JP2005316738A - Plant-wide optimum process controller - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an optimum process controller easily implementing an optimum target value and an optimum operation quantity according to calculation of a practical approximate optimum solution. <P>SOLUTION: A process optimization part calculating an actuator optimum target value or optimum operation quantity includes a process simulation part for a process dynamic model and an optimum computing part performing optimization calculation. In the process simulation part, one or more operation quantity values for the actuator are generated within an operation range of the operation quantity to be inputted to the dynamic model for generating a convergent value of the dynamic model. In this way, an operable initial solution candidate satisfying a physical quantity conservation restriction of a restriction condition setting part is found, and from the candidates, an operable initial solution candidate satisfying all the restriction conditions set by the restriction condition setting part is used as an operable initial solution. In the optimum computing part, one or more local optimum solutions are generated by a mathematical programming by using the operable initial solution as an initial value, and from the local optimum solutions, a solution optimizing an evaluation function is used as an approximation optimum solution. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a process controller for a material or energy flow, for example, a process system controller such as a sewage treatment process, a wastewater treatment process, a water purification process, a chemical process, etc., and particularly a plant-wide optimum process of a highly nonlinear process The present invention relates to a control device.

  PID control and sequence control, which are a kind of classical control, have been widely used as control methods for industrial processes such as water and sewage water quantity processes and water quality processes. These PID control and sequence control are balanced control methods from the viewpoints of reliability, control performance, and ease of design and tuning. As the most standard control methods in the industry, even today, It is used to control over 80% of industrial processes.

  However, in sewage treatment, for example, in recent years, regulations for the release of nitrogen and phosphorus into rivers have become mandatory due to the conventional requirement for removing only organic substances, and further, economic costs such as operating costs have been reduced. There is also a demand for energy saving and energy saving, and the purpose of control is diversified. In order to meet such demands, various advanced controls have recently been used in addition to conventional PID control and sequence control.

  Advanced control is roughly classified into two approaches, rule-based control and model-based control, and these are used properly according to the purpose. If there is a model that can simulate the behavior of the target process, adopting model-based control has the potential to perform better control. In particular, in the so-called process control field such as chemical process control and water and sewage process control, there is a strong demand for reducing the operating costs of chemicals and electric power, etc., as the amount of operation, as well as demands for control performance. Model predictive control, which is one of model-based controls that have been taken into consideration, has been developed.

  Model predictive control is a control method that has been put to practical use in the industry by DMC (Dynamic Matrix Control), independent of Shell Oi1, from the late 1970s to the early 1980s. Widely used (see Non-Patent Document 1 below). While the effectiveness of model predictive control is well verified in actual processes, on the other hand, when more complex processes involving metabolic reactions of organisms such as sewage treatment processes are targeted, higher dimensions are required. However, a nonlinear model predictive control system must be constructed, and although it is a very promising technology in the future, it is not always easy to realize it at the current technical level.

  In order to deal with such a problem, the inventors have proposed a “process simulator application control apparatus and method” (see the following Patent Document 1), and an unpublished “plant-wide optimal process control apparatus” ( Japanese Patent Application No. 2000-354865) has been proposed. In particular, the “plant-wide optimal process control device” is directed to optimization of the entire process in consideration of the ease of realization in an actual process control system. Here, the process control system is divided into a hierarchical structure of a superordinate system (supervisory control system: hereinafter abbreviated as SVC) and a subordinate system (local control part: abbreviated as Local Control Loops, LC), The upper system calculates the optimal target value (or optimal operation amount) that optimizes the entire process, and the lower system uses a one-loop controller consisting mainly of a PI controller so that the process follows the optimal target value. It was something to control.

The “Plant Wide Optimal Process Control Device” shows a detailed embodiment for the sewage treatment process, and the control system in this case is an optimal system considering economic cost and environmental load. The following constraint conditions are incorporated using a non-linear evaluation function for optimization variables as an optimization index.
a. Nonlinear algebraic equations representing mass balance (such as biochemical reactions and hydraulic flows).
b. An inequality representing the upper and lower limits of the manipulated variable.
c. An inequality representing output constraints.
d. Non-linear inequality that restricts required sludge residence time, residence time, water area load, etc.

Therefore, this SVC can be formulated as a “nonlinear optimization problem” with nonlinear equations and inequality constraints and a nonlinear evaluation function. This formulation is intended for a sewage treatment process. Of the above constraints, the three constraints a, b, and c are arbitrary physical processes such as petrochemical processes, water purification processes, and power processes. Is a constraint that usually appears in For a, the energy balance may be considered instead of the material balance. Another constraint condition d usually appears in some form in other processes as a condition to be satisfied between a physical (civil engineering) structure and a substance to be produced, consumed, processed, or converted in the structure. It is also common practice to choose an optimization index as an economic index. Therefore, this form of SVC formulation is fairly universal that can be applied to many process systems.
J. Richalet, “Industrial Implementation of Predictive Control”, Preprint of Japan Batch Forum (2002) JP 2002-373002 A

  As described above, after formulating, it is necessary to actually solve the nonlinear optimization problem, but in general, solving the nonlinear optimization problem is considerably more than solving the linear optimization problem. Further, the nonlinear equality constraints corresponding to the mass balance contained in this SVC make it more difficult to solve this nonlinear optimization problem. This is because when a combination of independent variables to be optimized satisfies the equation constraint, a combination that slightly changes the value of one independent variable no longer satisfies the equation condition. . Moreover, since this material balance is a fundamental and essential requirement, it cannot be ignored or relaxed, and nonlinear equation constraints must be taken into account. Therefore, in order to actually implement such an SVC with a plant-wide control device, a method for obtaining a practical solution of this optimization problem (an approximate optimal solution that can withstand practical use even if not optimal) is required. Become.

  Furthermore, when this SVC is to be executed online, there is a possibility that it will not be used as a result unless a practical optimal solution can be executed in a time shorter than the online control period. For that purpose, it is also required to solve a practical solution at high speed.

  The present invention has been made to solve such problems, and a first object is to provide practical approximate optimum solutions of optimum target values and optimum manipulated variables when performing plant-wide optimum control. It is an object of the present invention to provide a plant-wide optimum process control device that can be calculated and can be easily realized.

  A second object of the present invention is to provide a plant-wide optimum process control apparatus that can calculate a practical approximate optimum solution of an optimum target value and an optimum manipulated variable at high speed and can be used online. is there.

The invention according to claim 1
At least one process sensor capable of measuring several states of the target process, with the inflowing substance or energy as an external input and the processing step of the substance or energy flow as the target process;
At least one actuator capable of changing the state of the subject process;
A constraint condition setting unit including a storage amount constraint setting unit capable of setting at least a static constraint condition represented by an algebraic equation relating to a conservation rule of a storage amount such as an externally input substance amount and energy amount;
An optimal evaluation function setting unit that can set an evaluation function that can evaluate the operating cost and multiple control performances;
A local controller that operates the actuator based on the process measurements obtained by the process sensor to compensate for dynamic variations in the process to have the desired characteristics;
Enter the constraint condition set by the constraint condition setting unit and supply the optimal target value of the local control unit based on the evaluation function set by the optimal evaluation function setting unit, or directly calculate the optimal operation amount of the actuator A process optimization unit to
In a plant-wide optimum process control device equipped with
The process optimization unit performs optimization calculation with a process simulation unit represented by a differential equation that is a dynamic model of a process corresponding to an algebraic equation that is a steady state model of a process set by a storage amount constraint setting unit Including an optimal calculation unit,
The process simulation unit generates one or more actuator operation amount values within the operation range of the operation amount, inputs them to the dynamic model, and generates a convergence value (steady value) of the dynamic model. One or more feasible initial solution candidates of the optimum computing unit satisfying the substance amount storage constraint of the setting unit are obtained, and one of the feasible initial solution candidates satisfying all the constraint conditions set in the constraint condition setting unit The above feasible initial solution candidates are adopted as feasible initial solutions, and the optimal computing unit generates one or more local optimal solutions by mathematical programming with one or more feasible initial solutions as initial values, An algorithm that incorporates the one that optimizes the evaluation function as an approximate optimal solution is incorporated.
It is characterized by this.

  The invention according to claim 2 is the plant-wide optimum process control device according to claim 1, wherein the process optimization unit has substantially the same range that can be taken by the independent variables to be optimized such as the manipulated variables and state variables that appear. It normalizes so that it becomes.

  The invention according to claim 3 is the plant-wide optimum process control device according to claim 1 or 2, wherein the combination of manipulated variable values for generating executable initial solution candidates in the process simulation unit of the process optimization unit is a gene. And an evaluation function value calculated from the gene through an optimal calculation unit or some index converted from the evaluation function value is regarded as a goodness of fit, and a genetic algorithm is used in combination.

  The invention according to claim 4 is characterized in that, in the plant-wide optimum process control apparatus according to claim 3, an immune algorithm or metaheuristics is applied instead of the genetic algorithm of the process optimization unit.

