JP2005346714A - Parameter adjustment support apparatus and method of process model - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To allow parameters to be adjusted as easily as possible and to adjust parameters while discriminating identifiabilities of parameters of a given white box model. <P>SOLUTION: A parameter adjustment support apparatus and method of a process model is provided; with a process data collection means 2 consisting of a process external input collection means 21 and a process output collection means 22, which collects and holds measurement data of process external inputs of a process, which is such a limited biological chemical process that dynamics are represented by mutual variations between a plurality of materials, and the plurality of materials; a process simulation means 3 which performs simulation by entering process external input data to the process model consisting of material balance and reaction rates; a process output display means 10; a process state transition diagram display means 11; and a parameter value input means for entering parameter values. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a parameter adjustment support device for a process model that simulates the behavior of a general process such as a biological process such as a sewage treatment process and a food processing process, and a chemical process such as petroleum refining, and Regarding the method.

Process models / process simulators to simulate process behavior in recent years in chemical processes such as water treatment processes such as water and sewage water treatment processes, biological processes using microbial metabolism such as food processing processes, and oil refining processes The use of is increasing. (Here, the term process simulator often refers to a process model that is implemented on a computer so that it can be calculated on a computer, etc., such as differential equations and partial differential equations. However, (Here, the process model and the process simulator are identified, and more appropriate language is used depending on the context.)
This is because, for example, the following can be performed if there is a process simulator that can simulate an actual process in detail.

(1) An optimum process is designed by using a process simulator for a preliminary examination when designing a process (design support).

(2) In the case of process operation management and maintenance management, a study for more appropriate operation management / maintenance management is performed by behavior analysis using a process simulator (operation management support).

(3) Predicting a dangerous phenomenon that may occur in advance by performing scenario simulation assuming various situations (scenario simulation).

(4) Using a process simulator, estimate the value of a variable that cannot be measured online from software that can be measured online (software sensor, observer).

(5) Design an automatic controller incorporating a process simulator (model-based control).

(6) The process simulator is used to optimize the economic cost and quality (in the case of a chemical process) and the economic cost and environmental cost (in the case of an environmental process such as a water treatment process) (short-term process optimization).

(7) Use process simulator to calculate life cycle cost (LCC) from the process design stage to the process of depreciating and discarding the process, and to perform life cycle assessment (LCA) (long-term process optimization) ).

(8) It is used for education and training of process engineers by simulating and visualizing internal phenomena of processes that are difficult to understand except by experts who are familiar with the process (simulation / operation training).

  In this way, the process model / process simulator potentially has various uses, but whether or not the process model can actually be applied to various real problems depends on the parameters of the process model in the actual process. It depends on whether or not it can be adjusted well so that it conforms to. In other words, a process model that is a reduction of some form of a phenomenon actually occurring in the real world (real world) is meaningful only if it can accurately approximate the real world.

  By the way, how to create a process model is roughly divided into a black box approach and a whiteboard approach.

  The black box approach is a method of mathematically constructing a model by analyzing data obtained from an actual process, and this model is often called a black box model. Typical examples of the black box approach include system identification, which is a field of control theory, neural networks that mathematically model the work of the human brain, and multivariate analysis used in the field of statistics. There is. The black-box approach is a promising approach for modeling processes with a complex internal structure compared to mechanical or electromagnetic systems, such as chemical and biological processes. In this approach, a model is constructed from real data so as to match the actual phenomenon, so it can be expected to approximate the real world with high accuracy. However, there are the following problems.

(A) If time series data of an actual process does not exist, modeling cannot be performed. Therefore, it cannot be used for prior examination before process construction such as process design. (B) Basically, modeling is performed to match actual data, so it is often difficult to simulate phenomena that have not occurred in the past. (There is a limit to the accuracy of generalization ability and prediction ability).

(C) The parameters of the black box model (although they have mathematical meanings and system-theoretic meanings) usually do not have physicochemical meanings, so it is difficult to associate them with physical quantities or chemical quantities. For this reason, it is difficult to explain the model to non-experts and it is unacceptable to ordinary people who are not experts. Therefore, it is difficult to spread.

  On the other hand, the white box approach is a method of constructing a process model based on physicochemical insight, and a model constructed by this method is called a white box model, a physical model, a physicochemical model, or the like. The white box model is an attractive method in the sense that the above (A) to (C), which are disadvantages of the black box model, can be solved. However, despite the fact that the white box model is built according to the laws of physicochemistry, it is often very difficult to adapt it to the real process (especially when the model is complex). This is because a unified procedure for adjusting a plurality of parameters of the white box model so as to be adapted to an actual process is often not clear. Therefore, process engineers often make adjustments by trial and error using empirical knowledge. However, such a method has the following problems.

(A ') A person who does not have detailed knowledge about the internal structure of the process cannot provide a parameter adjustment guideline. In addition, it is difficult to understand the meaning of the guidelines. Therefore, as a result, the white box model has a high possibility of becoming “a spider written in a picture”.

(B ′) Before concrete parameter adjustment, it is not clear whether it is possible to determine the values of the parameters of the process model constructed by the white box approach in principle (structural Identifiability issue). (Here, the concept of whether or not a parameter value can be uniquely determined in principle is called identifiability. In particular, it does not include a model and noise that can completely (exactly) represent an actual process. A model that can uniquely determine the value of a parameter given complete data is called a structurally identifiable model, and the problem of structural identifiability is generally the parameter space Θ⊆R of the model. The problem is whether or not a reverse mapping of the mapping from ^{p 1} , (p: number of parameters) to model output (observed by sensor) space Y⊆R ^p , (p: number of outputs) exists uniquely. However, it is not easy to examine structural identifiability for individual process models.)
Therefore, in a model that is not structurally identifiable, the same response (output) may be obtained even for combinations of different values of a plurality of parameters, and it may be difficult to provide a guideline for parameter identification (adjustment). is there.

(C ′) If the response data actually obtained from the target process is disturbed by noise or the response data does not change much (because the process input does not change), parameter adjustment is performed. Is not enough information. This limits the number of parameters that can be adjusted (practical identifiability problem). (Here, even if the model is structurally identifiable, if the data obtained is limited or the data contains noise, the model parameters may not be identified. In this way, identifiability that depends on the quality of the data (with the model structure) is not practical identifiability, but depends on the model structure alone. Is desired.)
In order to adapt the white box model to the real world, the person who handles the white box model (especially the creator of the white box model) despite the problems (A ') to (C'). Sometimes focused their attention only on physicochemical properties such as mass balance, energy balance, reaction process and reaction rate, and sometimes neglected the problem of parameter identification / adjustment as a final tuning problem. . For this reason, a person using the white box model has repeated trial and error for adjusting the parameters, and spends a lot of time on this. In some cases, parameter adjustment has failed, and as a result, the white box model is often not used.

  In view of the above problems, an object of the present invention is to provide a practical adjustment method that can adapt a white box model to a real world.

  In particular, in connection with the above (A ′), it is possible to provide an adjustment support device that enables a process engineer who is not familiar with the internal structure of the process to perform parameter adjustment as easily as possible, and (B ′ ) And (C ′), to provide a parameter adjustment device that adjusts parameters while determining the identifiability of the parameters of a given white box model.

