models the causal relationships for each water treatment unit within the water treatment facility, such as the coagulation tank, reverse osmosis membrane separation device, and electrodeionization device, is created, and the unit model is used to diagnose the status of each water treatment unit.

[0004] However, situation diagnosis using only the causal model of a single water treatment unit was unable to express the causal relationships of the on-site conditions caused by the numerous water treatment units that make up the water treatment facility, making it difficult to accurately diagnose the situation at the water treatment facility.

[0005] Creating a model for each water treatment facility makes it possible to diagnose the situation at that facility, but this requires creating a new model each time the number of facilities to be diagnosed increases, which is costly. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-62729 Summary of the Invention [Problem to be solved by the invention]

[0007] An object of the present invention is to provide a status diagnosis support system for a water treatment plant that can easily express the causal relationships of the on-site status of a water treatment plant. [Means for solving the problem]

[0008] [1] A status diagnosis support system for a water treatment facility having multiple water treatment units, a storage unit that stores a plurality of unit models, which are Bayesian network models corresponding to each of the plurality of water treatment units; a base station model generation unit that generates a Bayesian network model of the water treatment base; a calculation unit that calculates a probability distribution based on a conditional probability distribution set for a node in the unit model and a probability distribution input from another node; Equipped with each of the plurality of unit models has an input port serving as an assignment point of a probability distribution calculation result and an output port serving as an extraction point of the probability distribution calculation result; the base station model generation unit generates a Bayesian network model of the water treatment base station by connecting an output port of a first unit model and an input port of a second unit model; A situation diagnosis support system in which the probability distribution calculation result calculated by the first unit model is substituted into the second unit model.

[0009] [2] The situation diagnosis support system according to [1], wherein the input ports and output ports of the plurality of unit models have a common configuration. [Effects of the Invention]

[0010] According to the present invention, the causal relationships of the on-site conditions of a water treatment plant can be easily expressed. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a schematic configuration diagram of a water treatment plant status diagnosis support system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a schematic diagram of a unit model. [Figure 3] FIG. 1 is a schematic diagram of a unit model connected to ports. [Figure 4] FIG. 10 is a diagram illustrating an example of probability calculation for a port-connected unit model. [Figure 5] FIG. 10 is a diagram illustrating an example of probability calculation for a port-connected unit model. [Figure 6] FIG. 10 is a diagram illustrating an example of probability calculation of a unit model according to a comparative example. DETAILED DESCRIPTION OF THE INVENTION

[0012] The present invention will be described in detail below with reference to the drawings.

[0013] As shown in FIG. 1, a situation diagnosis support system 2 according to this embodiment provides situation diagnosis support for a water treatment facility 1 in which a plurality of water treatment units are installed.

[0014] The water treatment facility 1 has a plurality of water treatment units. The water treatment units are not particularly limited, but may be, for example, a coagulation tank, a turbidity membrane device, an MF membrane (microfiltration membrane) separation device, an RO membrane (reverse osmosis membrane) separation device, an electrodeionization device, etc. The water treatment units are equipped with various sensors that measure the amount of chemicals injected, the amount of water supplied, the amount of treated water, the quality of the treated water, etc.

[0015] The situation diagnosis support system 2 includes a calculation device 20, a storage unit 21, and a display unit 22. The calculation device 20 includes a base model generation unit 200 and a probability distribution calculation unit 201. The calculation device 20 is a computer including a CPU, RAM, ROM, etc.

[0016] The situation diagnosis support system 2 may include a communication unit that communicates with the sensors at the water treatment plant 1. The situation diagnosis support system 2 can acquire sensor measurement values ​​in real time using the communication unit.

[0017] The display unit 22 is a liquid crystal display or the like, and displays the results of calculations by the arithmetic unit 20.

