Water treatment chemical injection control system, water treatment chemical injection control method, and program

Abstract

To improve the explanatory property when correcting a recommended chemical injection rate (recommended value) for a chemical injection operation using a machine learning model.SOLUTION: A water treatment chemical injection control system 100 includes: a chemical injection model construction unit 200 that constructs a chemical injection model by calculating a chemical injection rate through machine learning of chemical injection operation result data of a waterworks facility, thereby reproducing a chemical injection operation; a chemical injection rate calculation unit 300 that calculates a chemical injection rate using a chemical injection model, based on a current value of measurement data of the waterworks facility; and a chemical injection rate correction unit 400 that calculates a correction amount for the chemical injection rate using the chemical injection model, by referring to a water purification knowledge database 500 including correction information for the chemical injection operation expressed in an explicit format.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a water purification chemical injection control system, a water purification chemical injection control method, and a program for performing chemical injection operations in a water purification process in a waterworks facility. [Background technology]

[0002] Waterworks facilities are an important part of urban infrastructure, and the stable supply of clean tap water is required. The water purification process is one of the most important processes, as it directly affects the quality of tap water.

[0003] Many waterworks facilities use coagulation and sedimentation, sand filtration, and disinfection treatment as water purification processes. These treatments involve the injection of coagulants to coagulate turbid components in the raw water, and disinfectants to inactivate pathogenic microorganisms. Properly performing these chemical injection operations is important for meeting the water quality standards set out in the Water Supply Act, as well as achieving the voluntary water quality targets of each waterworks facility.

[0004] A variety of control technologies have been developed to ensure appropriate chemical injection operations. In addition to sanitary engineering approaches based on physical, chemical, and biological knowledge related to coagulation, sedimentation, and disinfection in water purification processes, many approaches using artificial intelligence (AI) that utilize machine learning have been seen in recent years.

[0005] Changes in the workforce at urban infrastructure facilities have led to a decline in skilled staff and the resulting loss of on-site know-how, and waterworks facilities are no exception. Therefore, it is expected that the need for machine learning, which can utilize data on past excellent operational performance and reproduce the operations of skilled staff, will continue to grow in the future.

[0006] The AI ​​approach to machine learning described above is characterized by its ability to reproduce, to a certain extent, operations performed by personnel in the past by building a model (machine learning model) that has learned from operational performance data. This makes it possible to present recommended injection operation values ​​under various operating conditions, which is expected to be useful for control and operational guidance.

[0007] However, several issues have been pointed out when applying machine learning models alone, and in some cases these have become barriers to adoption. One of the issues concerns the robustness of machine learning models. The scope of application of machine learning models is generally limited to the range of actual data that was learned, and there is an issue in that the validity of the model calculation values ​​cannot be fully guaranteed under conditions outside that range (so-called extrapolation conditions).

[0008] Another issue is the black-box nature of machine learning models. Because the learning results of machine learning models are expressed using a large number of model coefficients (commonly referred to as "weighting coefficients"), they are less readable than physical and chemical models. This makes the recommended injection values ​​of machine learning models less explainable and convincing. Although previous attempts have been made to alleviate the black-box nature of the process of determining recommended injection values ​​(e.g., Enbutsu et al., "Research on Extracting Plant Operation Rules Using Neural Networks," IEEJ Transactions on Plant Operation, Vol. 111, No. 1), the current situation is that this issue has not yet been fully resolved.

[0009] To address these issues, for example, Patent Document 1 discloses a control method that corrects the injection rate recommended by a chemical injection model constructed by machine learning based on the results of treated water quality tests (jar tests, etc.). Because this method corrects the recommended value based on jar tests, it is expected to improve both persuasiveness and validity, and alleviate issues related to black-box nature and robustness.

[0010] Furthermore, Patent Document 2 discloses a control method that extracts knowledge related to the setting of control target values ​​for operation and the like through statistical analysis and machine learning using facility process data, etc., and utilizes that knowledge. This method is also expected to alleviate the black-box nature of the system by improving explainability through knowledge. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] Japanese Patent Application Publication No. 2023-147019 [Patent Document 2] Japanese Patent Application Laid-Open No. 2011-060135 Summary of the Invention [Problem to be solved by the invention]

[0012] However, the method disclosed in Patent Document 1 requires additional treated water quality tests, which increases the workload at the facility where it is installed. This poses the problem that this method may be difficult to apply in cases where the raw water quality changes suddenly (such as a sudden rise in raw water turbidity), putting operational pressure on the facility.