A process model parameter adjustment support apparatus according to the present invention includes a process external input including disturbance and an operation amount, a process state amount, a process observation output that is a function of the process external input and the process state amount, and the process In an arbitrary process having an external input sensor for measuring an external input and a process output sensor for measuring the process observation output, the process external input and the process state in the process are composed of a plurality of substances, and the process Is a limited biological and chemical process in which dynamics are expressed by mutual changes between the plurality of substances,
A process external input collecting means for collecting measurement data of the external input sensor and holding the data for a predetermined period, and a process for collecting measurement data of the process output sensor and holding the data for a predetermined period Process data collection means comprising output collection means;
A process external input supplied through the process data external input collection means to a process model having a parameter set (parameter vector) θεRp (p: number of parameters) composed of a plurality of parameters and configured by a material balance and reaction rate A process simulation means for simulating the temporal and spatial behavior of the process by inputting data;
Process output display means for displaying a process output simulation value by the process simulation means and the process output value;
A process state transition diagram display means for representing a state interaction and a state change of the process model;
Parameter value input means for inputting a parameter value of the parameter set attached to the process state transition diagram;
It is characterized by comprising.

  According to the present invention, the parameter adjuster can adjust the parameters while considering the physicochemical meaning of the parameters of the process model based on the process state transition diagram, thereby facilitating the parameter adjustment. be able to.

  Further, the present invention is characterized in that the parameter set input means has a help function for explaining a physicochemical meaning of each parameter which is an element of the parameter set.

  According to the present invention, a parameter adjustment guideline can be supported by presenting the physicochemical meaning of a parameter of the process model in detail and in an easy-to-understand manner to a parameter adjuster. As a result, the process model can be made closer to an actual process.

  In addition, the present invention is characterized in that a link is established from the process state transition display means to the mathematical expression, and the process state transition display means and the mathematical expression representing the process model are associated with each other.

  According to the present invention, the parameter adjuster can easily understand the correspondence between the mathematical description of the process model and the process state transition diagram, and can deepen the understanding of the meaning of the process model. This makes it possible for parameter adjusters who have mathematical knowledge about the process model to understand the meaning of parameters more deeply, which in turn enables the parameter adjuster to more accurately adapt the process model to the actual process. can do.

  Further, according to the present invention, the process state transition diagram display means includes a diagram showing the process state variable to be displayed and the size and color of a line connecting the process states of the parameter values input by the parameter set input means. It is modified so as to be variable depending on the size.

  According to the present invention, it is possible to visually understand the meaning of each parameter of the parameter set. For this reason, for example, by displaying the process state transition diagram when the default value (initial value) of the parameter is given and the process state transition diagram after parameter adjustment, the parameter adjuster adjusts which parameter is adjusted to what extent. It becomes possible to understand visually easily. Therefore, parameter adjustment of a process simulator that includes a large number of parameters can be made simpler.

  In the present invention, the process state transition display means and the parameter set input means are mounted on a device different from the process simulation means, and a parameter set can be input remotely from a WWW (World Wide Web) browser. It is modified as follows.

  According to the present invention, as in the case of designing a process controller (for example, IMC: Internal Model Control) incorporating a process simulator, the process controller is installed at a site where an actual process exists, and the parameters of the process model are monitored. In such a situation, the parameter adjustment can be easily performed remotely using the Internet technology. Thereby, process controller tuning by remote control is attained.

The process model parameter adjustment support method includes a process external input (disturbance and manipulated variable), a process state quantity, a process observation output that is a function of the process external input and the process state quantity, and the process external input. In any process having an external input sensor for measuring the process output sensor for measuring the process observation output,
The process external input and the process state in the process are composed of a plurality of substances, and the process is a limited biological / chemical process in which dynamics are expressed by mutual changes between the plurality of substances. Yes,
Process for collecting measurement data of the external input sensor and holding the data for a predetermined period, and process for collecting the measurement data of the process output sensor and holding the data for a predetermined period A process data collection process comprising an output collection process;
Process external input supplied through the process data external input collection step for a process model having a parameter set (parameter vector) θεRp (p: number of parameters) composed of a plurality of parameters and configured by a material balance and reaction rate A process simulation step that simulates the temporal and spatial behavior of the process by entering data; and
A process output display step for displaying a process output simulation value by the process simulation means and the process output value;
A process state transition diagram display step representing a state interaction and a state change of the process model;
A parameter value input step for inputting a parameter value of the parameter set attached to the process state transition diagram;
It is characterized by comprising.

  As described above, the effects obtained by the configuration of the present invention are summarized as at least the following two points.

(1) In a process simulator having a plurality of parameters represented by ordinary differential equations or partial differential equations, a parameter set that can be automatically identified according to the quality of available data is extracted and identified. it can.

(2) Parameter adjustment of a process model representing a biological process or chemical process such as a sewage treatment process, a water treatment process, or a petrochemical process can be facilitated by providing visual information.

  With these two effects, the physicochemical model of the target process can be finally adapted to the actual target process. As described in [Prior Art], various processes using the process model can be performed. Consideration becomes possible.

  FIG. 1 shows a basic configuration of an embodiment of the present invention together with a sewage treatment process on the assumption that the target process is a sewage treatment process.

The sewage treatment process parameter adjustment apparatus shown in FIG. 1 includes a sewage treatment process 1 as a control target, disturbance data such as the amount of sewage flowing into the sewage treatment process 1 and the quality of inflow sewage, and the operation amount such as the amount of aeration air and the amount of returned sludge Process external input data consisting of data, and observation output such as dissolved oxygen (DO), suspended solids (MLSS), pH, or COD (chemical oxygen demand) measured by sensors installed in each reactor The sewage treatment process 1 using process data collection means 2 that collects data at a predetermined cycle and stores the data in a process data server, and a process model configured based on physicochemical knowledge of the sewage treatment process 1 the COD (chemical oxygen demand) amount of organic matter measured by indicators such as, DO, MLSS, ammonia nitrogen _(NH 4 -N), And Santai nitrogen (NO 3 _-N) and phosphorus Santai phosphorus (PO 4 _-P) process simulation means 3 capable of simulating the behavior of various water quality, such as, with the process model is used in the process simulation means 3 Initial parameters to set initial values of parameters such as kinetic parameters such as maximum specific growth rate, half-saturation constant, hydrolysis rate constant related to reaction rate, and stoichiometric parameters such as yield related to material balance Initial values such as the kinetic parameters and stoichiometric parameters are set by the parameter initial value setting means 4 to the parameter sensitivity function model that can be derived from the process model used in the value setting means 4 and the process simulation means 3. Parameter sensitivity to analyze the sensitivity of parameters near the default setting Analysis means 5, parameter sensitivity display means 6 for displaying the time change of the sensitivity of the parameter analyzed by the parameter sensitivity analysis means 5, and a highly sensitive parameter set among all the parameters analyzed by the parameter sensitivity analysis means 5 A high-sensitivity parameter set extraction means 7 for extracting the parameters, and each parameter of the high-sensitivity parameter set extracted by the high-sensitivity parameter set extraction means 7, the process output data by the process data collection means 2 and the process output by the process simulation means 3 Parameter identification means 8 for adjusting / identifying so that the error in the value of the simulation data becomes small, process external input data and process output data of the process data collection means 2, and process output stain calculated by the process simulation means 3 Error calculation and storage means 9 used to evaluate and identify parameter identification means 8, process output data supplied from process output data collection means 22, and process calculated by process simulation means 3. And a process value / process simulation value display means 10 for simultaneously displaying output simulation data.

