Abnormality diagnosis system for water treatment plant and abnormality diagnosis method for water treatment plant

Abstract

An abnormality diagnosis system (1000) comprises: a measurement value collection unit (101) which collects a plurality of types of measurement values in a water treatment process; a measurement value DB (102) in which the measurement values collected by the measurement value collection unit (101) are accumulated; an abnormality detection unit (103) which determines, on the basis of the degree of abnormality which is calculated from the measurement values, whether or not an abnormality is occurring in the water treatment plant; an identification information storage unit (105) in which clustering information (CLM) for identification that indicates a relation between the degree of abnormality and an abnormality mode is stored; and an abnormality mode identification unit (106) which, when it is determined that the abnormality is occurring, identifies, by comparing the degree of abnormality that has been calculated by the abnormality detection unit (103) and the clustering information (CLM) for identification, the abnormality mode of the abnormality that is occurring, wherein the abnormality mode is associated with the degree of importance with respect to the water treatment plant.

Description

Abnormality diagnosis system for water treatment plant and abnormality diagnosis method for water treatment plant

[0001] The present disclosure relates to an abnormality diagnosis system for a water treatment plant and an abnormality diagnosis method for a water treatment plant.

[0002] In a water treatment process performed at a water treatment plant, it is necessary to achieve both predetermined target performance and energy conservation during operation. To this end, it is necessary to monitor the process status related to the target performance, quickly detect state changes and abnormalities that may hinder the achievement of the target performance, and take proactive measures. However, there are many different causes of abnormalities in water treatment processes, making it difficult to monitor all measured values ​​that may indicate the occurrence of an abnormality. Therefore, there is a process status monitoring device that calculates statistics that serve as indicators of the degree of abnormality in the process using a multivariate statistical process control (MSPC) technique and compares these statistics with a threshold value to determine the presence or absence of an abnormality (see, for example, Patent Document 1).

[0003] JP 2012-138044 A

[0004] Anomalies in water treatment processes can be serious and adversely affect the reaction system of a water treatment plant, while other anomalies are minor and do not disrupt the normal progress of the water treatment process. Operators prioritize addressing serious anomalies. In this regard, the process status monitoring device of Patent Document 1 allows operators to detect the occurrence of an anomaly without monitoring a large number of measured values, but does not provide information on the severity of the anomaly. Patent Document 1 also discloses a method for detecting anomalies triggered by key variables that the operator is well aware of, thereby detecting only the anomaly symptoms that the operator wants to know about. However, the index of anomaly severity in a water treatment process and the severity of the anomaly are not necessarily simply proportional to each other, and it is not easy to accurately identify key variables that correspond to anomalies with high severity.

[0005] The present disclosure discloses a technique for solving the above-mentioned problems, and aims to provide an abnormality diagnosis system and an abnormality diagnosis method for a water treatment plant that can detect the occurrence of an abnormality in a water treatment process and diagnose the severity of the detected abnormality.

[0006] The water treatment plant abnormality diagnosis system disclosed herein is an abnormality diagnosis system for diagnosing abnormalities in a water treatment plant that is a monitored object, and includes a measurement value collection unit that collects multiple types of measurement values ​​in the water treatment process performed by the water treatment plant, a measurement value memory unit that accumulates the measurement values ​​collected by the measurement value collection unit, an abnormality detection unit that determines whether an abnormality has occurred in the water treatment plant based on the degree of abnormality calculated from the measurement values, an abnormality mode discrimination information memory unit that stores abnormality mode discrimination information that indicates the relationship between the degree of abnormality and the abnormality mode, an abnormality mode discrimination unit that, when it is determined that an abnormality has occurred, discriminates the abnormality mode of the abnormality that has occurred by comparing the degree of abnormality calculated by the abnormality detection unit with the abnormality mode discrimination information, and a diagnosis result display unit that displays the results of the discrimination, and the abnormality mode is associated with the severity of the abnormality for the water treatment plant.

[0007] In addition, the abnormality diagnosis method for a water treatment plant disclosed herein is a method for diagnosing abnormalities in a water treatment plant that is a monitored object, and includes a measurement value collection process that collects multiple types of measurement values ​​in the water treatment process by the water treatment plant, a measurement value accumulation process that accumulates the measurement values ​​collected in the measurement value collection process, an abnormality detection process that determines whether or not an abnormality has occurred in the water treatment plant based on the degree of abnormality calculated from the measurement values, an abnormality mode discrimination process that, if it is determined that an abnormality has occurred, discriminates the abnormality mode of the abnormality that has occurred by comparing the degree of abnormality calculated in the abnormality detection process with abnormality mode discrimination information that indicates the relationship between the degree of abnormality and the abnormality mode, and a diagnosis result display process that displays the results of the discrimination, and the abnormality mode is associated with the severity for the water treatment plant.

[0008] According to the water treatment plant abnormality diagnosis system or water treatment plant abnormality diagnosis method disclosed herein, it is possible to detect the occurrence of an abnormality in a water treatment process and diagnose the severity of the detected abnormality.

[0009] 1 is a block diagram showing the configuration of an abnormality diagnosis system for a water treatment plant according to embodiment 1. FIG. 2 is a block diagram showing the configuration of an abnormality detection unit according to embodiment 1. FIG. 3 is a diagram showing an example of discrimination clustering information according to embodiment 1, and a diagram showing a discrimination clustering map. FIG. 4 is a diagram showing an example of a relationship between an abnormality mode and the severity of an abnormality according to embodiment 1. FIG. 5 is a diagram showing an example of the hardware configuration of an abnormality diagnosis device according to embodiment 1. FIG. 6 is a flow chart showing the operation of the abnormality diagnosis system for a water treatment plant according to embodiment 1. FIG. 7 is a flow chart showing an abnormality detection process according to embodiment 1. FIG. 8 is a block diagram showing the configuration of an abnormality diagnosis system for a water treatment plant according to embodiment 2. FIG. 9 is a block diagram showing the configuration of a past abnormality learning device according to embodiment 2. FIG. 10 is a flow chart showing the operation of the abnormality diagnosis system for a water treatment plant according to embodiment 2. FIG. 11 is a flow chart showing a learning process according to embodiment 2. FIG. 12 is a block diagram showing the configuration of an abnormality diagnosis system for a water treatment plant according to embodiment 3. FIG. 13 is a block diagram showing the configuration of a simulation device according to embodiment 3. FIG. 14 is a flow chart showing the operation of the abnormality diagnosis system for a water treatment plant according to embodiment 3.

[0010] Embodiment 1. Embodiment 1 will be described with reference to FIGS. 1 to 7. FIG. 1 is a block diagram showing the configuration of an abnormality diagnosis system for a water treatment plant according to Embodiment 1. Abnormality diagnosis system 1000, i.e., an abnormality diagnosis system for a water treatment plant, includes an abnormality diagnosis device 100 that detects an abnormality occurring in a water treatment process (not shown) of a monitored water treatment plant and determines the abnormality mode of the occurring abnormality, and a measuring device 1001 and a sensor 1002 that are connected to abnormality diagnosis device 100 via a network and measure various characteristic values ​​of the monitored water treatment plant. Abnormality diagnosis device 100 detects an abnormality in the water treatment plant and determines the abnormality mode of the occurring abnormality using the measurement values ​​acquired by measuring device 1001 and sensor 1002, respectively. Note that, although the water treatment plant in Embodiment 1 is basically assumed to be a sewage treatment plant, the system can also be applied to water treatment plants other than sewage treatment plants, such as desalination plants.

[0011] The measuring device 1001 includes various meters (not shown). More specifically, the measuring device 1001 includes meters such as a flow meter indicating the flow rate of each facility and each tank in the sewage treatment plant, a DO meter (dissolved oxygen meter) indicating the water quality of each tank, an MLSS (mixed liquor suspended solid) meter, an NH4 meter (ammonia meter), an ORP meter (oxygen reduction potential meter), a TN meter (total nitrogen meter), and a TP meter (total phosphorus meter). The measuring device 1001 transmits measurement value data D, which is data on various measurement values, to the abnormality diagnosis device 100 via a network. The measurement value data D includes data on multiple types of measurement values. The measurement value data D may also include data indicating the measurement time.

