Weighting coefficient determination device, weighting coefficient determination method, and program

Abstract

To provide a weighting factor determination device, a weighting factor determination method and a program, that can set a weighting factor of an evaluation function used for solving an optimization problem to a value suitable for the optimization problem.SOLUTION: A weighting factor determination device 200 comprises: calculating a calculated value di'_sim when a value of an evaluation function J is maximum or minimum for variables in an evaluation function that define an evaluation index ei' included in the evaluation function J by algorithm of a first optimization problem; calculating the value of a difference evaluation index |di'_sim-di'_opt| for evaluating the difference between the calculated value di'_sim of the variable in the evaluation function and a preset optimal value di'_opt of the variable in the evaluation function; and determining an optimal solution wi_opt for the weighting factor based on a calculated difference evaluation index value.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a weighting coefficient determination device, a weighting coefficient determination method, and a program, and is particularly suitable for use in determining weighting coefficients included in evaluation functions used to solve optimization problems. [Background technology]

[0002] An optimization problem is a problem of determining design variables that maximize the evaluation of each of multiple evaluation indexes. Specifically, multiple evaluation indexes are constructed using variables, including design variables (also called decision variables) that are the targets of the optimization problem. An evaluation function is formulated using the multiple evaluation indexes and weighting coefficients for each evaluation index, and the values ​​of the variables, including the design variables, that minimize or maximize the value of the evaluation function are determined. In this way, the optimization problem is expressed using an evaluation function and weighting coefficients. The weighting coefficient represents the balance of evaluation among the multiple evaluation indexes and is determined in advance before calculating the value of the evaluation function. The magnitude relationship of the weighting coefficient values ​​is determined according to the importance of the multiple evaluation indexes, with the absolute value being larger for more important evaluation indexes. The evaluation function is also referred to as an objective function or cost function. The weighting coefficient is also referred to as a cost coefficient.

[0003] Patent Document 1 discloses a technology that utilizes an optimization problem to control the amount of heat input to a coke oven. Specifically, Patent Document 1 discloses a method of calculating the amount of coke temperature change for each oven when the value of an evaluation function expressed as a weighted linear sum of an evaluation index using the coke temperature deviation between ovens and an evaluation index using the amount of coke temperature change for each oven is minimized, and converting the amount of coke temperature change for each oven into the amount of change in the aperture of a control valve that adjusts the flow rate of combustion gas and air. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 9-302351 [Non-patent literature]

[0005] [Non-Patent Document 1] Yuta Tsuboi and Takaki Makino, "Application of Inverse Reinforcement Learning and Imitation Learning in Natural Language Processing," Measurement and Control, Vol. 52, No. 10, October 2013, pp. 922-927 Summary of the Invention [Problem to be solved by the invention]

[0006] However, including the technology described in Patent Document 1, weighting coefficients in an evaluation function are generally determined based on the experience of the designer of the optimization problem. For example, in the technology described in Patent Document 1, if controlling the calorific value input to a coke oven results in a failure to obtain coke of the desired quality, the weighting coefficients in the evaluation function are repeatedly adjusted. Therefore, it is not easy to set an appropriate value for the weighting coefficient in the evaluation function. This problem is not limited to optimization problems for performing process control, such as controlling the calorific value input to a coke oven, but is a common problem in various optimization problems.

[0007] The present invention has been made in view of the above problems, and has as its object to set the weighting coefficients of an evaluation function used to solve an optimization problem to values ​​suitable for that optimization problem. [Means for solving the problem]

[0008] The weight coefficient determination device of the present invention is a weight coefficient determination device that determines weight coefficients for a plurality of evaluation indexes included in an evaluation function used to solve an optimization problem by inverse reinforcement learning, and includes a calculation means that calculates a calculated value when a value of the evaluation function becomes maximum or minimum for a variable in the evaluation function that is a variable that determines the evaluation index, and calculates a value of a difference evaluation index that evaluates the difference between the calculated value of the variable in the evaluation function calculated by the calculation means and a preset optimal value of the variable in the evaluation function, By solving an optimization problem using a differential evaluation function, which is an evaluation function including the differential evaluation index,and a weighting factor determining means for determining the value of the weighting factor.

[0009] The weight coefficient determination method of the present invention is a weight coefficient determination method for determining weight coefficients for a plurality of evaluation indexes included in an evaluation function used to solve an optimization problem by inverse reinforcement learning, and includes the steps of: calculating a calculated value when a value of the evaluation function is maximized or minimized for a variable in the evaluation function that is a variable that determines the evaluation index; and calculating a value of a difference evaluation index that evaluates the difference between the calculated value of the variable in the evaluation function calculated in the calculation step and a preset optimal value of the variable in the evaluation function; By solving an optimization problem using a differential evaluation function, which is an evaluation function including the differential evaluation index, and a weighting coefficient determining step of determining the value of the weighting coefficient.

[0010] The program of the present invention causes a computer to function as each of the means of the weighting coefficient determination device. [Effects of the Invention]

[0011] According to the present invention, a weighting coefficient for an evaluation index determined by a variable in an evaluation function is determined based on the result of evaluating the difference between the calculated value of the variable in the evaluation function calculated using the algorithm for the optimization problem and the optimal value, so that the weighting coefficient of the evaluation function used to solve the optimization problem can be set to a value appropriate for that optimization problem. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 10 is a diagram illustrating an example of a method for determining a weighting coefficient. [Figure 2] FIG. 2 is a diagram illustrating an example of a functional configuration of a weighting coefficient determination device. [Figure 3] 10 is a flowchart illustrating an example of a weighting coefficient determination method. [Figure 4] 1 illustrates an example of a coke oven and coke making process. FIG. [Figure 5A] FIG. 2 is a diagram illustrating an example of furnace battery temperature. [Figure 5B] FIG. 10 is a diagram showing an example of coke being pushed out of a carbonization chamber. [Figure 6] FIG. 2 illustrates an example of a functional configuration of a processing device. [Figure 7] FIG. 1 is a diagram showing an example of the relationship between coke temperature, furnace battery temperature, input heat amount, and carbonization time and time. [Figure 8] FIG. 2 is a diagram illustrating an example of a target furnace temperature trajectory. [Figure 9A] FIG. 10 is a diagram illustrating an example of a target furnace temperature trajectory determined at the start time of unsteady operation. [Figure 9B] FIG. 10 is a diagram illustrating an example of a target furnace temperature trajectory determined at the timing when the operating conditions are changed. [Figure 10] 10 is a flowchart illustrating an example of a processing method. [Figure 11] FIG. 2 is a diagram showing an example of a target furnace temperature trajectory and an actual furnace temperature trajectory. [Figure 12] FIG. 10 is a diagram illustrating an application example of the functional configuration of a weighting coefficient determination device. [Figure 13] 3A to 3C are diagrams illustrating an example of an operator operation history, an operating condition, and a control history. [Figure 14] 10A and 10B are diagrams illustrating an example of a target furnace temperature trajectory, an estimated furnace temperature trajectory, and an operator intervention furnace temperature trajectory. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. [overview] First, an example of the outline of this embodiment will be described. In this embodiment, weighting coefficients for a plurality of evaluation indexes included in an evaluation function used to solve an optimization problem are determined by inverse reinforcement learning. FIG. 1 is a diagram illustrating an example of a method for determining weighting coefficients. Note that the present invention is not limited to the one shown in FIG. 1.

[0014] As explained in the Background Art section, in an optimization problem, multiple evaluation indexes are constructed using variables, including the design variable to be solved for the optimization problem. An evaluation function is formulated using the multiple evaluation indexes and weighting coefficients for each evaluation index, and the values ​​of the variables, including the design variable, that minimize or maximize the evaluation function value are found. In the following explanation, variables, including the design variable, are also referred to as variables within the evaluation function. Figure 1 illustrates an example in which the evaluation function J is expressed as a weighted linear sum of I evaluation indexes ei (I is an integer equal to or greater than 2) (see J = Σwi · ei shown in calculation process C1). The evaluation index ei may be expressed as a function of the variables d i within the evaluation function, or may be equal to the variables d i within the evaluation function. Furthermore, one evaluation index ei may be expressed as a function of multiple variables d i within the evaluation function. In each equation in Figure 1, the variable i is a variable that identifies the evaluation index ei (variable d i within the evaluation function).

[0015] In calculation process C1 in Fig. 1, the design variables when the value of evaluation function J is maximized or minimized are calculated as the optimal solution. The design variables are variables in the evaluation function that are the targets of solving the optimization problem. Note that the value of evaluation function J being maximized or minimized means that the value of evaluation function J is maximized or minimized within a range that satisfies the conditions (and constraints, if necessary) determined by the algorithm for the optimization problem (this also applies to the following explanations). An example of a condition determined by the algorithm for the optimization problem is that the change in the current value of the design variable from the previous value is within a predetermined value.

[0016] In this way, in the optimization problem, the weighting coefficient wi for each evaluation index ei in the evaluation function J is set in advance, and the variable di in the evaluation function is calculated when the value of the evaluation function J is maximized or minimized. The inventors have noticed that it is possible to identify the variable di in the evaluation function when a desired result is obtained from past performance. For example, when processing an item, if the processing cost and processing time of the item are used as evaluation indexes (variables in the evaluation function), past processing performance for each item may be obtained in which the processing cost for each item is equal to or less than the desired cost and the processing time for each item is equal to or less than the desired time. The inventors have found that by using an inverse reinforcement learning technique and assuming that the variable di in the evaluation function when such a desired result is obtained is the given optimal value, it is possible to search for an optimal solution for the evaluation function J (weighting coefficient wi).

[0017] As described in Non-Patent Document 1, inverse reinforcement learning is a method that assumes that an optimal training action sequence is given and minimizes an objective function defined by the action sequence under a reward function and the optimal training action sequence. In other words, inverse reinforcement learning assumes that the optimal action / behavior is given, and determines a reward function that minimizes the difference between the action / behavior determined under the reward function and the preset optimal action / behavior.

[0018] In FIG. 1, the reward function corresponds to the evaluation function J. The optimal training action sequence corresponds to the optimal evaluation function variable di'_opt. In the following description, this optimal evaluation function variable is also referred to as the optimal value of the evaluation function variable. Here, variable i', like i, is a variable that specifies the evaluation index ei' (variable di' in the evaluation function), and the number of evaluation indexes ei' (variable di' in the evaluation function) is assumed to be I' (I' is a positive integer equal to or less than I). In other words, the evaluation index ei' (variable di' in the evaluation function) indicates at least one evaluation index ei (variable di' in the evaluation function) among the evaluation indexes ei (variable di' in the evaluation function) included in the evaluation function J. The evaluation index ei' (variable di' in the evaluation function) may completely match the evaluation index ei (variable di' in the evaluation function) included in the evaluation function J, or may be only a part of the evaluation index ei (variable di' in the evaluation function) included in the evaluation function J. In addition, in Figure 1, the behavior / action defined under the reward function corresponds to the variable di'_sim in the evaluation function when the value of the evaluation function J is maximum or minimum. In the following explanation, this variable in the evaluation function is also referred to as the calculated value of the variable in the evaluation function.

[0019] 1, the objective function defined by the action sequence under the reward function and the optimal training action sequence is expressed using the differential evaluation index |di'_sim-di'_opt| calculated in calculation process C2 in FIG. 1. Here, the differential evaluation index |di'_sim-di'_opt| only needs to be calculated for at least one evaluation function variable di among the evaluation function variables di included in the evaluation function J, and does not necessarily have to be calculated for all of the evaluation function variables di included in the evaluation function J. As described above, the evaluation index ei (variable di within the evaluation function) is the evaluation index ei (variable di within the evaluation function) included in the evaluation function J in the optimization problem. The evaluation index ei' (variable di within the evaluation function) indicates the evaluation index ei (variable di within the evaluation function) used in inverse reinforcement learning (used to calculate the differential evaluation index) among the evaluation indexes ei (variable di within the evaluation function) included in the evaluation function J in the optimization problem. Of the variables di_sim in the evaluation function when the value of the evaluation function J is maximum or minimum, the variable in the evaluation function used in reinforcement learning (used to calculate the differential evaluation index) is di'_sim.

[0020] In addition, in Figure 1, minimizing the objective function defined by the sequence of actions under the reward function and the optimal training sequence of actions corresponds to calculating the evaluation function J (weighting coefficient wi) that minimizes the differential evaluation index |di'_sim-di'_opt| for each variable di' in the evaluation function.

[0021] In FIG. 1, to determine the weight coefficient wi using inverse reinforcement learning, first, in calculation process C1, the initial value of the weight coefficient wi is applied to the right-hand side of the evaluation function J to calculate the calculated value di_sim (optimal solution) of the variable in the evaluation function. As mentioned above, the optimal value di'_opt of the variable in the evaluation function is determined based on the past performance of the behavior and action, and is given. Therefore, in calculation process C2, the value of the differential evaluation index |di'_sim - di'_opt| is calculated using the calculated value di_sim and the optimal value di'_opt of the variable in the evaluation function, and the optimal solution wi_opt of the weight coefficient is determined based on the calculated differential evaluation index value. If the value of the differential evaluation index |di'_sim - di'_opt| does not satisfy a predetermined condition, the weight coefficient wi is updated. Then, calculation processes C1 and C2 are executed again using the updated weight coefficient wi. That is, calculation processes C1 and C2 are repeatedly executed until the value of the differential evaluation index |di'_sim - di'_opt| satisfies the predetermined condition. Then, the calculated value di'_sim of the variable in the evaluation function when the value of the difference evaluation index |di'_sim - di'_opt| satisfies a predetermined condition is identified. Then, the weighting coefficient wi in the evaluation function J that is the calculation source of the calculated value di'_sim of the identified variable in the evaluation function is determined as the optimal solution wi_opt of the weighting coefficient. Note that if the value of the difference evaluation index calculated in the first calculation process C2 satisfies the predetermined condition, there is no need to re-execute the calculation processes C1 and C2.