In addition, the sewage treatment process 1 includes an initial settling basin 11, a biological reaction tank 12, a final settling basin 13, a blower 141, a return pump 142, and an excess sludge extraction pump, which are operation amounts to each facility of the sewage treatment process 1. Three actuators 143, a process disturbance sensor 151 for measuring the amount of influent sewage considered to be a process disturbance and various influent sewage qualities, an air volume sensor 152 for measuring aeration air volume, a sensor 153 for measuring the amount of returned sludge, and excess sludge extraction A sensor 154 that measures the amount of water, and process output sensors 161 to 163 that measure various water quality values such as DO, NH ₄ -N, NO ₃ -N, and PO ₄ -P of each facility of the sewage treatment process 1 Is done.

  The process data collection means 2 includes a process external input data collection means 21 for collecting external input data composed of disturbance and manipulated variable data of the sewage treatment process, and a process output data collection means for collecting various water quality data of the sewage treatment process. 22.

The operation of this embodiment will be described with reference to FIG.
In this embodiment, in order to simplify the explanation, explanation will be made using a simple process model that simulates the behavior of a sewage treatment process. However, the application target of the present invention is limited to such a simple process model. is not.

  First, a parameter sensitivity analysis model used for the parameter sensitivity analysis unit 5 is derived in advance from the process model used by the process simulation unit 3.

  A process model in the case of a stirred continuous reaction tank (CSTR) in which sewage flows directly into the biological reaction tank 12 ignoring the presence of the first settling tank 11 and the final settling tank 13 is given as follows, for example. .

ここで、Ｙhは収率、μmaxは最大比増殖速度、Ｋは酸素濃度に関する半飽和定数、Ｋhは加水分解定数、Ｋxは微生物−難分解性有機物の濃度比に関する半飽和定数、ＫＬは総括酸素移動容量に関する係数（ＫＬ・Ｑbが総括酸素移動容量係数）、ｂは死滅速度、を表すパラメータである。Ｖは反応槽の水量であり一定値であるとする。また、Ｓo2(t)は溶存酸素濃度（ＤＯ）、Ｓc(t)は溶解性有機物濃度、Ｘc(t)は難分解性有機物濃度、Ｘn(t)は微生物濃度、を表す状態変数であり、Ｑ(t)、ＱB(t)は、各々流入量と曝気量を表す入力変数である。各状態変数に添え字のＩＮをつけたものは、各状態変数に対応する流入水質である。また、Ｙ(t)は、溶存酸素濃度（ＤＯ）と溶解性のＣＯＤに対応する出力変数であるが、ＤＯとＭＬＳＳを出力変数とした場合は、以下のように変更される。

Where Yh is the yield, μmax is the maximum specific growth rate, K is the half-saturation constant relating to the oxygen concentration, Kh is the hydrolysis constant, Kx is the half-saturation constant relating to the concentration ratio of the microorganism-refractory organic matter, and KL is the overall oxygen. A coefficient relating to a transfer capacity (KL · Qb is an overall oxygen transfer capacity coefficient), and b is a parameter representing a death rate. V is the amount of water in the reaction tank and is a constant value. In addition, So2 (t) is a dissolved oxygen concentration (DO), Sc (t) is a soluble organic matter concentration, Xc (t) is a persistent organic matter concentration, and Xn (t) is a microbial concentration. Q (t) and QB (t) are input variables representing the inflow amount and the aeration amount, respectively. Each state variable with a subscript IN is the quality of the inflow water corresponding to each state variable. Y (t) is an output variable corresponding to dissolved oxygen concentration (DO) and soluble COD, but when DO and MLSS are output variables, it is changed as follows.

なお、参考のため、このプロセスモデルの反応部分を参考文献［１］［LAWQ Task Group on Mathematical Modeling for Design and Operation of Biological Wastewater Treatment"Activated Sludge Model No.2", IAWQ Scientific Technical Report No.3,(1995)］の記法に従って記述すると、表１のようになる。

For reference, the reaction part of this process model is referred to [1] [LAWQ Task Group on Mathematical Modeling for Design and Operation of Biological Wastewater Treatment “Activated Sludge Model No. 2”, IAWQ Scientific Technical Report No. 3, (1995)] is described in Table 1.

また、表記の簡単化のため、次式の様にプロセスモデルをベクトル形式で表現しておく。

In order to simplify the notation, the process model is expressed in a vector format as shown in the following equation.

ここで、調整の対象となるパラメータの候補は、θ＝［Ｙh，μmax，Ｋ，Ｋh，Ｋｘ，ＫＬ，ｂ］の７つである。

Here, there are seven parameter candidates to be adjusted, θ = [Yh, μmax, K, Kh, Kx, KL, b].

The parameter sensitivity function Z _θ (t) for the state variable is defined as follows:

ここで、Ｚ_θ(t)のｉ行ｊ列の要素（Ｚ_θ(t)）ｉｊは、次式で定義される。

Here, Z _theta i-th row and j-th column element (Z _theta (t)) ij of (t) is defined by the following equation.

また、センサーで計測している計測値に対してのみパラメータの感度を考える場合は、出力変数に対するパラメータ感度関数Ｙθ(t)を次式で定義する。

Further, when considering the sensitivity of the parameter only for the measurement value measured by the sensor, the parameter sensitivity function Yθ (t) for the output variable is defined by the following equation.

また、パラメータ調整によってプロセスシミュレーションの出力値と実際にセンサーで計測している観測値とを適合させたい出力変数が複数ある場合には、出力変数に対するパラメータ感度関数（１９）の汎関数（Ｒ^ｑ−Ｒ^１への写像、ｑは出力変数の数）に対するパラメータ感度関数を定義する。

Further, when there are a plurality of output variables for which the output value of the process simulation and the observation value actually measured by the sensor are to be matched by parameter adjustment, the functional (R ^q of the parameter sensitivity function (19) for the output variable is set. Define a parameter sensitivity function for -R ¹ mapping, q is the number of output variables.

  For example, in the example of the equations (1) to (4), when it is desired to adapt the process simulation values of So2 (t) and Sc (t) to the actual values, the following evaluation function J is used as the functional. And a sensitivity function for this evaluation function can be defined.

ここで、（・）^ｒｅａｌは実際にセンサーなどで計測した計測値を表しｗ_１やｗ_２は重みを表す。評価関数Ｊに関する感度は、次式で定義できる。

Here, (·) ^real represents a measured value actually measured by a sensor or the like, and w ₁ and w ₂ represent weights. The sensitivity regarding the evaluation function J can be defined by the following equation.

他の汎関数の選び方として、例えば、

とすることもできる。この場合の感度関数は、

となる。

For example, how to select other functionals:

It can also be. The sensitivity function in this case is

It becomes.

  Note that Equations (24) and (28) are calculated from Equation (15), the weight, the observation output, and the simulation output. Therefore, if equation (15) can be calculated, equations (24) and (28) can be calculated using known information.

When some of the elements of the parameter vector θ are not to be adjusted in advance, some elements in θ , for example, θ : = [μmax Kh b KL] are newly regarded as parameter vectors. A parameter sensitivity function may be defined.