[0012] The sensor 1002 includes various sensors (not shown) that detect various parameters that indicate the water quality of each tank, such as various sensors that detect the ammoniacal nitrogen concentration, nitrate nitrogen concentration, total nitrogen concentration, phosphate-phosphorus concentration, and total phosphorus concentration of the water in each tank. The sensor 1002 transmits data of measurement values ​​acquired by the various sensors as measurement value data D to the abnormality diagnosis device 100 via the network. Note that although the measurement targets of the measuring device 1001 and the sensor 1002 are different, the data of their respective measurement values ​​is both referred to as measurement value data D.

[0013] The abnormality diagnosis device 100 includes a measurement value collection unit 101 that collects measurement value data D, a measurement value database 102 (hereinafter referred to as measurement value DB102) that stores the collected measurement value data D, i.e., a measurement value storage unit, an abnormality detection unit 103 that detects abnormalities occurring in the sewage treatment plant based on the measurement value data D, a contribution factor calculation unit 104 that calculates the contribution degree described below, a discrimination information storage unit 105 that stores discrimination clustering information CLM used to discriminate abnormality modes, i.e., abnormality mode discrimination information storage unit, an abnormality mode discrimination unit 106 that discriminates abnormality modes using the discrimination clustering information CLM, a diagnosis result display unit 107 that displays the results of abnormality mode discrimination to the operator, and a discrimination information generation unit 108 that generates and updates the discrimination clustering information CLM.

[0014] The measurement value collection unit 101 receives measurement value data D from the measuring device 1001 and the sensor 1002, collects the measurement value data D, and outputs the collected measurement value data D to the measurement value DB 102. Note that in the first embodiment, the measurement value data (measurement value data D) is acquired by the measuring device 1001 and the sensor 1002 and transmitted to the abnormality diagnosis device 100 via a network, but the measurement value data may be acquired from the results of a manually performed water quality analysis and input by an operator or the like to the abnormality diagnosis device 100. In this case, the measurement value data is input via, for example, an HMI (Human Machine Interface).

[0015] The measurement value DB 102 is a database that accumulates the measurement value data D by organizing and storing the measurement value data D in a fixed format. The measurement value DB 102 accumulates the measurement value data D by acquiring the measurement value data D from the measurement value collector 101 and sequentially storing the acquired measurement value data D. One method of accumulating the measurement value data D is to store the flow rate and water quality at the same time as the same series. Note that, when there are multiple sensors that each measure the same influent, there is a possibility that there is a difference in measurement time between the multiple sensors. In consideration of this, it is also possible to correct the difference in measurement time between the multiple sensors based on the residence time of the influent in the sewage treatment plant and store the measurement values ​​of each tank for the same influent as the same series.

[0016] The abnormality detection unit 103 detects the occurrence of an abnormality by calculating the degree of abnormality by the MSPC analysis method using the measurement value data D accumulated in the measurement value DB 102. The abnormality detection unit 103 outputs an abnormality presence / absence determination result W indicating the presence or absence of an abnormality to the contribution factor calculation unit 104, the abnormality mode determination unit 106, the diagnosis result display unit 107, and the determination information generation unit 108.

[0017] The anomaly detection unit 103 will be further described. Fig. 2 is a block diagram showing the configuration of the anomaly detection unit according to the first embodiment. The anomaly detection unit 103 includes a dataset acquisition unit 1031 that acquires a dataset DS from the measurement value DB 102, a preprocessing unit 1032 that performs preprocessing on the dataset DS and outputs it as a preprocessed dataset DS*, a principal component analysis unit 1033 that performs PCA analysis (Principal Component Analysis) on the preprocessed dataset DS* and generates principal component axis information PA, an anomaly degree calculation unit 1034 that calculates an anomaly degree WD using the principal component axis information PA and the preprocessed dataset DS*, a threshold setting unit 1035 that sets a threshold value Th for determining an anomaly, and an anomaly occurrence determination unit 1036 that determines whether or not an anomaly exists based on the relationship between the anomaly degree WD and the threshold value Th.

[0018] The dataset acquisition unit 1031 acquires, for example, a combination of measurement value data D of the same series (a combination of measurement value data D measured at the same time) as one dataset DS from the measurement value data D accumulated in the measurement value DB 102. The dataset DS acquired by the dataset acquisition unit 1031 differs between the case of PCA analysis (described later) and the case of calculating the degree of anomaly. The dataset DS for PCA analysis is desirably a dataset of a certain period or more that does not include data when an abnormality occurs. As an example in the case of a sewage treatment plant, only measurement value data D measured on sunny days during one month of steady-state operation is extracted from the measurement value DB 102 to create a dataset for PCA. When calculating the degree of anomaly, all datasets DS whose measurement times fall within the period targeted for anomaly detection are acquired.

[0019] The preprocessing unit 1032 acquires the dataset DS from the dataset acquisition unit 1031 and performs preprocessing on the dataset DS. This preprocessing includes processes such as complementing missing values ​​in the dataset DS, excluding outliers, standardization, and moving averaging. The preprocessing unit 1032 outputs the preprocessed dataset DS*, which has been subjected to the above-described preprocessing, to the principal component analysis unit 1033 and the degree of anomaly calculation unit 1034.

[0020] The principal component analysis unit 1033 performs PCA analysis on the dataset DS*. PCA analysis is a multivariate analysis method aimed at extracting data features and reducing the dimensionality of data. In PCA analysis, multiple variables are linearly combined to generate composite variables called principal components in order to capture correlations between variables. Because principal components are linear combinations of the original variables, when considering the data of the original variables, principal component data can be obtained from the original data. The first principal component is the direction in which the variance of the principal component data is maximized, and the first principal component axis is set in this direction. Note that the first principal component is the direction that best expresses the variation in the data. The second principal component is the direction in which the variance of the principal component data is maximized among directions perpendicular to the first principal component, and the second principal component axis is set in this direction. The third and subsequent principal components are the directions in which the variance of the principal component data is maximized among directions perpendicular to all of the previously set principal component axes. By sequentially determining the first principal component, the second principal component, the third principal component, and so on as described above, N principal component axes (the first to N-th principal component axes) are set for N-dimensional data. The principal component analysis unit 1033 outputs information about each principal component axis, such as a conversion formula from the original variables, as principal component axis information PA. Note that, although PCA analysis is used in the first embodiment, methods other than PCA analysis may be used as long as the data processing enables calculation of the degree of anomaly WD, which will be described later.

[0021] The anomaly degree calculation unit 1034 acquires the principal component axis information PA from the principal component analysis unit 1033, acquires the data set DS* to be subjected to anomaly detection from the pre-processing unit 1032, and calculates the anomaly degree WD using the principal component axis information PA and the data set DS*. The "anomaly degree" in the first embodiment is calculated by using two types of statistics described below, namely, T ２ Statistics (hereinafter T ２ The anomaly calculation unit 1034 calculates the T ２ The abnormality degree calculation unit 1034 calculates the calculated T ２ The data of the value and the Q value is output to the abnormality occurrence determination unit 1036 as the abnormality degree WD.

[0022] T ２The value is calculated by (Equation 1). In (Equation 1), t ｒ indicates the principal component score of the rth component, and σt ｒ is ｒ is the standard deviation of T shown in (Equation 1). ２ The value indicates the square of the distance from the origin of the data in the principal component space normalized by the square of the standard deviation of each principal component score. Note that the principal component space is a space formed by the principal component axes of the principal components under consideration among the principal components obtained by principal component analysis. In the case of (Equation 1), R principal components are considered, and the principal component space is formed by the principal component axes from the first principal component to the Rth principal component. The principal component score is the coordinate value on the principal component axis in the principal component space.

[0023] The Q value is calculated by (Equation 2). In (Equation 2), x ｐ indicates the p-th component of the measurement data D, which is the original data, and x ｐ is x ｐ This is the p-th component when dimensionality reduction by PCA analysis and reconstruction on P dimensions (inverse mapping from the space after dimensionality reduction) are performed on the original data. Note that if the original data is N-dimensional, P = N. The Q value indicates the deviation from the original data that occurs even when inverse mapping is performed from the space after dimensionality reduction, and is a value that represents the information lost as a result of dimensionality reduction by PCA analysis as a residual. This represents the part of the original data that cannot be expressed by the PCA model.