  The invention according to claim 5 is the plant-wide optimum process control device according to claim 1 or 2, as a combination of manipulated variable values for generating executable initial solution candidates in the process simulation unit of the process optimization unit, The actual operation amount at the time of calculation of the target process is adopted.

  The invention according to claim 6 is the plant-wide optimum process control apparatus according to claim 1 or 2, wherein the optimization is performed by regarding some of the independent variables to be optimized in the process optimization unit as fixed values. It is characterized in that optimization is performed only for the remaining independent variables.

  The invention according to claim 7 is the plant-wide optimum process control device according to claim 1 or 2, wherein the process optimization unit simultaneously solves the relaxation problem of the formulated optimization problem, thereby reducing the lower bound of the optimum value. It is characterized by estimating.

The invention according to claim 8 is the plant-wide optimum process control device according to claim 1 or 2, wherein the combination of manipulated variable values for generating executable initial solution candidates in the process simulation unit of the process optimization unit is a gene. The evaluation function value calculated from the gene through the optimal computation unit or any index converted from the evaluation function value is regarded as the fitness, and the genetic algorithm is used together.
In the process optimization section, the lower bound of the optimum value is estimated by simultaneously solving the relaxation problem of the formulated optimization problem,
The termination condition of the genetic algorithm is determined based on a threshold level provided for the difference between the approximate optimum solution and the lower bound of the optimum solution.

  The invention according to claim 9 solves this optimization problem when no feasible solution is found among the feasible initial solution candidates in the plant-wide optimum process control device according to claim 1 or 2. It is characterized by having a function of presenting that there is a high possibility that there is no.

  The invention according to claim 10 is the plant-wide optimum process control device according to claim 1 or 2, further comprising a local control unit corresponding to each of a plurality of target processes, a constraint condition setting unit, an optimum evaluation function setting unit, and Each function of the process optimization unit is provided in a single higher-level control device, and elements corresponding to the higher-level control device are connected via a data transmission path.

  According to an eleventh aspect of the present invention, in the plant-wide optimum control apparatus according to the tenth aspect, the functions of the constraint condition setting, the optimum evaluation function setting unit, and the process optimization unit are divided into a plurality of control devices. Are connected by a data transmission line.

  The present invention configured as described above makes it possible to concretely calculate a practical approximate optimum solution of the optimum target value and the optimum manipulated variable when performing plant-wide optimum control. A plant-wide optimum process control device that can be easily realized is provided.

  In addition, a plant-wide optimum process control apparatus that can calculate a practical approximate optimum solution of an optimum target value and an optimum manipulated variable at high speed and can be used online is provided.

  Hereinafter, the present invention will be described in detail based on preferred embodiments shown in the drawings.

First embodiment

<Configuration of First Embodiment>
FIG. 1 is a block diagram showing the configuration of a control system for an advanced sewage treatment process aimed at removing nitrogen and phosphorus as a first embodiment of a plant-wide optimum process control apparatus according to the present invention. . However, the present invention is not limited to such a sewage treatment control system, and can be applied to any process such as a chemical process.

  The sewage treatment process shown in FIG. 1 is a sewage advanced treatment process 1 configured as a two-stage circulation nitrification denitrification process. The treated water is first introduced into the sedimentation basin 101, and from here on, an anaerobic tank 102, Water is discharged through the first aerobic tank 103, the anaerobic tank 104, the second aerobic tank 105, and the final sedimentation basin 106.

  As an actuator capable of changing the state of this advanced sewage treatment process 1, the first sedimentation basin 101 is equipped with an initial sedimentation basin excess sludge extraction pump 111, and similarly, between the outlet of the first sedimentation basin 101 and the anoxic tank 104. A step inflow pump 112, a blower 113 for supplying oxygen to the first aerobic tank 103, a blower 114 for supplying oxygen to the second aerobic tank 105, a carbon source input pump 115 for supplying a carbon source to the anaerobic tank 104, A circulation pump 116 that recirculates water to be treated from the second aerobic tank 105 to the anaerobic tank 104, a return sludge pump 117 that returns sludge from the final sedimentation tank 106 to the first sedimentation tank 101, and a flocculant in the second aerobic tank 105 And a final sedimentation basin excess sludge extraction pump 119 for extracting sludge from the final sedimentation basin 106. These actuators 111 to 119 operate at a predetermined cycle.

  Furthermore, in the advanced sewage treatment process 1, as a process sensor for measuring the state, an ammonia sensor 121 for measuring the ammonia concentration in the first aerobic tank 103, a nitric acid sensor 122 for measuring the nitric acid concentration in the anaerobic tank 104, a tank MLSS sensor 123 that measures the amount of activated sludge (hereinafter abbreviated as MLSS) in at least one tank 102 to 105 (the oxygen-free tank 104 in the illustrated example), and ammonia that measures the ammonia concentration in the second aerobic tank 105. Sensor 124, phosphoric acid sensor 125 for measuring the phosphoric acid concentration in the second aerobic tank 105, SS sensor 126 for measuring the SS concentration of the amount of sludge withdrawn from the final sedimentation basin 106, excess sludge amount withdrawn from the final sedimentation basin 106 A flow sensor 127 for measuring the DO, a DO sensor 128 for measuring the dissolved oxygen concentration in the second aerobic tank 105, ORP sensor 129 that measures the oxidation-reduction potential difference of tank 105, discharge sewage sensor 1210 that can measure the amount of discharged sewage and a plurality of water quality elements such as COD, TN, TP, and the amount of inflow sewage and COD, TN, A sewage inflow sensor 1211 capable of measuring a plurality of water quality elements such as TP is provided. These process sensors 121 to 129, 1210, and 1211 measure at a predetermined cycle.

  The plant-wide optimum process control apparatus includes a process measurement data collection unit 2 that collects and holds process data obtained from the process sensors 121 to 129 and 1210 in a predetermined cycle, and an inflow as an external input measured by a sewage inflow sensor 1211. Based on the process measurement data supplied at a predetermined cycle from the external measurement data collection unit 3 that collects and holds the time-series data of the amount of water and the quality of the inflow sewage water, the process sensors 121 to 129, Based on the process sensor abnormality diagnosis unit 4 that diagnoses the abnormality of 1210 and switches the contact of the switch unit SW2 and the external input measurement data supplied from the external input measurement data collection unit 3 at a predetermined cycle, An external input sensor that diagnoses the abnormality of the sensor 1211 and performs the switching operation of the switch unit SW1 A diagnosis unit 5, and a manual analysis data storage unit 6 to be mainly save manual analysis data of the sewage inflow water.

  In addition, the plant-wide optimum process control device includes a storage amount constraint setting unit 71 for setting a constraint condition related to a storage amount such as process mass balance or energy balance, and a constraint condition such as a water quality or a control value of the water amount that is the output of the process. A constraint condition setting unit 7 including an output constraint setting unit 72 for setting, and an operation amount constraint setting unit 73 for setting a constraint condition such as a limit of an operation amount that is a process input and a limit (Δ limit) of an operation amount change rate. And an optimum evaluation function setting unit 8 that quantitatively defines the optimality of the process, various constraint values set by the constraint condition setting unit 7 and evaluation functions set by the optimum evaluation function setting unit 8 The data of the inflow sewage amount of the sewage inflow sensor 1211 obtained through the external input measurement data collection unit 3 and the data of various inflow sewage qualities or the manual analysis data storage unit The process optimization unit 9 for calculating the optimum target values of the optimum operation amounts of the actuators 111 to 119 and the process controlled amounts measured by the process sensors 121 to 129 and 1210, and the process measurement A local control unit 10 that receives the process data collected in the data collection unit 2 and the optimum target value supplied from the process optimization unit 9 as input and calculates the operation amounts of the various actuators 111 to 119 is provided.

  The process optimization unit 9 includes a process simulation unit 91 for generating an executable solution for the optimization operation and an optimization operation unit 92 for performing the optimization operation using the executable solution generated thereby as an initial value. It is configured.

<Operation of First Embodiment>
The operation of the first embodiment will be described below. First, the nitric acid concentration data of the anoxic tank 104 measured by the nitric acid sensor 122 is stored in the process measurement data collection unit 2 at a predetermined cycle. Here, the reason why the output signal line of the nitric acid sensor 122 is branched into three and input to the process measurement data collecting unit 2 is that the process measurement data collecting unit 2 selectively collects a plurality of data. This is because a data collection unit (not shown) is provided. Similarly, the ammonia concentration data of the first aerobic tank 103 and the second aerobic tank 105 respectively measured by the ammonia sensor 121 and the ammonia sensor 124 are stored in the process measurement data collection unit 2 at a predetermined cycle. Exactly the same, MLSS concentration data measured by the MLSS sensor 123, phosphoric acid concentration data measured by the phosphoric acid sensor 125, surplus sludge SS concentration data measured by the SS sensor 126, surplus measured by the flow sensor 127 Sludge extraction flow rate data is stored in the process measurement data collection unit 2 at a predetermined cycle.