[0022] By performing the calculation processes C1 and C2 as described above, the weight coefficient wi of the evaluation function J can be determined by inverse reinforcement learning. The evaluation index ei constituting the evaluation function J is not particularly limited. For example, the evaluation index ei includes an evaluation index related to the item being processed. In this case, the variable di in the evaluation function that determines the evaluation index ei includes a variable indicating a physical quantity whose value changes as the item is processed. Examples of such physical quantities include at least one of a variable indicating a physical quantity representing the state of a device used to process the item and a variable indicating a physical quantity representing the state of the item. Examples of item processing include transportation (logistics) of items and product manufacturing. Examples of physical quantities representing the state of a device used to process items include a physical quantity representing the state of a device that transports items and a physical quantity representing the state of a product manufacturing or inspection device. An example of a physical quantity representing the state of an item is the temperature of the item. When the variable di in the evaluation function other than the design variable is such a physical quantity, the variable di in the evaluation function may be predicted (estimated) using a machine learning model or a physics model (differential equations representing physical phenomena). However, the evaluation indexes ei that make up the evaluation function J are not limited to evaluation indexes related to the items being processed, and can be any evaluation indexes as long as there are multiple evaluation indexes ei (multiple variables di within the evaluation function) that can construct the evaluation function J. Based on the contents described in this section, an example of a weight coefficient determination device and a weight coefficient determination method for determining the weight coefficients wi of the evaluation function J by inverse reinforcement learning will be described below.

[0023] [Weighting coefficient determination device and weighting coefficient determination method] Fig. 2 is a diagram showing an example of the functional configuration of the weighting coefficient determination device 200. Fig. 3 is a flowchart illustrating an example of a weighting coefficient determination method executed using the weighting coefficient determination device 200. The hardware of the weighting coefficient determination device 200 is realized, for example, by using an information processing device including a central processing unit, a main storage device, an auxiliary storage device, an input device, and an output device. The hardware of the weighting coefficient determination device 200 may also be realized, for example, by dedicated hardware such as an ASIC (Application Specific Integrated Circuit).

[0024] The weighting coefficient determination device 200 is a device that executes a process for determining weighting coefficients for multiple evaluation indexes included in an evaluation function used to solve an optimization problem by inverse reinforcement learning. The weighting coefficient determination device 200 includes a calculation unit 210, a weighting coefficient determination unit 220, an evaluation index determination unit 230, a storage unit 240, and an output unit 250. Note that the storage unit 240 may be provided outside the weighting coefficient determination device 200.

[0025] <<Calculation Unit 210, Weighting Coefficient Determination Unit 220>> The calculation unit 210 calculates a calculated value di_sim (optimal solution) of multiple variables in the evaluation function, including the design variable that is the target of solving the optimization problem, using an algorithm for the optimization problem. The calculated value di_sim of the multiple variables in the evaluation function is the value of the variable di in the evaluation function when the value of the evaluation function J is minimum or maximum. In the following description, the optimization problem for calculating the calculated value di_sim of the variable in the evaluation function when the value of the evaluation function J is minimum or maximum is also referred to as a first optimization problem. An example of the algorithm for the first optimization problem is an algorithm based on a metaheuristics method such as a genetic algorithm. In this embodiment, an example is given of a case where a metaheuristics method such as a genetic algorithm is used as the algorithm for the first optimization problem. However, the algorithm for the first optimization problem may be a known algorithm other than an algorithm based on a metaheuristics method. Note that the metaheuristics method such as a genetic algorithm itself can be realized using known technology, and therefore a detailed description thereof will be omitted.

[0026] The weighting coefficient determination unit 220 calculates a difference evaluation index value that evaluates the difference between a calculated value di'_sim of at least one evaluation function variable among the multiple evaluation function variables di and a preset optimal value di'_opt of the evaluation function variable, and determines the value of the weighting coefficient wi based on the calculated difference evaluation index value. In the example shown in FIG. 1, the difference evaluation index corresponds to |di'_sim - di'_opt|. In this embodiment, the weighting coefficient determination unit 220 determines the value of the weighting coefficient wi using an optimization problem algorithm that uses the sum of the difference evaluation indexes |di'_sim - di'_opt| for each evaluation function variable di' as the evaluation function J_dif. In the following description, this optimization problem is also referred to as a second optimization problem. In addition, to distinguish it from the evaluation function J, the evaluation function J_dif is also referred to as a difference evaluation function J_dif. Weighting coefficient determination unit 220 determines whether or not the calculated value di'_sim of the variable in the evaluation function calculated by calculation unit 210 by varying the value of weighting coefficient wi is the calculated value di'_sim of the variable in the evaluation function when the value of the difference evaluation function J_dif is minimum or maximum. Note that the value of the difference evaluation function J_dif being minimum or maximum means that the value of the difference evaluation function J_dif is maximum or minimum within a range that satisfies the conditions (and constraint equations, if necessary) determined by the algorithm of the second optimization problem (this also applies to the following explanation).

[0027] Then, the weighting coefficient determination unit 220 determines the weighting coefficient wi set for the evaluation function J that is the source of calculation of the calculated value di'_sim of the variable in the evaluation function when the value of the difference evaluation function J_dif is minimum or maximum as the optimal solution wi_opt of the weighting coefficient. In this embodiment, a case where a metaheuristic method such as a genetic algorithm is used as the algorithm for the second optimization problem is exemplified. However, like the algorithm for the first optimization problem, the algorithm for the second optimization problem may also be a known algorithm other than an algorithm using a metaheuristic method. Note that the algorithm for the first optimization problem and the algorithm for the second optimization problem do not have to be the same algorithm.

[0028] <<Evaluation Index Determination Unit 230>> When the absolute value of the optimal solution wi_opt of the weighting coefficient determined by the weighting coefficient determination unit 220 is equal to or less than a positive threshold, the evaluation index determination unit 230 determines that the evaluation index ei multiplied by the optimal solution wi_opt of the weighting coefficient has little influence on the value of the evaluation function J, and determines that the evaluation index ei is an evaluation index ei not to be included in the evaluation function J. On the other hand, when the absolute value of the optimal solution wi_opt of the weighting coefficient determined by the weighting coefficient determination unit 220 is greater than a positive threshold, the evaluation index determination unit 230 determines that the evaluation index ei is an evaluation index ei to be included in the evaluation function J. A value close to 0 (zero) or 0 (zero) is set in advance as the positive threshold. For example, when the absolute value of the optimal solution wi_opt of the weighting coefficient determined by the weighting coefficient determination unit 220 is equal to or less than a positive threshold, the evaluation index determination unit 230 sets (changes) the weighting coefficient to 0 (zero) and thereby determines that the evaluation index ei multiplied by the optimal solution wi_opt of the weighting coefficient is an evaluation index ei not to be included in the evaluation function J. On the other hand, if the absolute value of the optimal solution wi_opt of the weighting coefficient determined by the weighting coefficient determination unit 220 exceeds a positive threshold, the evaluation index determination unit 230 does not change the optimal solution wi_opt of the weighting coefficient, and determines the evaluation index ei multiplied by the optimal solution wi_opt of the weighting coefficient as the evaluation index ei to be included in the evaluation function J.

[0029] <<Storage section 240>> The storage unit 240 stores information that needs to be set in advance before the calculation unit 210 and the weighting coefficient determination unit 220 start processing, as information necessary for the processing by the calculation unit 210, the weighting coefficient determination unit 220, and the evaluation index determination unit 230. The storage unit 240 stores, for example, mathematical expressions representing the evaluation function J and the differential evaluation function J_dif, the optimal value di'_opt of the variable in the evaluation function, and the weighting coefficient wi'_dif in the differential evaluation function J_dif (see equation (3) described later).

[0030] As described above, the optimal value di'_opt of the variable in the evaluation function is determined based on the past performance of the action / movement. The storage unit 240 stores the actual value of the physical quantity, which is obtained when the optimal value di'_opt of the variable in the evaluation function is determined and is used to calculate the evaluation index ei of the evaluation function J, in association with the optimal value di'_opt of the variable in the evaluation function. This actual value of the physical quantity is used when the calculation unit 210 calculates the evaluation function J (evaluation index ei). In this way, the calculation unit 210 can calculate the calculated value di'_sim of the variable in the evaluation function when the value of the difference evaluation index |di'_sim - di'_opt| satisfies a predetermined condition under the same conditions as when the optimal value di'_opt of the variable in the evaluation function was obtained. In addition, the storage unit 240 stores, for example, a threshold value to be compared with the absolute value of the optimum solution wi_opt of the weighting coefficient. The information stored in the storage unit 240 may be acquired, for example, by at least one of an operator operating an input device, receiving from an external device, and reading from a portable storage medium.

[0031] <<Output section 250>> The output unit 250 outputs information indicating the contents determined by the weighting coefficient determination unit 220 and the evaluation index determination unit 230. For example, when the evaluation index determination unit 230 sets (changes) the optimal solution wi_opt of the weighting coefficient determined by the weighting coefficient determination unit 220 to 0 (zero), the output unit 250 outputs information of the optimal solution wi_opt of the weighting coefficient determined by the weighting coefficient determination unit 220 for the optimal solution wi_opt of the weighting coefficient that was not changed by the evaluation index determination unit 230, and outputs information indicating that the weighting coefficient should be set to 0 (zero) for the optimal solution wi_opt of the weighting coefficient that was changed by the evaluation index determination unit 230. Note that the output unit 250 may output other information as long as it is information indicating the contents determined by the weighting coefficient determination unit 220 and the evaluation index determination unit 230. For example, the output unit 250 may output information on the optimal solution wi_opt of the weighting coefficients determined by the weighting coefficient determination unit 220, and information on the evaluation indexes ei determined by the evaluation index determination unit 230 as the evaluation indexes ei to be included in the evaluation function J and the evaluation indexes ei determined as the evaluation indexes ei not to be included in the evaluation function J. Examples of the form of output include at least one of display on a computer display, transmission to an external device, and storage in a storage medium inside or outside the weighting coefficient determination device 200.

[0032] The following describes the weighting coefficient determination method according to the flowchart in Fig. 3. It is assumed that information that needs to be set in advance before the calculation unit 210, weighting coefficient determination unit 220, and evaluation index determination unit 230 start processing is stored in the storage unit 240 before the flowchart in Fig. 3 starts.

[0033] In step S301 of FIG. 3, the weighting coefficient determination unit 220 reads out and acquires the optimal value di′_opt of the variable in the evaluation function from the storage unit 240. Next, in step S302, the weighting coefficient determination unit 220 generates a candidate group (plurality of candidates) for the weighting coefficient wi. The candidates included in the candidate group for the weighting coefficient wi each include I weighting coefficients wi (i = 1 to I). In this embodiment, a metaheuristic method such as a genetic algorithm is used to calculate an optimal solution wi_opt for the weighting coefficients. Therefore, each candidate included in the candidate group for the weighting coefficient wi is generated according to a metaheuristic method such as a genetic algorithm.

[0034] Next, in step S303, the calculation unit 210 calculates, as a calculated value di_sim of the variables in the evaluation function, the variables in the evaluation function including the evaluation function when the value of the evaluation function J in the following equation (1) is minimized, using the initial values ​​of the I weighting coefficients wi generated in step S302 or the I weighting coefficients wi updated in step S306 (described later). The evaluation index ei is expressed, for example, by the following equation (2). Here, di_ref is a target value for the evaluation index di_sim. In this embodiment, the variables in the evaluation function when the value of the evaluation function J in the equation (1) is minimized are calculated according to a metaheuristic method such as a genetic algorithm. Furthermore, an optimization problem may be constructed using a constraint equation in addition to the evaluation function J in the equation (1). In this case, the calculation unit 210 calculates the variables in the evaluation function when the value of the evaluation function J is minimized within a range that satisfies the constraint equation. It is preferable that at least one of the constraint equations is expressed using a variable di in the explanatory variables. For example, if the evaluation function J is calculated by multiplying each term (wi·ei for each of i = 1 to I) on the right side of equation (1) by (-1), the calculation unit 210 will calculate the variables in the evaluation function, including the design variables when the value of the evaluation function J is maximized, as the calculated value di_sim of the variables in the evaluation function.

[0035]

number

[0036] Next, in step S304, the weighting coefficient determination unit 220 performs a differential evaluation of the following equation (3) using the optimal value di'_opt of the variable in the evaluation function acquired in step S301 and the calculated value di'_sim of the variable in the evaluation function identified by the variable i' among the calculated values ​​di_sim of the variable in the evaluation function calculated in step S303: Price The value of the number J_dif is calculated for each candidate group of weighting coefficients wi. Note that the i' under Σ in equation (3) indicates that an accumulation is performed for each variable i' among the variables i.