[0013] Furthermore, the method of utilizing knowledge disclosed in Patent Document 2 does not correct the recommended value of the machine learning model. Therefore, this method has the problem that there are cases where the validity of the recommended value of the machine learning model cannot be improved.

[0014] The present invention has been made in consideration of the above situation, and aims to improve the explainability at least when correcting the chemical injection rate (recommended value) of a chemical injection operation using a machine learning model. [Means for solving the problem]

[0015] In order to solve the above problems, one embodiment of the water purification chemical injection control system of the present invention comprises a chemical injection model construction unit that calculates the chemical injection rate through machine learning of the water facility's chemical injection operation performance data and constructs a chemical injection model that reproduces the chemical injection operation; a chemical injection rate calculation unit that calculates the chemical injection rate using the chemical injection model based on the current values ​​of the water facility's measurement data; and a chemical injection rate correction unit that calculates the correction amount of the chemical injection rate using the chemical injection model by referring to a water purification knowledge database that contains correction information for chemical injection operations expressed in an explicit format. [Effects of the Invention]

[0016] According to at least one aspect of the present invention, the chemical injection rate (recommended value) of a chemical injection operation based on a machine learning model trained on injection operation performance data is corrected using explicit knowledge. This configuration significantly improves the explainability (black-box resistance) of the correction of the chemical injection rate of a chemical injection operation presented by a water purification chemical injection control system. Furthermore, improving the explainability of the correction makes the basis for the correction amount clearer, which is expected to increase validity (robustness) and a sense of acceptability. Problems, configurations, and effects other than those described above will become apparent from the following description of the preferred embodiments of the invention. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a diagram showing an example of the configuration of a water purification chemical injection control system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram illustrating an example of a hardware configuration of a water purification chemical injection control system. [Figure 3] 10 is a flowchart showing an example of a processing procedure of a medicine injection model constructing unit. [Figure 4] FIG. 1 is a diagram illustrating an example of the structure of a machine learning model. [Figure 5] 10 is a flowchart showing an example of a processing procedure of a chemical injection rate calculation unit. [Figure 6] 10 is a flowchart showing an example of a processing procedure of a chemical injection rate correction unit. [Figure 7] FIG. 2 is a diagram illustrating an example of knowledge stored in a water purification knowledge DB. [Figure 8] FIG. 10 is a diagram showing an example of a screen display of the water purification chemical injection control system. DETAILED DESCRIPTION OF THE INVENTION

[0018] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, examples of modes for carrying out the present invention (hereinafter referred to as "embodiments") will be described with reference to the accompanying drawings. In this specification and the accompanying drawings, identical or similar components are denoted by the same reference numerals, and redundant explanations may be omitted or only explanations focusing on the differences may be given. The number of each component may be singular or plural unless otherwise specified.

[0019] <One embodiment> In this embodiment, we will describe an example of a water purification chemical injection control system that uses control technology for performing chemical injection operations in the water purification process at water supply facilities, in particular control technology that can appropriately correct the operating amount (calculated value or recommended value) calculated by a chemical injection model based on machine learning of operational performance. In order to overcome the issues of black-box nature and robustness while taking advantage of the advantages of machine learning models that effectively utilize excellent driving operation performance data, it is effective to use useful information (knowledge) based on insights obtained through data analysis and customer know-how. The present invention overcomes the above-mentioned issues by correcting the calculated value (recommended value) of the chemical injection operation by the machine learning model using explicit knowledge. The present invention includes multiple implementation means, examples of which are as follows.

[0020] [Configuration of water purification chemical injection control system] Figure 1 is a diagram showing an example of the configuration of a water purification chemical injection control system according to one embodiment of the present invention. As shown in Figure 1, the components of water purification chemical injection control system 100 are a chemical injection model construction unit 200 that constructs a model using machine learning, a chemical injection rate calculation unit 300 that calculates the chemical injection rate using the constructed chemical injection model, a chemical injection rate correction unit 400 that calculates a correction value for the calculated chemical injection rate, and a water purification knowledge database (DB) 500 that stores information referenced when calculating the correction value. The information referenced when calculating the correction value for the chemical injection rate includes rules, functions, and judgment information, and is referred to as "knowledge" in this specification.