[0024] In MSPC, T ２ The T value and the Q value are managed independently, and if either value exceeds the control limit, it is determined that an abnormality has occurred. ２ The T value and Q value represent the degree of abnormality. ２ The Q value and the Q value are complementary to each other, and by managing each independently, it is possible to manage both the parts that have correlation and the parts that do not have correlation in the variable space when the measurement value data is used as a variable.

[0025] The threshold value setting unit 1035 sets a threshold value Th for determining an abnormality and outputs the threshold value Th to the abnormality occurrence determination unit 1036. The threshold value Th is generally set as T２ The value of the Q value is given as "m+3σ", where m and σ are the anomaly levels (T ２ The threshold Th indicates the average value and standard deviation of the degree of abnormality when the degree of abnormality (Q value and Q value) is calculated, and is the average value and standard deviation obtained when the degree of abnormality is calculated for a set of normal data that does not include data when an abnormality occurs. ２ Threshold value Th_T ２ and a threshold value Th_Q for the Q value, both of which are collectively referred to as the threshold value Th. Note that the threshold value Th is not limited to the value determined as described above, and may be another value calculated independently depending on the purpose.

[0026] The abnormality occurrence determination unit 1036 determines whether or not an abnormality has occurred based on the relationship between the abnormality degree WD at the current time calculated from the data set DS* at the current time (data set DS* configured by the measurement value data D at the current time). ２ If either the value or the Q value exceeds the threshold value Th, that is, if ２ value)>(T ２ Threshold value Th_T ２ ) and (current Q value) > (Q ２ If any one of the thresholds Th_Q for the abnormality value is satisfied, the abnormality occurrence determination unit 1036 determines that an abnormality has occurred. That is, the abnormality occurrence determination unit 1036 determines whether or not an abnormality has occurred by comparing the abnormality degree WD with the threshold Th. The abnormality occurrence determination unit 1036 outputs the principal component axis information PA, which is the result of the principal component analysis, the calculated abnormality degree WD, and the result of the abnormality occurrence determination together as an abnormality presence / absence determination result W. That is, the abnormality presence / absence determination result W also includes the principal component axis information PA and the abnormality degree WD.

[0027] It should be noted that the method is not limited to using only the abnormality degree WD at the current time for the determination, and it is also possible to perform the determination using the threshold value Th as described above for the abnormality degree WD for a certain period of time in the recent past (for example, the most recent 5 minutes). In this case, the time (T ２It is also possible to determine that an abnormality has occurred if the degree of abnormality WD exceeds the threshold value Th for a certain period of time.

[0028] The contribution factor calculation unit 104 calculates the contribution of each measurement value to the occurring abnormality. The "contribution" indicates the degree of influence that each measurement value has on the fluctuation of the abnormality degree WD, and is T ２ The contribution factor calculation unit 104 acquires a data set DS from the measurement value DB 102, acquires an abnormality presence / absence determination result W from the abnormality detection unit 103, and, if the abnormality presence / absence determination result W indicates the occurrence of an abnormality, calculates the contribution factor using the measurement value data D for the current time included in the data set DS.

[0029] T ２ The contribution CpT to the value is calculated by the following (Equation 3). On the left side of (Equation 3), the parenthesized T ２ Cp with T ２ On the right side of (Equation 3), t represents the principal component vector. Σ represents the variance-covariance matrix, and the "-1" in the upper right corner indicates that it is an inverse matrix. ｐ is the data of factor p, that is, the pth component of the measurement value data D, which is the original data. ｐ is the transpose vector of the p-th row vector of the loading matrix. ２ The greater the contribution CpT to the value, the greater the T ２ The value is the threshold value Th_T ２ When the measured value (measured value of the p-th component) exceeds the threshold, there is a high possibility that the measured value (measured value of the p-th component) is the cause of the abnormality.

[0030] The contribution CpQ to the Q value is calculated by the following (Equation 4). On the left side of (Equation 4), Cp with Q in parentheses indicates the value of the contribution CpQ to the Q value. ｐ indicates the p-th component of the measurement data D, which is the original data, and x ｐ is x ｐThis is the p-th component when dimension reduction by PCA analysis and reconstruction on P dimensions (inverse mapping from the space after dimension reduction) are performed on the Q-value. The greater the contribution CpQ to the Q-value, the greater the possibility that the measurement value (measurement value of the p-th component) is the cause of an abnormality when the Q-value exceeds the threshold Th_Q.

[0031] The contribution factor calculation unit 104 identifies the factor p that ranks highest in the order of magnitude of the contribution CpT and contribution CpQ calculated for each factor p, and sets this as the causal possibility factor p*. The number of factors p identified as the causal possibility factor p* is arbitrary. The contribution factor calculation unit 104 outputs the data on the contribution CpT and contribution CpQ, and the data on the causal possibility factor p* as contribution data Cp.

[0032] The discrimination information storage unit 105 stores discrimination clustering information CLM. ２ This is information that associates the abnormality degree values ​​such as the Q value and the T value with the abnormality mode of the abnormality that actually occurs. ２ The horizontal axis is the T value and the vertical axis is the Q value. ２ The three plots shown in Fig. 3, i.e., square, diamond, and triangle plots, represent the T ２ The discrimination clustering information CLM is generated and updated by the discrimination information generating unit 108, which will be described later.

[0033] In FIG. 3, the plot of the star indicates the current time T ２ The graph plots the Q-value and the Q-value. In the example shown in FIG. 3, the square plot area is cluster CL1, the diamond plot area is cluster CL2, and the triangular plot area is cluster CL3. Clusters are generated from each plot using a clustering method such as the k-means method. The dotted line boundary BD1 indicates the boundary between cluster CL1 and cluster CL2. Similarly, boundary BD2 indicates the boundary between cluster CL2 and cluster CL3, and boundary BD3 indicates the boundary between cluster CL3 and cluster CL1.

[0034] 4 is a diagram illustrating an example of the relationship between the abnormality mode and the severity of the abnormality according to the first embodiment. ２ When the abnormal mode indicated by cluster CL1, which has a large value and a small Q value, is defined as mode 1, the abnormal mode of cluster CL1 (mode 1) is an abnormal mode that has a large degree (severity) of adverse impact on the sewage treatment plant. ２ When the abnormal mode indicated by cluster CL2, which has a medium value and a large Q value, is called mode 2, the abnormal mode (mode 2) of cluster CL2 is an abnormal mode with a low degree of severity. ２ When the abnormality mode indicated by cluster CL3, in which the Q value and Q value are small, is designated as mode 3, the abnormality mode (mode 3) of cluster CL3 is an abnormality mode of medium severity. Note that the relationship between each cluster and severity as described above is merely an example. For example, it is conceivable to designate an abnormality mode of high severity only when the Q value exceeds a predetermined threshold. The setting of the abnormality mode is also an example. The abnormality mode can be set arbitrarily as long as it corresponds to the severity, and the number of levels of severity (the number of clusters) to be set can also be set arbitrarily.

[0035] In a sewage treatment plant, examples of abnormalities with a high severity include severe process abnormalities such as abnormalities in the flow rate or components of influent water, abnormal microbial activity in a biological reactor, abnormalities in operating equipment such as pumps and agitators, and abnormalities in water quality meters subject to legal regulations for treated water. Examples of abnormalities with a low severity include abnormalities in water quality meters that are not subject to legal regulations, abnormalities in water quality meters that can be replaced by manual analysis, and deviations in measurement values ​​due to sensor contamination. In the first embodiment, a clustering map such as that shown in FIG. 3 is generated as the discrimination clustering information CLM, but the discrimination clustering information CLM can also be generated by interviewing operators or extracting know-how from operational information.