The optimum evaluation function setting unit 8 sets an optimum evaluation function for evaluating the optimality of the sewage advanced treatment process 1. In the advanced sewage treatment process 1, it is required to reduce the operating cost and improve the removal efficiency of nitrogen and phosphorus. However, since these are in a trade-off relationship, an optimal evaluation function is set in consideration of these trade-offs. It is preferable. For example, according to the idea of imposing a levy according to the load of nitrogen or phosphorus of the effluent water, the improvement of the effluent water is converted into a monetary amount as the effluent water cost, and this effluent water quality cost and the operating cost are evaluated simultaneously, The following evaluation function J is set.
J = EC + OC (1)

Here, EC is the discharged water quality cost, OC is the operating cost, and is defined by the following equation, for example.

Here, COD, BOD, SS, TN, and TP represent the chemical oxygen demand, the biochemical oxygen demand, the amount of suspended solids, the total nitrogen concentration, and the total phosphorus concentration, respectively, of the discharged water. All are [kg / m ³ ]. Q _ef [m ³ / d] represents the amount of discharged water. Q _b , Q _ret , Q _circ , Q _ex , Q _pac , and Q _cb are aeration air amount, return amount, circulation amount, surplus sludge extraction amount, flocculant input amount, carbon source input amount, respectively. [m ³ / d]. Q _sludge [kg / d] is the amount of generated sludge. T0 and T represent a cost evaluation start time and a cost evaluation period, respectively. w _i is a weight corresponding to the cost conversion coefficient.

  Of course, the optimum evaluation function setting unit 8 is not limited to such an evaluation function, and an evaluation function based on various criteria such as user needs or administrative judgment can be used. However, the optimum evaluation function setting unit 8 needs to be able to quantitatively evaluate the criterion of optimality. This concept of the optimum evaluation function is effective as an evaluation function considering such a trade-off since the trade-off of the advanced sewage treatment process 1 can be quantitatively taken into consideration.

  Next, as another setting condition, in the constraint condition setting unit 7, each constraint condition is set by the storage amount constraint setting unit 71, the output constraint setting unit 72, and the operation amount constraint setting unit 73.

  The storage amount constraint setting unit 71 is described by an algebraic equation derived from a differential equation considering mass balance in the processing of the sewage altitude processing process 1.

  First, for example, the process model (process simulator) of advanced sewage treatment process 1 is specifically described in the document “IAWQ Task Group on Mathematical Modeling for Design and Operation of Biological Wastewater Treatment Processes,“ Activated Sludge Models ASM1, ASM2, ASM2d. and ASM3 ”, IAWQ Scientific Technical”, as shown in the following equation, biological reaction tank model using activated sludge models ASM1 to ASM3, and sedimentation basin using sedimentation models such as layer model and Takacs model It is expressed as a connected model.

<Biological reaction tank model>

Here, z _* (t) [g / m ³ ] ∈ ⁿ ₊ is a vector representing the water quality element of the reaction tank, n is the number of state variables, z _{* in} (t) [g / m ³ ] ∈R ⁿ ₊ Is a vector representing the quality element of the sewage flowing into the target reaction tank, V _* [m ³ ] ∈R ₊ is the volume of the target reaction tank, Q _* (t) [m ³ / day] ∈R ₊ Represents the inflow. SεR ^{p × n} is a matrix representing stoichiometry, r (·) εR ^p ₊ is a vector representing the reaction rate, and p is the number of element reaction processes. S _{o2 *} (t) [g / m ³ ] ∈R ₊ is one element of z _* (t) in the dissolved oxygen concentration (assigned as the second element in ASM1 to ASM3). S _o2 [g / m ³ ] ∈R ₊ is the amount of saturated dissolved oxygen, KLa [1 / day] ∈R ₊ is the overall oxygen transfer capacity coefficient, QB (t) [m ³ / day] ∈R ₊ is the aeration amount, K [1 / m ³ ] ∈R ₊ represents a constant. * Represents a character string of each reaction tank, for example, *: = aerobic aerobic tank. R ⁿ represents a set of n-th order real numbers, and R ⁿ ₊ represents a set of n-th order positive real numbers.

<Settling pond model>

Here, s _sd (t) [g / m ³ ] ∈R ^s ₊ is a vector representing the soluble water quality element of the settling basin, and s is the number of soluble water quality elements. _{^{x b sd (t) [g}} / m 3] ∈R n-3 + is a vector representing the buoyant water quality component of the precipitation portion of the sedimentation _{^{tank, x u sd (t) [}} g / m 3] ∈R ^n-s ₊ is a vector representing the buoyant water quality component of the precipitation pond supernatant portion. α is a constant smaller than 1. Q ^w _sdout (t), Q ^r _sdout (t), and Q ^c _sdout (t) represent the excess sludge extraction flow rate, the return flow rate, and the discharge flow rate, respectively.

By appropriately connecting these biological reaction tank model and sedimentation pond model according to the process configuration of the advanced sewage treatment process 1, a sewage advanced treatment process simulator is constructed. This is a process simulator used in the process simulation unit 91 of FIG. In order to simplify the following explanation, this process simulator is described formally as follows.

  Here, x is a vector representing water quality elements such as ammonia, nitric acid, phosphoric acid, COD elements in the reaction tank, SS, MLSS, etc. u is an operation such as aeration amount, circulation amount, return amount, carbon source input amount, etc. D is a vector representing disturbances such as inflow sewage water quality factor and inflow sewage amount, and θ is a vector representing various parameters such as maximum specific growth rate, half-saturation constant, hydrolysis constant, death constant, etc. is there.

Now, if only the steady state is considered without considering the transient state of the process behavior, the differential value on the left side of equation (13) describing this process model (simulator) can be set to zero. The mass balance equation in the state is obtained as follows.
f (x, u, θ, d) = 0 (14)

  This equation (14) is an example of the constraint condition expression of the storage amount constraint setting unit 71. In the present invention, in order to satisfy the expression (14) as the constraint condition, an algorithm using the expression (13) is developed. ing.

Next, in the output restriction setting unit 72, for example, a restriction value of water quality that is an output can be set, or a restriction condition can be set such that the water quality concentration is never negative. For example, in addition to the equation (14), the following constraint condition is set.
0 ≦ x ≦ x _max (15)

Here, x _max is an upper limit value of water quality corresponding to a water quality regulation value or the like. This expression (15) is a constraint condition expression of the output constraint setting unit 72.

Next, in the operation amount constraint setting unit 73, a limit condition within the range that can be taken by the actuator of the process can be set, for example, a constraint condition such as the following equation is set.
u _min ≤ u ≤ u _max (16)

Here, u _min and u _max correspond to the upper limit value and the lower limit value of the manipulated variable. This equation (16) is a constraint condition equation of the operation amount constraint setting unit 73.

  Accordingly, the constraint condition formulas (14), (15), and (16) are set in the constraint condition setting unit 7.

  In the present embodiment, after setting by the above-described optimal evaluation function setting unit 8 and constraint condition setting unit 7, process information is collected and controlled. The measured process data and external input data are stored in either the process measurement data collection unit 2 or the external input measurement data collection unit 3.

  Next, since the process measurement data collection unit 2 is not a part directly related to the present invention, the detailed configuration of the process measurement data collection unit 2 is omitted. The process measurement data collection unit 2 includes first to seventh data collection units. The first data collection unit collects nitric acid concentration data of the anoxic tank 104 measured by the nitric acid sensor 122. The second data collection unit collects ammonia concentration data of the first aerobic tank 103 and the second aerobic tank 105 measured by the ammonia sensors 121 and 124. The third data collection unit collects nitric acid concentration data of the anoxic tank 104 measured by the nitric acid sensor 122. The fourth data collection unit collects nitric acid concentration data of the anoxic tank 104 measured by the nitric acid sensor 122. The fifth data collection unit collects MLSS concentration data of the anoxic tank 104 measured by the MLSS sensor 123. The sixth data collection unit collects the phosphate concentration data of the second aerobic tank 105 measured by the phosphate sensor 125. The seventh data collection unit uses the MLSS concentration data of the anoxic tank 104 measured by the MLSS sensor 123, the SS concentration data of sludge extracted from the final sedimentation basin 106 measured by the SS sensor 126, and the flow rate sensor 127. The surplus sludge amount data extracted from the final sedimentation basin 106 is collected. The measurement data collected by the process measurement data collection unit 2 is sent to the process sensor abnormality diagnosis unit 4 and the local control unit 10.

  Next, the process sensor abnormality diagnosis unit 4 performs abnormality diagnosis of measurement values by the process sensors 121 to 129 and 1210 based on the time series data of the process values collected by the process measurement data collection unit 2. This abnormality diagnosis can be performed by a method using a statistical method such as principal component analysis based on correlation by a plurality of sensors, for example.