From the above, if the equation (15) can be calculated, the equations (24) and (28) can be calculated. Therefore, the calculation method of the equation (15) may be considered. By differentiating _Zθ with time t and using equation (7), the following differential equation for the parameter sensitivity function Z _θ (t) can be derived.

（３０）式は、プロセスモデルとパラメータ（のデフォルト）値が与えられれば計算可能である。従って、（２４）式や（２８）式も計算可能である。

Equation (30) can be calculated if a process model and parameter (default) values are given. Therefore, equations (24) and (28) can also be calculated.

  The parameter sensitivity analysis model used in the three-parameter sensitivity analysis means described below is the one obtained by substituting the equation (30) into the equations (19), (24), (28) or the equation (30) itself.

As a specific example of the expression (30), for example, the sensitivity function of the expression (30) when the parameter vector θ : = [μmax Kh b KL] is given by the following expression.

このような計算を行うことによって、パラメータ感度解析モデルを導出することができる。

By performing such calculations, a parameter sensitivity analysis model can be derived.

The important points to be particularly noted regarding this parameter sensitivity analysis model are as follows.

(I) ∂f / ∂Z and ∂f / ∂θ used in equation (30) can be analytically calculated directly from the model if there is a process model used in the three-process simulation means. For this reason, in the sensitivity calculation of parameters, it is not necessary to perform a troublesome procedure of repeatedly performing simulation by changing the value of each parameter.

(II) When the parameter sensitivity function is defined by equation (24) or (28), the sensitivity function of equation (24) or (28) is a vector having the same number of elements as the number of parameters. Therefore, when considering extracting a high sensitivity parameter set using the sensitivity function defined in this way, a high sensitivity parameter set equal to or less than the number of parameters can always be extracted.

(III) Since ∂f / ∂Z and ∂f / ∂θ used in (30) are functions depending on process state variables and inputs, if the process input value changes, these values also change accordingly. Change. As a result, the parameter sensitivity function becomes a function dependent on the process input value, and changes according to the type of process input and the waveform of the input. The high-sensitivity parameter set extraction means described below can change the number of parameters extracted as a high-sensitivity parameter set in accordance with such a change in input value. Therefore, identifiable parameters can be extracted without being conscious of the structural and practical identifiability of the process model.

  Based on the above preparation, the operation and effect of the embodiment will be described.

  First, the inflow amount Q (t) measured by the disturbance sensor 151 in the sewage treatment process 1 and a plurality of influent water quality components, for example, dissolved oxygen concentration DO, soluble organic matter Sc (t) (measured as soluble COD). ), Floating organic matter Xc (t) (measured as a component of total COD-soluble COD multiplied by α <1), heterotrophic microorganism Xh (t) (component of total COD-soluble COD) Obtained by multiplying 1−α by 1) and stored in a predetermined area of the process data server by the process disturbance input data collecting means 21 at a predetermined measurement cycle. At the same time, the measured values of the operation amount collected by the operation amount sensors 152 to 154 in the sewage treatment process 1 are stored in a predetermined area of the process data server by the process external input data collecting means 21 at a predetermined measurement cycle. Further, water quality values such as DO and MLSS measured by the sensors 161 to 163 are stored in a predetermined area of the process data server by the process output data collecting means 22.

  At the same time or in advance, the parameter initial value setting means 4 sets parameter initial values possessed by the process model and the parameter sensitivity analysis model used by the process simulation means 3 and the parameter sensitivity analysis means 5. Here, it is preferable to set a parameter value in a range assumed in advance based on physicochemical considerations (because the sensitivity of the parameter also depends on the initial parameter value). For example, as a biological reaction model of a sewage treatment process model, IWA (International Water Association) has published activated sludge models No. 1 to No. 3 (ASM1 to ASM3). In such a case, a default value may be set. In addition, when such a default value is not given, for example, the parameter of the content rate in which the substance A is contained in the substance B is a value within a theoretically possible range such that the parameter is within a range of 0 to 1. Set the value of.

  Next, the state value of the process is calculated by inputting the external input data collected by the process external input data collecting means 21 to the process model of the process simulation means 3. For example, when considering the process model of equations (1) to (5), input So2IN (t), ScIN (t), XcIN (t), XhIN (t), Q (t), etc. t), Sc (t), Xc (t), Xh (t) and other values are calculated.

  Next, the process state value calculated by the process simulation unit 3 and the external input data collected by the process external input data collection unit 21 are input to the parameter sensitivity analysis unit 5. At this time, data that can be measured online among the process state values may be substituted for the process state value calculated by the process simulation means 3. When such data is input, the parameter sensitivity analysis means 5 can calculate the equation (15). This is calculated, and further the equations (24) and (28) are calculated. For example, when considering the process model of equations (1) to (5), first, a sensitivity analysis model such as equations (31) to (33) is obtained. Here, the equation (31) is changed from the equations (32) to (33) to the initial parameter values of μmax, Kh, b, KL, and So2 (t), Sc (t), Xc (t), Xh (t ) Process state calculated values or measured values, and process external input values such as QB (t) can be calculated. And if (31) Formula can be calculated, (24) Formula and (28) Formula can also be calculated. In this way, the 5-parameter sensitivity analysis means is executed.

に対する感度解析結果は、図２に示すようなグラフとして表示される。 Next, the parameter sensitivity display means 6 displays the time series data of the parameter sensitivity analyzed by the parameter sensitivity analysis means 5. When all the sensitivity analysis results of the equation (15) are displayed, the sensitivity analysis results corresponding to the number of parameters (to perform sensitivity analysis) × the number of state variables are displayed. Further, when displaying the sensitivity analysis results of the equations (24) and (28), the sensitivity analysis results for the number of parameters are displayed. At this time, external input data and calculated or measured state variable data may be displayed simultaneously. For example, when considering the process model of (1) to (5),

The sensitivity analysis result for is displayed as a graph as shown in FIG.

  Next, the high sensitivity parameter set extraction unit 7 extracts a high sensitivity parameter set based on the results of the parameter sensitivity analysis unit 5 and the parameter sensitivity display unit 6.

  As a specific example, when using the invention described in claim 2 at the time of filing a parent application, for example, the parameter sensitivity function of FIG. 2 is first sampled at a predetermined period T and converted to time series data. This conceptual diagram is shown in FIG. As shown in FIG. 3, by using the equations (24) and (28), it is possible to generate sensitivity function vectors for the number of parameters. For example, when considering the process model of the equations (1) to (5), four sensitivity function vectors corresponding to μmax, Kh, b, and KL are generated. These are set to eμmax, eKn, eb, and eKL, respectively. Next, by arranging these generated vectors, a sensitivity function matrix E as shown below is created.

次にこの感度関数行列Ｅに対して、主成分分析、あるいはほぼ同じことであるが、Ｅ^γＥの特異値分解を行う。（なお、ここで、Ｅ^γＥに対する特異値分解は主成分分析の計算途中で現れ、この特異値分解が計算できれば、主成分分析に必要な情報は全て得られる。）
すると次式の様な形に変換することができる。

Next, for this sensitivity function matrix E, principal component analysis, or approximately the same, but singular value decomposition of E ^γ E is performed. (Here, the singular value decomposition for E ^γ E appears during the calculation of principal component analysis. If this singular value decomposition can be calculated, all the information necessary for principal component analysis can be obtained.)
Then, it can be converted into the following form.