[0036] The abnormality mode determination unit 106 determines the abnormality mode of the abnormality when an abnormality occurs based on the abnormality presence / absence determination result W and the clustering information for determination CLM. In the example of FIG. 3, as described above, the plot of the star symbol indicates the current time T２ The plots of stars indicate the Q value and the Q value, and since the star plots are included in the area of ​​cluster CL2, the abnormality mode of the current abnormality is determined to be the abnormality mode of cluster CL2 (an abnormality mode with a low severity). In this way, by comparing the calculated abnormality degree of the current abnormality with the discrimination clustering information CLM, it is possible to discriminate the abnormality mode of a newly detected abnormality at the current time. The abnormality mode discrimination unit 106 outputs the result of the abnormality mode discrimination as an abnormality mode discrimination result M.

[0037] Furthermore, the abnormality mode discrimination unit 106 of the first embodiment narrows down the abnormality modes based on the above-mentioned contribution degree, and after narrowing down the abnormality modes with high possibility, it is also possible to perform discrimination using the discrimination clustering information CLM. More specifically, the severity is roughly predicted from the magnitude of the cause possibility factor p* and the contribution degree CpT and contribution degree CpQ of the cause possibility factor p*. For example, if it is known that an abnormality in which factor 1 (p=1) contributes greatly is of high severity, and if p=1 is also included in the cause possibility factor p*, it is known that the abnormality is likely to be of high or medium severity. After narrowing down the abnormality modes in this way, the current time T ２ The abnormal mode is determined by comparing the Q value and the Q value with the discrimination clustering information CLM. Note that, although the abnormal mode is narrowed down by a combination of the causal possibility factor p* and its contribution degree in the above, narrowing down may be performed by the contribution degree value alone. Also, narrowing down the abnormal mode is not essential. If narrowing down is not performed, the contribution factor calculation unit 104 may be omitted.

[0038] When narrowing down the candidates based on the contribution level as described above, the accuracy of abnormal mode discrimination can be improved by combining this with abnormal mode discrimination based on comparison with the discrimination clustering information CLM. When comparing only with the discrimination clustering information CLM, it may be difficult to determine the cluster to which the candidate belongs, depending on the current abnormality level value. For example, in the example of FIG. 3, if a star symbol is located on or near the boundary BD1, it may be difficult to determine whether the candidate belongs to cluster CL1 or cluster CL2. This problem is more likely to occur when abnormal modes are classified more finely. Here, by combining narrowing down the candidates based on the contribution level, it is possible to accurately determine the abnormal mode even when it is difficult to determine the abnormal mode based on comparison with the discrimination clustering information CLM alone, thereby improving the accuracy of abnormal mode discrimination.

[0039] The diagnostic result display unit 107 is configured with a display device such as a liquid crystal display, and displays the determination result of whether or not an abnormality has occurred, the abnormality mode of the occurring abnormality, and its severity, depending on the contents of the abnormality presence / absence determination result and the abnormality mode determination result M. The diagnostic result display unit 107 may also display values ​​calculated in the process of determining whether or not an abnormality has occurred and determining the abnormality mode, such as an alert about the abnormality, the abnormality degree, contribution rate, and causal probability factors. If the abnormality mode determination determines that the occurring abnormality is a severe process abnormality, the diagnostic result display unit 107 may display a message requesting the operator to prioritize addressing the abnormality. Note that, although the abnormality diagnosis device 100 is configured to include the diagnostic result display unit 107 in the first embodiment, the diagnostic result display unit 107 may be configured to be included in a device separate from the abnormality diagnosis device 100. In this case, the abnormality diagnosis device 100 may include a transmission unit that transmits the abnormality mode determination result M to the separate device, for example, via a network.

[0040] The discrimination information generation unit 108 generates discrimination clustering information CLM and updates the discrimination clustering information CLM as necessary. The generation and updating of the discrimination clustering information CLM is performed based on the relationship between the abnormality mode of an abnormality that has occurred in actual operation and the abnormality degree WD at the time of the abnormality occurrence. The discrimination information generation unit 108 acquires an abnormality presence / absence determination result W from the anomaly detection unit 103, and generates discrimination clustering information CLM based on the relationship between the abnormality degree WD included in the abnormality presence / absence determination result W and the abnormality mode of the abnormality that has occurred. Here, since the abnormality mode corresponds to the severity of the abnormality, the discrimination information generation unit 108 generates discrimination clustering information CLM based on the relationship between the abnormality degree WD acquired in actual operation and the severity of the abnormality having this abnormality degree WD.

[0041] When the amount of data in actual operation is small, such as immediately after the start of actual operation, the accuracy of the discrimination clustering information CLM is insufficient. In such cases, the abnormal mode is determined using another method, or a temporary discrimination clustering map is used to accumulate data on abnormal mode discrimination results, and the discrimination clustering information CLM is generated when a certain number of results have been obtained. Once the discrimination clustering information CLM is generated, the discrimination information generation unit 108 stores the generated discrimination clustering information CLM in the discrimination information storage unit 105. Thereafter, the abnormal mode is determined using the discrimination clustering information CLM stored in the discrimination information storage unit 105. However, the discrimination information generation unit 108 acquires the abnormal mode discrimination result M each time, and updates the discrimination clustering information CLM accordingly if an update is necessary, such as when a problem is found in the accuracy of the abnormal mode discrimination result M. Note that the input of the abnormal mode discrimination result M to the discrimination information generation unit 108 is not shown in FIG. 1 .

[0042] The update of the clustering information for discrimination CLM will be described. Generally, when discrimination is performed using clustering information, the more data that is used to generate each cluster, the higher the accuracy of discrimination. Conversely, when the number of collected data is small, the accuracy of discrimination is relatively low. In the first embodiment, as described above, T ２The abnormal mode is determined by comparing the Q value and the Q value with the discrimination clustering information CLM. However, there are cases where the result of determination using the discrimination clustering information CLM differs from the abnormal mode determination result based on interviews with operators or know-how extracted from operation information, and the abnormal mode determination result based on interviews with operators, etc. is evaluated to be more correct. In such cases, there is a problem with the accuracy of the abnormal mode determination result M, and it is also evaluated that there is a problem with the accuracy of the discrimination clustering information CLM. For this reason, the discrimination clustering information CLM is updated so that it matches the abnormal mode determination result based on interviews with operators, etc. As in the case of generating the discrimination clustering information CLM described above, the updating of the discrimination clustering information CLM is performed based on the relationship between the abnormality degree WD obtained in actual operation and the severity of the abnormality having this abnormality degree WD.

[0043] Next, a hardware configuration for realizing each functional unit of the abnormality diagnosis device 100 will be described. FIG. 5 is a diagram illustrating an example of the hardware configuration of the abnormality diagnosis device according to the first embodiment. The abnormality diagnosis device 100 is mainly composed of a processor 91, a memory 92 serving as a main storage device, and an auxiliary storage device 93. The processor 91 is composed of, for example, a central processing unit (CPU), an application specific integrated circuit (ASIC), a digital signal processor (DSP), or a field programmable gate array (FPGA). The memory 92 is composed of a volatile storage device such as a random access memory, and the auxiliary storage device 93 is composed of a nonvolatile storage device such as a flash memory or a hard disk. The auxiliary storage device 93 stores predetermined programs to be executed by the processor 91. The processor 91 reads and executes these programs as needed to perform various arithmetic operations. At this time, the predetermined program is temporarily saved from the auxiliary storage device 93 to the memory 92, and the processor 91 reads the program from the memory 92. The arithmetic processing by each functional unit shown in Figures 1 and 2 is realized by the processor 91 executing the predetermined program as described above. The results of the arithmetic processing by the processor 91 are temporarily stored in the memory 92, and then stored in the auxiliary storage device 93 according to the purpose of the executed arithmetic processing.

[0044] The abnormality diagnosis device 100 also includes a receiving circuit 94 that receives measurement value data D from the measuring device 1001 and the sensor 1002, and a display device 95 that displays the abnormality mode determination result, etc. The processor 91, memory 92, auxiliary storage device 93, receiving circuit 94, and display device 95 are connected via a system bus 96. If the abnormality mode determination result, etc. is to be displayed on another device, the receiving circuit 94 can be replaced with a transmitting / receiving circuit, or a separate transmitting circuit can be added, and the abnormality mode determination result M, etc. can be transmitted via the transmitting / receiving circuit or the transmitting circuit.