  That is, some of the process sensors often have a very strong correlation. For example, the ammonia sensor 124, the DO sensor 128, and the ORP sensor 129 are often correlated with each other, and the correlation is particularly strong under aerobic conditions. Therefore, principal component analysis is performed on these three time series data or processed time series data. Then, when the sensor is normal, the value of the first principal component is, for example, about 0.9 (the sum of the first to third principal components is about 1). Therefore, for example, 0.75 is set as the threshold level, and if the value of the first principal component is less than 0.75, it is determined that one of the sensors is abnormal. For example, if the sensor is normal, an FB (feedback) flag is set, and if the sensor is abnormal, an FF (feed forward) flag is set to store the sensor state.

  In addition, when the process sensor abnormality diagnosis unit 4 determines that the process sensor state is FB and is normal, the process sensor abnormality diagnosis unit 4 connects the contact of the switch unit SW2 to the d side. On the other hand, when the process sensor state is FF and it is determined to be abnormal, the contact of the switch unit SW2 is switched and connected to the c side.

  Next, the external input sensor abnormality diagnosis unit 5 performs sensor value abnormality diagnosis using time-series data of the inflow sewage quality and the inflow sewage amount collected by the external input measurement data collection unit 3. Various methods are conceivable as specific methods, for example, as follows. As an example, the ammonia concentration in the influent sewage is taken up.

  First, the typical value of ammonia concentration in the influent sewage is investigated. Normally, the value of ammonia concentration is about 15 to 40 [mg / L] for ordinary municipal sewage. Therefore, an upper limit value and a lower limit value that the ammonia concentration can take are set. For example, the lower limit is 10 [mg / L], and the upper limit is 40 [mg / L].

  In addition, the maximum width of the change in ammonia concentration is set in advance. For example, assuming that collection is performed at a cycle of 10 minutes, the maximum change rate of the ammonia concentration in 10 minutes is set to 10 [mg / L]. Furthermore, the autocorrelation function of the time series data of ammonia concentration is calculated, and it is examined how much past data and current data are correlated. Then, for example, a time lag in which the correlation coefficient is 0.8 (when 1 is set in the case of complete correlation) is observed. For example, it is assumed that the data up to 30 minutes ago usually has a correlation coefficient of 0.8 or more. Based on these preliminary investigations, the normality / abnormality of the ammonia sensor is determined.

  Then, normality / abnormality such as inflow sewage quality and inflow sewage amount is diagnosed. If it is determined that the sewage inflow sensor 1211 is normal, the external input sensor abnormality diagnosis unit 5 sets the contact of the switch unit SW1 to the a side. The data collected by the external input measurement data collection unit 3 is input to the process simulation unit 91 of the process optimization unit 9. On the other hand, when it is determined that the sewage inflow sensor 1211 is abnormal, the external input sensor abnormality diagnosis unit 5 switches the contact of the switch unit SW1 to the b side and optimizes the data stored in the manual analysis data storage unit 6. Input to the process simulation unit 91 of the unit 9. In the manual analysis data storage unit 6, for example, data such as COD, ammonia nitrogen, and phosphoric acid phosphorus analyzed once a day are stored in advance.

  If it is impossible to determine whether the sewage inflow sensor 1211 is normal or abnormal, the operator or administrator determines whether the sewage inflow sensor 1211 is normal or abnormal, and manually operates the switch unit SW1 to input externally. Either one of the measurement data collection unit 3 or the manual analysis data storage unit 6 is selected and input to the process simulation unit 91.

  Next, in the process optimization unit 9, based on the optimum evaluation function set by the optimum evaluation function setting unit 8 and various constraint conditions set by the constraint condition setting unit 7, the optimum target value and the optimum operation of the process are set. Calculate the quantity. The present invention proposes an algorithm for obtaining a specific solution to this optimization problem.

First, the optimization problem is written formally as follows.

  By solving this optimization problem, the optimum target value and the optimum operation amount can be calculated. As described above, when any one of the process sensors is diagnosed as abnormal by the process sensor abnormality diagnosis unit 4, the optimum operation amount corresponding to the abnormality is usually calculated, and otherwise, the optimum target value is calculated. . More optimally, the optimal target value and the optimal operation amount may be appropriately used according to time and circumstances.

In the case of the optimum target value, NO3-N, NH4-N, MLSS, PO4-P, A-SRT, etc. may be calculated using the answer to this optimization problem. When written formally,
Optimal Reference Set Points = arg _{NO3-N, NH4-N, MLSS, PO4-P, A-SRT} min Cost function subject to (Mass balance, Actuator limit, Output Limit).

  The calculated optimum target value is transferred to the local control unit 10, where the operation amounts of various actuators are determined. The operation of the local control unit 10 will be described later.

On the other hand, the optimal amount of operation is the answer to the optimization problem itself.
Optimal Actuator Set Points = arg _{Qstep, Qb, Qpac, Qcb, Qcirc, Qret, Qwaste} min Cost function subject to (Mass balance, Actuator limit, Output Limit).

  The present invention relates to the question of how to obtain an approximate global optimal solution for this optimization problem.

  In general, in an optimization problem with a constraint, if a point that does not satisfy the constraint condition (this is called a non-executable solution) is solved as an initial value, it becomes difficult to find an optimal solution that satisfies the constraint condition. That is, in a constrained optimization problem, it is preferable to obtain an appropriate solution that satisfies all the constraint conditions (referred to as an executable solution) before obtaining an optimal solution.

  As described above, in the present invention, various constraint conditions are set by the constraint condition setting unit 7. In particular, the storage amount constraint setting unit 71 sets the equation (14) as an equation constraint condition. Yes. Even if a certain point (a combination of variables) satisfies the equality constraint condition, it is rare that the equality condition is satisfied anymore if the value of the variable at that point deviates even slightly. It is difficult to handle in the sense of being. This difficulty can be avoided by reducing the number of variables to be optimized by solving the simultaneous equations when the number of equations in the equation constraints is less than or equal to the number of variables to be optimized. However, here, the inequality constraint condition is also considered in the output constraint setting unit 72 and the operation amount constraint setting unit 73, and the equality constraint condition and the inequality constraint condition are mixed. Since there is no guarantee that the answer to solve the equality constraint satisfies the inequality constraint, such a variable elimination method cannot be easily applied. For example, even if the equality constraint in equation (14) is solved, it is rare that the inequality condition that all variables are positive is satisfied. Therefore, it is considered that it is quite difficult to obtain a feasible solution satisfying both the inequality constraint condition and the equality constraint condition algebraically.

  However, the present invention finds that this problem can be avoided relatively easily by returning to the origin of the idea from which these constraints are derived, and proposes an optimization algorithm using the problem. The idea is explained below.

  The equality constraint in Eq. (14) is originally a static model (static) obtained as a steady-state solution of a differential equation that is a dynamic model (dynamic mass balance model) of the sewage treatment process expressed by Eq. (13). Therefore, in order to satisfy this equality constraint condition, a constant amount of operation and an external input are input to the dynamic model of Equation (13), and simulation is performed over a predetermined period. Another advantage of using such a method is that the simulation is actually performed, so if you input the amount of operations that can be taken in the actual process, the steady solution will usually be in reality. Therefore, the constraint value of the output constraint setting unit 72 and the operation amount constraint setting unit 73 is normally satisfied, and if it is not satisfied, the input operation amount can be changed. If the simulation is repeated and the simulation is repeated again, the inequalities can be satisfied, so this method can satisfy the equality and inequality constraints at the same time and obtain an executable solution. it can.

  In accordance with the above basic concept, a specific operation centered on the process optimization unit 9 will be described using the flowchart of FIG.

  First, in step 101, inflow sewage water quality data or inflow sewage that is measured by the sewage inflow sensor 1211 and is collected by the external input measurement data collection unit 3 and includes a plurality of water quality elements such as COD, TN, TP, and water quantity data. Retain water volume data as time-series data.

  Next, in step 102, averaging operation is performed on the held time series data at a predetermined cycle, a median value is taken out, or the worst load (inflow sewage amount + inflow sewage water quality) is set. The time series data is represented by a single representative value, such as taking out the maximum value.

  Next, in step 103, the first sedimentation pond excess sludge extraction pump 111, the step inflow pump 112, the blower 113 for supplying oxygen to the first aerobic tank 103, and the blower 114 for supplying oxygen to the second aerobic tank 105. The operation amount of the carbon source charging pump 115 for supplying the carbon source to the anaerobic tank 104, the circulation pump 116, the return sludge pump 117, the flocculant charging pump 118, the final sedimentation tank surplus sludge extraction pump 119, etc. can be controlled. Within a certain range, that is, one value (fixed value) within the range satisfying equation (16).