このような変換を行った場合、もしパラメータ感度関数同士のお互いの相関が強い場合、例えば図２において、μmaxとｂとＫＬのように相関が強い場合は、λ_ｉ ^２のいくつかは、０に近い非常に小さな値になる。図２に示したような感度関数グラフのような場合は、λ_３ ^２とλ_４ ^２は非常に小さな値になる。従って、この場合λ_１ ^２とλ_２ ^２の２つに対応する２個の高感度パラメータを抽出することができる。この場合主成分分析により、パラメータベクトル自身も以下の様に変換される。

When such a conversion is performed, if the correlation between the parameter sensitivity functions is strong, for example, in FIG. 2, if the correlation is strong such as μmax, b, and KL, some of λ _i ² are 0. Very close to the value. In the case of the sensitivity function graph as shown in FIG. 2, λ ₃ ² and λ ₄ ² are very small values. Therefore, in this case, two high sensitivity parameters corresponding to two of λ ₁ ² and λ ₂ ² can be extracted. In this case, the parameter vector itself is converted as follows by principal component analysis.

そのため、この場合は、高感度パラメータセットはθmoaの第１成分と第２成分の２つのパラメータからなるベクトルとして選ばれる。つまり、

が高感度パラメータとして選択される。自動的にこの操作を行う場合は、λ_ｉ ^２に対する閾値λ ^２ を設定しておき、この閾値以上のものを高感度パラメータセットとして抽出することにすればよい。これが、高感度パラメータセット抽出手段７の一例である。

Therefore, in this case, the high sensitivity parameter set is selected as a vector composed of two parameters of the first component and the second component of θmoa. That means

Is selected as the high sensitivity parameter. If automatically perform this operation, may be set a threshold value lambda ² with respect to lambda _i ^2, it may be to extract more than the threshold value as a highly sensitive parameter set. This is an example of the high sensitivity parameter set extraction unit 7.

  Extracting a high sensitivity parameter set in this way provides the following effects. (1) A highly sensitive parameter set can be extracted by a quantitative standard that can be numerically calculated.

(2) A highly sensitive parameter set can be extracted without being conscious of the structural identifiability or practical identifiability of the process model. If the process model is not structurally identifiable, some of the singular values in the above singular value decomposition (principal component analysis) will always be 0, and inevitably only a sensitive parameter set with less than the number of parameters is obtained. I can't. In addition, although the practical identifiability depends on the quality of the obtained data, as can be seen from the above procedure, when the data quality is improved, the number of singular values having a value equal to or greater than the threshold λ ² increases. The number of parameters that can be adjusted as increases. Conversely, if the quality of the data is poor, only a small number of parameters can be adjusted accordingly. Such an operation can be performed quantitatively and automatically.

(3) A conversion parameter selected when extracting a high sensitivity parameter set is represented by a linear combination of all parameters before conversion, and the number of conversion parameters selected is equal to or less than the number of parameters before conversion. When the number of conversion parameters to be selected is smaller than the number of parameters before conversion, the value of the parameter before conversion is not uniquely determined. For this reason, when identifying parameters after extracting a high-sensitivity parameter set, together with physicochemical considerations, only a certain parameter value is physically obtained using the degree of freedom that the parameter value before conversion is not uniquely determined. Parameter adjustment can be performed so as not to have an extreme value that is chemically impossible.

  As a specific example of the other high-sensitivity parameter set extraction means 7, for example, when the invention described in claim 3 is used, until eμmax, eKh, eb, eKL, which are elements of E, are extracted, Do the same as in the described invention. The subsequent algorithm can be executed as follows, for example.

Step1
The elements of E, eμmax, eKh, eb, eKL are arranged in order of increasing sensitivity. For example, the high sensitivity is evaluated using a norm such as ‖e * ‖i (* = μmax, Kh, b, KL, ‖ · ‖i is i-norm, i = 1,..., ∞). You should arrange in order from the largest. For example, e1, e2, e3, and e4 are set in descending order of sensitivity.

Step 2
Singular value decomposition is performed with E2: = [e1, e2]. If the threshold value lambda ² or more even two values of the singular values, we adopt this as a highly sensitive parameter. If one of the threshold lambda ² or less of the singular values, e2 is not employed as a highly sensitive parameter set, re-defined as E2 = e1.

Step 3
Return to Step 2 as Ei + 1: = [Ei, ei + 1]. When i becomes equal to the number of parameters, the process ends.

  However, the algorithm in the case of the invention described in claim 3 at the time of filing the parent application is not limited to the above algorithm, and the definition of the high sensitivity parameter is made accurately, and the independence among them is Any algorithm can be used as long as it selects a strong algorithm.

  When the high sensitivity parameter set is extracted in this way, in addition to the effects (1) and (2) of the effect of the high sensitivity parameter set extraction means according to the invention described in claim 2 at the time of filing the parent application, there is a physical meaning. Can be directly adjusted parameters. As a result, the effect of (3) above is lost, but highly independent parameters and high sensitivity are extracted as a high sensitivity parameter set. Is not selected, and a high-sensitivity parameter set that allows efficient parameter adjustment can be extracted. For example, parameters that have an inverse relationship such as the growth rate and death rate of microorganisms are rarely used together, and parameters that represent highly sensitive growth rates are usually selected as elements of a high sensitivity parameter set. Will be.

  As another example of the high-sensitivity parameter set extraction means 7, when the invention described in claim 4 at the time of filing the parent application is used, the same number of parameters as the equations (24) and (28) are used. The sensitivity function is displayed on the display by the 6-parameter sensitivity display means, and the one with high sensitivity is extracted from the sensitivity function. Alternatively, the sensitivity function corresponding to (number of parameters x number of state variables) in equation (15) is displayed on the display by the 6-parameter sensitivity display means, and appropriate parameters are extracted from the range not exceeding the number of parameters. To do. For example, when the number of state variables is equal to or less than the number of parameters, it is conceivable that the most sensitive parameter for each state variable is selected one by one. As a specific example, when considering the models of equations (1) to (4), for example, KL for So2 (t), b for So (t), μmax for Xh (t), Xc (t ) Can be selected as Kh or the like. For example, when the value of Xh (t) is not measured and is not known, the KL, b, and Kh may be adjusted.

  Such a high-sensitivity parameter set extraction method is not efficient because it uses a human system, but has the advantage that human knowledge can be reflected in the extraction of the high-sensitivity parameter set. For example, humans have the knowledge that “the parameter KL represents the overall oxygen transfer capacity coefficient and directly affects the dissolved oxygen concentration So2 (t) and does not directly affect other state variables”. . Therefore, it is possible to perform rational high sensitivity parameter set extraction such that KL is selected as an element of the high sensitivity parameter set when adjusting only So2 (t).

Next, the parameter identification unit 8 adjusts (identifies) the values of the plurality of high sensitivity parameter sets extracted by the high sensitivity parameter set extraction unit 7. For example, when using the invention according to claim 5 at the time of filing of the parent application, for example, as the evaluation function (or fitness), the sum of squares of the errors of the equations (24) and (28) is taken, and the value A heuristic search is performed so that is minimized. (Here, in the case of goodness of fit, it is often defined as the reciprocal of the sum of squares of the error, etc., so in that case it is the maximum.)
As a specific example, when an immune algorithm is used for parameter identification of μmax and Kh, the procedure is as follows.