[0037]

number

[0038] The weighting coefficient wi'_dif in equation (3) is a weighting coefficient for the differential evaluation index |di'_sim - di'_opt|, and is different from the weighting coefficient wi in the evaluation function J. The weighting coefficient wi'_dif can be determined based on the calculation accuracy of the calculated value di_sim of the variable in the evaluation function and the reliability of the optimal value di'_opt of the variable in the evaluation function, and does not need to take into account actual actions and behaviors, as is the case when determining the weighting coefficient wi.

[0039] Next, in step S305, the weighting coefficient determination unit 220 performs a differential evaluation for each candidate of the weighting coefficient wi. Price Based on the calculation result of the value of the number J_dif, it is determined whether or not the convergence condition is satisfied. The convergence condition may be any condition that is used in metaheuristic methods such as genetic algorithms. The convergence condition may be, for example, the difference evaluation value calculated for each candidate included in the candidate group of the weight coefficient wi. Price Even if the minimum value of the number J_dif is equal to or less than a predetermined value, the difference evaluation Price Alternatively, the condition may be that the number of calculations of the value of the number J_dif (the number of repetitive processes) is a predetermined value.

[0040] For example, the difference evaluation is performed by multiplying each term (wi'_dif·|di'_sim-di'_opt|) on the right side of equation (3) by (-1). Price When the number J_dif is used, the convergence condition described above is, for example, the difference evaluation value calculated for each candidate included in the candidate group of the weight coefficient wi. Price The condition used is that the maximum value of the number J_dif is equal to or greater than a predetermined value.

[0041] If the result of the determination in step S305 is that the convergence condition is not satisfied (NO in step S305), the process of step S306 is executed. In step S306, the weighting coefficient determination unit 220 updates the candidate group of weighting coefficients wi. Then, the processes of steps S303 to S305 are executed using the candidates for the updated weighting coefficients wi. In this way, the processes of steps S303 to S306 are repeatedly executed until the convergence condition is satisfied.

[0042] Then, in step S306, if it is determined that the convergence condition is satisfied (YES in step S306), the process of step S307 is executed. In step S307, the weighting coefficient determination unit 220 performs a differential evaluation for the plurality of weighting coefficient wi candidates calculated when the convergence condition is satisfied. Price The difference evaluation that shows the minimum value of the number J_dif Price The weighting coefficient determination unit 220 determines, as the optimal solution wi_opt of the weighting coefficient, the (candidate for) weighting coefficient wi set in the evaluation function J used when the calculation unit 210 calculated the calculated value di'_sim of the variable in the evaluation function as the optimal solution of the first optimization problem.

[0043] Next, in step S308, the evaluation index determination unit 230 determines whether the absolute value of the optimal solution wi_opt of the weighting coefficients determined by the weighting coefficient determination unit 220 is equal to or less than a positive threshold. If the result of this determination is that the absolute value of the optimal solution wi_opt of the weighting coefficients determined by the weighting coefficient determination unit 220 is equal to or less than the positive threshold, the evaluation index determination unit 230 determines the evaluation index ei as an evaluation index ei that is not to be included in the evaluation function J. On the other hand, if the absolute value of the optimal solution wi_opt of the weighting coefficients determined by the weighting coefficient determination unit 220 is greater than the positive threshold, the evaluation index determination unit 230 determines the evaluation index ei as an evaluation index ei that is to be included in the evaluation function J.

[0044] Next, in step S309, the output unit 250 outputs information indicating the details determined by the weighting coefficient determination unit 220 and the evaluation index determination unit 230. When the process of step S309 ends, the process according to the flowchart in FIG.

[0045] [summary] As described above, in this embodiment, the weighting coefficient determination device 200 uses the algorithm for the first optimization problem to calculate the calculated value di'_sim for the variables in the evaluation function that determine the evaluation index ei' included in the evaluation function J when the value of the evaluation function J is maximized or minimized. The weighting coefficient determination device 200 then calculates the value of the difference evaluation index |di'_sim-di'_opt| that evaluates the difference between the calculated calculated value di'_sim of the variable in the evaluation function and a preset optimal value di'_opt of the variable in the evaluation function, and determines the optimal solution wi_opt of the weighting coefficients based on the calculated difference evaluation index value. Therefore, the weighting coefficient wi of the evaluation function J used to solve the first optimization problem can be set to a value appropriate for the first optimization problem.

[0046] In this embodiment, the weighting coefficient determination device 200 uses a difference evaluation function J_dif Therefore, the optimal solution wi_opt for the weighting coefficients can be determined without relying on a trial-and-error method.

[0047] Furthermore, in this embodiment, the weighting coefficient determination device 200 calculates the calculated value di_sim of the variable in the evaluation function using a candidate value of the weighting coefficient wi according to an algorithm for the first optimization problem. Then, the weighting coefficient determination device 200 determines the candidate value of the weighting coefficient wi used in calculating the calculated value di'_sim of the variable in the evaluation function when the value of the difference evaluation index |di'_sim - di'_opt| satisfies a predetermined condition as the optimal solution wi_opt of the weighting coefficient. At this time, the weighting coefficient determination device 200 repeatedly changes the candidate value of the weighting coefficient wi until the value of the difference evaluation index satisfies the predetermined condition. Therefore, it is possible to search for the optimal solution wi_opt of the weighting coefficient using an algorithm for an optimization problem that is not dependent on a specific problem.

[0048] Furthermore, in this embodiment, the storage unit 240 stores the optimal value di'_opt of the variable in the evaluation function and the actual value of the physical quantity that is obtained when the optimal value di'_opt of the variable in the evaluation function was determined and is used to calculate the evaluation index ei, in association with each other. The weighting coefficient determination device 200 then calculates the calculated value di_sim of the variable in the evaluation function using the actual value of the physical quantity that is stored in association with the optimal value di'_opt of the variable in the evaluation function. Therefore, the calculation unit 210 can calculate the calculated value di'_sim of the variable in the evaluation function when the value of the difference evaluation index |di'_sim - di'_opt| satisfies a predetermined condition under the same conditions as when the optimal value di'_opt of the variable in the evaluation function was obtained.

[0049] In this embodiment, the variables di in the evaluation function include variables indicating physical quantities whose values ​​change as the product is processed. Therefore, when solving the first optimization problem to determine indicators of behavior and action when processing the product, the weighting coefficients of the evaluation function can be determined to values ​​appropriate for the optimization problem.

[0050] In this embodiment, the weighting coefficient determination device 200 performs a prediction calculation of the physical quantity of at least one of the variables di' in the evaluation function other than the design variables among the variables di in the evaluation function. Therefore, for example, it is possible to evaluate the physical quantity assumed (predicted) when an article is processed by comparing it with an optimal value determined by an expert or the like.

[0051] Furthermore, in this embodiment, the weighting coefficient determination device 200 determines, based on the value of the optimal solution wi_opt of the weighting coefficient, whether or not to include the evaluation index ei for the optimal solution wi_opt of the weighting coefficient in the evaluation function J, and determines the evaluation index ei to be included in the evaluation function J. Therefore, it is possible to include multiple (many) candidates that can be considered as the evaluation index ei in the evaluation function J, and select an evaluation index ei that is suitable for the evaluation function J from among these evaluation indexes ei.

[0052] [Application example] In the following, an example of control of the input heat quantity during unsteady operation of a coke oven is given, and a case where an evaluation function J is constructed and a calculated value di_sim of a variable in the evaluation function is calculated is illustrated. Here, an example of control of the input heat quantity during unsteady operation of a coke oven is described.

[0053] <Outline of coke ovens and the coke manufacturing process> First, an example of the schematic configuration of a coke oven 1 and an overview of an example of a coke production process will be described with reference to Figs. 4, 5A, and 5B. Fig. 4 is a diagram illustrating an example of a coke oven and a coke production process. Fig. 5A is a diagram illustrating an example of a furnace battery temperature. Fig. 5B is a diagram illustrating an example of a state in which coke is pushed out of a coke chamber. Note that Figs. 5A and 5B show a see-through view of the interior. As shown in Figures 4 and 5A, in a coke oven 1, carbonization chambers (kilns) 2 and combustion chambers 3 are arranged alternately with furnace walls 4 interposed therebetween. The carbonization chambers 2 carbonize charged coal to obtain coke. The combustion chambers 3 burn fuel gas to keep the carbonization chambers 2 at a high temperature.

[0054] As explained in the Background Art section, in the coke production process using a coke oven 1, the so-called block unloading method is used for the unloading coal loading work. The unloading coal loading work is a work in which coke is pushed out of the coke chamber 2 by an extrusion ram 7 as shown in Figure 5B, and then coal is supplied to the coke chamber 2. In the block unloading method, all coke chambers 2 are divided into Da numbers (Da is an integer greater than or equal to 2), and the unloading coal loading work is carried out in units of the divided numbers. Each coke chamber 2 is assigned to one of the numbers so that multiple coke chambers 2 every Da numbers in the arrangement order of the coke chambers 2 belong to the same number. In this section, an example is shown in which the unloading coal loading work is carried out using the block unloading method with Da set to 5. In this case, for example, the carbonization chambers 2 No. 1, 6, 11, 16, etc. are assigned to street 1, the carbonization chambers 2 No. 2, 7, 12, 17, etc. are assigned to street 2, the carbonization chambers 2 No. 3, 8, 13, 18, etc. are assigned to street 3, the carbonization chambers 2 No. 4, 9, 14, 19, etc. are assigned to street 4, and the carbonization chambers 2 No. 5, 10, 15, 20, etc. are assigned to street 5. The carbonization chambers 2 are assigned in order of number, starting with the lowest number, and the carbonization chambers 2 are loaded into the kiln. For example, the unloading and loading work for the coking chambers 2 assigned to Route 1 is carried out in the following order: coking chamber 2 of coking chamber No. 1, coking chamber 2 of coking chamber No. 6, coking chamber 2 of coking chamber No. 11, coking chamber 2 of coking chamber No. 16, etc. To prevent a sudden drop in temperature, the unloading and loading order is, for example, Route 1, Route 3, Route 5, Route 2, Route 4. The time from the end of the unloading and loading work on one route to the end of the unloading and loading work on the next route is referred to as the "running time." The running time is generally approximately 3 to 6 hours. Note that this section is not limited to the block unloading method. For example, if the following explanation treats each route (block) as an individual coking chamber 2, it can also be applied to cases where unloading and loading work is performed on a single coking chamber 2 basis.

[0055] In addition, in the coke oven manufacturing process, furnace battery control is performed to simultaneously adjust the heat input to all combustion chambers 3 and control the average carbonization state for each. Specifically, the heat input to the coke oven 1 is controlled by operating a single regulating valve 5 installed for each combustion chamber 3. The regulating valve 5 adjusts the flow rate of a mixture of fuel gas and combustion air. The regulating valve 5 is operated via an actuator (not shown) under the control of a processing device 600 (described later). The representative temperature of all combustion chambers 3 is referred to as the furnace battery temperature. For example, as shown in FIG. 5A, thermometers 6 are installed in multiple combustion chambers 3 to measure the ambient temperature of each chamber 3. The average temperature of the combustion chambers 3 where the thermometers 6 are installed is referred to as the furnace battery temperature. This section illustrates an example in which the furnace temperature, which is the temperature in the combustion chambers 3 of the coke oven 1, is the furnace battery temperature. Note that the method described in this section is not limited to simultaneously adjusting the heat input to all combustion chambers 3. For example, when carrying out the unloading and loading work for each carbonization chamber 2, an adjustment valve and an actuator may be installed in each combustion chamber 3, and the carbonization state (amount of input heat) may be controlled for each carbonization chamber 2.

[0056] Furthermore, the thermometer 6 may be installed in each of all combustion chambers 3, or in only some of the combustion chambers 3. For example, the thermometer 6 may be installed in all combustion chambers 3, and the temperature of each combustion chamber 3 may be taken as the temperature of that combustion chamber 3 (furnace temperature). As described above, the coke is pushed out of the coke chamber 2 by the pusher ram 7. In the example shown in FIG. 5B, the coke pushed out of the coke chamber 2 by the pusher ram 7 is discharged via the guide car 9 to a fire engine (not shown) located below the guide car 9, and then transported to a downstream process by the fire engine. The guide car 9 moves to the position of the coke chamber 2 where the unloading and loading work is performed. In FIG. 5B, the coke produced in the coke chamber 2 located at the bottom of FIG. 5B is discharged via the guide car 9 to a fire engine (not shown) to complete the unloading and loading work. The guide car 9 moves to the coke chamber 2 located at the top of FIG. 5B, as shown by the dashed-dotted line. In addition, FIG. 5B illustrates a case where a thermometer 8 for non-contact measurement of the coke temperature is installed inside the guide car 9. The thermometer 8 is installed so as to view the coke passage path inside the guide car 9 through a window provided in the guide car 9. In this manner, this embodiment illustrates a case where the temperature of the coke is measured immediately after it leaves the coke chamber 2 during (at the time of extrusion) the coke extrusion operation (unloading operation). Note that the coke temperature does not necessarily have to be measured in this manner as long as the temperature of the coke discharged from the coke chamber 2 is measured. In the following description, the temperature of the coke discharged from the coke chamber 2 is also referred to as the coke temperature.

[0057] <Outline of Processing Device 600> 6 is a diagram showing an example of the functional configuration of the processing device 600. The hardware of the processing device 600 is realized by using, for example, an information processing device including a central processing unit, a main storage device, an auxiliary storage device, an input device, and an output device. The hardware of the processing device 600 may be realized by a PLC (Programmable Logic Controller) or dedicated hardware such as an ASIC (Application Specific Integrated Circuit).