[0018] The memory unit 21 stores a model (causal relationship model) that expresses, in a graph structure, the qualitative causal relationship between the status of various water treatment units installed at the water treatment facility 1 and the observation information. For example, using a Bayesian network, the relationship between the situation (condition) to be detected and the observation information is modeled for each water treatment unit. The observation information includes values ​​measured by sensors installed in the water treatment equipment and the results of analyzing sampled water with an analyzer. Hereinafter, the causal relationship model for the water treatment unit will be referred to as a unit model.

[0019] A Bayesian network expresses probability distributions using frequency distributions based on discretization of numerical values. For example, discretization is expressed using five state variables: LL, L, D, H, and HH. When expressing numerical information, LL has the engineering meaning of very low, L is low, D is design assumption, H is high, and HH is very high.

[0020] When expressing the occurrence of an event, LL has the engineering meaning of being extremely unlikely to have occurred compared to the design assumptions, L being unlikely to have occurred, D being an occurrence at the level of the design assumptions, H being highly likely to have occurred compared to the design assumptions, and HH being extremely likely to have occurred.

[0021] The above discretization is based on an engineering meaning, so there is an order relationship, and it is possible to implement discretization of normalized numerical intervals. For example, the numerical range of 0.0 to 2.0 is taken as the normalized numerical range, and this numerical range is divided into five intervals, with LL corresponding to 0.0 to 0.4, L to 0.4 to 0.8, D to 0.8 to 1.2, H to 1.2 to 1.6, and HH to 1.6 to 2.0.

[0022] Input nodes, situation nodes, observation nodes, etc. are used as logical node components based on the discrete representation of the Bayesian network.

[0023] An input node represents the input information itself, is a node that does not have input from other nodes, and defines the input boundary of the model.

[0024] A situation node is a node that represents the situation to be detected, and represents a situation that can occur in on-site equipment (a trouble event that should not occur in principle). When the values ​​of HH and H in the probability distribution are large, it indicates that the situation is likely to actually occur on-site.

[0025] An observation node is a node that represents observed information. Observation information includes measurements from sensors installed in the water treatment unit and the results of analyzing sampled water using an analytical device. Sensor measurement values ​​include those transmitted online and those visually confirmed on-site.

[0026] For example, in the observation node for treated water flow rate, when the values ​​of HH and H in the probability distribution are large, it means that there is a high probability that the treated water flow meter will indicate a high value. Furthermore, when actual observation information is obtained, a reasonable five-level state variable is determined in accordance with the threshold information set for each node, and a probability distribution is given in which the probability of that state variable is 100%.

[0027] In model construction, a probability distribution (called a "prior probability distribution") for the input node is set. In addition, a conditional probability distribution that expresses the probabilistic relationship with the input node is set for the situation node, observation node, etc.

[0028] The prior probability distribution and the conditional probability distribution may be values ​​based on expert experience and knowledge including subjective opinions, or may be values ​​based on frequency information from a large amount of actual data.

[0029] As shown in Figure 2, a unit model stored in memory unit 21 has an input port that serves as an assignment point (accepts input) for a probability distribution calculated by another unit model, and an output port that serves as an extraction point (outputs) for the probability distribution calculated by the unit model.

[0030] For example, the input port defines "feedwater flow rate," "feedwater pH," "feedwater organic matter amount," etc. The output port defines "treated water flow rate," "treated water pH," "treated water organic matter amount," etc.

[0031] One unit model can be connected to another unit model via ports. For example, as shown in Figure 3, the output port of unit model A is connected to the input port of unit model B. Unit models A and B have a common input and output port configuration, but the model parts other than the input and output ports have their own unique configuration.

[0032] The "treated water flow rate" of the output port of unit model A is connected to the "feed water flow rate" of the input port of unit model B. The "treated water organic matter amount" of the output port of unit model A is connected to the "feed water organic matter amount" of the input port of unit model B. The "treated water pH" of the output port of unit model A is connected to the "feed water pH" of the input port of unit model B.

[0033] In the example unit model B shown in Figure 3, there is no node that uses the probability distribution of feedwater pH in the probability calculation, so the feedwater pH at the input port is directly connected to the treated water pH at the output port.