[0021] Furthermore, various processes executed by each processing unit of the water purification chemical injection control system 100 refer to and write data from a water purification facility DB 20 that stores measurement data such as water quality, process state quantities, and operation quantities acquired by sensors 15 installed in the target water purification facility 10. A water purification facility is a waterworks facility that has a water purification process.

[0022] The water quality may be, for example, turbidity, pH, water temperature, alkalinity, etc. The water quality may be the quality of raw water or the quality of water at the outlet of a settling tank during the water purification process. Here, a sensor capable of measuring water quality continuously in real time is used as the water quality sensor that measures water quality. The process state quantity is, for example, the opening degree of the valve of the chemical injection pump, the operating state of the equipment of the water purification chemical injection control system 100, and the like. The manipulated variable is, for example, the chemical injection rate or the water flow rate.

[0023] The execution conditions and results of each process are visualized by a display device 600. The chemical injection rate calculated by this water purification chemical injection control system 100 is reflected in the operation of control elements such as chemical injection pumps as a control value (=SV+ΔSV) for the chemical injection operation of the water purification facility 10.

[0024] The above is the outline and general flow of execution of the water purification chemical injection control system 100. Below, specific examples of embodiments will be described regarding each processing unit that realizes the technical features of the present invention and the data used therein.

[0025] [Hardware configuration of water purification chemical injection control system] Here, the hardware configuration of the water purification chemical injection control system 100 will be described with reference to FIG. Figure 2 is a diagram showing an example of the hardware configuration of the water purification chemical injection control system 100. The calculator 30 is an example of hardware used as a computer. The water purification chemical injection control system 100 according to this embodiment realizes a water purification chemical injection control system in which the blocks shown in Figure 1 work together by having the calculator 30 (computer) execute a computer program.

[0026] The computer 30 includes a CPU (Central Processing Unit) 31, a ROM (Read Only Memory) 32, and a RAM (Random Access Memory) 33, all connected to a system bus. The system bus also includes a display device 600, an input device 35, a non-volatile storage 36, and a communication interface 37.

[0027] The CPU 31 reads out the program code of the software that realizes each function of the water purification chemical injection control system 100 from the ROM 32, loads it into the RAM 33, and executes it. Variables, parameters, etc. generated during the calculation processing of the CPU 31 are temporarily written to the RAM 33, and these variables, parameters, etc. are read out by the CPU 31 as appropriate. The functions of the blocks of the water purification chemical injection control system 100 are realized by the CPU 31 executing the program code read out from the ROM 32. However, other processors such as an MPU (Micro Processing Unit) may be used instead of the CPU 31. The CPU 31 is an example of a calculation device, and the RAM 33 and non-volatile storage 36 are examples of storage devices.

[0028] The display device 600 (an example of a display unit) is a monitor such as a liquid crystal display, and displays a GUI (Graphical User Interface) screen, the results of arithmetic processing by the CPU 31, etc. The input device 35 generates an input signal according to a user's operation and outputs it to the CPU 31. The input device 35 may be, for example, a mouse, a keyboard, a touch sensor, or the like, and the user can input information and instructions by operating the input device 35. The display device 600 and the input device 35 may be integrated as a touch panel.

[0029] The nonvolatile storage 36 is an example of a non-transitory recording medium, and is capable of storing data used by a program, data obtained by executing a program, and the like. The nonvolatile storage 36 may also store an operating system (OS) or a program executed by the CPU 31. The water purification facility DB 20 may be configured by the nonvolatile storage 36, or may be configured separately from the water purification chemical injection control system 100 as shown in FIG. 1. The nonvolatile storage 36 may be a hard disk drive (HDD), a solid state drive (SSD), an optical or magnetic disk medium, a semiconductor memory card, or the like.