[0045] Next, the operation will be described. Fig. 6 is a flow diagram showing the operation of the anomaly diagnosis system for a water treatment plant in embodiment 1. It is assumed that the discrimination clustering information CLM has already been generated. First, measurement value data D, which is data on various measurement values ​​indicating the state of the sewage treatment plant to be monitored, is acquired by the measuring device 1001 and the sensor 1002, thereby collecting the measurement value data D (step ST101: measurement value collection step).

[0046] Next, the collected measurement value data D is stored while being organized in a certain format, thereby accumulating the measurement value data D (step ST102: measurement value accumulation step). The measurement value data D is accumulated by sequentially storing the collected measurement value data D. The method for accumulating the measurement value data D is as described above.

[0047] Next, the degree of anomaly is calculated from the accumulated measurement value data D to detect the occurrence of an anomaly (step ST103: anomaly detection step). In the anomaly detection step, the occurrence of an anomaly is detected by the MSPC analysis technique. FIG. 7 is a flow diagram showing the anomaly detection step according to the first embodiment, and shows the MSPC analysis flow according to the first embodiment. First, a combination of measurement value data D of the same series is acquired as one data set DS (step ST1031: data set acquisition step), and preprocessing is performed on the data set DS to obtain a preprocessed data set DS* (step ST1032: preprocessing step). The preprocessing of the data set DS is as described above.

[0048] Next, PCA analysis is performed on the data set DS* (step ST1033: principal component analysis step). The PCA analysis is as described above, and principal component axis information PA is obtained as information on each principal component axis.

[0049] Next, the degree of abnormality is calculated using the data set DS* that is the target of abnormality detection and the principal component axis information PA (step ST1034: abnormality degree calculation step). The degree of abnormality is as described above. In the first embodiment, T ２ The value and Q value were calculated, and the calculated T ２ The data of the value and the Q value is defined as the abnormality degree WD.

[0050] Next, a threshold value Th for determining an abnormality is set (step ST1035: threshold value setting step). The specific method for setting the threshold value Th is as described above.

[0051] Next, the presence or absence of an abnormality is determined based on the relationship between the abnormality degree WD at the current time and the threshold value Th, and an abnormality determination result W is obtained (step ST1036: abnormality occurrence determination step). ２ It is determined that an abnormality has occurred when either the value or the Q value exceeds the threshold value Th. As described above, it is also possible to determine whether an abnormality has occurred based on whether the period of time during which the degree of abnormality WD exceeded the threshold value Th or did not exceed it was longer within the most recent fixed period of time, or whether the degree of abnormality WD has exceeded the threshold value Th for a fixed period of time in succession.

[0052] When determining whether an abnormality has occurred in real time in an operating sewage treatment plant, it is possible to use different execution periods for the principal component analysis process and threshold setting process and the other processes (dataset acquisition process, preprocessing process, anomaly degree calculation process, and anomaly occurrence determination process). For example, the former process is performed in a first execution period (e.g., 24 hours), and the latter process is performed in a second execution period (e.g., 1 minute) that is shorter than the first execution period. In this case, once the calculation of principal components by principal component analysis and the setting of thresholds are performed, the calculation of anomalies and the determination of whether an abnormality has occurred are performed at each time (every 1 minute, which is the second execution period) until the next execution timing (24 hours after the first execution period). In this way, by varying the execution period depending on the content of the process, the amount of calculation in the anomaly detection process can be reduced, and the analysis by MSPC can be performed faster.

[0053] As a result of the above-described anomaly detection process, if it is determined that "anomaly exists," the process proceeds to step ST105, and if it is determined that "no anomaly exists," the process proceeds to step ST107 (step ST104).

[0054] If it is determined in the anomaly detection step that "anomaly exists," the contribution of each measurement value is calculated for the anomaly that has occurred, and contribution data Cp is obtained (step ST105: contribution factor calculation step). The "contribution" is as described above, and indicates the degree of influence that each measurement value has on the fluctuation of the degree of anomaly WD. In the contribution factor calculation step, the cause possibility factor p*, which is a factor p with a large contribution, is also identified. The method for identifying the cause possibility factor p* is as described above, and T ２ After calculating the contribution of each factor p for each of the Q value and the Q value, the factor p ranked highest in the order of the magnitude of the contribution value is designated as the possible cause factor p*.

[0055] Next, if an abnormality has occurred, the abnormality mode of the abnormality is determined based on the abnormality presence / absence determination result W and the discrimination clustering information CLM, and an abnormality mode discrimination result M is obtained (step ST106: abnormality mode discrimination step). As described above, in the first embodiment, the abnormality mode is discriminated using the discrimination clustering map shown in FIG. 3. The specific discrimination method is as described above. Also, as described above, the abnormality modes may be narrowed down based on the contribution degree before discrimination in the abnormality mode discrimination step. If narrowing down based on the contribution degree is performed, it is performed between steps ST105 and ST106.

[0056] Next, the presence or absence of an abnormality, the abnormality mode of the abnormality that has occurred, and its severity are displayed according to the contents of the abnormality presence / absence determination result W and the abnormality mode determination result M (step ST107: diagnostic result display step). As described above, in the diagnostic result display step, an alert about the abnormality, the abnormality degree, the contribution degree, the possible causal factors, etc. may be displayed.

[0057] Next, the discrimination clustering information CLM is updated as necessary (step ST108: discrimination information updating step). The discrimination clustering information CLM is updated as described above. Note that although Fig. 6 shows that the discrimination information updating step is performed after the diagnosis result displaying step, the order of the two steps may be reversed, or these two steps may be performed simultaneously.

[0058] According to the first embodiment, it is possible to detect the occurrence of an abnormality in a water treatment process and diagnose the severity of the detected abnormality. More specifically, the system includes a measurement value collector that collects multiple types of measurement values ​​in the water treatment process at the water treatment plant, a measurement value memory that accumulates the measurement values ​​collected by the measurement value collector, an abnormality detector that determines whether an abnormality has occurred in the water treatment plant based on an abnormality degree calculated from the measurement values, a discrimination information memory that stores discrimination clustering information indicating the relationship between the abnormality degree and an abnormality mode, an abnormality mode discriminator that, when it is determined that an abnormality has occurred, discriminates the abnormality mode of the occurring abnormality by comparing the abnormality degree calculated by the abnormality detector with the discrimination clustering information, and a diagnosis result display unit that displays the result of the abnormality mode discrimination, and the abnormality mode is associated with a severity for the water treatment plant.

[0059] When it is determined that an abnormality has occurred, the abnormality mode of the occurring abnormality is determined by comparing the abnormality degree calculated by the abnormality detection unit with the discrimination clustering information, and the abnormality mode is associated with the severity of the abnormality for the water treatment plant. By determining the abnormality mode, the severity, which is the extent of the adverse effect that the occurring abnormality has on the water treatment plant, is also diagnosed.

[0060] Furthermore, the user who is notified of the abnormality detection through the display of the diagnosis result by the diagnosis result display unit can also understand the severity of the abnormality and can quickly determine whether or not the abnormality should be addressed as a priority, thereby improving the efficiency of maintenance management of the water treatment plant.

[0061] Second Embodiment Next, a second embodiment will be described with reference to FIGS. 8 to 11. Components identical to or corresponding to those shown in FIGS. 1 to 7 are designated by the same reference numerals, and their description will be omitted. The second embodiment differs from the first embodiment in that discrimination clustering information is generated by machine learning before the start of actual operation. FIG. 8 is a block diagram showing the configuration of an abnormality diagnosis system for a water treatment plant in the second embodiment. The abnormality diagnosis system 2000 includes an abnormality diagnosis device 100 similar to that of the first embodiment, as well as a past anomaly learning device 900 connected to the abnormality diagnosis device 200 via a network. The past anomaly learning device 900 generates discrimination clustering information CLM_P from information on past anomalies and transmits the generated discrimination clustering information CLM_P to a discrimination information storage unit 105 of the abnormality diagnosis device 200. The discrimination information storage unit 105 stores the discrimination clustering information CLM_P acquired from the past anomaly learning device 900. The discrimination clustering information CLM_P is used as discrimination clustering information CLM in actual operation to discriminate abnormal modes.