[0058] In this section, the processing device 600 determines the target furnace temperature, which is the target value for the furnace temperature during non-steady operation. As mentioned above, this section illustrates the case where the furnace temperature is the kiln battery temperature. In steady operation, the aforementioned kiln discharge loading operation is repeatedly performed at approximately regular intervals corresponding to the run time. In contrast, in non-steady operation, the kiln discharge loading operation is temporarily suspended due to equipment maintenance, etc. Therefore, during non-steady operation, for example, the kiln battery temperature is temporarily lowered to perform equipment maintenance, etc., and then the kiln battery temperature is raised. Non-steady operation is an operation that includes a period during which the kiln discharge loading operation is temporarily suspended and a period during which at least one subsequent run of the kiln discharge loading operation (loading and pushing) is performed. In this section, non-steady operation is defined as an operation that includes a period during which the kiln discharge loading operation is temporarily suspended and a period during which at least one subsequent run of the kiln discharge loading operation is performed.

[0059] In this section, the start and end of unsteady operation are assumed to be the end of the unloading and loading work (coke extrusion (unloading) work). The end of the coke extrusion work is the timing after all of the coke in the coke chamber 2 has been pushed (discharged) from the coke chamber 2. The end of the coke extrusion work may be, for example, the timing when all of the coke is discharged from the coke chamber 2, the timing when the door of the coke chamber 2 is closed thereafter, the timing when the temperature of the coke discharged from the coke chamber 2 is measured, or the timing when the coke extruded from the coke chamber 2 starts to be transported to the downstream process. It may also be the timing when the coke extrusion work is deemed to end in the operation manual of the coke plant. In the following description, the period when the unloading and loading work is temporarily suspended is also referred to as the suspension period. In addition, if an operational abnormality occurs, the start of unsteady operation may also be the timing when the operational abnormality is detected.

[0060] In FIG. 6, a processing device 600 that executes processing including determining a target oven temperature in a coke oven 1 includes an acquisition unit 610, a predicted value calculation unit 620, a target oven temperature determination unit 630, and a control unit 640. The acquisition unit 610 acquires various data used by the processing device 600. The data acquired by the acquisition unit 610 includes past and present performance values ​​of operations, future operation schedule values, target values ​​for operations, and various setting values ​​used for calculations in the processing device 600. Examples of data acquisition methods include at least one of an operator's operation on an input device, reception from an external device, and reading from a portable storage medium. Note that individual pieces of data are input to the processing device 600 at any timing and acquired by the acquisition unit 610. Therefore, individual pieces of data do not necessarily need to be input to the processing device 600 and acquired by the acquisition unit 610 at the same timing.

[0061] The predicted value calculation unit 620 calculates a predicted value of the furnace battery temperature during unsteady operation based on a first influencing factor, which is a factor that influences the predicted value of the furnace battery temperature during unsteady operation. The predicted value calculation unit 620 also calculates a predicted value of the physical quantity that represents the coke carbonization state based on a second influencing factor, which is a factor that influences the predicted value of the physical quantity that represents the coke carbonization state during unsteady operation.

[0062] The carbonization state of the coke indicates the degree to which the coal in the produced coke has been carbonized (pyrolyzed), and is an index of the quality of the coke. The first influencing factor is not particularly limited as long as it is a factor that affects the predicted value of the furnace battery temperature during unsteady operation. In this section, an example is given in which the first influencing factor includes the predicted value of the heat input to the combustion chamber 3 during unsteady operation and the furnace battery temperature at a timing prior to the predicted value of the furnace battery temperature during unsteady operation. Here, the furnace battery temperature at a timing prior to the predicted time of the predicted value of the furnace battery temperature may be a predicted value or an actually measured value.

[0063] The second influencing factor is not particularly limited as long as it is a factor that affects the predicted value of a physical quantity that represents the carbonization state of the coke during unsteady operation (in this embodiment, the coke temperature at the end of unsteady operation). In this embodiment, an example is shown in which the second influencing factor includes the predicted value of the furnace battery temperature and the carbonization time. The carbonization time is equal to the time required to perform one coal loading operation at the kiln for all Da (=5) combinations.

[0064] Examples of physical quantities that represent the carbonization state of the coke include the temperature of the coke produced during non-steady operation and the temperature of the furnace wall 4 during non-steady operation. The physical quantity that represents the carbonization state of the coke during non-steady operation is used to bring the carbonization state of the coke closer to a target state when the steady state is restored. From this perspective, the physical quantity that represents the carbonization state of the coke during non-steady operation is preferably the one that is closest to the timing of restoration from non-steady operation to the steady state. Therefore, in this section, we will exemplify a case where the physical quantity that represents the carbonization state of the coke during non-steady operation is the coke temperature of the coal that is first charged into the coke chamber 2 after the end of the rest period. Note that, as shown in FIG. 7 described later, in this embodiment, the end time of the coke extrusion operation in the sixth case after the end of the rest period is set to the end of the non-steady operation (the end time t e ), the coke temperature of the coal that is first loaded into the coke chamber 2 after the end of the shutdown period is the coke temperature at the end of the unsteady operation. Therefore, in the following description, the coke temperature of the coal that is first loaded into the coke chamber 2 after the end of the shutdown period is also referred to as the coke temperature at the end of the unsteady operation. In addition, this section illustrates a case where the coke temperature is a representative value for each run. Examples of the representative value include the arithmetic mean, median, mode, and minimum value. As mentioned above, the physical quantity representing the carbonization state of the coke during unsteady operation is not limited to the coke temperature, and may be, for example, the temperature of the oven wall 4 during unsteady operation.

[0065] The coke temperature is the temperature of the coke discharged from the coke chamber 2 and is calculated, for example, from the value measured by the thermometer 8 shown in FIG. 5B. When the extrusion ram 7 extrudes the coke from the coke chamber 2, the temperature of the coke sequentially discharged from the coke chamber 2 is measured by the thermometer 8, and the representative value of the temperatures measured at each time and each position is defined as the temperature of the coke produced in the coke chamber 2. Examples of the representative value include the arithmetic mean value (the sum of the temperatures measured at each time and each position divided by the number of temperature measurements), the median, the mode, and the minimum value. The representative value of the temperatures of the coke produced in the coke chambers 2 belonging to one batch is defined as the coke temperature (batch representative value). As described above, the batch representative value is, for example, the batch average value (the sum of the temperatures of the coke produced in the coke chambers 2 belonging to one batch divided by the number of coke chambers 2 belonging to that batch). Note that, since the coke temperature is preferably the temperature of the coke immediately after being discharged from the coke chamber 2, the coke temperature is determined as shown in FIG. 5B as an example. However, the thermometer for measuring the coke temperature and the method for determining the coke temperature themselves may be, for example, those employed in a coke plant, and are not limited to the above. Note that, even when the physical quantity representing the carbonization state of the coke during unsteady operation is the temperature of the oven wall 4 during unsteady operation, the temperature of the oven wall 4 is set as a representative value (run representative value) of the temperature of the oven wall 4 when coke is produced in the coke chamber 2 belonging to one run. The temperature of the oven wall 4 is measured, for example, by a thermometer (not shown) embedded in the oven wall 4.

[0066] The factors adopted as the first influence factor and the second influence factor are determined, for example, using a known method for selecting explanatory variables (first influence factor, second influence factor) for the dependent variable (predicted value of the furnace battery temperature during unsteady operation, predicted value of the coke temperature at the end of unsteady operation) (for example, calculating the correlation coefficient between the dependent variable and the explanatory variables using training data, or eliminating explanatory variables with multicollinearity). Alternatively, the dependent variable and at least one of the explanatory variables may be the same physical quantity, and the predicted value or actual value at a time before the predicted time of the predicted value of the dependent variable may be used as the explanatory variable (first influence factor, second influence factor). The method itself for calculating the dependent variable from the explanatory variables may be realized, for example, by a known machine learning method such as regression analysis.

[0067] FIG. 7 is a diagram showing an example of the relationship between the coke temperature, the furnace battery temperature, the input heat amount, and the carbonization time. In this section, the coke temperature and carbonization time are representative values ​​for each run, and are obtained when the unloading and loading work for one run is performed. In other words, the coke temperature and carbonization time are obtained in a cycle of the run time. In the graphs of coke temperature and carbonization time shown in Figure 7 (top and bottom graphs), the distance on the time axis between two adjacent plots (●) is the run time t t In Figure 7, at time t e As shown in time t t (t e ) is shown as an example. t is generally a constant time, but may vary.

[0068] As mentioned above, the carbonization time is equal to the time required to perform each of the five (Da = 5) coke unloading and loading operations. Therefore, in the graph of coke temperature and carbonization time shown in Figure 7, the distance between the two ends of the six adjacent plots (●) on the time axis is the carbonization time. In Figure 7, the time t e Dry distillation time t k (t e ) is shown below.

[0069] Furthermore, since the kiln unloading and loading work is not performed during the shutdown period, the coke temperature and carbonization time cannot be obtained (see Figure 7, where there is no plot (●) on the graph of coke temperature and carbonization time during the shutdown period).

[0070] On the other hand, the oven battery temperature and input heat amount are obtained regardless of the coal loading operation (see Figure 7, where plots (●) are added to the graph of oven battery temperature and input heat amount even during idle periods). This section illustrates a case where predicted and actual values ​​of oven battery temperature and input heat amount are obtained in the control cycle of the coke oven 1 (the output cycle of the control signal in the control unit 640, described later), predicted and actual values ​​of coke temperature are obtained in the pass-through cycle, and scheduled and actual values ​​of carbonization time are obtained in the pass-through cycle. The control cycle of the coke oven 1 is also assumed to be 1 hour (hr).

[0071] In Figure 7, time t s is an example of the start time of unsteady operation, and time t e is an example of the end time of unsteady operation. Specifically, in this section, the end time of the coke extrusion work Db times (Db is an integer equal to or greater than 1) before the start of the suspension of the kiln loading work (charging and extrusion into the coke chamber 2) is defined as the start time t s More specifically, this section illustrates the case where Db is 2. Therefore, in FIG. 7, the end time of the coke pushing operation in the second run before the start of the suspension period (the time of the second coke temperature plot counting backward from the start time of the suspension period) is the start time t sWhen the shutdown period is known in advance, such as when equipment maintenance is performed, Db may be 1 or an integer equal to or greater than 1. On the other hand, when the shutdown period is not known in advance, such as when an operational abnormality occurs, Db is preferably 1. In FIG. 7 , for convenience of notation, some plots overlap with the shutdown period in the graph of coke temperature and carbonization time, but these plots are obtained by the kiln unloading and loading work for each run immediately before and after the shutdown period. Therefore, in the above explanation, the run two runs before the start of the shutdown period corresponds to the time of the second coke temperature plot counting backward from the start time of the shutdown period, including the plot overlapping with the start time of the shutdown period.

[0072] In addition, in this section, the end time of the coke extrusion work in the Da+1th case after the end of the suspension of the coal loading work (charging into the coke chamber 2 and extrusion) is the end time of the non-steady operation t e Here, the start of coke extrusion work in a street refers to the start of coke extrusion work in the coke chamber 2 that is the first to be performed among the coke chambers 2 that belong to that street, and the end of coke extrusion work in a street refers to the end of coke extrusion work in the coke chamber 2 that is the last to be performed among the coke chambers 2 that belong to that street. As mentioned above, this section illustrates the case where Da is 5. Therefore, in Figure 7, the end time of the coke extrusion work in the sixth street after the end of the suspension period (the time of the sixth coke temperature plot counting forward from the end time of the suspension period) is the end time t e After the rest period ends, in the first through Da (fifth) passes, coke is produced from the coal present in the coke chamber 2 during the rest period.

[0073] On the other hand, in the Da+1th (6th) case after the end of the quiescent period, coal is charged into the coke chamber 2 after the end of the quiescent period. It is preferable that the carbonization state of the coke charged into the coke chamber 2 after the end of the quiescent period approaches the carbonization state in the steady state as quickly as possible. Therefore, in this section, the end time of the coke extrusion work in the Da+1th (6th) case after the end of the quiescent period of the coal loading work (charging and extruding into the coke chamber 2) is defined as the end time t e That is, the end time of the unsteady operation t e is the end time of the coke extrusion operation on the first run of unloading coal loading operation after the end of the shutdown period. However, the end time of unsteady operation is not limited to the end time of the coke extrusion operation on the Da+1-th run after the end of the shutdown of the unloading coal loading operation (loading coal into the coke chamber 2 and extrusion). For example, the end time of unsteady operation may be the end time of the coke extrusion operation on the Da+x-th run after the end of the shutdown of the unloading coal loading operation (loading coal into the coke chamber 2 and extrusion), and the value of x may be selected from integers equal to or greater than 1. The values ​​of x and Db may be adjusted as appropriate, for example, as described below, so that coke of the desired quality is obtained as a result of actually controlling the input heat amount in accordance with the deviation of the actual value of the oven battery temperature from the target oven temperature trajectory. As described above, the period of non-steady operation (non-steady operation) is from time t s ~t e This will be the period.