[0030] A communication device such as a network interface card (NIC) is used as the communication interface 37. The communication interface 37 can transmit and receive various data to and from external devices via a communication network such as a LAN connected to a terminal of the NIC or via a dedicated line. Communication between the water purification chemical injection control system 100 and the water purification facility 10 is realized by the communication interface 37.

[0031] [Chemical injection model construction processing] 3 is a flowchart showing an example of the processing procedure of the chemical injection model construction unit 200. This processing unit uses data related to past chemical injection operations of the water purification facility 10 to construct a chemical injection model that can reproduce past chemical injection operations through machine learning.

[0032] In the first chemical injection data preprocessing step S1, the chemical injection model construction unit 200 processes various data stored in the water purification facility DB 20 so that it can be used as training data for machine learning in subsequent steps. Specifically, it checks for abnormal values ​​due to sensor 15 failure or maintenance inspection, and if an abnormal value is found, it excludes the data at that time from the training data. Abnormal values ​​are confirmed by determining whether they are within the measurement range specified by the sensor 15 specifications or within predetermined upper and lower limits. Furthermore, if there is missing data, it performs an interpolation process. The interpolated value may be the most recent data value or the average value of the previous and following data.

[0033] Next, in the machine learning training data preparation step S2, the chemical injection model construction unit 200 prepares training data from the data preprocessed in the chemical injection data preprocessing step S1, to be used in machine learning to construct a chemical injection model in the chemical injection model construction unit 200. Here, training data refers to a data set consisting of the chemical injection rate, which is the objective variable, and various data items, which are explanatory variables. The explanatory variables are typically selected from the data items included in the water purification facility DB 20, such as raw water quality, which affects the chemical injection operation. The number of training data sets to be prepared may cover the entire data period included in the water purification facility DB 20, or may be selected at a predetermined time interval. Furthermore, the training data may be created according to the content to be reflected in the chemical injection rate model, such as by preparing training data for cases where the raw water turbidity is high or low.

[0034] Next, in the machine learning condition setting step S3, the medicine injection model construction unit 200 sets the type of machine learning model to be used, the number of times of learning, and the learning end condition.

[0035] [Example of a machine learning model] An example of a typical machine learning model structure is shown in Figure 4. The machine learning model shown here is a hierarchical neural network model, and it uses explanatory variables (a set of explanatory variables x 11 ~x 1I ) input layer, the response variable (the response variable set y31 ~y 3k ), and an intermediate layer that connects the input layer and the output layer (Figure 4 shows one layer, but there can be multiple). Each layer is equipped with neuron elements marked with a circle in the figure. Each neuron element has the function of converting input values ​​and outputting them. The neuron number subscripts for each layer are as follows: 1i: Input layer neuron number subscript (capital letter I = number of input layer neurons) 2j: Hidden layer neuron number subscript (capital J = number of hidden layer neurons) 3k: Output layer neuron number subscript (capital letter K = number of output layer neurons)

[0036] It is common to place bias neurons in the intermediate and output layers, which have no input and always output a value of 1. By placing these bias neurons, it is possible in some cases to improve learning performance with fewer learning iterations.

[0037] The neuron elements arranged in each layer are mutually connected to the neuron elements of the adjacent layer, and are assigned model parameters (called weight coefficients w) that are indices of the connection strength. In Figure 4, the weight coefficient set w between the hidden layer and the input layer 2j , 1j , the weight coefficients between the output layer and the hidden layer w 3k , 2j An example is shown.

[0038] Returning to the explanation of the flowchart in Figure 3, in machine learning calculation step S4, the chemical injection model construction unit 200 performs machine learning using the training data prepared in step S2 and the learning conditions preset in step S3. In this embodiment, machine learning refers to calculations that adjust the weighting coefficients according to a predetermined algorithm to suit the model application.

[0039] The model's purpose in the present invention is to quantitatively reproduce the chemical injection operations contained in the training data, i.e., the operational performance of the operator of the water purification facility 10, and utilize them for chemical injection control and operational support guidance. Therefore, in the machine learning here, when the values ​​of the data items serving as explanatory variables are input to the input layer, the weighting coefficient w is adjusted so that the chemical injection rate, which is the objective variable, from the output layer reproduces the training data (reducing the error between the model output value and the training data value). A well-known technique, such as the Error Backpropagation Method (BP method), is applied as the machine learning algorithm. In this way, the learning algorithm of the hierarchical neural network model used in the chemical injection model construction unit 200 adjusts the model parameters (weighting coefficient w) based on the error between the training data values ​​to be machine-learned and the calculated values ​​of the hierarchical neural network model.