[0062] The abnormality diagnosis device 200 has a configuration basically similar to that of the abnormality diagnosis device 100 of the first embodiment, but differs from the abnormality diagnosis device 100 in that it does not have the discrimination information generation unit 108. That is, in the second embodiment, it is not necessary to generate and update the discrimination clustering information CLM during actual operation. Another difference from the first embodiment is that the discrimination clustering information CLM is also sent to the abnormality detection unit 103. Strictly speaking, the discrimination clustering information CLM is sent to the threshold setting unit 1035 of the abnormality detection unit 103. The discrimination clustering information CLM of the second embodiment is discrimination clustering information CLM_P based on past data, and includes information on clusters corresponding to various abnormalities that have occurred in the past. Therefore, it is possible to set a threshold value Th corresponding to each cluster. In other words, it is possible to set a threshold value Th corresponding to each abnormality. This enables more accurate abnormality detection. Furthermore, since the threshold value Th can be set based on the discrimination clustering information CLM that has already been generated, it is possible to set a threshold value Th corresponding to each abnormality as in the first embodiment. ２There is no need to calculate the value of "m+3σ" for each of the Q value and the Q value. This reduces the calculation load when detecting an anomaly.

[0063] The past anomaly learning device 900 applies the MSPC method to data from past anomalies to calculate the degree of anomaly (T ２ The past anomaly learning device according to the second embodiment calculates the degree of anomaly (Q value and Q value) and generates the discrimination clustering information CLM by applying a machine learning technique to the calculated degree of anomaly. The past anomaly learning device 900 includes a past anomaly DB 901 that accumulates data from past anomaly occurrences, i.e., a past anomaly storage unit; a dataset acquisition unit 902 that acquires a dataset DS_P from the past anomaly DB 901; a preprocessing unit 903 that performs preprocessing on the dataset DS_P and outputs it as a preprocessed dataset DS_P*; a principal component analysis unit 904 that performs PCA analysis on the preprocessed dataset DS_P* and generates principal component axis information PA_P; a past anomaly degree calculation unit 905 that calculates a degree of past anomaly WD_P using the principal component axis information PA_P and the preprocessed dataset DS_P*; and a past anomaly clustering unit 906 that performs machine learning clustering on the degree of past anomaly WD_P to associate the degree of past anomaly WD_P with an abnormality mode and generate discrimination clustering information CLM_P.

[0064] The past abnormality DB 901 stores various measurement value data D_P (not shown) when abnormalities have occurred in the past. The past abnormality DB 901 also associates each stored measurement value data D_P with the abnormality mode of the abnormality that occurred. In other words, each piece of measurement value data D_P stored in the past abnormality DB 901 is labeled with an abnormality mode.

[0065] The dataset acquisition unit 902 acquires, as one dataset DS_P, for example, a combination of measurement value data D_P of the same series (a combination of measurement value data D_P measured at the same time) from the measurement value data D_P stored in the past abnormality DB 901. Since the measurement value data D_P constituting the dataset DS_P is labeled with an abnormality mode, the dataset DS_P is also similarly labeled with an abnormality mode.

[0066] The preprocessing unit 903 acquires the dataset DS_P from the dataset acquisition unit 902 and performs preprocessing on the dataset DS_P. The contents of the preprocessing are the same as those in embodiment 1. The preprocessing unit 903 outputs the preprocessed dataset DS_P* to the principal component analysis unit 904 and the past anomaly degree calculation unit 905.

[0067] The principal component analysis unit 904 performs PCA analysis on the data set DS_P*. The PCA analysis is the same as in the first embodiment, and determines a first principal component axis and a second principal component axis from the multiple measurement value data D_P that make up the data set DS_P*. The principal component analysis unit 904 outputs information about each principal component axis as principal component axis information PA_P.

[0068] The past abnormality degree calculation unit 905 acquires the principal component axis information PA_P from the principal component analysis unit 904, acquires the data set DS_P* to be subjected to abnormality detection from the pre-processing unit 903, and calculates the past abnormality degree WD_P using the principal component axis information PA_P and the data set DS_P*. The calculation of the "past abnormality degree" in the second embodiment is performed by calculating T for the measurement value data D_P, i.e., the measurement value data at the time of the occurrence of an abnormality in the past, in the same manner as the calculation of the "abnormality degree" in the first embodiment. ２ The past abnormality degree calculation unit 905 calculates the calculated past abnormality degree (T ２ The data (value and Q value) is output as the degree of past abnormality WD_P to the past abnormality clustering unit 906. Note that, as can be seen from the fact that the degree of past abnormality WD_P is calculated from the measurement value data D_P (the original data of the degree of past abnormality WD_P is the measurement value data D_P), the degree of past abnormality WD_P is also associated with the abnormality mode.

[0069] The past anomaly clustering unit 906 acquires the past anomaly degree WD_P from the past anomaly degree calculation unit 905, and performs machine learning such as unsupervised learning represented by the k-means method or supervised learning represented by decision tree analysis using the past anomaly degree WD_P as an input, thereby obtaining the T ２ The past abnormality clustering unit 906 clusters the relationship between the Q value and the past abnormality degree WD_P, i.e., the relationship between the past abnormality degree WD_P and the abnormality mode, to generate clustering information for discrimination CLM_P. The past abnormality clustering unit 906 transmits the generated clustering information for discrimination CLM_P to the discrimination information storage unit 105 of the abnormality diagnosis device 200. Note that the abnormality modes associated with the past abnormality degree WD_P in generating the clustering information for discrimination CLM_P are abnormality modes of abnormalities that occurred in the past, and their severity levels are also known. In other words, in generating the clustering information for discrimination CLM_P, the past abnormality degree WD_P is associated with known past abnormalities (whose severity levels are also known).

[0070] Note that the contribution degree of data when an abnormality occurred in the past may be calculated and the calculated contribution degree may be associated with the abnormality mode using a method similar to that described in embodiment 1. Information regarding the association between the contribution degree of data when an abnormality occurred in the past and the abnormality mode is added to the discrimination clustering information CLM_P.

[0071] The hardware configuration of the past abnormality learning device 900 is the same as that of the abnormality diagnosis device 100 .

[0072] Various measurement values ​​in water treatment plants often have daily or seasonal periodicity. Therefore, when generating the discrimination clustering information CLM_P using data from past abnormality occurrences, the accuracy of anomaly detection and abnormality mode discrimination in actual operation can be improved by generating the discrimination clustering information CLM_P using data from the same time period within a day or the same season within a year. For example, when anomaly detection and abnormality mode discrimination in actual operation are performed in the spring, it is desirable to generate the discrimination clustering information CLM_P using data from an abnormality occurrence in the spring, such as one year prior.

[0073] Next, the operation will be described. FIG. 10 is a flow diagram showing the operation of the water treatment plant anomaly diagnosis system according to the second embodiment. First, discrimination clustering information CLM_P is generated by machine learning using data from past anomaly occurrences (step ST201: learning process). Next, the discrimination clustering information CLM_P generated in the learning process is used to detect an anomaly occurring in the water treatment plant to be monitored and determine the anomaly mode of the anomaly (step ST202: actual operation process). In the actual operation process, the discrimination clustering information CLM_P generated in the learning process is treated as discrimination clustering information CLM, and an anomaly mode determination is performed. Furthermore, during the actual operation process, it is not necessary to update the discrimination clustering information CLM. Therefore, the actual operation process is similar to the flow shown in the first embodiment ( FIG. 6 ), except that the discrimination information update process in step ST108 is omitted.

[0074] The learning process will now be described. FIG. 11 is a flow diagram showing the learning process according to the second embodiment. In the learning process, discrimination clustering information CLM_P is generated using the MSPC technique and machine learning described in the first embodiment. First, data from past abnormality occurrences is accumulated (step ST2011: past abnormality data accumulation process). Next, for the data from past abnormality occurrences, a combination of measurement value data D_P of the same series is acquired as one data set DS_P (step ST2012: data set acquisition process), and preprocessing is performed on the data set DS_P to obtain a preprocessed data set DS_P* (step ST2013: preprocessing process). The preprocessing of the data set DS_P is as described above.