Step1
μmax and Kh are regarded as discrete data, and an initial antibody group composed of a plurality of discrete data is generated. Specifically, a data set obtained by appropriately discretizing real values within a physicochemically meaningful range with respect to μmax and Kh is regarded as a memory cell, and an initial antibody group is generated from the memory cell.

Step 2
Calculate the affinity between antibodies and the affinity between antibody and antigen. In the calculation of the affinity between antibodies, an evaluation value such as a square error between a simulation output value for each antibody (candidate parameter value) and a value measured by an actual sensor is used. This is done based on the 9 error calculation and the value supplied from the storage means.

Step 3
An antibody with high affinity for the antigen is added to the memory cell.

Step4
Calculate the expected value of the antibody that will remain in the next generation, and eliminate the antibody with low expected value.

Step5
A new antibody to replace the extinguished antibody is calculated and mutated with an appropriate probability.

Step 6
End if criteria are met. If not, return to Step 2.

  By such an operation, the parameter identification unit 8 can be executed.

  Since the parameter identification means 8 using such metaheuristics can be executed relatively fast and can be applied to mathematically non-convex search problems, the parameters of the process model, which is a complex function regarding parameters, can be adjusted online. This is a particularly effective method.

  As another execution method of the parameter identification means 8, for example, when applying the invention described in claim 6 at the time of filing of the parent application, it can be performed as follows. First, it is defined as h (Z, θ): = Y (t). This is because Y (t) defined by the equations (5) and (6) is an output of the equations (1) to (4) including θ and is a function of Y (t) itself θ. Thus, it is possible to write h (Z, θ). At this time, if the evaluation function is defined by equation (20), for example, the following can be adopted as an iterative algorithm for nonlinear parameter identification.

ここで、ｉは繰り返し回数を示し、ｉ＝１，２，…である。初期値であるθ^０は、物理化学的に意味のある値を設定しておく。また、Ｋiは繰り返し回数がｉ回目である場合の勾配ゲインを表す行列であり、正定行列である。Ｋiの選び方には、多くの自由度があるが、例えば、Gauss-Newton法を利用する場合は、

の様なアルゴリズムとなる。但し、Ｑは正定行列である。この中で利用する式は（１）〜（５）式を用いれば、全て計算可能であるため、これは実行可能なアルゴリズムである。

Here, i indicates the number of repetitions, and i = 1, 2,. The initial value θ ⁰ is set to a value that is physicochemically meaningful. Ki is a matrix representing the gradient gain when the number of repetitions is the i-th, and is a positive definite matrix. There are many degrees of freedom in choosing Ki. For example, when using the Gauss-Newton method,

The algorithm is as follows. Where Q is a positive definite matrix. This is an executable algorithm because all of the formulas used in this can be calculated by using formulas (1) to (5).

  Since such parameter identification means is based on a nonlinear least square method, a maximum estimation method, or the like, an optimal parameter value can be obtained in the sense of the least square error or the meaning of luminous intensity. In addition, when the default value of a parameter is given in advance, such as ASM2, for example, the vicinity is searched, so that the possibility of an abnormal value that is impossible in physicochemical is reduced. effective.

  As another method, when the invention according to claim 7 at the time of filing of the parent application is adopted for another execution of the parameter identification means 8, the parameter with the highest sensitivity extracted by the high sensitivity parameter set extraction means 7 is used. In order, the adjustment is performed with reference to the display result of the process value / process simulation value display means 10. In this case, it is possible to easily calculate how the output changes when the parameter value is changed by using the equation (15) indicating the ratio of the change amount of the state variable to the change amount of the parameter. Therefore, the display of the 10 process value / process simulation value display means when the parameter value is changed can be easily changed. Therefore, along with the change of the parameter value, the value of the state variable is corrected using the equation (15), and the parameter can be adjusted (identified) while displaying the result.

  Such parameter identification means has an effect that only high-sensitivity parameters can be adjusted as efficiently as possible while collating with physicochemical knowledge.

[Other Examples]
The sewage treatment process parameter adjustment apparatus shown in FIG. 4 includes a sewage treatment process 1 as a control target, disturbance data such as an inflow sewage amount and an inflow sewage quality into the sewage treatment process, and operation amount data such as an aeration air amount and a return sludge amount. The process external input data consisting of and the observation output data such as dissolved oxygen (DO), suspended solids (MLSS), pH, or COD measured by the sensors installed in each reaction tank, at a predetermined cycle COD (chemical oxygen demand of the sewage treatment process 1) using the process data collection means 2 collected in the process data server and stored in the process data server and a process model configured based on the physicochemical knowledge of the sewage treatment process 1 ), Etc., organic substances, DO, MLSS, ammonia nitrogen (NH ₄ -N), nitrate nitrogen (NO ₃ -N) ) And phosphate phosphorus (PO ₄ -P) and the like, and the process simulation means 3 capable of simulating the behavior of various water qualities, and the process simulation value calculated and output by the process simulation means 3 and the process data collection means 2 Process value / process simulation value display means 10 for displaying the collected observation output data, and process state transition diagram display showing a process state transition according to the reaction of the process model used in the process simulation means 3 Means 11 and parameter value setting means 12 attached to the process state transition diagram display means 11.

  The sewage treatment process 1 is configured in the same manner as in FIG. The process data collection means 2 is also configured in the same way as in FIG.

Next, the operation and effect of the present embodiment will be described.
The effect | action of a present Example is demonstrated using FIG. In this embodiment, the activated sludge model ASM2 described in the above-mentioned reference [1] will be described as an example of a process model. Of course, the same concept can be used in other biological and chemical processes.

First, similarly to the operation of the embodiment in the case of using FIG. 1, the inflow amount Q (t) measured by the disturbance sensor 151 in the sewage treatment process 1 and a plurality of influent water quality components, for example, dissolved oxygen concentration DO, dissolved Organic Sc (t) (measured as soluble COD). Suspended organic matter Xc (t) (measured as a component of total COD-soluble COD multiplied by α <1), heterotrophic microorganism Xh (t) (measured as a component of total COD-soluble COD) Multiplied by 1-α), phosphorous phosphorus (measured with a PO ₄ -P sensor), ammonia nitrogen (measured with an NH ₄ -N sensor), nitrate nitrogen (NO ₃ -Measured by the N sensor) by the process external input data collecting means 21 and stored in a predetermined area of the process data server at a predetermined measurement cycle. At the same time, the measured values of the operation amount collected by the operation amount sensors 152 to 154 in the sewage treatment process 1 are stored in a predetermined area of the process data server by the process external input data collecting means 21 at a predetermined measurement cycle. Further, water quality values such as DO, MLSS, PO ₄ -P, NH ₄ -N, and NO ₃ -N measured by the sensors 161 to 163 are stored in a predetermined area of the process data server by the process output data collecting unit 22.

  Next, by supplying the process external input data from the process external input data collecting unit 21 to the process simulation unit 3, the temporal and spatial changes in the process state are simulated.

  Next, the process output value supplied from the process output data collecting unit 22 and the process output simulation value calculated by the process simulation unit 3 are simultaneously displayed on the process value / process simulation value display unit 10.