[0040] Next, in the learning performance evaluation step S5, the chemical injection model construction unit 200 compares the calculated chemical injection rate value from the chemical injection model constructed by machine learning with the actual chemical injection rate value of the training data to evaluate the learning performance. The evaluation here is based on average error, maximum error, etc. If the learning performance does not achieve the preset performance target (NO in S5), the process returns to the machine learning training data preparation step S2. Then, the training data, learning conditions, etc. are changed, and the series of steps (S2 to S5) are repeated until the learning performance achieves the predetermined performance target.

[0041] If the learning performance achieves the predetermined performance target (YES judgment in S5), in the next machine learning model saving step S6, the chemical injection model construction unit 200 saves the chemical injection model in the water purification knowledge DB 500 so that model calculations will be possible in the next step and thereafter. Note that the chemical injection model construction unit 200 may output an overview or detailed information of the chemical injection model to the display device 600.

[0042] In the above-mentioned learning performance evaluation step S5, if the learning performance cannot be achieved even after repeating the series of steps (S2 to S5) a preset number of times, an instruction to omit the repetition is given. The medicine injection model construction unit 200 stops the repetition in accordance with the instruction to omit the repetition, and executes the machine learning model saving step S6. After the machine learning model saving step S6 is completed, the processing of this flowchart ends. The above is an explanation of the contents of the chemical injection model construction unit 200.

[0043] [Chemical injection rate calculation processing] 5 is a flowchart showing an example of the processing procedure of the chemical injection rate calculation unit 300. This processing unit calculates the chemical injection rate using a chemical injection model that is constructed by machine learning in the chemical injection model construction unit 200 and stored in the water purification knowledge DB 500.

[0044] In the first machine learning model loading step S11, the chemical injection rate calculation unit 300 loads the model parameters of the chemical injection model stored in the water purification knowledge DB 500 so that they can be used for calculations in the subsequent steps. One model may be loaded, but multiple predefined types (such as for when raw water is highly turbid, when raw water is low turbid, etc.) may also be loaded so that calculations can be performed by switching between models depending on the raw water quality conditions.

[0045] Next, in the explanatory variable data reading step S12, the chemical injection rate calculation unit 300 reads the current values ​​of the data items that serve as explanatory variables for the chemical injection model from the water purification facility DB 20 so that they can be used in calculations from the next step onwards.

[0046] Next, in the machine learning model calculation step S13, the chemical injection rate calculation unit 300 inputs the explanatory variable data into the input layer of the chemical injection model, and obtains the calculated value of the coagulant injection rate as the output from the output layer.

[0047] Then, in the calculation result output step S14, the chemical injection rate calculation unit 300 outputs the obtained calculation value of the flocculant injection rate to the water purification facility DB 20, the water purification knowledge DB 500, and the display device 600. After the calculation result output step S14 is completed, the processing of this flowchart ends. The above is an explanation of the contents of the chemical injection rate calculation unit 300.

[0048] [Chemical injection rate correction processing] 6 is a flowchart showing an example of the processing procedure of the chemical injection rate correction unit 400, which is one of the features of the present invention. This processing unit calculates a correction value for the chemical injection rate using explicit correction logic stored in the water purification knowledge DB 500.

[0049] In the first correction knowledge setting step S21, the chemical injection rate correction unit 400 sets the correction knowledge to be used in calculating the correction value of the chemical injection rate. Correction knowledge is information related to the correction of the chemical injection operation expressed in an explicit format. Knowledge setting may be any one of the above, or multiple knowledge may be set to be used depending on the conditions.

[0050] [Examples of knowledge stored in the water purification knowledge database] An example of the knowledge assumed to be applied here is shown in Figure 7. In order to mitigate the black-box nature of the knowledge, it is important that the knowledge is expressed in an explicit format.