[0074] Returning to the explanation of FIG. 6, in this section, the predicted value calculation unit 620 has a furnace state calculation unit 621 and an input heat amount calculation unit 622. As described above, in this column, the first influencing factor includes a predicted value of the heat input to the combustion chamber 3. Therefore, the input heat amount calculation unit 622 calculates a predicted value of the heat input to the combustion chamber 3 during unsteady operation. The furnace state calculation unit 621 calculates a predicted value of the furnace temperature during unsteady operation based on the first influencing factor described above, and calculates a predicted value of the physical quantity representing the coke carbonization state during unsteady operation based on the second influencing factor described above. In this column, the furnace temperature is the furnace battery temperature, and the physical quantity representing the coke carbonization state is the coke temperature.

[0075] The target furnace temperature determination unit 630 determines a target furnace temperature trajectory, which is the time change of the target furnace temperature, which is the target value of the furnace temperature during unsteady operation, based on the predicted value of the furnace temperature during unsteady operation calculated by the predicted value calculation unit 620 and the predicted value of the physical quantity representing the carbonization state of the coke during unsteady operation calculated by the predicted value calculation unit 620. In this section, the target furnace temperature is the target furnace battery temperature, which is the target value of the furnace battery temperature. The control unit 640 generates and outputs a control signal to set the amount of heat input to the combustion chamber 3 to a heat amount corresponding to the difference between the target furnace temperature trajectory determined by the target furnace temperature determination unit 630 and the actual value of the furnace temperature.

[0076] Here, we will explain specific examples of the processing in the furnace state calculation unit 621, input heat amount calculation unit 622, target furnace temperature determination unit 630, and control unit 640. In the following explanation, as necessary, the input heat amount to the combustion chamber 3 will be abbreviated as input heat amount, the coke temperature at the end of unsteady operation will be abbreviated as coke temperature, and the furnace battery temperature during unsteady operation will be abbreviated as furnace battery temperature.

[0077] <Reactor State Calculation Unit 621> First, a specific example of the processing in the reactor state calculation unit 621 will be described. In this section, the furnace state calculation unit 621 calculates the predicted value of the furnace battery temperature by performing calculations using a linear time series model as the process model, with the input heat amount (GJ / hr) as input. The linear time series model is one of the statistical analysis models that calculates the predicted value of the process state at predetermined time intervals. In this section, the linear time series model is used as a second-order lag model, and the regression equations shown in the following equations (4) to (6) are exemplified.

[0078]

number

[0079] Here, Tr is the furnace battery temperature (°C). Q is the input heat quantity (GJ / hr). ΔTr(t+1), ΔTr(t), and ΔTr(t-1) are the change (°C) in the furnace battery temperatures Tr(t+1), Tr(t), and Tr(t-1) at times t+1, t, and t-1, respectively, relative to the furnace battery temperatures Tr(t), Tr(t-1), and Tr(t-2) at times t, t-1, and t-2. That is, the relationships ΔTr(t+1) = Tr(t+1) - Tr(t), ΔTr(t) = Tr(t) - Tr(t-1), and ΔTr(t-1) = Tr(t-1) - Tr(t-2) hold. ΔQ(t+1) is the change (GJ / hr) in the input heat quantity Q(t+1) at time t+1 relative to the input heat quantity at time t. That is, the relationship ΔQ(t+1) = Q(t+1) - Q(t) holds. As mentioned above, t+1, t-1, and t-2 are the times one hour after, one hour before, and two hours before time t, respectively.

[0080] Q(t) is the input heat quantity calculated by the input heat quantity calculation unit 622. ΔQ(t+1) is calculated based on the input heat quantity calculated by the input heat quantity calculation unit 622. When time t is the start time t of the unsteady operation, s , Tr(t), Tr(t-1), and Tr(t-2) are the time series of the s , t s -1, t s ΔTr(t) and ΔTr(t-1) are calculated based on these actual values. Time t is the start time of unsteady operation t s If the time is later than t, ΔTr(t) and ΔTr(t-1) are calculated based on the calculation results (ΔTr(t+1)) of equations (4) to (6) at a time before the time t. s A time later than t(>t s ) is the predicted value of the furnace battery temperature.

[0081] The coefficients a1, a2, and b1 are coefficients for ΔTr(t), ΔTr(t-1), and ΔQ(t+1), respectively. The coefficients a1, a2, and b1 are separately determined so that the form of equation (4) best fits the past operating results of the coke oven 1. For example, a set of data for ΔTr(t+1), ΔTr(t), ΔTr(t-1), and ΔQ(t+1) obtained from the past operating results of the coke oven 1 can be used as training data to create a large amount of training data, and the coefficients a1, a2, and b1 can be determined by performing multiple regression analysis using the training data.

[0082] In this section, the furnace state calculation unit 621 calculates the predicted value of the coke temperature by performing calculations using a linear model as a process model (physical model) with the furnace battery temperature (°C) and carbonization time (hr) as input. The linear model is one of the statistical analysis models for calculating the predicted value of the process state, and in this section, the multiple regression equations shown in the following equations (7) to (12) are exemplified as the linear model.

[0083]

number

[0084] where Tc is the coke temperature (°C) and t k is the carbonization time (hr). Tc(t s ) is the start time of the unsteady operation t s is the actual value of the coke temperature at Tc(t e ) is the end time of the unsteady operation t e Therefore, as shown in equation (8), ΔTc is the predicted value of the coke temperature at the start time t s From the end time of the unsteady operation t e The change in coke temperature during the period from Tr(t s ) is the start time of the unsteady operation t s This is the actual value of the furnace temperature at Tr(t e -Δt1), Tr(t e -Δt2), Tr(t e-Δt3) are the end times of the unsteady operation t e Δt1 hours before, Δt2 hours before, Δt3 hours before time t e -Δt1, t e -Δt2, t e -Δt3 is the predicted value of the furnace temperature. Tr(t e -Δt1), Tr(t e -Δt2), Tr(t e -Δt3) is calculated by equations (4) to (6). Here, Δt1, Δt2, and Δt3 are Δt1≧0, Δt1<Δt2<Δt3, and t e -Δt3>t s The relationship is set so that t k (t s ) is the start time of the unsteady operation t s is the actual value of the dry distillation time at t k (t e ) is the end time of the unsteady operation t e The coefficients c1, c2, c3, and d1 are ΔTr1, ΔTr2, ΔTr3, and Δt k The coefficients c1, c2, c3, and d1 are separately calculated to find the coefficients that best fit the form of equation (7) to the past operation results of coke oven 1. For example, a set of ΔTr1, ΔTr2, ΔTr3, and Δt k The data above can be used as one training data to create a large number of training data, and the coefficients c1, c2, c3, and d1 can be found by performing multiple regression analysis using the training data.

[0085] <Input heat amount calculation unit 622> The input heat quantity calculation unit 622 calculates a predicted value of the input heat quantity during unsteady operation based on the candidate target furnace temperature trajectory and the predicted value of the furnace battery temperature calculated by the furnace state calculation unit 621. In this section, the case where the candidate target furnace temperature trajectory is output from the target furnace temperature determination unit 630 is illustrated as an example. The input heat quantity calculation unit 622 calculates a predicted value Q(t+1) one hour after time t as the predicted value of the input heat quantity that reduces the difference (preferably becomes 0 (zero)) between the candidate target furnace temperature trajectory and the predicted value of the furnace battery temperature calculated by the furnace state calculation unit 621. The method for calculating such a predicted value of the input heat quantity itself can be realized by known technology. For example, the input heat quantity calculation unit 622 uses a control simulator that computer-simulates PID control to calculate a predicted value of the input heat quantity according to the deviation of the predicted value of the furnace battery temperature calculated by the furnace state calculation unit 621 from the candidate target furnace temperature trajectory by PID control. Moreover, other control such as PI control may be used instead of PID control.

[0086] Alternatively, the input heat quantity calculation unit 622 may calculate the predicted value Q(t+1) of the input heat quantity using the following equations (13) and (14).

[0087]

number

[0088] Here, Tr_ref(t+m) is a candidate for the target furnace temperature trajectory (target value of the furnace battery temperature) at time t+m, and this section illustrates the case where it is provided by the target furnace temperature determination unit 630 (see the arrow line from the target furnace temperature determination unit 630 to the predicted value calculation unit 620 in Figure 6). Tr(t+m) is the predicted value of the furnace battery temperature at time t+m, and in this section it is calculated by the furnace state calculation unit 621. Therefore, Tr_err, which is the value obtained by subtracting Tr(t+m) from Tr_ref(t+m), is the prediction error of the furnace battery temperature at time t+m. Q(t+1)_old is the input heat amount at time t+1 before the update, and Q(t+1)_new is the input heat amount at time t+1 after the update. Furthermore, G in equation (14) is a predetermined gain by which Tr_err is multiplied. m is a positive integer of 2 or greater, and is set appropriately depending on how far ahead in time from time t the prediction error of the furnace battery temperature is to be calculated.

[0089] When performing the calculations of equations (13) and (14), the input heat quantity calculation unit 622 outputs the candidate initial value of the input heat quantity Q(t+1) at time t+1 to the furnace state calculation unit 621. The candidate initial value of the input heat quantity Q(t+1) at time t+1 may be determined in any manner. For example, the input heat quantity calculation unit 622 may determine the candidate initial value of the input heat quantity Q(t+1) at time t+1 using a random number or a preset value. The furnace state calculation unit 621 uses the candidate initial value of the input heat quantity Q(t+1) at time t+1 to calculate the predicted value of the furnace battery temperature as described in the section <Furnace state calculation unit 621>. In this case, the furnace state calculation unit 621 calculates at least the predicted value of the furnace battery temperature for each hour from time t to t+m. The input heat calculation unit 622 calculates the prediction error Tr_err of the furnace battery temperature at time t+m using equation (13) based on the predicted value of the furnace battery temperature at time t+m calculated by the furnace state calculation unit 621 and a candidate for the target furnace temperature trajectory (target value of the furnace battery temperature) at time t+m.

[0090] Then, the input heat quantity calculation unit 622 determines whether the prediction error Tr_err of the furnace battery temperature at time t+m is less than or equal to a predetermined value. The smaller the prediction error Tr_err of the furnace battery temperature at time t+m, the better. Therefore, for example, 0 (zero) or a value close to 0 is set as the predetermined value. If the prediction error Tr_err of the furnace battery temperature at time t+m is not less than or equal to the predetermined value, the input heat quantity calculation unit 622 assigns the initial value of the candidate for the input heat quantity Q(t+1) at time t+1 to the first term (Q(t+1)_old) on the right side of equation (14), and assigns the prediction error Tr_err of the furnace battery temperature at time t+m to the second term on the right side of equation (14), and calculates the updated input heat quantity Q(t+1)_new at time t+1 using equation (14). The input heat quantity calculation unit 622 sets the updated input heat quantity Q(t+1)_new at time t+1 calculated in this way as a new candidate for the input heat quantity Q(t+1) at time t+1. Then, using the new candidate for the input heat quantity Q(t+1) at time t+1, the following steps are repeatedly performed: output of the candidate for the input heat quantity Q(t+1) at time t+1 to the furnace state calculation unit 621; calculation of the predicted value of the furnace battery temperature by the furnace state calculation unit 621; calculation of the prediction error Tr_err of the furnace battery temperature at time t+m; and calculation of the updated input heat quantity Q(t+1)_new at time t+1, until the prediction error Tr_err of the furnace battery temperature at time t+m becomes equal to or less than a predetermined value.

[0091] If the prediction error Tr_err of the furnace battery temperature at time t+m is less than or equal to a predetermined value, the input heat amount calculation unit 622 determines the updated input heat amount Q(t+1)_new at time t+1 used when calculating the prediction error Tr_err of the furnace battery temperature at time t+m as the predicted value of the input heat amount Q(t+1) at time t+1. The furnace state calculation unit 621 calculates the predicted value of the furnace battery temperature Tr(t+1) at time t+1 using the predicted value of the input heat amount Q(t+1) at time t+1 determined by the input heat amount calculation unit 622, and determines the calculated predicted value as the predicted value of the furnace battery temperature Tr(t+1) at time t+1.

[0092] The predicted value calculation unit 620 outputs the predicted values ​​of the input heat quantity Q(t+1) and the furnace battery temperature Tr(t+1) determined as described above to the target furnace temperature determination unit 630 (see the arrow line (furnace battery temperature, input heat quantity) from the predicted value calculation unit 620 to the target furnace temperature determination unit 630 in Figure 6). Then, time t is time t s +1~t e The predicted values ​​of the input heat quantity Q(t+1) and the furnace temperature Tr(t+1) are determined in the same manner as described above, and the time t s +1~t e The predicted values ​​of input heat Q(t+1) and furnace temperature Tr(t+1) are determined for each hour up to the end of unsteady operation t e Furnace temperature Tr(t e ) is determined, the furnace state calculation unit 621 calculates the coke temperature Tc(t e The predicted value calculation unit 620 calculates and determines the predicted value of the coke temperature Tc(t e ) to the target furnace temperature determination unit 630 (see the arrow (coke temperature) from the predicted value calculation unit 620 to the target furnace temperature determination unit 630 in FIG. 6).

[0093] The number of candidates for the target reactor temperature trajectory Tr_ref(t+m) at time t+m may be one or more, but in the section <Target reactor temperature determination unit 630> described later, a case where there are multiple candidates for the target reactor temperature trajectory Tr_ref(t+m) at time t+m will be exemplified. When there are multiple candidates for the target reactor temperature trajectory Tr_ref(t+m) at time t+m, the predicted value calculation unit 620 calculates the target reactor temperature trajectory Tr_ref(t+m) at time t for each of the multiple candidates. s +1~t e The predicted values ​​of the input heat quantity Q(t+1) and the furnace battery temperature Tr(t+1) at each time every hour are determined in the above manner.