On the other hand, process state transition diagrams as shown in FIGS. 5 to 7 are displayed on the process state transition diagram display means 11. (Here, the process state transition diagrams of FIGS. 5 to 7 show state transitions related to digestion, denitrification, dephosphorization, and organic substance removal of ASM2. Only FIG. 5 will be described below.) (In order to emphasize the complexity of the reaction of the process considered in ASM2, FIG. 6 and FIG. 7 are also included for reference.)
At the same time, as shown in FIG. 5, an input unit capable of setting parameter values is provided on the process state transition diagram. This is an input unit for setting the parameter value setting means 12. In the case where the process state transition diagram display means 11 is constructed on a personal computer Windows, the input unit related to the stoichiometric parameter of the parameter value setting means 12 can be input in, for example, a text box. It may be possible to select a value in a menu format. Further, since the diagram is too complicated, not all display is performed. However, as shown on the right side of FIG. 5, dynamic parameters can be set in the arrow portions representing all state transitions. This dynamic parameter setting unit may be set in a text box or menu format. When the display area on a Windows such as a personal computer is small, a screen for setting dynamic parameters may be displayed as a pop-up window, for example, by clicking the arrow portion of the state transition (see FIG. 5). (See right part).

With the above preparation, first, the difference (error) between the process output value displayed on the process value / process simulation value display means 10 and the process output simulation value is confirmed. For example, when a simulation of a sewage treatment process is performed using ASM2, the output value of the concentration of state variables such as phosphate phosphorus PO ₄ -P, ammonia nitrogen NH ₄ -N, nitrate nitrogen NO ₃ -N, etc. Check the error of the simulation value. Then, the state variable with the largest error is selected. Here, for example, it is assumed that the error of the nitrate nitrogen concentration is the largest.

  Next, the process state transition diagram display means 11 searches for the state variable with the largest error, in this case, the parameter that affects the nitrate nitrogen concentration. For example, it can be seen that the stoichiometric parameters Yaut and Yh affect the nitrate nitrogen concentration. Although not shown in FIG. 5, in ASM2, the reaction rate of the nitrification part (A part) and the denitrification part (B1 and B2 part) in FIG. 5 is expressed by the following equation. .

ここで、動力学パラメータは、μaut，μh，ηno3，Ｋo2，Ｋnh4，Ｋno3，Ｋpo4，Ｋalk，Ｋf，Ｋaであるため、これらのパラメータは硝酸態窒素濃度に影響を与えることが分かる。

Here, since the kinetic parameters are μaut, μh, ηno3, Ko2, Knh4, Kno3, Kpo4, Kalk, Kf, Ka, it can be seen that these parameters affect the nitrate nitrogen concentration.

  Therefore, using the 11 process state transition diagram display means, when adjusting the value of nitrate nitrogen concentration, the values of Yaut, Yh, μaut, μh, ηno3, Ko2, Knh4, Kno3, Kpo4, Kalk, Kf, Ka You can adjust some of them. If you are a parameter adjuster who understands the physicochemical and mathematical meanings of chemical / biological reaction models such as ASM2, adjust them by trial and error in order from the most likely to affect them. Good.

For parameter adjusters who do not have physicochemical knowledge of the process model, a help function for displaying the meaning of the parameters is prepared according to the invention described in the present application. Below are some examples of help information display for the help function. For Yaut, “a stoichiometric parameter representing the yield for nitrifying bacteria. Increasing this value increases the nitrate nitrogen concentration and decreases the ammonia nitrogen concentration and oxygen concentration. It can be adjusted by adjusting the kinetic parameters. It is desirable to change this parameter value when it is difficult to set.The parameter value should be set in the range of 0 to 1. " With respect to Ko2, “a kinetic parameter representing a half-saturation constant relating to the dissolved oxygen concentration. If this value is reduced, even if the dissolved oxygen concentration is relatively low, the aerobic condition is likely to occur and the reaction under the aerobic condition is likely to proceed. Conversely, reactions under anaerobic conditions are less likely to proceed, and all state variables related to aerobic / anaerobic conditions are affected, so adjustment is necessary. ” As for η _no3 , “the kinetic parameter for correcting the maximum rate of denitrification under anoxic conditions, the concentration of nitrate nitrogen related to denitrification, etc. because it is related only to the denitrification under anoxic conditions, etc. It is good to use it for adjustment. " If there is such a help message, for example, the process output simulation value and the process output value of the nitrate nitrogen concentration NO ₃ -N and SF concentration and the SA concentration related to the denitrification reaction greatly deviate from each other. It can be seen that the value of the variable may be adjusted using η _no3 when the process output simulation value and the process output value are substantially the same.

  Such a help function is effective for selecting the most suitable parameter from the parameter set to be adjusted.

  Further, for parameter adjusters who have some mathematical knowledge and physicochemical knowledge of models, the process state transition diagrams of FIGS. 5 to 7 and FIG. It is useful to associate mathematical expressions that represent process models. For example, as shown in FIG. 5, when “model display” is clicked, a mathematical expression representing a process model as shown in FIG. 8 may be displayed.

  By having a function for associating process state transition diagrams with mathematical expressions in this way, parameter adjusters who have physicochemical and mathematical knowledge about chemical reactions and biological reactions make adjustments. With reference to both, it is possible to select parameters for adjustment.

  In addition, if you want to display the correspondence between the process reaction mechanism expressed in the process model and the actual process more visually, you can use the invention described in this application to set the parameter values and process state transition diagram. It is good to keep them linked. For example, with respect to the stoichiometric parameter, as shown in FIG. 9, the display size of the state variable and the value of the stoichiometric parameter may be linked in the process state transition diagram. Although FIG. 9 shows the heterotrophic microorganism that is a state variable, when the yield Yh of the heterotrophic microorganism is increased, the size of the state variable representing the heterotrophic microorganism is increased accordingly. That's fine. Specifically, a proportional relationship may be established between the magnitude of the yield Yh and the radius of the circle representing the heterotrophic microorganism. As for kinetic parameters, the reaction rate usually changes depending on the values of multiple state variables.For example, prepare multiple arrows with colors for each state variable. The size of the part may be made to correspond to the dynamic parameter. For example, as shown in FIG. 10, colors such as “black for oxygen concentration, red for nitrate nitrogen concentration,...” Are determined, and state transitions related to these concentrations are shown in FIGS. A plurality of arrows in the process state transition diagram shown in FIG. 7 are displayed in parallel for one reaction. For example, the half-saturation constant, which is one of the kinetic parameters, is a parameter indicating how quickly the reaction proceeds. When the half-saturation constant is large, as shown in FIG. If the “arrow” part is enlarged, and the half-saturation constant is small, the “arrow” part may be reduced. For example, the value of the half-saturation constant Ko2 related to the oxygen concentration is associated with the “arrow” part of the black arrow. Moreover, since the maximum specific growth rate, which is one of the kinetic parameters, is a parameter that gives the maximum value of the reaction rate, when the maximum specific growth rate is large, an arrow is displayed as shown in FIG. On the contrary, when the maximum specific growth rate is small, it is only necessary to make correspondence such as displaying an arrow thinly.