[0051] (rule format) Knowledge [1] in the figure is an example of rule format. Figure 7 shows an example of rule format knowledge as follows: The raw water turbidity time difference △Tu is the amount of change in raw water turbidity over a certain time interval. IF raw water turbidity Tu ≥ X degree AND raw water turbidity time difference △Tu ≥ Y degree THEN Injection rate correction amount △SV=K×△Tu”

[0052] Rule-based knowledge is basically expressed as an IF-THEN type, and if the conditions in the IF section are met, the correction value defined in the THEN section is adopted. The IF-THEN type allows for a clear definition of the relationship between the premise or condition and the conclusion or action to be taken when this premise or condition is met. These rules can be created using the empirical know-how of the operation manager of the water purification facility 10, or the results of statistical analysis of performance data stored in the water purification facility DB 20. In either case, the rules must be based on a good operational record that indicates that the chemical injection operation was performed appropriately. The appropriateness of the injection operation can be determined, for example, by whether the treated water quality, such as the sediment turbidity, meets a predetermined value.

[0053] (Functional form) Another way to express this knowledge is as a function, as in knowledge [2]. Figure 7 shows an example of functional knowledge: "Injection rate correction amount ΔSV = α × raw water quality M + β × raw water quality N." As mentioned above, raw water quality M and N are turbidity, pH, water temperature, alkalinity, etc.

[0054] The explanatory variables of the function here can be set based on information related to chemical injection rate correction, such as knowledge that correction was necessary when raw water quality item values ​​were higher or lower than usual. When creating a function for the injection rate correction amount, it is important to use differential data between the values ​​calculated using the machine learning chemical injection model's arithmetic formula and the training data (i.e., actual values). This is because when this difference is large, the chemical injection model's accuracy in reproducing the actual values ​​is low, and chemical injection rate correction is necessary.

[0055] (natural language format) Another expression format may be a natural language format such as knowledge [n]. Figure 7 shows an example of natural language format knowledge: "If alarm B regarding treated water quality A occurs, respond by adding C [mg / L] (equivalent to ΔSV) to the injection rate." The content of the alarm may be that the water quality has deteriorated and does not meet the standard. An alarm may be set to occur not only when the measurement data value is too high, but also when it is too low.

[0056] In this natural language format, the chemical injection rate correction unit 400 uses a natural language analysis program for knowledge calculation. By using the natural language format, various types of knowledge, such as IF-THEN types, can be set in a natural, conversational format. Note that the knowledge expression format is not limited to the example given here. Any explicit knowledge expression format can be adopted as long as there is a program that can calculate that knowledge.

[0057] Incidentally, it is desirable that the correction information (knowledge) for chemical injection operations expressed in an explicit format include at least information on the turbidity of the raw water to be treated by the waterworks facility.Turbidity is an index that indicates the degree of cloudiness of water and is commonly used in water quality management.

[0058] Returning to the explanation of the flowchart in Figure 6, in correction knowledge reading step S22, the chemical injection rate correction unit 400 reads the correction knowledge set in the previous correction knowledge setting step S21, making it available for use in subsequent steps. The chemical injection rate correction unit 400 reads and analyzes the correction knowledge, and applies (reflects) the results of the analysis to the model parameters of the chemical injection model. That is, the chemical injection rate correction unit 400 adjusts the model parameters of the chemical injection model based on the results of the analysis of the correction knowledge.

[0059] Next, in a correction knowledge input data reading step S23, the chemical injection rate correction unit 400 reads from the water purification facility DB 20 the values ​​of the data items that will be input into the correction knowledge (referred to as "input data").

[0060] Next, in the correction knowledge calculation step S24, the chemical injection rate correction unit 400 applies the input data to the correction knowledge to calculate the correction value of the chemical injection rate. Here, the order in which the correction knowledge to be used to calculate the correction value of the chemical injection rate from among the multiple knowledge [1] to [n] shown in FIG. 7 is applied may be fixed and set in advance. Also, the correction knowledge to be used may be set in advance depending on the water quality (e.g., the raw water turbidity value). Furthermore, the operation manager may operate the input device 35 to manually select the correction knowledge to be used to calculate the correction value of the chemical injection rate.