[0034] The base information in the storage unit 21 stores information such as the connection order of the multiple water treatment units that make up the water treatment base 1. The base model generation unit 200 references the base information, connects corresponding unit models to each other via ports, and generates a causal relationship model (base model) of the water treatment base 1.

[0035] The probability distribution calculation unit 201 modifies the probability distribution for each unit model in the base model based on the prior probability distribution and the new observation information each time observation information is obtained, and calculates the posterior probability distribution. The probability distribution calculation unit 201 also calculates the probability distribution of each node using Bayes' theorem (called the "posterior probability distribution" by inference) based on the probability distribution of the input information and the conditional probability distribution of each node. The probability distribution calculation unit 201 also performs a process (called "Bayesian update") to reflect the posterior probability distribution of the input information in the prior probability distribution.

[0036] The result of probability calculation performed by the probability distribution calculation unit 201 for a certain unit model is propagated to another unit model connected via a port.

[0037] Examples of the propagation of probability calculation results are shown in Figures 4 and 5. In this example, unit model A has an observation node "Amount of organic matter in treated water from water treatment unit A" and a situation node "Poor treatment status of organic matter", and "Amount of organic matter" is defined as the input port and output port.

[0038] Unit model B connected to unit model A has a situation node "large organic matter load in water supply," and "amount of organic matter" is defined for the input port and output port.

[0039] As shown in Figure 5, when the observation node "Amount of organic matter in water treatment unit A treated water" of unit model A shows a high value, the probability of occurrence of the situation node "Poor organic matter treatment status" becomes high (the probability of the state variables H and HH becomes high). The occurrence probability propagates through the output port of unit model A and the input port of unit model B, and as a result, the probability of occurrence of the situation node "High organic matter load in water supply" of unit model B becomes high (the probability of LL to D becomes low, and the probability of H and HH becomes high).

[0040] Figure 6 shows a calculation example when unit model A and unit model B are not connected by port. As can be seen from this figure, even if the occurrence probability of the situation node "Poor organic matter treatment status" of unit model A increases, the occurrence probability of the situation node "High organic matter load in water supply" of unit model B does not increase (the probability distribution does not change). Unit model B is not affected by the diagnosis result (probability distribution) of unit model A.

[0041] In this embodiment, by creating a model (componentizing) on ​​a water treatment unit basis and connecting ports, the causal relationships of the on-site situation at a water treatment facility can be easily expressed and the on-site situation at the facility scale can be diagnosed.

[0042] Multiple unit models are reusable and can be combined to easily represent base models for other water treatment plants.

[0043] The present invention is not limited to the above-described embodiments, and the components can be modified and embodied in practice without departing from the spirit of the invention. Furthermore, various inventions can be created by appropriately combining multiple components disclosed in the above-described embodiments. For example, some components may be omitted from all the components shown in the embodiments. Furthermore, components from different embodiments may be appropriately combined. [Explanation of symbols]

[0044] 1. Water treatment facility 2. Situation diagnosis support system 20 Arithmetic unit 200 Base Model Generation Unit 201 Probability distribution calculation unit

Claims

1. A status diagnosis support system for a water treatment facility having a plurality of water treatment units, a storage unit that stores a plurality of unit models, which are Bayesian network models corresponding to each of the plurality of water treatment units; a base station model generation unit that generates a Bayesian network model of the water treatment base station; a calculation unit that calculates a probability distribution based on a conditional probability distribution set for a node in the unit model and a probability distribution input from another node; Equipped with each of the plurality of unit models has an input port serving as an assignment point of a probability distribution calculation result and an output port serving as an extraction point of the probability distribution calculation result; the base station model generation unit generates a Bayesian network model of the water treatment base station by connecting an output port of a first unit model and an input port of a second unit model; A situation diagnosis support system, wherein the probability distribution calculation result calculated by the first unit model is substituted into the second unit model.

2. 2. The situation diagnosis support system according to claim 1, wherein the input ports and the output ports of the plurality of unit models have a common configuration.

Images