[0094] <Target Furnace Temperature Determination Unit 630> The target furnace temperature determination unit 630 determines the target furnace temperature trajectory based on the predicted values ​​of the input heat quantity, the predicted value of the furnace battery temperature, and the predicted value of the coke temperature calculated (confirmed) by the predicted value calculation unit 620.

[0095] FIG. 8 is a diagram illustrating an example of the target furnace temperature trajectory. The graph shown at the top of FIG. 8 shows an example of the time change in coke temperature, and the graph shown at the bottom shows an example of the time change in furnace battery temperature. In FIG. 8, the positions (times) on the time axis indicated by white circles are the times at which the coke temperature and furnace battery temperature are obtained. For convenience of notation, in FIG. 8, the start time t s From the end time t e Although open circles are shown only for the period up to the time of the coke temperature and the furnace battery temperature, the coke temperature and the furnace battery temperature can also be obtained outside this period.

[0096] In the graph shown at the top of Figure 8, Tc(t e ) is the coke temperature Tc(t e ) is a predicted value of the coke temperature at the end of unsteady operation, and is set according to the required quality of the coke. In the following description, this target value Tc_sv is also referred to as the target coke temperature.

[0097] In the graph shown at the bottom of Figure 8, Tr(t s ) is the furnace battery temperature Tr(t s) is the actual value. ΔTr1 is the change in furnace battery temperature (°C) from the furnace battery temperature in the steady state immediately before the start of unsteady operation (= furnace battery temperature at the start of unsteady operation) to the lowest furnace battery temperature during unsteady operation. In the following explanation, ΔTr1 is also referred to as the furnace temperature change at the start of unsteady operation. ΔTr2 is the change in furnace battery temperature (°C) from the lowest furnace battery temperature during unsteady operation to the furnace battery temperature in the steady state immediately after the end of unsteady operation (= furnace battery temperature at the end of unsteady operation). In the following explanation, ΔTr2 is also referred to as the furnace temperature change at the end of unsteady operation. time1 is the time (hr) required for the furnace battery temperature to change from the furnace battery temperature in the steady state immediately before the start of unsteady operation to the lowest furnace battery temperature during unsteady operation. In the following explanation, this time time1 is also referred to as the time to reach the minimum furnace temperature time1. Time2 is the time (hr) required for the furnace battery temperature to change from the lowest furnace battery temperature during unsteady operation to the furnace battery temperature in the steady state immediately after the end of unsteady operation. In the following explanation, this time time2 is also referred to as the minimum furnace temperature maintenance time2. Time0 is the duration (hr) of unsteady operation. The start time t of unsteady operation is s From the end time t e The graph shown by the solid line to the target reactor temperature trajectory Tr_ref.

[0098] This section illustrates an example in which the target furnace temperature determination unit 630 calculates an optimal solution for the target furnace temperature trajectory Tr_ref as shown in the bottom of Fig. 8 by solving an optimization problem such as a combinatorial optimization problem, and determines the target furnace temperature trajectory Tr_ref. In this case, the furnace temperature change amount ΔTr1 at the start of the unsteady state, the furnace temperature change amount ΔTr2 at the end of the unsteady state, the time time1 to reach the minimum furnace temperature, and the time time2 to maintain the minimum furnace temperature are design variables (variables to be solved).

[0099] First, the target reactor temperature determination unit 630 generates a group of candidates (plurality of candidates) for the target reactor temperature trajectory Tr_ref and outputs them to the prediction value calculation unit 620. In this section, an example is shown in which an optimal solution for the target reactor temperature trajectory Tr_ref is calculated using a metaheuristics method such as a genetic algorithm. Therefore, each candidate included in the group of candidates for the target reactor temperature trajectory Tr_ref is generated according to a metaheuristics method such as a genetic algorithm. Since the metaheuristics method such as a genetic algorithm itself is realized by a known technique, a detailed description thereof will be omitted.

[0100] The predicted value calculation unit 620 calculates the predicted value of the input heat quantity Q(t), the predicted value of the furnace battery temperature Tr(t), and the predicted value of the coke temperature Tc(t) as described in the sections <Oven state calculation unit 621> and <Input heat quantity calculation unit 622>. e The predicted values ​​of the input heat quantity Q(t) and the furnace battery temperature Tr(t) are calculated (confirmed) at the start time t s From the end time of the unsteady operation t e The value is the hourly value up to the coke temperature Tc(t e ) is the value at the end of the unsteady operation.

[0101] The target furnace temperature determination unit 630 determines the candidate for the target furnace temperature trajectory Tr_ref output to the predicted value calculation unit 620, and the predicted values ​​of the input heat quantity Q(t), the predicted value of the furnace battery temperature Tr(t), and the coke temperature Tc(t) calculated (confirmed) by the predicted value calculation unit 620 for the candidate for the target furnace temperature trajectory Tr_ref. e ) and the value of the evaluation function J (fitness function) of the following equation (15) when the constraint equation of the following equation (16) is satisfied is calculated using the predicted value of

[0102]

number

[0103] The integration range of t in the first and third terms on the right side of equation (15) is t s From t e (t s ≦t≦te ) The first term on the right side of equation (15) is an example of a first evaluation index that evaluates the difference between the candidate target furnace temperature trajectory Tr_ref and the predicted value of the furnace battery temperature Tr(t). The second term on the right side of equation (15) is the difference between the target coke temperature Tc_sv, which is an example of a target physical quantity, and the coke temperature Tc(t), which is an example of a predicted value of a physical quantity that represents the carbonization state of the coke. e ) and the predicted value of w(t). The third term on the right side of equation (15) is an example of a third evaluation index for evaluating the input heat quantity Q(t). w1, w2, and w3 are weighting coefficients w1, w2, and w3, which are determined by the weighting coefficient determination device 200 (this will be described later in the section "Determination of Weighting Coefficient wi in an Application Example"). In the example shown in equation (15), the smaller the value of each term on the right side of equation (15), the higher the evaluation by the respective evaluation index. Therefore, the closer the value of the evaluation function J is to 0, the better. In other words, the target furnace temperature determination unit 630 searches for the design variables that minimize the value of the evaluation function J (within a range greater than or equal to 0) within the range that satisfies the constraint equation in equation (16) as the optimal solution. As mentioned above, in this section, the design variables are the furnace temperature change amount ΔTr1 at the start of the unsteady state, the furnace temperature change amount ΔTr2 at the end of the unsteady state, the time to reach the minimum furnace temperature, time1, and the time to maintain the minimum furnace temperature, time2 (see FIG. 8). Alternatively, the design variables that maximize the value of the evaluation function may be searched for as the optimal solution. In this case, for example, the evaluation function is obtained by multiplying each term on the right side of equation (15) by (-1). The target furnace temperature determination unit 630 searches for the design variables that maximize the value of the evaluation function as the optimal solution within the range that satisfies the constraint equation (16).

[0104] If the generated candidate for the target reactor temperature trajectory Tr_ref does not satisfy the constraint equation (16), the target reactor temperature determination unit 630 discards the candidate for the target reactor temperature trajectory Tr_ref without outputting it to the predicted value calculation unit 620.

[0105] The target reactor temperature determination unit 630 determines whether a convergence condition is satisfied based on the calculation result of the value of the evaluation function J for each candidate included in the candidate group of the target reactor temperature trajectory Tr_ref. The convergence condition may be any condition used in a metaheuristic method such as a genetic algorithm. The convergence condition may be, for example, a condition that the minimum value of the values ​​of the evaluation function J calculated for each candidate included in the candidate group of the target reactor temperature trajectory Tr_ref is equal to or less than a predetermined value, or a condition that the number of calculations of the value of the evaluation function J (the number of iterations) is a predetermined value.

[0106] If the convergence condition is not satisfied, the target reactor temperature determination unit 630 regenerates and updates the group of candidates for the target reactor temperature trajectory Tr_ref. Then, the target reactor temperature determination unit 630 repeatedly calculates the value of the evaluation function J, determines whether the convergence condition is satisfied, and regenerates the group of candidates for the target reactor temperature trajectory Tr_ref, as described above, until the convergence condition is satisfied.

[0107] The target furnace temperature determination unit 630 determines the target furnace temperature trajectory Tr_ref, which is determined by the design variables (furnace temperature change amount ΔTr1 at the start of the unsteady state, furnace temperature change amount ΔTr2 at the end of the unsteady state, time time1 to reach the minimum furnace temperature, and time time2 to maintain the minimum furnace temperature) when the value of the evaluation function J is the smallest among the values ​​of the evaluation function J for multiple candidate target furnace temperature trajectories Tr_ref calculated when the convergence conditions are satisfied, as the optimal solution of the target furnace temperature trajectory Tr_ref.

[0108] <Control unit 640> The control unit 640 calculates the amount of heat input to the combustion chamber 3 at the time one hour after the current time using the same method as the method used by the input heat amount calculation unit 622 to calculate the predicted value of the input heat amount. For example, when the input heat amount calculation unit 622 calculates the predicted value of the input heat amount at the time one hour after the current time using a control simulator that computer-simulates PID control, the control unit 640 has a PID controller. The control unit 640 uses the PID controller to calculate the value of the target furnace temperature trajectory Tr_ref determined by the target furnace temperature determination unit 630 at the time one hour after the current time. The control unit 640 then generates and outputs a control signal to set the amount of heat input to the combustion chamber 3 to a heat amount corresponding to the difference between the calculated amount of heat input to the combustion chamber 3 at the time one hour after the current time and the value of the target furnace temperature trajectory Tr_ref determined by the target furnace temperature determination unit 630 at the time one hour after the current time. An example of an output destination of the control signal is a control device for an actuator that operates the adjustment valve 5. The control device instructs the actuator to operate the regulating valve 5 so that the opening degree corresponds to the control signal.

[0109] Furthermore, when the input heat amount calculation unit 622 calculates the predicted value Q(t+1) of the input heat amount using equations (13) and (14), for example, the control unit 640 may calculate the predicted value Q(t+1) of the input heat amount using equations (13) and (14), and generate and output a control signal for adjusting the input heat amount to the combustion chamber 3 to the predicted value Q(t+1) of the input heat amount. In this case, it is sufficient to calculate the predicted value Q(t+1) of the input heat amount up to the time one hour after the current time. That is, in this case, in the description of the <input heat amount calculation unit 622> section, time t is replaced with time t s +1~t e There is no need to repeatedly execute the process of shifting the time to a later time by one hour up to -1.

[0110] <When operating conditions change during non-steady operation> The above processing in the furnace state calculation unit 621, input heat amount calculation unit 622, and target furnace temperature determination unit 630 is performed from the start time t sWhen the schedule value of the carbonization time is changed, the target furnace temperature trajectory is determined, and the target furnace temperature trajectory is then determined. After that, during the unsteady operation, the operating conditions that affect the determination of the target furnace temperature trajectory may be changed. In such a case, it is preferable to re-determine the target furnace temperature trajectory from the timing when the operating conditions are changed. An example of such an operating condition is the scheduled value of the carbonization time. When the scheduled value of the carbonization time is changed, the end time t e Therefore, in the following, the schedule value of the carbonization time is changed at the start time t s An example of a method for redetermining the target furnace temperature trajectory when operating conditions are changed during unsteady operation will be described, taking as an example a case where the scheduled value of the coke carbonization time in a coke chamber 2 belonging to a certain street is changed when, for example, the unloading coal loading work in a coke chamber 2 belonging to a street where the unloading coal loading work is performed earlier than the street in question is delayed or accelerated.

[0111] Figure 9A shows the start time of the unsteady operation, t s FIG. 9B is a diagram illustrating an example of a target furnace temperature trajectory determined at the timing when the operating conditions are changed. In FIG. 9A and FIG. 9B, Tr_ref_old is the start time t s 9A and 9B, Tr_mes indicates the actual furnace temperature trajectory, which is the actual value of the furnace battery temperature Tr(t). In FIG. 9A, the target furnace temperature trajectory is determined at the start time t s In Figure 9A, the design variables corresponding to the target furnace temperature trajectory Tr_ref_old (furnace temperature change amount at the start of unsteady state ΔTr1, furnace temperature change amount at the end of unsteady state ΔTr2, time to reach minimum furnace temperature time1, and time to maintain minimum furnace temperature time2) are denoted as ΔTr1_old, ΔTr2_old, time1_old, and time2_old.

[0112] If the operating conditions that affect the determination of the target furnace temperature trajectory are changed during unsteady operation, the start time of unsteady operation t sFrom the end time of the unsteady operation t e The time from the time the operating conditions are changed to the end time t e During the period up to , the predicted value calculation unit 620 (furnace state calculation unit 621 and input heat amount calculation unit 622) predicts the input heat amount Q(t), the predicted value of the furnace battery temperature Tr(t), and the predicted value of the coke temperature Tc(t e ) and the target furnace temperature determination unit 630 determines the target furnace temperature trajectory Tr_ref. In this case, among the design variables, those that cannot be changed at the time when the operating conditions are changed are set to values ​​on the (latest) target furnace temperature trajectory Tr_ref already determined by the target furnace temperature determination unit 630. Therefore, when the target furnace temperature determination unit 630 calculates the evaluation function J, in addition to the constraint equation shown in equation (16), a constraint equation is added that indicates a constraint condition that the design variables that cannot be changed at the time when the operating conditions are changed are values ​​on the (latest) target furnace temperature trajectory Tr_ref already determined by the target furnace temperature determination unit 630.