[0061] Then, in the final injection rate correction value output step S25, the chemical injection rate correction unit 400 outputs the calculated correction value to the water purification facility DB 20, the water purification knowledge DB 500, and the display device 600. After the injection rate correction value output step S25 is completed, the processing of this flowchart ends. The above is an explanation of the chemical injection rate correction unit 400.

[0062] [Example of display on a display device] Next, an example of a screen display output from the water purification chemical injection control system 100 to the display device 600 will be described with reference to FIG. 8 is a diagram illustrating an example of a screen display output from the water purification chemical injection control system 100 to the display device 600. This screen display example is configured with a menu display area 610, a water purification facility process data trend display area 620, a water purification knowledge display area 630, and a chemical injection rate display area 640, thereby improving visibility.

[0063] The water purification knowledge display area 630 clearly displays the correction knowledge (see FIG. 7) used by the chemical injection rate correction unit 400, which is one of the features of the present invention. This allows the operation manager to visually confirm the logic used to correct the chemical injection rate of the chemical injection model. In the example of FIG. 8, knowledge [1] is displayed in the water purification knowledge display area 630.

[0064] The chemical injection rate display area 640 clearly shows both the recommended value (mg / L) and the current value (mg / L) of the chemical injection rate. For the recommended value, the calculated value of the chemical injection model (machine learning calculated value) and the corrected value using correction knowledge are displayed side by side, allowing the operation manager to understand what corrections have been made. The layout position of each area on the screen may be changed as appropriate, other than the example screen display in Figure 8.

[0065] As described above, the water purification chemical injection control system 100 of this embodiment comprises a chemical injection model construction unit 200 that calculates the chemical injection rate through machine learning of the water facility's chemical injection operation performance data and constructs a chemical injection model that reproduces the chemical injection operation; a chemical injection rate calculation unit 300 that calculates the chemical injection rate using the chemical injection model based on the current values ​​of the water facility's measurement data; and a chemical injection rate correction unit 400 that refers to a water purification knowledge database 500 containing correction information for chemical injection operations expressed in an explicit format and calculates the correction amount for the chemical injection rate using the chemical injection model.

[0066] According to this embodiment, the chemical injection rate (recommended value) of a chemical injection operation is corrected using explicit knowledge based on a machine learning model that has been trained on injection operation records. This configuration significantly improves the explainability (black-box resistance) of the correction of the chemical injection rate of a chemical injection operation presented by the water purification chemical injection control system 100. Furthermore, this configuration improves the explainability of the correction of the chemical injection rate of a chemical injection operation using a machine learning model without increasing the workload on-site. Furthermore, improving the explainability of the correction makes the basis for the correction amount clearer, which is expected to increase the validity (robustness) and convincingness of the correction.

[0067] In the water purification chemical injection control system 100 described above, by executing a series of processing units, it is expected that the chemical injection model, which is a machine learning model, can simultaneously solve two problems: robustness, which makes it difficult to ensure accuracy under extrapolation conditions outside the range of training data, and black box nature, which makes the model difficult to read.

[0068] Furthermore, in this embodiment, the correction information (correction knowledge) for chemical injection operations expressed in an explicit format in the water purification knowledge DB 500 is defined separately from the chemical injection model. In this way, by defining the correction information for chemical injection operations expressed in an explicit format separately from the chemical injection model, it is expected that the explainability will improve and the black-box nature will be alleviated. The above is a description of an example of a representative embodiment relating to the functions and operations of the water purification chemical injection control system 100, which is a feature of the present invention.

[0069] A generation AI may be applied to the chemical injection rate correction unit 400 of the water purification chemical injection control system. For example, the generation AI model may be trained on the chemical injection operation performance data, the chemical injection rate, and the correction value, and then the generation AI model may create an outline of the correction knowledge. Alternatively, for example, the generation AI model may select correction knowledge to use from multiple correction knowledge based on the chemical injection operation performance data and the current chemical injection rate.

[0070] As described above, the present invention is not limited to the above-described embodiments, and it goes without saying that various other modifications and applications are possible as long as they do not deviate from the gist of the invention described in the claims. For example, the above-described embodiments have been described in detail and specifically to clearly explain the present invention, and are not necessarily limited to those including all of the components described. Furthermore, it is also possible to add, replace, or delete other components from part of the configuration of each embodiment.