  Thus, by associating the parameter values with the visual display of the process state transition diagram, the meaning of each parameter can be easily understood visually. Furthermore, if the process state transition diagram when the default value of the parameter is given and the process state transition diagram after adjustment are displayed at the same time, the parameter adjuster, especially in the case of a process model with a large number of parameters You can easily understand visually which of the many parameters you have already changed from the default value.

  Based on the parameter adjustment support information as described above, the parameter to be adjusted is selected, the parameter value is actually determined, and the parameter value setting means is attached to the 11 process state transition diagram display means. This value is reflected in the parameters of the process model of the three process simulation means. Then, the process output simulation value updated by the three process simulation means is calculated using the newly set parameter value.

  By repeating the above operations and adjusting the parameters for the state variable with the largest error, the same operations are repeated in order of increasing error between the process output value and the process output simulation value (mainly chemical and biological reactions). Process parameter adjustment (support).

  In addition, when the purpose of process simulation is to calculate the amount of operation that ultimately controls the process appropriately, such as Internal Model Control, it is possible to have three process simulation means inside the controller. preferable. In such a case, according to the invention described in the present application, the process 5 state transition diagram display means as shown in FIGS. 5 to 7 is installed, for example, in a monitoring room for monitoring the target process or a place far away from the target process. It is built on a WWW browser on a personal computer or workstation in an office, etc., and the value can be reflected in the parameter value of the process simulator inside the controller where the target process exists via the Internet or a dedicated line. Good.

  Thereby, the parameters of the process simulator can be adjusted by remote operation. Since the process simulator is incorporated in the controller, the parameters of the controller are also adjusted by remote operation.

  The above embodiments can be applied to many chemical processes and biological processes. This is because many chemical processes and biological processes are described in a model called a “Compartmental model” that represents the interaction between multiple state variables corresponding to multiple substances (amounts and concentrations) as shown in FIG. The parameters of these models are basically stoichiometric parameters that represent the quantitative relationship between substances, and kinetic parameters related to the reaction rate of the interaction between the state variables (chemical and biological reactions). It is because it consists of. Therefore, the method described above can be applied to a fairly general purpose in chemical processes and biological processes.

The figure which wrote together the block diagram in the case of implementing a parameter adjustment apparatus, and the sewage treatment process which is an example of an object process. The figure which shows the example of the time change of a parameter sensitivity function. The conceptual diagram which shows the method of comprising a parameter sensitivity function vector. The figure which wrote together the block diagram in the case of implementing a parameter adjustment assistance apparatus, and the sewage treatment process which is an example of object process. The figure which shows the example of a process state transition diagram display means and a parameter value setting means. The figure which shows the example 2 of a process state transition diagram display means. The figure which shows Example 3 of a process state transition diagram display means. The figure which shows the example of the process model (mathematical model) linked | related with the process state transition diagram display. The figure which shows the example of matching of the stoichiometric parameter and the display method of a process state transition diagram display means. The figure which shows the example of matching of the display method of the state variable relevant to reaction rate, and a process state transition diagram display means. The figure which shows the example 1 (in the case of a half-saturation constant) of matching of a dynamics parameter and the display method of a process state transition diagram display means. The figure which shows the example 2 (in the case of the maximum specific growth rate) of matching of a dynamic parameter and the display method of a process state transition diagram display means.

Explanation of symbols

1 Sewerage process 11 First sedimentation tank 12 Biological reaction tank 13 Final sedimentation tank 141 Blower (actuator)
142 Return pump (actuator)
143 Excess sludge extraction pump (actuator)
151 Process disturbance sensor (Process input sensor)
152 Aeration air volume sensor (process input sensor)
153 Return sludge sensor (process input sensor)
154 Excess sludge extraction sensor (process input sensor)
161 First sedimentation tank state sensor (process output sensor)
162 Bioreactor state sensor (process output sensor)
163 Final sedimentation tank state sensor (process output sensor)
2 Process data collection means 21 Process external input data collection means 22 Process output data collection means 3 Process simulation means 4 Parameter initial value setting means 5 Parameter sensitivity analysis means 6 Parameter sensitivity display means 7 High sensitivity parameter set extraction means 8 Parameter identification means 9 Error calculation and storage means 10 Process value / process simulation value display means 11 Process state transition diagram display means 12 Parameter value setting means

Claims (6)

Process external input including disturbance and manipulated variable, process state quantity, process observation output that is a function of the process external input and the process state quantity, an external input sensor that measures the process external input, and the process observation In any process having a process output sensor that measures output,
The process external input and the process state in the process are composed of a plurality of substances, and the process is a limited biological / chemical process in which dynamics are expressed by mutual changes between the plurality of substances. Yes,
A process external input collecting means for collecting measurement data of the external input sensor and holding the data for a predetermined period, and a process for collecting measurement data of the process output sensor and holding the data for a predetermined period Process data collection means comprising output collection means;
A process external input supplied through the process data external input collection means to a process model having a parameter set (parameter vector) θεRp (p: number of parameters) composed of a plurality of parameters and configured by a material balance and reaction rate A process simulation means for simulating the temporal and spatial behavior of the process by inputting data;
Process output display means for displaying a process output simulation value by the process simulation means and the process output value;
A process state transition diagram display means for representing a state interaction and a state change of the process model;
Parameter value input means for inputting a parameter value of the parameter set attached to the process state transition diagram;
A process model parameter adjustment support device characterized by comprising: 2. The process model parameter adjustment support apparatus according to claim 1, wherein the parameter set input unit has a help function for explaining a physicochemical meaning of each parameter which is an element of the parameter set. 2. The process model parameter adjustment support according to claim 1, further comprising a function of establishing a link from the process state transition display means to a mathematical expression and associating the process state transition display means with a mathematical expression representing the process model. apparatus. The process state transition diagram display means includes a diagram showing a process state variable to be displayed and the size and color of a line connecting the process states depending on the size of the parameter value input by the parameter set input means. 2. The process model parameter adjustment support apparatus according to claim 1, wherein the process model parameter adjustment support apparatus is modified so as to be variable. The process state transition display means and the parameter set input means are mounted on a device different from the process simulation means, and are modified so that a parameter set can be input remotely from a WWW (World Wide Web) browser. The process model parameter adjustment support apparatus according to claim 1, wherein: Process external input including disturbance and manipulated variable, process state quantity, process observation output that is a function of the process external input and the process state quantity, an external input sensor that measures the process external input, and the process observation In any process having a process output sensor that measures output,
The process external input and the process state in the process are composed of a plurality of substances, and the process is a limited biological and chemical process in which dynamics are expressed by mutual changes between the plurality of substances. Yes,
Process for collecting measurement data of the external input sensor and holding the data for a predetermined period, and process for collecting the measurement data of the process output sensor and holding the data for a predetermined period A process data collection process comprising an output collection process;
Process external input supplied through the process data external input collection step for a process model having a parameter set (parameter vector) θεRp (p: number of parameters) composed of a plurality of parameters and configured by a material balance and reaction rate A process simulation step that simulates the temporal and spatial behavior of the process by entering data; and
A process output display step for displaying a process output simulation value by the process simulation means and the process output value;
A process state transition diagram display step representing a state interaction and a state change of the process model;
A parameter value input step for inputting a parameter value of the parameter set attached to the process state transition diagram;
A process model parameter adjustment support method characterized by comprising:

Images