[0113] In addition, the start time of the unsteady operation t s For the input heat quantity and furnace battery temperature during the period from the time when the operation conditions are changed to the time when the operation conditions are changed, actual measured values ​​are used instead of predicted values. Therefore, when the target furnace temperature determination unit 630 calculates the evaluation function J, in addition to the constraint equation shown in equation (16), the start time t s A constraint equation is added that indicates the constraint that the input heat quantity Q(t) and furnace battery temperature Tr(t) during the period from the time when the operating conditions are changed to the time when the operating conditions are changed are actual measured values ​​(not predicted values).

[0114] The changed operating conditions are used in the processing of the furnace state calculation unit 621, the input heat amount calculation unit 622, and the target furnace temperature determination unit 630. The end time t e Dry distillation time t k (t e ) schedule value is changed, t in Eq. (12) k (t e ) is changed. Also, the period of unsteady operation, time0, is changed.

[0115] In FIG. 9B, time t m In this example, the operating conditions are changed at time t m The actual value of the furnace battery temperature Tr(t) (actual furnace temperature trajectory Tr_mes) is obtained up to the time t when the operating conditions are changed. m In this case, the time t when the operating conditions were changed among the design variables m The design variables that cannot be changed in this calculation are the furnace temperature change amount ΔTr1 at the start of the unsteady state and the time to reach the minimum furnace temperature time1. Therefore, when the target furnace temperature determination unit 630 calculates the evaluation function J, in addition to the constraint equation shown in equation (16), the furnace temperature change amount ΔTr1 at the start of the unsteady state and the time to reach the minimum furnace temperature time1 are set to the start time t s A constraint equation (ΔTr1=ΔTr1_old, time1=time1_old) is added in which the furnace temperature change amount ΔTr1_old at the start of the unsteady state and the time to reach the minimum furnace temperature time1_old on the target furnace temperature trajectory Tr_ref_old determined by is defined as the furnace temperature change amount ΔTr1_old at the start of the unsteady state and the time to reach the minimum furnace temperature time1_old. In addition, the unsteady operation period time0 in equation (16) is changed from the period before change time0_old shown in FIG. 9A to the period after change time0_new shown in FIG. 9B. Note that the change in the scheduled value of the carbonization time is recognized, for example, during the coal loading operation before unloading from the kiln (coke extrusion operation).

[0116] After the above changes are made, the processing in the furnace state calculation unit 621, the input heat amount calculation unit 622, and the target furnace temperature determination unit 630 is executed, and the time t when the operating conditions are changed is calculated. m In Figure 9B, the design variables that can be changed at time t m The design variables that can be changed in the process (the furnace temperature change amount at the end of the unsteady state ΔTr2, and the minimum furnace temperature maintenance time time2) are denoted as ΔTr2_new and time2_new. Also, in FIG. 9B, the time t mThe part at the time after this is changed to the target furnace temperature trajectory Tr_ref_new determined by the furnace temperature change amount ΔTr2_new at the end of the unsteady state and the minimum furnace temperature maintenance time time2_new. Note that, for comparison with the target furnace temperature trajectory Tr_ref_new, in FIG. 9B, the time t m The parts at subsequent times are also shown.

[0117] <Flowchart> Next, an example of a processing method using the processing device 600 will be described with reference to the flowchart of FIG.

[0118] First, in step S1001, the acquisition unit 610 acquires data to be used in the processing device 600. In this section, the acquisition unit 610 acquires the target coke temperature Tc_sv (°C), which is the target value of the coke temperature at the end of the unsteady operation (see FIG. 8). The acquisition unit 610 also acquires the pass time t t and the number of paths Da (items). The path time t t is the scheduled or actual value. The acquisition unit 610 also acquires the start time t s Dry distillation time t k (t s ) actual value (hr) and the end time of the unsteady operation t e Dry distillation time t k (t e ) schedule value (hr) and get.

[0119] The acquisition unit 610 also acquires the start time t s and the end time of the unsteady operation t e (See Figure 8.) The start time of the unsteady operation, t s is the actual value. The end time of the unsteady operation t e is the scheduled time. The acquisition unit 610 acquires the start time t s At the start time of the unsteady operation t e The carbonization time (t e) is added to the schedule value and the value is set at the end time t e Instead of doing so, the acquisition unit 610 may calculate the start time t s and the street time t t The sum of the two values ​​is calculated at the end of the unsteady operation t e The acquisition unit 610 may calculate the start time t s and the end time of the unsteady operation t e Based on these, the period of unsteady operation, time0 (hr), is calculated (see FIG. 8).

[0120] The acquisition unit 610 also acquires the start time t s Furnace temperature Tr(t s ) and obtain the actual temperature (℃) (see Figure 8). In addition to the aforementioned actual values, the acquisition unit 610 acquires various actual values ​​(operational performance data) of the coke oven 1, such as the oven battery temperature Tr(t), the coke temperature Tc(t) at the end of the coke pushing operation, and the amount of heat Q(t) input to the combustion chamber 3. As described above with reference to Fig. 7, the oven battery temperature Tr and the amount of heat Q input to the combustion chamber 3 are acquired at the control cycle of the coke oven 1. In addition, the coke temperature Tc at the end of the coke pushing operation is acquired at the pass time cycle.

[0121] The acquisition unit 610 also acquires the weighting coefficients w1, w2, w3, coefficients a1 to a2, b1, c1 to c3, d1, times Δt1, Δt2, Δt3, gain G, and the target value Tc_sv of the coke temperature at the end of the unsteady operation as constants used in the processing in the predicted value calculation unit 620 and the target furnace temperature determination unit 630. The acquisition unit 610 also acquires the operating conditions (target coke temperature Tc_sv and carbonization time (t e ) schedule value, etc.) may be acquired. The timing of data acquisition by the acquisition unit 610 is not limited to the timing exemplified in FIG. 10, and may be any timing in FIG.

[0122] Next, in step S1002, the acquisition unit 610 determines whether it is time to determine the target furnace temperature trajectory Tr_ref. In this section, the acquisition unit 610 determines that it is time to determine the target furnace temperature trajectory Tr_ref when the unsteady operation starts and when the operating conditions are changed. For example, in step S1001, the acquisition unit 610 determines whether it is time to determine the target furnace temperature trajectory Tr_ref when the unsteady operation starts and when the operating conditions are changed. s In step S1001, the acquisition unit 610 determines that the unsteady operation has started when the end time t e Dry distillation time t k (t e If a schedule value different from the already acquired schedule value is acquired as the schedule value of the operation condition, it is determined that the operation condition has been changed.

[0123] If it is determined in step S1002 that it is not time to determine the target reactor temperature trajectory Tr_ref (NO in step S1002), the process of step S1001 is executed again. Then, if it is determined in step S1002 that it is time to determine the target reactor temperature trajectory Tr_ref (YES in step S1002), the process of step S1003 is executed.

[0124] In step S1003, the target reactor temperature determination unit 630 generates initial values ​​of candidates for the target reactor temperature trajectory Tr_ref. Next, in step S1004, the processing device 600 sets the current processing time t to the time t for determining the target reactor temperature trajectory Tr_ref. c In this section, the time t c is the start time of the unsteady operation t s or the time t when the operating conditions change m The current processing time t is not the actual time but the simulation time in the computer simulation.

[0125] Next, in step S1005, the input heat quantity calculation unit 622 calculates the predicted value of the input heat quantity Q(t+1) at time t+1 for each of the multiple candidates included in the group of candidates for the target furnace temperature trajectory Tr_ref, based on the candidates for the target furnace temperature trajectory Tr_ref and the predicted value of the furnace battery temperature calculated by the furnace state calculation unit 621. As described above, the calculation of the predicted value of the input heat quantity Q(t+1) at time t+1 is realized, for example, by a control simulator that computer-simulates PID control, or by calculation using equations (13) and (14).

[0126] Next, in step S1006, the furnace state calculation unit 621 calculates the predicted value of the furnace battery temperature Tr(t+1) using the predicted value of the input heat quantity Q(t+1) at time t+1 calculated in step S1005 and the predicted values ​​and / or measured values ​​of the furnace battery temperatures Tr(t), Tr(t-1), Tr(t-2) at times t, t-1, t-2 before time t+1 (see equations (4) to (6)). s If time t+1 is the time t, the actual measured values ​​of the furnace battery temperatures Tr(t), Tr(t-1), and Tr(t-2) are used. s From +4 onwards, the predicted values ​​of the furnace battery temperatures Tr(t), Tr(t-1), Tr(t-2) are used. If time t+1 is a time between these, both the predicted and measured values ​​of the furnace battery temperatures Tr(t), Tr(t-1), Tr(t-2) are used.

[0127] Next, in step S1007, the processing device 600 determines whether the current processing time t is time t e -1. As a result of this determination, the current processing time t is time t e If it is not -1 (NO in step S1007), the process of step S1008 is executed.

[0128] In step S1008, the processing device 600 adds one hour to the current processing time t to update the current processing time t. Then, in steps S1005 to S1006, the predicted value of the input heat quantity Q(t+1) and the predicted value of the furnace battery temperature Tr(t+1) at time t+1, one hour after the updated current processing time t, are calculated for each of the multiple candidates included in the candidate group of the target furnace temperature trajectory Tr_ref. As described above, in step S1007, the current processing time t is updated to time t e The processes of steps S1005 to S1008 are repeatedly executed until it is determined that the count becomes -1.

[0129] In step S1007, the current processing time t is set to time t e If it is determined that the target reactor temperature trajectory Tr_ref is -1 (YES in step S1007), the time t c (Start time of non-steady operation t s or the time t when the operating conditions change m ) to the end time of the unsteady operation t e The predicted values ​​of the input heat quantity Q(t+1) and the furnace battery temperature Tr(t+1) at each time t+1 for each hour up to t+1 are calculated for each of the multiple candidates included in the candidate group of the target furnace temperature trajectory Tr_ref. In this case, the processing of step S1009 is executed.

[0130] In step S1009, the furnace state calculation unit 621 calculates the start time t s Furnace temperature Tr(t s ) and the end time of the non-steady operation t e Furnace temperature Tr(t e ) predicted value, the start time of unsteady operation t s Dry distillation time t k (t s ) and the end time of the unsteady operation t e Dry distillation time t k (t e ) based on the schedule value, the end time of the non-steady operation t e Coke temperature Tc(t e) is calculated for each of the multiple candidates included in the candidate group of the target furnace temperature trajectory Tr_ref (see equations (7) to (12)). e Coke temperature Tc(t e ) is calculated using the predicted value of the furnace battery temperature during unsteady operation, calculated using equations (9) to (11). e -Δt1), Tr(t e -Δt2), Tr(t e -Δt3) is also used.

[0131] Next, in step S1010, the target reactor temperature determination unit 630 calculates the value of the evaluation function J of equation (15) when the constraint equations including equation (16) are satisfied, for each of the multiple candidates included in the candidate group of the target reactor temperature trajectory Tr_ref. Next, in step S1011, the target reactor temperature determination unit 630 determines whether the convergence condition is satisfied. If the result of this determination is that the convergence condition is not satisfied (NO in step S1011), the process of step S1012 is executed. In step S1012, the target reactor temperature determination unit 630 updates the group of candidates for the target reactor temperature trajectory Tr_ref. Then, the processes of steps S1004 to S1011 are executed using the candidates for the updated target reactor temperature trajectory Tr_ref. In this way, the processes of steps S1004 to S1012 are repeatedly executed until the convergence condition is satisfied.

[0132] Then, in step S1011, when it is determined that the convergence condition is satisfied (YES in step S1011), the processing of step S1013 is executed. In step S1013, the target furnace temperature determination unit 630 determines the target furnace temperature trajectory Tr_ref, which is determined by the design variables (furnace temperature change amount at the start of unsteady state ΔTr1, furnace temperature change amount at the end of unsteady state ΔTr2, minimum furnace temperature arrival time time1, and minimum furnace temperature maintenance time time2) when the value of the evaluation function J is the smallest among the values ​​of the evaluation function J for multiple candidates for the target furnace temperature trajectory Tr_ref calculated when the convergence condition is satisfied, as the optimal solution of the target furnace temperature trajectory Tr_ref. Next, in step S1014, the control unit 640 calculates the amount of heat input to the combustion chamber 3 at the time one hour after the current time using the same method as that used by the input heat amount calculation unit 622 to calculate the predicted value of the input heat amount Q(t+1) in step S1006. Then, the control unit 640 generates and outputs a control signal to set the amount of heat input to the combustion chamber 3 to a heat amount corresponding to the difference between the calculated amount of heat input to the combustion chamber 3 at the time one hour after the current time and the value of the target furnace temperature trajectory Tr_ref at the time one hour after the current time determined by the target furnace temperature determination unit 630. Note that the current time is the actual time when step S1014 is executed. When the processing of step S1014 is completed, the processing according to the flowchart of FIG. 10 is terminated.

[0133] <Calculation example> FIG. 11 is a diagram showing an example of the target furnace temperature trajectory Tr_ref and the actual furnace temperature trajectory Tr_mes. The actual furnace temperature trajectory Tr_mes shown in FIG. 11 is the furnace battery temperature when coke satisfying the desired quality was produced based on the results of past operations. The target furnace temperature trajectory Tr_ref is the time t c (Start time of non-steady operation t s or the time t when the operating conditions change m ) is the time t at which the target reactor temperature trajectory Tr_ref is determined without using the actual values ​​at times later than c The furnace battery temperature was determined using the method explained in this section using the actual values ​​up to that point.