[0071] Furthermore, the above-described configurations, functions, processing units, etc. may be partially or entirely realized in hardware, for example, by designing them as integrated circuits, etc. As the hardware, a broad processor device such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit) may be used.

[0072] Furthermore, each component of the water purification chemical injection control system according to the above-described embodiment may be implemented in any hardware as long as the respective hardware can transmit and receive information to each other via a network. Furthermore, the processing performed by a certain processing unit may be realized by a single piece of hardware, or may be realized by distributed processing using multiple pieces of hardware. [Explanation of symbols]

[0073] 10...water purification facility, 15...sensor, 20...water purification facility DB, 30...computer, 31...CPU, 32...ROM, 36...non-volatile storage, 100...water purification chemical injection control system, 200...chemical injection model construction unit, 300...chemical injection rate calculation unit, 400...chemical injection rate correction unit, 500...water purification knowledge DB, 600...display device, 630...water purification knowledge display area, 640...chemical injection rate display area

Claims

1. A chemical injection model construction unit that calculates chemical injection rates through machine learning of data on the performance of chemical injection operations at waterworks facilities and constructs a chemical injection model that reproduces chemical injection operations; a chemical injection rate calculation unit that calculates a chemical injection rate using the chemical injection model based on current values ​​of the measurement data of the water facility; a chemical injection rate correction unit that calculates a correction amount for the chemical injection rate using the chemical injection model by referring to a water purification knowledge database that includes correction information for the chemical injection operation expressed in an explicit format. Water purification chemical injection control system.

2. The correction information of the chemical injection operation expressed in an explicit format in the water purification knowledge database is defined separately from the chemical injection model. The water purification chemical injection control system according to claim 1 .

3. a display unit that displays correction information of the chemical injection operation expressed in an explicit format, which information is referred to when the chemical injection rate calculation unit calculates the correction amount of the chemical injection rate. The water purification chemical injection control system according to claim 2 .

4. The calculated value of the chemical injection rate by the chemical injection rate calculation unit and the corrected value of the chemical injection rate by the chemical injection rate correction unit are displayed together on the display unit. The water purification chemical injection control system according to claim 3 .

5. The correction information of the chemical injection operation expressed in an explicit format used in the chemical injection rate correction unit is in an IF-THEN rule format. The water purification chemical injection control system according to claim 2 .

6. The correction information of the chemical injection operation expressed in an explicit format used in the chemical injection rate correction unit is in a functional format. The water purification chemical injection control system according to claim 2 .

7. The correction information of the chemical injection operation expressed in an explicit format used by the chemical injection rate correction unit is in a natural language format. The water purification chemical injection control system according to claim 2 .

8. The correction information for the chemical injection operation includes at least information regarding the turbidity of the raw water to be treated by the water facility. The water purification chemical injection control system according to claim 2 .

9. The model used for machine learning in the chemical injection model construction unit is a hierarchical neural network model. The water purification chemical injection control system according to claim 2 .

10. A learning algorithm of the hierarchical neural network model used in the chemical injection model construction unit adjusts model parameters based on the error between the teacher data values ​​to be machine-learned and the calculated values ​​of the hierarchical neural network model. The water purification chemical injection control system according to claim 9.

11. A water purification chemical injection control method in which a computer having an arithmetic device that executes a program and a storage device that stores the program executes water purification chemical injection control, The computer We have constructed a chemical injection model that calculates chemical injection rates using machine learning of data on chemical injection operations at water facilities and reproduces chemical injection operations. calculating a chemical injection rate using the chemical injection model based on the current value of the measurement data of the water facility; A water purification knowledge database including correction information for the chemical injection operation expressed in an explicit format is referenced, and a correction amount for the chemical injection rate is calculated using the chemical injection model. Water purification chemical injection control method.

12. A procedure for constructing a chemical injection model that calculates chemical injection rates and reproduces chemical injection operations through machine learning of data on the chemical injection operations of waterworks facilities; calculating a chemical injection rate using the chemical injection model based on current values ​​of measurement data from the waterworks facility; a step of calculating a correction amount for the chemical injection rate using the chemical injection model by referring to a water purification knowledge database including correction information for the chemical injection operation expressed in an explicit format; A program for a computer to run.

Images