  Next, in step 104, the set fixed value manipulated variable and the above-described representative value are input to a process model like the expression (13) of the process simulation unit 91, and the process simulation is performed over a predetermined period. And the convergence value is taken out. Note that this convergence value satisfies the nonlinear algebraic equation (14).

  Next, in step 105, the value obtained by the convergence value, for example, the TN (total nitrogen) concentration, TP (total phosphorus) concentration, COD concentration, etc. of the discharged water quality is expressed by the expression (15) in the output constraint setting unit 72. Substitute into water quality inequality constraints. In step 106, it is checked whether or not this inequality is satisfied. If other inequality constraints, for example, inequality constraint conditions for securing the required A-SRT of the sewage treatment process or inequality constraint conditions for securing the required water area load in the settling basin are considered as the constraint conditions. If so, check whether those inequality constraints are met.

  If any one of these inequalities does not satisfy the constraint condition, the value of the manipulated variable is changed in step 107. This can be changed relatively easily with prior knowledge of the process behavior. For example, when the restriction condition regarding the TN of the discharged water quality is violated, the blower 113 for supplying oxygen to the first aerobic tank 103, the blower 114 for supplying oxygen to the second aerobic tank 105, and a carbon source are supplied. It can be easily understood that if the value of the operation amount related to nitrification (ammonia → nitric acid) and denitrification (nitric acid → nitrogen gas) of the carbon source charging pump 115 or the like is increased, TN is reduced and the constraint condition is easily satisfied. If there is no such prior knowledge, the value is appropriately changed within the operable range. Then, this value is input again to the process model of the process simulation unit 91 to perform process simulation.

  Conversely, if all inequality constraint conditions are satisfied, the equality constraint conditions have already been satisfied from the calculation procedure so far, and therefore all the constraint conditions of the constraint condition setting unit 7 are satisfied. Therefore, in this case, since this value is an executable solution when the optimization calculation unit 92 performs an optimization calculation with restrictions, it is adopted as an initial value at which the optimization calculation can be executed in step 108.

  Next, in step 109, the optimum computation unit 92 uses, as an initial value, the executable solution obtained in step 108, for example, nonlinear programming such as sequential quadratic programming (hereinafter abbreviated as SQP). To perform optimization operations. In this case, since the initial value is already selected as an executable solution, the solution obtained by this is at least an executable solution, which is a local optimum solution in the vicinity of the manipulated variable. It should be noted that if this initial value is selected as a non-executable solution by selecting a random number, for example, an executable solution is rarely obtained even when SQP or the like is executed.

  With the above procedure, a local optimum solution that satisfies all the constraint conditions can be obtained. However, another difficulty of the nonlinear optimization problem is that the optimization problem is generally a mathematical non-convex problem, so that the answer obtained by mathematical programming such as SQP is not necessarily a global optimal solution. It is.

  Therefore, an algorithm for obtaining a globally optimal solution approximately will be described with reference to FIGS.

  First, in step 111, a memory for storing the optimum value (= minimum value) of the global optimum solution and the optimum variable that gives the optimum value is prepared. Leave no big value.

  Next, in step 112, the operation amount is fixed within a range that satisfies the constraint condition of the operation amount constraint setting unit 73. To simplify the following description, reference is made to FIG. FIG. 4 shows the minimum value corresponding to the equation (16) with the abscissa of the blower 114 supplying oxygen to the second aerobic tank 105 as the operation amount, and the return amount by the return sludge pump 117 as the ordinate. And the maximum value for each side, the return amount and the aeration amount are separated in a grid pattern. First, an end point of this lattice, for example, the point “1” in FIG. 4 is first determined as a fixed value of the operation amount. Next, in step 113, a local optimum value is obtained for the manipulated variable by the above-described procedure for obtaining the local optimum solution.

  Next, in step 114, the local optimum value obtained in step 113 is compared with the stored optimum value. If the local optimal value is smaller than the stored optimal value, this is regarded as a new optimal value, reserved in the memory together with the corresponding optimal variable, and in the next step 116, the local optimal value is obtained. Replace the solution as a new global optimal solution. If the local optimum value is larger than the stored optimum value, nothing is done and the next operation amount is determined in the next step 115 and the steps 113 and 114 are executed. In other words, the same procedure is repeated by shifting the grid point of FIG. 4 by one and fixing the operation amount at the point “2”, for example. The end condition in this case is a case where the lattice point reaches the “END” point at the bottom right of the figure, and until then, the same procedure is repeated. Next, in step 117, it is determined whether or not the termination condition has been reached, and the optimum variable corresponding to the optimum value secured in the memory at the time of arrival is regarded as the approximate global optimum solution.

This optimal solution may have a better optimal solution if the grid points shown in FIG. 4 are rough, and if the grid points are fine, the optimal solution is likely to be a true optimal solution. . However, if the grid points are taken very finely, it takes a long time to calculate and the meaning is lost because it is not different from searching the entire region by simulation. Therefore, the method of taking this grid point should be divided by the fineness of the range that the calculation time allows. That is, assuming that a time is required for one local search and the calculation time is to be finished within z hours, the number of manipulated variables is n, and the number of divisions of each manipulated variable for constituting a lattice point is m. Then, the maximum m that satisfies a × m ⁿ <z may be set as the division number. In FIG. 4, the description has been made using two variables of the return amount and the aeration amount. However, it is actually necessary to take grid points in the hypercube for all the manipulated variables.

  If the diagnosis result of the process sensor abnormality diagnosis unit 4 or the external input sensor abnormality diagnosis unit 5 is not abnormal from the approximate global optimal solutions obtained according to the above procedure, for example, the nitric acid in the anoxic tank 104 The concentration target value, the ammonia concentration target value of the second aerobic tank 105, the phosphoric acid concentration target value, and the like are delivered to the local control unit 10. If any one of the sensors is abnormal, the optimum operation amount is directly designated for at least the operation amount corresponding to the sensor.

  According to such a procedure, the optimization operation of the process optimization unit 7 is executed.

  The plant-wide optimum sewage treatment process control can be specifically implemented according to the series of operations as described above.

  In the algorithm for obtaining an approximate global optimum solution in the process optimization operation of the process optimization unit 9, a solution satisfying all the inequality constraints among the process simulation convergence values that always satisfy the equation constraints. If is not found, this optimization problem is very likely not to have a solution. Such a situation can occur, for example, when the inflow sewage quality load is extremely high with respect to the regulation value of the discharged water quality, or when the load of the sewage amount is extremely high such as rainy weather. Or it may occur when a reaction inhibitor such as an extremely acidic substance flows in. In such a case, it is necessary to change the operation itself.

  Therefore, for example, the following message is displayed on the screen of the monitoring room of the sewage treatment process, for example, “This plant-wide optimum controller may not have a solution. The load of sewage flow rate or sewage inflow water quality is high. "Please check the cause and switch the operation mode."

  Similarly, in the algorithm for obtaining an approximate global optimal solution in the process optimization calculation of the process optimization unit 9, the carbon source amount by the normal carbon source input pump 115 and the flocculant input amount by the final sedimentation pond excess sludge extraction pump 119 are: For example, it takes a numerically smaller value than the aeration amount, return amount, circulation amount, and the like. For this reason, the influence of calculation errors tends to be relatively large on the carbon source input amount and the flocculant input amount. Therefore, it is preferable to perform variable conversion so that these manipulated variables are numerically within substantially the same range.

  Next, the local control unit 10 that operates the actuators 111 to 119 of the sewage altitude treatment process 1 according to the optimum target value obtained according to the algorithm described above will be described. Since the local control unit 10 is not a part directly related to the present invention, the detailed configuration thereof is omitted, but the first to seventh local control circuits are provided. Here, what is written as n (n = 1 to 7) in o corresponds to the output of the nth local control circuit.

  The first local control circuit captures the nitric acid concentration data collected and held in the first data collecting unit of the process measurement data collecting unit 2, and at the same time, the target value of the optimum nitric acid concentration calculated by the process optimizing unit 9. Capture. Then, the optimum step inflow amount is calculated based on the nitric acid concentration data and the optimum nitric acid concentration target value, and is output as the operation amount of the step inflow pump 112.

  The second local control circuit captures the ammonia concentration data collected and held in the second data collection unit of the process measurement data collection unit 2, and at the same time, the target of the optimum ammonia concentration calculated in the process optimization unit 9 Capture the value. Then, based on the ammonia concentration data and the optimum ammonia concentration target value, the optimum aeration air volume in the first aerobic tank 103 and the second aerobic tank 105 is calculated and output as the operation amount of the blower 113 and the blower 114. .

  The third local control circuit takes in the nitric acid concentration data collected and held in the third data collecting unit of the process measurement data collecting unit 2 and at the same time, the target of the optimum nitric acid concentration calculated in the process optimizing unit 9 Capture value. Then, an optimal carbon source input amount is calculated based on the nitric acid concentration data and the target value of the optimal nitric acid concentration, and is output as an operation amount of the carbon source input pump 115.