[0075] Next, PCA analysis is performed on the data set DS_P* (step ST2014: principal component analysis step). The PCA analysis is as described above, and principal component axis information PA_P is obtained as information on each principal component axis for data from past abnormality occurrences.

[0076] Next, the past abnormality degree is calculated using the data set DS_P*, which is the original data for calculating the past abnormality degree, and the principal component axis information PA_ (step ST2015: past abnormality degree calculation step). The past abnormality degree is as described above. In the second embodiment, T is calculated for data at the time of the past abnormality occurrence. ２ The value and Q value were calculated, and the calculated T ２ The data of the value and the Q value is defined as the past abnormality degree WD_P.

[0077] Next, data from past anomaly occurrences is classified by a clustering method based on the past anomaly degree WD_P to generate discrimination clustering information CLM_P (step ST2016: past anomaly clustering step). The clustering in the past anomaly clustering step is as described above, and the past anomaly degree WD_P (T ２ The relationship between the variance (value and Q value) and the abnormal mode is clustered to generate clustering information for discrimination CLM_P.

[0078] According to the second embodiment, the same effects as those of the first embodiment can be achieved. Furthermore, the computational load during actual operation can be reduced, enabling faster anomaly detection and anomaly mode discrimination. More specifically, a past anomaly DB is provided for storing data on measurement values ​​when past anomalies have occurred, and in a learning process prior to the actual operation process, discrimination clustering information is generated using the data on measurement values ​​when past anomalies have occurred. During actual operation, the anomaly mode can be discriminated using the already generated discrimination clustering information. That is, unlike the first embodiment, calculations for generating and updating the discrimination clustering information are not required. This reduces the amount of calculation required for discriminating anomaly modes during actual operation. While generating and updating the discrimination clustering information requires a configuration with advanced computing capabilities, the second embodiment makes it possible to omit such a configuration from the anomaly diagnosis device. Furthermore, by setting the threshold for anomaly detection from already generated discrimination clustering information, more accurate anomaly detection can be achieved and the computational load for anomaly detection during actual operation can be reduced.

[0079] Furthermore, if discrimination clustering information is generated using past data with high autocorrelation, the accuracy of discrimination of abnormal modes can be improved.

[0080] Embodiment 3. Next, embodiment 2 will be described with reference to FIGS. 12 to 15. Components identical to or corresponding to those shown in FIGS. 1 to 11 are designated by the same reference numerals, and their description will be omitted. In embodiment 3, in generating the discrimination clustering information CLM_P in embodiment 2, not only data from past anomalies but also simulation results from a water quality simulator are used. FIG. 12 is a block diagram showing the configuration of an anomaly diagnosis system for a water treatment plant in embodiment 3. The anomaly diagnosis system 3000 includes an anomaly diagnosis device 200 and a past anomaly learning device 900 similar to those in embodiment 2, as well as a simulation device 800 connected to the past anomaly learning device 900 via a network. The simulation device 800 transmits simulation results RES from the water quality simulator, which virtually reproduces anomalies, to a past anomaly DB 901 of the past anomaly learning device 900. The past anomaly DB 901 stores the simulation results RES as well as measurement value data D_P, which is data from past anomalies. In the past abnormality learning device 900, the simulation result RES is treated in the same way as the measurement value data D_P.

[0081] 13 is a block diagram showing the configuration of a simulation device according to embodiment 3. The simulation device 800 includes a calculation condition receiving unit 801 that receives input of calculation conditions COND for a water quality simulation, an abnormal condition generating unit 802 that generates abnormal conditions COND* for virtually reproducing an abnormality based on the calculation conditions COND, and a water quality simulator unit 803 that performs a water quality simulation using the abnormal conditions COND* as input and outputs a simulation result RES.

[0082] The calculation condition receiving unit 801 is an interface that receives calculation conditions COND input by the user and outputs the received calculation conditions COND to the abnormality condition generating unit 802. The calculation conditions COND include the operating conditions of the process in the abnormality to be reproduced and conditions such as the quality of the influent water.

[0083] The abnormal condition generation unit 802 generates abnormal conditions COND* for virtually reproducing abnormalities in the water quality simulation based on the calculation conditions COND input by the user. The abnormal condition generation unit 802 outputs the abnormal conditions COND* to the calculation unit 803 a of the water quality simulator unit 803.

[0084] The water quality simulator unit 803 includes a calculation unit 803a that performs calculations associated with the simulation and a model storage unit 803b that stores various simulation models MDL required for the water quality simulator. In the water quality simulation performed by the water quality simulator unit 803, the simulation model MDL used in the water quality simulation is read from the model storage unit 803b to the calculation unit 803a. The simulation is then executed based on the read simulation model MDL and the abnormality condition COND*, for example, by calculating the time evolution from the initial condition, and a simulation result RES is generated. The water quality simulator unit 803 transmits the simulation result RES to the past abnormality DB 901 of the past abnormality learning device 900. Examples of the simulation model MDL include various physical reaction models, such as a microbial reaction activity model. If the model storage unit 803b stores multiple simulation models MDL, the simulation model MDL to be used in the water quality simulation can be specified using the calculation condition COND.

[0085] The abnormalities reproduced by the water quality simulator unit 803 are those that can occur under the conditions specified in the calculation conditions COND. Since the calculation conditions COND can be arbitrarily specified by the user, the water quality simulator unit 803 can also reproduce abnormalities that have not actually occurred in the past. In other words, the water quality simulator unit 803 virtually reproduces abnormalities that have actually occurred or that may occur.

[0086] By reproducing anomalies using the simulation device 800 and using the data of the reproduced anomalies for machine learning in the past anomaly learning device 900, even if there is a lack of data to use for machine learning in the past anomaly learning device 900, such as a lack of data from when anomalies actually occurred in the past, it is possible to ensure the data size necessary for machine learning and generate the discrimination clustering information CLM_P with high accuracy. Furthermore, by reproducing anomalies that have not actually occurred in the past, there is no longer any restriction as to whether or not anomalies have actually occurred, so that anomalies that have not actually occurred but may occur can also be reflected in the discrimination clustering information CLM_P, thereby expanding the range of discriminable abnormality modes. This also allows for an expansion of the scope of application as an anomaly diagnosis system.

[0087] Next, the operation will be described. FIG. 14 is a flow diagram showing the operation of the anomaly diagnosis system for a water treatment plant in the third embodiment. First, a water quality simulation is performed using a water quality simulator to obtain a simulation result RES (step ST301: simulation step). Next, discrimination clustering information CLM_P is generated by machine learning using data from past anomaly occurrences and the simulation result RES (step ST302: learning step). Next, using the discrimination clustering information CLM_P generated in the learning step, an anomaly occurring in the water treatment plant to be monitored is detected, and the anomaly mode of the occurring anomaly is determined (step ST303: actual operation step). Details of the learning step and actual operation step are the same as those in the second embodiment. In the learning step, the simulation result RES obtained in the simulation step is used for machine learning in the same way as data from past anomalies.

[0088] The simulation process will now be described. Fig. 15 is a flow diagram showing the simulation process according to the third embodiment. First, calculation conditions COND input by the user are accepted (step ST3011: calculation condition acceptance process). As described above, the calculation conditions COND include the operating conditions of the process in the abnormality to be reproduced and conditions such as the quality of the influent water.

[0089] Next, based on the calculation conditions COND, abnormal conditions COND* are generated to virtually reproduce abnormalities in the water quality simulation (step ST3012: abnormal condition generation step).

[0090] Next, a water quality simulation is performed based on the abnormality condition COND* and the simulation model MDL, and the simulation result RES is obtained (step ST3013: water quality simulation step). The simulation model MDL is as described above. The simulation result RES is stored in the past abnormality DB 901.

[0091] In the third embodiment, the simulation results RES generated by the simulation device 800 are temporarily stored in the past abnormality DB 901, but since it is sufficient that the results of the water quality simulation by the water quality simulator unit 803 are reflected in the generation of the discrimination clustering information CLM_P, it is not essential to store the simulation results RES in the past abnormality DB 901, and it is sufficient that the simulation results RES are at least reflected in the calculation of the past abnormality degree WD_P. For example, it is also possible to send the simulation results RES directly from the simulation device 800 to the data set acquisition unit 902.