  The fourth local control circuit captures the nitric acid concentration data collected and held in the fourth data collection unit of the process measurement data collection unit 2 and at the same time, the target of the optimum nitric acid concentration calculated by the process optimization unit 9 Capture the value. Then, the optimum circulation amount is calculated based on the nitric acid concentration data and the target value of the optimum nitric acid concentration, and is output as the operation amount of the circulation pump 116.

  The fifth local control circuit captures the MLSS concentration data collected and held in the fifth data collection unit of the process measurement data collection unit 2, and at the same time, the target of the optimum MLSS concentration calculated in the process optimization unit 9 Capture the value. Based on the MLSS concentration data and the target value of the optimum MLSS concentration, the optimum return amount is calculated and output as the operation amount of the return sludge pump 117.

  The sixth local control circuit captures the phosphoric acid concentration data collected and held in the sixth data collecting unit of the process measurement data collecting unit 2, and at the same time, the optimum phosphoric acid concentration calculated by the process optimizing unit 9 The target value of is taken in. Based on the phosphoric acid concentration data and the target value of the optimal phosphoric acid concentration, the optimum flocculant charging amount is calculated and output as the operation amount of the flocculant charging pump 118.

  The seventh local control circuit includes the MLSS concentration data in the second aerobic tank 105 collected and held in the seventh data collection unit of the process measurement data collection unit 2, and the SS in the extracted sludge of the final sedimentation basin 106. The concentration data and the amount of sludge extracted by the flow rate sensor 127 are taken in, and at the same time, the optimum SRT (solids residence time) or A-SRT (solids residence time in the aerobic tank) calculated by the process optimization unit 9 is obtained. take in. Then, based on the MLSS concentration data, the SS concentration data, the extracted sludge amount data, and the optimum target value of SRT or A-SRT, the optimum excess sludge extraction amount is calculated, and the final sedimentation tank excess sludge extraction pump 119 is calculated. Is output as the operation amount.

  According to the series of operations as described above, optimal plant-wide sewage treatment process control can be performed.

<Effect of the first embodiment>
As is apparent from the above description, according to the first embodiment of the plant-wide optimum process control apparatus according to the present invention, the process simulation unit among the process simulation unit and the optimum calculation unit constituting the process optimization unit. Finds one or more feasible initial solution candidates of the optimal computing unit, and adopts one or more feasible initial solution candidates satisfying all the constraint conditions set in the constraint condition setting unit as feasible initial solutions In the optimal computing unit, one or more feasible initial solutions are used as initial values, one or more local optimal solutions are generated by mathematical programming, and the evaluation function is optimized (minimum) and adopted as an approximate optimal solution. Therefore, it is possible to specifically construct a plant-wide optimum process control apparatus.

  Further, by properly considering the handling of numerical errors in the optimization calculation, the optimization result of the plant-wide optimal process control device can be made reliable.

  Furthermore, by using the case where the solution is not found in the optimization calculation, it is shown that there is no answer that satisfies the constraints on the water quality regulation and the operation quantity regulation, and measures such as process operation change are taken. Can be announced.

Second embodiment

<Configuration of Second Embodiment>
The second embodiment differs from the first embodiment in that the process optimization unit 9 in FIG. 1 incorporates an algorithm for speeding up the calculation of optimization. 1 is the same.

<Operation of Second Embodiment>
In the following, the operation of the second embodiment will be described, particularly with respect to the parts different in configuration, together with the relationship with the first embodiment.

  FIG. 5 summarizes the algorithm incorporated in the process optimization unit 9 that performs optimization calculation as described in the first embodiment. In the figure, (a) shows the aeration air volume (air magnification) and the return quantity (return rate) on the bottom axis, and the evaluation value of equation (1) at that time is shown on the vertical axis. ) Is the same as FIG. 4 in which the return amount and the aeration amount are divided in a grid pattern. What is being performed in the first embodiment is basically the following four operations.

  First, an operating point that is a lattice point in FIG. 5B is selected. Next, the initial point to be searched is selected by performing simulation for the operating point. Speaking of the evaluation values in FIG. 5A, this corresponds to placing the initial point of the search on the value on the vertical axis (a point on the situation) with respect to a certain grid point. Next, a mathematical programming method such as sequential quadratic programming is used to search from that phase toward the valley bottom with a lower evaluation value. As can be easily seen from FIG. 5, the valley bottom finally reached (local minimum point = local optimum solution) differs depending on where the initial initial point is placed.

  In order to avoid this, in the first embodiment, initial points are placed on the phases corresponding to all the lattice points as shown in FIG. 5B, and valley bottoms (local optimal solutions) obtained from those initial points are obtained. ) And comparing these values, the idea that the smallest local optimal solution is regarded as a global optimal solution is specifically algorithmized.

  However, as can be easily guessed from FIG. 5, such a search method is inefficient. In the case where there is no restriction on the calculation time, such a method is also allowed. However, in practice, such a plant-wide optimum process control device is executed online, so that it is within the time that can be executed online. It is necessary to perform calculation. Here, a simple trial calculation will explain that the method as in the first embodiment is quite difficult to implement online.

The calculation time required to search for one local optimal solution is 30 seconds (this is the approximate time when the inventor actually performed the calculation). As operation amounts, a blower 113 for supplying oxygen to the first aerobic tank 103, a blower 114 for supplying oxygen to the second aerobic tank 105, a carbon source input pump 115 for supplying a carbon source, a circulation pump 116, and a return sludge pump Consider seven of 117, flocculant charging pump 118, and final sedimentation basin excess sludge extraction pump 119. Then, five points are taken between the upper and lower limits of each manipulated variable. Then, the number of grid points corresponding to FIG. 5B is 5 ⁷ = 78125. Accordingly, when an approximate global optimum solution is obtained using all of these lattice points as initial points, 30 × 78125 = 2343750 seconds, that is, a calculation time of about 27 days is required. From this, it can be seen that in order to actually realize this plant-wide optimum process control apparatus, a speed-up method for obtaining an approximate global optimum solution is important.

  The second embodiment is intended to increase the speed, and for that purpose, consider using a heuristic search method called a metaheuristic. In particular, the operation will be described below using genetic algorithms (hereinafter abbreviated as GA), which is a typical metaheuristic technique.

  In GA, a strategy is to prepare multiple individuals consisting of gene sequences, evaluate these individuals with an evaluation index called fitness, and survive these individuals with good fitness, and deceive individuals with poor fitness It is a heuristic optimization method based on One feature of GA is that it can perform relatively high-speed calculations.

  Here, in order to use the GA for the optimization calculation of the present plant-wide optimum process control device, it is first necessary to define “individual” and “fitness”. Therefore, here, it is assumed that “individual = each grid point in FIG. 5B”. That is, a combination of operation amounts corresponding to each lattice point is regarded as one individual. Therefore, preparing a plurality of individuals means considering a plurality of lattice points simultaneously. Next, regarding “goodness”, the goodness-of-fit is simply evaluated by the value of the evaluation function of equation (1).

  However, in GA algorithm, it is usually considered that a good fit is better (it is natural to define it in this way, considering the meaning of the word fit), so the meaning of this word In order to match, the goodness of fit should be evaluated by the reciprocal of the value of the evaluation function in equation (1), but in both cases it is essentially the same (it is just a matter of definition). More importantly, in the second embodiment, not the value of the point on the curved surface of FIG. 5A corresponding to the operation amount of each grid point, but the diagram corresponding to the operation amount of each grid point. The point of 5 (a) is that the evaluation value of the local minimum solution obtained by SQP or the like is adopted as the initial value of the point on the curved surface as the fitness.

  That is, the fitness of each individual (= combination of the operation amount corresponding to each grid point) is evaluated by the evaluation value of the local optimum solution when that grid point is the initial solution. By making such an association, GA can be applied independently of a specific algorithm in the middle of obtaining a local optimum solution as shown in the first embodiment (a heuristic algorithm based on GA). A global search and an operation for obtaining a local optimal solution for an individual (= an operation amount corresponding to a grid point) can be discussed separately).

By defining “individual” and “fitness” in this way, an approximate global optimum solution can be obtained in accordance with normal GA processing. GA processing consists of the following series of operations.
Step1 Generation of initial individuals Step2 Evaluation of fitness of each individual Step3 Selection of excellent individuals (淘汰)
Step 4 (If necessary) Propagation of individuals Step 5 Generation of children by crossover of individual individuals Step 6 Generation of new individuals by mutation Step 7 Return to Step 2 until the end condition is met.

  What is necessary is just to materialize this series of flows.

  First, in Step 1, an initial individual must be generated, and several methods are conceivable for this. A typical method is to assign a number to each lattice point in FIG. 5B, generate a random number including all the numbers, and select a plurality of lattice points as initial individuals. Another simple method is to select the corner of the lattice point as the initial individual in order to search the entire region as much as possible. Either method may be used, or another method may be used, but a plurality of lattice points are generated as initial individuals.