[0092] According to the third embodiment, the same effects as those of the second embodiment can be obtained. Furthermore, the system further includes a water quality simulator unit that virtually reproduces an abnormality that has actually occurred or may occur, and generates the discrimination clustering information using data from past abnormality occurrences and data from abnormality occurrences virtually reproduced by the water quality simulator unit. Therefore, even if there is insufficient data from past abnormality occurrences, the discrimination clustering information can be generated with high accuracy. Furthermore, since it is possible to reproduce abnormalities that have not actually occurred, the range of abnormality modes that can be distinguished can be expanded, thereby expanding the scope of application of the anomaly diagnosis system.

[0093] Although various exemplary embodiments and examples are described in this disclosure, the various features, aspects, and functions described in one or more embodiments are not limited to the application of a particular embodiment, but may be applied to the embodiments alone or in various combinations. Therefore, countless variations not illustrated are anticipated within the scope of the technology disclosed in this specification. For example, this includes cases where at least one component is modified, added, or omitted, or where at least one component is extracted and combined with components of another embodiment.

[0094] 100, 200 Abnormality diagnosis device, 101 Measurement value collection unit, 102 Measurement value DB, 103 Abnormality detection unit, 1034 Abnormality degree calculation unit, 104 Contribution factor calculation unit, 105 Discrimination information storage unit, 106 Abnormality mode discrimination unit, 107 Diagnosis result display unit, 108 Discrimination information generation unit, 800 Simulation device, 803 Water quality simulator unit, 900 Past abnormality learning device, 901 Past abnormality DB, 905 Past abnormality degree calculation unit, 906 Past abnormality clustering unit, 1000, 2000, 3000 Abnormality diagnosis system, CLM, CLM_P Discrimination clustering information, Cp Contribution degree data, D, D_P Measurement value data, M Abnormality mode discrimination result, W Abnormality presence / absence determination result, WD Abnormality degree, WD_P Past abnormality degree

Claims

1. An abnormality diagnosis system for a water treatment plant that diagnoses abnormalities in a water treatment plant to be monitored, comprising: a measurement value collection unit that collects multiple types of measurement values ​​in the water treatment process of the water treatment plant; a measurement value storage unit that accumulates the measurement values ​​collected by the measurement value collection unit; an abnormality detection unit that determines whether or not an abnormality has occurred in the water treatment plant based on a degree of abnormality calculated from the measurement values; an abnormality mode discrimination information storage unit that stores abnormality mode discrimination information indicating the relationship between the degree of abnormality and an abnormality mode; an abnormality mode discrimination unit that, when it is determined that an abnormality has occurred, discriminates the abnormality mode of the abnormality that has occurred by comparing the degree of abnormality calculated by the abnormality detection unit with the abnormality mode discrimination information; and a diagnosis result display unit that displays the result of the discrimination, wherein the abnormality mode is associated with a severity for the water treatment plant.

2. An abnormality diagnosis system for a water treatment plant as described in claim 1, further comprising a contribution factor calculation unit that calculates a contribution factor indicating the degree of influence of each of the measured values ​​on the fluctuation of the degree of abnormality, and the abnormality mode discrimination unit narrows down the abnormality modes based on the contribution factor before making the discrimination by comparing the degree of abnormality with the abnormality mode discrimination information.

3. The abnormality diagnosis system for a water treatment plant as described in claim 2, wherein the contribution factor calculation unit identifies the measurement value that ranks highest in the hierarchy of the degree of contribution as a possible causal factor, and the abnormality mode discrimination unit narrows down the abnormality modes based on the degree of contribution and the possible causal factor.

4. An abnormality diagnosis system for a water treatment plant as described in any one of claims 1 to 3, further comprising a discrimination information generation unit that generates and updates the abnormality mode discrimination information based on the relationship between the degree of abnormality obtained during actual operation and the severity of the abnormality having that degree of abnormality.

5. An abnormality diagnosis system for a water treatment plant as described in any one of claims 1 to 3, further comprising: a past abnormality memory unit that accumulates data on the measurement values ​​when abnormalities occurred in the past; a past abnormality calculation unit that calculates the degree of abnormality as a past abnormality degree from the measurement values ​​accumulated in the past abnormality memory unit; and a past abnormality clustering unit that performs clustering using machine learning on the past abnormality degrees, associates the past abnormality degrees with the abnormality modes, and generates clustering information for discrimination, wherein the abnormality mode discrimination unit uses the discrimination clustering information generated in advance as the abnormality mode discrimination information to discriminate the abnormality mode of the abnormality currently occurring.

6. An abnormality diagnosis system for a water treatment plant as described in claim 5, further comprising a water quality simulator unit that virtually reproduces an abnormality that has actually occurred or an abnormality that may occur, wherein the past abnormality degree calculation unit calculates the past abnormality degree using the measurement values ​​stored in the past abnormality memory unit and the data of the abnormality virtually reproduced by the water quality simulator unit.

7. An abnormality diagnosis system for a water treatment plant as described in claim 5 or 6, wherein the generation of the discriminant clustering information uses the measurement values ​​from the past abnormality occurrence at the same time of day or season of year as the current time or season.

8. A method for diagnosing an abnormality in a water treatment plant to be monitored, comprising: a measurement value collecting step for collecting multiple types of measurement values ​​in a water treatment process by the water treatment plant; a measurement value accumulating step for accumulating the measurement values ​​collected in the measurement value collecting step; an abnormality detection step for determining whether or not an abnormality has occurred in the water treatment plant based on a degree of abnormality calculated from the measurement values; an abnormality mode discrimination step for discriminating the abnormality mode of the occurring abnormality by comparing the degree of abnormality calculated in the abnormality detection step with abnormality mode discrimination information indicating the relationship between the degree of abnormality and an abnormality mode when it is determined that an abnormality has occurred; and a diagnosis result display step for displaying the result of the discrimination, wherein the abnormality mode is associated with a severity for the water treatment plant.

9. The method for diagnosing an abnormality in a water treatment plant as described in claim 8, further comprising a contribution factor calculation step of calculating a contribution factor indicating the degree of influence of each of the measured values ​​on the fluctuation of the degree of abnormality, and in the abnormality mode discrimination step, narrowing down the abnormality modes based on the contribution factor before making the discrimination by comparing the degree of abnormality with the abnormality mode discrimination information.

10. A method for diagnosing an abnormality in a water treatment plant as described in claim 9, wherein, in the contribution factor calculation process, the measurement values ​​that are ranked high in terms of the magnitude of the contribution degree are identified as possible causal factors, and, in the abnormality mode discrimination process, the abnormality modes are narrowed down based on the contribution degree and the possible causal factors.

11. An abnormality diagnosis method for a water treatment plant described in any one of claims 8 to 10, in which the abnormality mode discrimination information is generated and updated based on the relationship between the degree of abnormality obtained during actual operation and the severity of the abnormality having that degree of abnormality.

12. A method for diagnosing an anomaly in a water treatment plant as described in any one of claims 8 to 10, further comprising: a past abnormality data accumulation process for accumulating data on the measurement values ​​when an abnormality occurred in the past; a past abnormality degree calculation process for calculating the degree of abnormality as a past abnormality degree from the measurement values ​​accumulated in the past abnormality data accumulation process; and a past abnormality clustering process for performing clustering by machine learning on the past abnormality degrees, matching the past abnormality degrees with the abnormality modes, and generating clustering information for discrimination, wherein in the abnormality mode discrimination process, the discrimination clustering information generated in advance is used as the abnormality mode discrimination information to discriminate the abnormality mode of the abnormality currently occurring.

13. The method for diagnosing an anomaly in a water treatment plant as described in claim 12, further comprising a water quality simulation process for virtually reproducing an anomaly that has actually occurred or may occur, wherein in the past anomaly calculation process, the past anomaly degree is calculated using the measurement values ​​accumulated in the past anomaly data accumulation process and the simulation results of the water quality simulation process.

14. A method for diagnosing an anomaly in a water treatment plant as described in claim 12 or 13, wherein the generation of the discriminant clustering information uses the measurement values ​​from the time of the past anomaly that occurred at the same time of day or season of year as the current time or season.

Images