  Next, in Step 2, the fitness of each individual is evaluated. If the algorithm for obtaining the local optimal solution in the first embodiment is applied, the fitness can be evaluated by the evaluation value or its reciprocal. it can.

  Next, in Step 3, an excellent individual can be easily selected by using the fitness in Step 2. For example, the fitness of all individuals may be arranged in order, and a predetermined number of individuals may be alive in order from the highest fitness, and other individuals with poor fitness may be selected. Alternatively, the degree of deterioration of the degree of fitness compared to the individual with the highest degree of fitness is evaluated (degradation level = the degree of fitness of the best individual / the degree of fitness of the target individual), for example, those with a degree of degradation exceeding 2 are hesitant You can also choose. Here, since the fitness is evaluated by the evaluation value of the local optimal solution corresponding to each grid point, the fitness for different grid points may completely match (when the same local optimal solution is reached) Note that this is an exact match). In that case, one having the same fitness can leave one individual and all the remaining can be deceived, or all the individuals can be left. When the former method is adopted, the diversity of individuals can be kept more and it works to find a better solution that exists in a completely different place, but if it goes too far, it will not be different from random search. On the contrary, when the latter method is adopted, the solution is found near the good solution found so far, so if the solution is really close to the global optimal solution, the solution can be found sooner. On the other hand, since we give up searching for a possibility that there is a better solution, there is also a problem that it is impossible to get out of the local optimum solution. Therefore, it should be tried and tried to determine which of these should be used.

  Next, in Step 4, if the number of individuals selected in Step 3 is small (for example, if it is less than half of the initial number of individuals), an operation that maintains a certain number of individuals by copying a plurality of selected individuals I do.

  Next, crossover is performed at Step5. Various methods can be considered as crossover methods. For example, each grid point is coded in binary, and two selected individuals are paired by an appropriate method. Then, as shown in FIG. 6, it is possible to employ a normal crossover method in which the code of the first half part and the code of the second half part of two individuals that are paired by determining an intersection with a random number are exchanged. Alternatively, when it is not desired to change the value of each manipulated variable, as shown in FIG. 6, the crossover point should be limited to only the joint of each manipulated variable coded in binary number. Good. When such crossover is performed, a child inheriting the parental character of the paired parent is generated. If the number of parents selected in Step 3 is set to be half that of the initial individual, the number of individuals can be kept constant together with the thus generated children.

  Next, with a very low probability, mutation is performed, for example, by inverting one bit of the binarized code so as to keep the diversity of the search.

  Unless the end condition is finally satisfied, the operation returns to Step 2 to continue this operation. The end condition is set by, for example, the number of generations (number of repetitions).

  As described above, the search method using GA in the second embodiment conceptually searches only in the vicinity where there is likely to be a global optimal solution as shown in FIG. The other is not to search. Any method can be used as long as it can efficiently search for a global optimal solution or a local optimal solution whose evaluation value is very close to the global optimal solution. In the field of metaheuristics, a technique such as an immune algorithm has been developed in addition to GA. These principles differ little by little, and the development of faster search methods is progressing. Therefore, if there is a faster search method capable of performing intensive search as shown in FIG. 7, it is natural to apply this instead of GA.

<Effects of Second Embodiment>
Thus, according to the second embodiment, in addition to the effects of the first embodiment, the optimization operation of the plant-wide optimum process control device can be speeded up, and this device can be implemented online. Will be able to.

Third embodiment

<Configuration of Third Embodiment>
The process control system is configured in the same manner as in the first embodiment shown in FIG. 1, and the plant-wide optimum process control apparatus captures the current operation information of the process in the process optimization unit 9 to simplify the calculation. The point of speeding up is different from the configuration of the first embodiment.

<Operation of Third Embodiment>
Hereinafter, the process optimization unit 9 having a configuration different from that of the above embodiment will be described with respect to the operation of the third embodiment. In the first and second embodiments, in order to obtain a global optimum solution, a local optimum solution is obtained for a plurality of points of grid points (= operating point, operating point) partitioned in a grid pattern, and these are obtained. By comparing, an attempt was made to find a global optimum solution. These do not use any information about the amount of operation being performed in the actual process. This is because it is completely unknown mathematically whether the manipulated variable operated in the actual process is optimal, and mathematically, it is unlikely that it is actually optimal. It is no wonder from the point of view of optimization not to use.

  However, “The amount of operation in the actual process operation is determined based on many years of experience, so it is empirically quite good, and even if it is not optimal, it may be near optimal. `` In the present invention, the optimum operation amount is determined based on the formula (1), but in the actual process, various conditions of the actual process that are not included in the evaluation of the formula (1) In many cases, the user does not want to significantly change the current operation amount for reasons such as “the operation amount is often determined in consideration of the situation”.

  In such a case, as shown in the first embodiment and the second embodiment, it is better to search near the actual driving conditions than to search many potential areas. There is also. Therefore, as an initial solution for the search, first, an operation amount in the current process operation is input. Then, a process simulation is performed, and a local search is performed by SQP or the like using the convergence value as an initial solution. Using the local optimal solution obtained in this way, the optimal target value and the optimal operation amount are supplied.

  Similarly, in an actual process, the value of a certain operation amount may be changed, but there are often situations in which it is not desired to change other operation amounts. For example, there are sewage treatment plants where the amount of aeration is greatly changed, but the amount of return and circulation is not changed significantly. In such a case, the current manipulated variable is given as a fixed value for items that do not want to be changed, such as the return amount and the circulation amount, and are not imitated as variables to be optimized. Then, it optimizes the manipulated variable and other state variables (such as water quality) that can be changed, such as aeration air volume, as variables to be optimized, and supplies the optimum target value and optimum manipulated variable. Can do.

<Effect of the third embodiment>
Thus, according to the third embodiment, in addition to the effects of the first embodiment, the following effects can be obtained.

  Instead of giving up global optimization, it is possible to provide a practical optimal solution that can further improve the current process operating conditions. At the same time, since the operating conditions are not changed too rapidly, the process operator can be given a sense of security and reliability, and is likely to be used in practice.

  In addition, since the actual process operation is often performed in consideration of various conditions that do not appear in the evaluation function, limiting the search to the vicinity of this operation condition results in an evaluation. It can be expected that optimization considering various conditions that do not appear in the function will be achieved.

  Furthermore, since the search is performed only once and the number of optimization variables may be reduced, the calculation can be performed extremely quickly and application on-line becomes easy.

Fourth embodiment

<Configuration of Fourth Embodiment>
The process control system is configured in the same manner as in the first embodiment shown in FIG. 1, and the plant-wide optimum process control device is just performing approximate global optimization operations in the process optimization unit 9. In addition, the configuration is different from the first embodiment in that the optimum degree is evaluated at the same time.

<Operation of Fourth Embodiment>
Hereinafter, the process optimization unit 9 having a configuration different from that of the above-described embodiment will be described regarding the operation of the fourth embodiment.

  Each of the above embodiments focuses on performing practical optimization calculations, and quantitatively estimates how much the resulting approximate optimal solution is optimal. There wasn't. Although the first and second embodiments are directed to obtaining an approximate global optimum solution, depending on the method of dividing lattice points, the number of GA search repetitions, and the like, as a result There was also a possibility of falling into a local optimal solution.

  The fourth embodiment is directed to the evaluation of the optimality of the optimal solution. In the optimization problem, the lower bound of the optimum solution as shown in FIG. 8 can be estimated by considering the relaxation problem for the optimization problem to be solved. The sewage treatment process model and the like taken up in the above embodiment include a fractional nonlinear function and a bilinear nonlinear function called mono-equation, but the equality and inequality constraints related to these are polynomials. It can be expressed as a type nonlinear function. For optimization problems that have such a polynomial-type nonlinear function as a constraint condition or objective function, it is known that the method of half positive definite relaxation is effective for estimating the lower bound. Therefore, as shown in each of the above embodiments, if the optimization problem is specifically solved by a practical method and the semi-definite relaxation problem is simultaneously solved and the lower bound is estimated simultaneously, how much is the maximum It can be seen at the same time whether the approximate optimal solution can have an error.

  If this fact is used, the algorithm for obtaining the approximate global optimum solution using the GA shown in the second embodiment can be further accelerated. In GA, the number of generations (number of iterations) is often set in advance, but if this lower bound is evaluated and an allowable error is set in advance, an iterative search is performed when the allowable error is entered. In this way, an approximate global optimum field within an allowable range can be searched very quickly.

<Effect of the fourth embodiment>
Thus, according to the fourth embodiment, in addition to the effects of the first embodiment, the following effects can be obtained.

  By obtaining a practical optimal solution and evaluating its reachable limit, the optimality of the obtained practical approximate optimal solution can be evaluated.

  In addition, by evaluating the limit of optimality and setting an allowable error, the approximate global optimum solution search can be accelerated more efficiently.

Fifth embodiment

  Each embodiment mentioned above is provided with the local control part 10 corresponding to one sewage altitude treatment process 1, and supplies the optimal target value of this local control part 10, or actuators 111-119. Although there are a constraint condition setting unit 7, an optimum evaluation function setting unit 8 and a process optimization unit 9 for directly calculating the optimum operation amount of the sewage, there are a plurality of target processes such as an advanced sewage treatment process 1 and the like. When there are a plurality of local control units 10 corresponding to the process, the processing capacity of the processing unit is directly calculated in order to directly calculate the optimum target values to be supplied to the local control units 10 or the optimum operation amounts of the actuators 111 to 119. One high-level control device is provided, and the optimal target value of the plurality of local control units 10 or the optimal operation of the actuator is provided in this high-level control device. Was calculated, it can be configured to be connected with each local control unit 10 or actuators 111 to 119 in the data transmission path.

  This makes it possible to use the higher-level control device in common for a plurality of target processes. As a result, when a plurality of plant-wide control devices are provided for a plurality of target processes, it is possible to reduce the number of higher-level control devices and control each target process as in an application provider service (ASP). There is also an effect that it can be introduced at low cost.

Sixth embodiment

  In the fifth embodiment, all the functions of the constraint condition setting unit 7, the optimum evaluation function setting unit 8, and the process optimization unit 9 are provided in a single upper control device for a plurality of target processes. However, it is also possible to divide these functions into a plurality of control devices and connect these control devices via a data transmission path.

  As a result, when the processing capacity of the existing control device is sufficient, the function of the host control device can be realized by connecting the control device via the data transmission path. As a result, it is possible to introduce a plant-wide control device without newly introducing a control device as a higher-level control device.

The block diagram which shows the structure of the control system which makes object the removal process of nitrogen and phosphorus as 1st Embodiment of the plant wide optimal process control apparatus which concerns on this invention. The flowchart which showed the algorithm for calculating | requiring the local optimal solution in the case of having a substance amount preservation | save equality constraint, in order to demonstrate the effect | action of 1st Embodiment. The flowchart which showed the algorithm for calculating | requiring the approximate global optimal solution in case it has a substance amount preservation | save equality constraint, in order to demonstrate the effect | action of 1st Embodiment. The figure for demonstrating the procedure for calculating | requiring the approximate global optimal solution in the case of having a substance amount preservation | save equality constraint, in order to demonstrate the effect | action of 1st Embodiment. Explanatory drawing which showed the fundamental view of the algorithm which calculates | requires the approximate global optimal solution relevant to 2nd Embodiment of the plant wide optimal process control apparatus which concerns on this invention. The figure explaining the concept of crossover in the high-speed calculation using GA in order to demonstrate operation | movement of 2nd Embodiment. The figure explaining the concept of the principle of speed-up of the high-speed calculation using GA in order to demonstrate operation | movement of 2nd Embodiment. The figure explaining the relationship between the optimal value in optimization calculation, and its lower bound, in order to demonstrate operation | movement of 4th Embodiment of the plant wide optimal process control apparatus which concerns on this invention.

Explanation of symbols

1 Sewage Advanced Processing Process 2 Process Measurement Data Collection Unit 3 External Input Measurement Data Collection Unit 4 Process Sensor Abnormality Diagnosis Unit 5 External Input Sensor Abnormality Diagnosis Unit 6 Manual Analysis Data Storage Unit 7 Constraint Condition Setting Unit 8 Optimal Evaluation Function Setting Unit 9 Process Optimization unit 10 Local control unit 71 Storage amount constraint setting unit 72 Output constraint setting unit 73 Operation amount constraint setting unit 91 Process simulation unit 92 Optimal operation unit 101 First sedimentation basin 102 Anaerobic tank 103 First aerobic tank 104 Anoxic tank 105 Second aerobic tank 106 Final sedimentation basin 111 First sedimentation basin excess sludge extraction pump 112 Step inflow pump 113, 114 Blower 115 Carbon source input pump 116 Circulation pump 117 Return sludge pump 118 Flocculant input pump 119 Final sedimentation tank excess sludge extraction pump 121,124 Ammonia sensor 122 Nitric acid sensor 123 MLSS sensor 125 Phosphoric acid sensor 126 SS sensor 127 Flow rate sensor 128 DO sensor 129 ORP sensor 1210 Outflow sewage sensor 1211 Sewage inflow sensor

Claims (11)

At least one process sensor capable of measuring several states of the target process, with the inflowing substance or energy as an external input and the processing step of the substance or energy flow as the target process;
At least one actuator capable of changing the state of the subject process;
A constraint condition setting unit including a storage amount constraint setting unit capable of setting at least a static constraint condition represented by an algebraic equation relating to a conservation rule of a storage amount such as the amount of substance and energy of the external input;
An optimal evaluation function setting unit that can set an evaluation function that can evaluate the operating cost and multiple control performances;
A local controller that operates the actuator based on process measurements obtained by the process sensor to compensate for dynamic variations in the process to have desirable characteristics;
The constraint condition set by the constraint condition setting unit is input, and the optimal target value of the local control unit is supplied based on the evaluation function set by the optimal evaluation function setting unit, or the actuator is optimally operated. A process optimization unit that directly calculates the quantity;
In a plant-wide optimum process control device equipped with
The process optimization unit includes a process simulation unit represented by a differential equation that is a dynamic model of a process corresponding to an algebraic equation that is a steady state model of a process set by the storage amount constraint setting unit, and an optimization calculation And an optimal calculation unit that performs
In the process simulation unit, one or more operation amount values of the actuator are generated within an operation range of the operation amount and input to a dynamic model to generate a convergence value of the dynamic model. One or more feasible initial solution candidates of the optimal computing unit satisfying the substance amount storage constraint of the setting unit are obtained, and all the constraint conditions set in the constraint condition setting unit are obtained from the feasible initial solution candidates. One or more feasible initial solution candidates to be satisfied are adopted as feasible initial solutions, and the optimum computing unit uses the one or more feasible initial solutions as initial values and uses one or more local optimum solutions by mathematical programming. An algorithm is built in which the one that generates and optimizes the evaluation function is adopted as an approximate optimal solution.
A plant-wide optimum process control device characterized by that.   The plant-wide according to claim 1, wherein the process optimization unit normalizes a range that can be taken by an independent variable to be optimized, such as an operation amount or a state variable that appears, to be substantially the same. Optimal process control device.   A combination of manipulated variable values for generating executable initial solution candidates in the process simulation unit of the process optimization unit is regarded as a gene, and is converted from an evaluation function value calculated from the gene through the optimal calculation unit or an evaluation function value The plant-wide optimum process control apparatus according to claim 1, wherein a genetic algorithm is used in combination with a certain index as a goodness of fit.   The plant-wide optimum process control device according to claim 3, wherein an immune algorithm or metaheuristics is applied instead of the genetic algorithm of the process optimization unit.   The actual operation amount at the time of calculation of the target process is adopted as a combination of operation amount values for generating executable initial solution candidates in the process simulation unit of the process optimization unit. The plant-wide optimum process control apparatus according to 2.   2. The process optimization unit considers some of the independent variables to be optimized as fixed values, does not perform optimization, and optimizes only the remaining independent variables. The plant-wide optimum process control apparatus according to 2.   The plant-wide optimum process control device according to claim 1, wherein the process optimization unit estimates a lower bound of the optimum value by simultaneously solving a relaxation problem of the formulated optimization problem. A combination of manipulated variable values for generating executable initial solution candidates in the process simulation unit of the process optimization unit is regarded as a gene, and is converted from an evaluation function value calculated from the gene through the optimal calculation unit or an evaluation function value A certain algorithm is considered as a goodness of fit, and a genetic algorithm is used together.
In the process optimization unit, the lower bound of the optimum value is estimated by simultaneously solving the relaxation problem of the formulated optimization problem,
3. The plant according to claim 1, wherein a termination condition of the genetic algorithm is determined based on a threshold level provided for a difference between a lower bound of the approximate optimal solution and the optimal solution. Wide optimal process control device.   A function is provided for presenting that there is a high possibility that there is no solution in the optimization problem when no executable solution is found among the executable initial solution candidates. The plant-wide optimum process control device according to 1 or 2.   The host controller is provided corresponding to each of a plurality of target processes, and each function of the constraint condition setting unit, the optimum evaluation function setting unit, and the process optimization unit is provided in a single host control device. The plant-wide optimum process control device according to claim 1, wherein elements corresponding to the side control device are connected by a data transmission path.   11. The plant according to claim 10, wherein functions of the constraint condition setting, the optimum evaluation function setting unit, and the process optimization unit are divided into a plurality of control devices, and these control devices are connected by a data transmission path. Wide optimal control device.

Images