[0134] Figure 11(a) shows the calculation results when the operating conditions are not changed during unsteady operation. As shown in Figure 11(a), it can be seen that the target furnace temperature trajectory Tr_ref can be made to follow the actual furnace temperature trajectory Tr_mes with high accuracy.

[0135] Figure 11(b) shows the calculation results when the operating conditions (carbonization time) are changed during unsteady operation. In Figure 11(b), as in Figure 9B, for comparison, the time t m Even after that, the start time of the non-steady operation t sIn Fig. 11(b), the target furnace temperature trajectory Tr_ref_old determined at the start time t s The time t when the operating conditions were changed m Until the start time of the unsteady operation t s The target furnace temperature trajectory Tr_ref_old determined at time t m From the end time of the unsteady operation t e Until the time t when the operating conditions are changed m The target reactor temperature trajectory Tr_ref_new determined in is used.

[0136] As shown in Fig. 11(b), at the time t m It can be seen that if the target furnace temperature trajectory Tr_ref_new is determined again in step S10, the target furnace temperature trajectory Tr_ref can be made to follow the actual furnace temperature trajectory Tr_mes with even higher accuracy.

[0137] <Summary> As described above, in this section, the processing device 600 calculates a predicted value of the furnace temperature during unsteady operation based on a first influencing factor, which is a factor that affects the predicted value of the furnace temperature during unsteady operation, and calculates a predicted value of the physical quantity that represents the carbonization state of the coke during unsteady operation based on a second influencing factor, which is a factor that affects the predicted value of the physical quantity.The processing device 600 then determines the target furnace temperature trajectory Tr_ref based on the predicted value of the furnace temperature and the predicted value of the physical quantity.Therefore, it is possible to predict what the furnace temperature and the carbonization state of the coke during unsteady operation will be like, and dynamically determine the target furnace temperature trajectory so that the predicted results are reflected.The target furnace temperature during unsteady operation can be determined with high accuracy.

[0138] In this section, the processing device 600 uses, as a first influencing factor, which is a factor that affects the predicted value of the furnace temperature during unsteady operation, an influencing factor including the amount of heat input to the combustion chamber 3 during unsteady operation and the furnace temperature at a timing before the predicted time of the predicted value (the time when the predicted value of the furnace temperature is obtained). Furthermore, the processing device 600 uses, as a second influencing factor, which is a factor that affects the physical quantity that represents the carbonization state of the coke during unsteady operation, an influencing factor including the predicted value of the furnace temperature during unsteady operation. Then, based on the first influencing factor, the processing device 600 calculates a predicted value of the furnace temperature at a prediction time that is later than the prediction time of the predicted value of the furnace temperature that has already been calculated (see Equations (4) to (6) as a specific example). Furthermore, based on the second influencing factor including the predicted value of the furnace temperature during unsteady operation calculated in this way, the processing device 600 calculates a predicted value of the physical quantity that represents the carbonization state of the coke during unsteady operation (see Equations (7) to (11) as a specific example). Therefore, the predicted values ​​of the furnace temperature and the physical quantities representing the coke distillation state can be calculated using influential factors that have a large effect on the predicted values ​​of the furnace temperature and the physical quantities representing the coke distillation state during non-steady operation.

[0139] In this section, the processing device 600 calculates the amount of change in the predicted furnace temperature for each predetermined time period determined based on the control period for the combustion chamber 3. Therefore, the amount of change in the predicted furnace temperature can be calculated at a timing synchronized with the control period for the combustion chamber 3. Therefore, for example, calculations for adjusting the time of the predicted furnace temperature value to a time that matches the control period for the combustion chamber 3 are not required, and the furnace temperature can be predicted by cumulatively adding the predicted values ​​of the amount of change in the furnace temperature.

[0140] In this section, the processing device 600 uses an influencing factor further including the carbonization time as the second influencing factor. Therefore, the prediction accuracy of the predicted value of the physical quantity representing the carbonization state of the coke during unsteady operation can be further improved.

[0141] In this section, the processing device 600 calculates the change in the physical quantity representing the carbonization state of the coke from the start to the end of the non-steady operation as a predicted value of the physical quantity representing the carbonization state of the coke during the non-steady operation. Therefore, by predicting the physical quantity representing the carbonization state of the coke at the start and end of the non-steady operation, the physical quantity representing the carbonization state of the coke can be predicted.

[0142] In this section, the processing device 600 executes the process by regarding the end of the non-steady operation as the end of the coke pushing operation in the Da+1th cycle after the suspension of the loading of coal into the coke chamber 2 and the pushing of the coke. Therefore, the end of the non-steady operation can be determined at the timing of the end of the batch-type operation in the coke oven. Also, for example, the end of the non-steady operation can be determined to coincide with the timing of measuring the coke temperature.

[0143] In this section, the processing device 600 executes the processing by setting the start time of the non-steady operation as the end time of the coke pushing operation in the run Db before the start of the suspension of the loading and pushing of coal into the coke chamber 2. Therefore, the end time of the non-steady operation can be determined at the timing of the end of the batch-type operation in the coke oven. Also, for example, the start time of the non-steady operation can be determined to coincide with the timing of measuring the coke temperature.

[0144] In addition, in this section, when the operating conditions are changed after the target furnace temperature trajectory is determined, the processing apparatus 600 changes the influencing factor that is changed by the change in the operating conditions among the first influencing factor and the second influencing factor according to the operating conditions, and then calculates the time t when the operating conditions are changed. m At subsequent times, the predicted values ​​of the furnace temperature during unsteady operation and the predicted values ​​of the physical quantities representing the carbonization state of the coke are recalculated, and the target furnace temperature trajectory is re-determined. Therefore, even if the operating conditions are changed after the target furnace temperature trajectory has been determined, the target furnace temperature during unsteady operation can be determined with high accuracy.

[0145] In addition, in this section, the processing device 600 determines the target furnace temperature trajectory based on the results of calculations that include calculating the difference between a candidate target furnace temperature trajectory and the predicted value of the furnace temperature (see the first term on the right side of equation (15) as a specific example), and the difference between a target physical quantity, which is the target value of the physical quantity representing the carbonization state of the coke, and the predicted value of the physical quantity representing the carbonization state of the coke (see the second term on the right side of equation (15) as a specific example). Therefore, it is possible to realize both a quantitative evaluation of the predicted value of the furnace temperature and a quantitative evaluation of the predicted value of the physical quantity representing the carbonization state of the coke. Therefore, the target furnace temperature in unsteady operation can be accurately determined based on these quantitative evaluations.

[0146] In this section, the processing device 600 determines the target furnace temperature trajectory based on the value of an evaluation function including a first evaluation index (see the first term on the right side of equation (15) as a specific example) that evaluates the difference between a candidate target furnace temperature trajectory and the predicted value of the furnace temperature, and a second evaluation index (see the second term on the right side of equation (15) as a specific example) that evaluates the difference between a target physical quantity, which is a target value of the physical quantity representing the coke carbonization state, and the predicted value of the physical quantity representing the coke carbonization state. Therefore, quantitative evaluation of the predicted value of the furnace temperature and quantitative evaluation of the predicted value of the physical quantity representing the coke carbonization state can be performed by solving an optimization problem.

[0147] In this section, the processing device 600 determines the target furnace temperature trajectory based on the value of an evaluation function that further includes a third evaluation index that evaluates the input heat amount (see the third term on the right side of Equation (15) as a specific example). Therefore, the optimization problem for determining the target furnace temperature in unsteady operation can be an optimization problem that can obtain an optimal solution with higher accuracy.

[0148] In this section, the processing device 600 calculates the value of the evaluation function by varying candidates for the target reactor temperature trajectory, and determines the target reactor temperature trajectory based on the calculated value of the evaluation function. Therefore, the target reactor temperature trajectory can be determined by a metaheuristic method without using differential equations or the like that describe physical phenomena.

[0149] In this section, the processing device 600 calculates the predicted value of the input heat amount based on the candidate target furnace temperature trajectory and the predicted value of the furnace temperature. Therefore, the predicted value of the input heat amount can be quantitatively obtained as the first influencing factor.

[0150] In this section, the processing device 600 uses the temperature of the coke produced during unsteady operation or the temperature of the oven wall 4 during unsteady operation as a physical quantity representing the carbonization state of the coke during unsteady operation. Therefore, a physical quantity measurable during unsteady operation can be used as a physical quantity representing the carbonization state of the coke during unsteady operation. This makes it possible to verify, for example, the predicted value of the physical quantity representing the carbonization state of the coke. As described with reference to FIG. 5B, the temperature of the coke produced during unsteady operation is measured, for example, during the coke extrusion operation. Similarly, the temperature of the oven wall 4 during unsteady operation is measured, for example, during the coke extrusion operation.

[0151] In this section, the processing device 600 uses the furnace battery temperature, which is a representative value of the temperatures in the multiple combustion chambers 3, as the furnace temperature. Therefore, it is possible to reduce the number of variables related to the furnace temperature. Therefore, for example, it is possible to further reduce the calculation load.

[0152] In this section, the processing device 600 generates and outputs a control signal for adjusting the amount of heat input to the combustion chamber 3 to a heat amount corresponding to the difference between the target furnace temperature trajectory and the actual furnace temperature value. Therefore, the processing device 600 can execute control for realizing the target furnace temperature trajectory.

[0153] In this section, the case where the furnace temperature is the furnace battery temperature has been exemplified. However, the furnace temperature is not limited to the furnace battery temperature. For example, if, as mentioned above, a control valve and an actuator are installed in all combustion chambers 3 and a thermometer 6 is installed in all combustion chambers 3, each equation can be changed to treat each block as an individual coke chamber 2, and the furnace temperature, which is the temperature in the combustion chamber 3 of the coke oven 1, can be the temperature of each combustion chamber 3 rather than the furnace battery temperature. In this case, for example, equations (4) to (6) and equations (13) to (14) can be changed to equations for each combustion chamber 3. In addition, when the start time t of unsteady operation is s and end time t e Depending on which coke chamber 2 the coke temperature in is, the temperatures of one or more combustion chambers 3 close to the coke chamber 2 may be used to determine ΔTr1 to ΔTr3, which are explanatory variables in equation (7). Also, for the first and third terms on the right side of equation (15), Σ|Tr(t)-Tr_ref| and Σ|Q(t)| may be calculated for each combustion chamber 3, and the sums of Σ|Tr(t)-Tr_ref| and Σ|Q(t)| for each combustion chamber 3 may be used as evaluation indexes to be multiplied by weighting coefficients w1 and w3, respectively.

[0154] [Determination of weighting coefficient wi in application example] In the above application examples, the weighting coefficients wi determined by the weighting coefficient determination device 200 are the weighting coefficients w1, w2, and w3 in equation (15). In this case, the correspondence between the [Application Example] column and the [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method] column is as follows:

[0155] The evaluation function J (equation (1)) in the [Application Example] column corresponds to equation (15) in the [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method] column. The evaluation index ei in the [Application Example] column is the first to third terms on the right side of equation (15), |Tr(t)-Tr_ref|, |Tc(t e )-Tc_sv|, |Q(t)|. The variables di in the evaluation function in the [Application Example] column correspond to Tr(t), Tr_ref, and Tc(t e), Q(t). Of these variables di in the evaluation function, the target furnace temperature trajectory Tr_ref (furnace temperature change amount at the start of unsteady state ΔTr1, furnace temperature change amount at the end of unsteady state ΔTr2, time to reach minimum furnace temperature time1, and time to maintain minimum furnace temperature time2) correspond to the design variables. Furnace battery temperature (Tr), input heat amount Q(t), and coke temperature (Tc) correspond to the variables indicating physical quantities whose values ​​fluctuate as the goods are processed, as described in the [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method] section. Furnace battery temperature (Tr) and input heat amount Q(t) correspond to the physical quantities indicating the state of the equipment (combustion chamber 3, control valve 5) used to process the goods, as described in the [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method] section. Coke temperature (Tc) corresponds to the physical quantity indicating the state of the goods, as described in the [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method] section.

[0156] In the [Application Example], the problem of finding an optimal solution for the target furnace temperature trajectory Tr_ref using the evaluation function J in equation (15) and the constraint equation in equation (16) corresponds to the first optimization problem described in the section [Weighting coefficient determination device and weighting coefficient determination method]. In the [Application Example], the predicted value calculation unit 620 calculates the predicted values ​​of the input heat quantity Q(t+1) and the furnace battery temperature Tr(t+1) and the coke temperature Tc(t e ) corresponds to the prediction calculation (estimation calculation) described in the section [Weighting coefficient determination device and weighting coefficient determination method]. Moreover, the differential evaluation function J_dif (Equation (3)) corresponds to, for example, the following equation (17). The problem of finding the optimal solution wi_opt for the weighting coefficients using the differential evaluation function J_dif corresponds to the second optimization problem described in the section [Weighting coefficient determination device and weighting coefficient determination method].

[0157]

number

[0158] The first to third terms on the right side of equation (17) are |Tr_sim(t)-Tr_ope(t)| and |Tc_sim(t e )-Tc_ope(t e)|, |Q(t)_sim(t)-Q_ope(t)| corresponds to the differential evaluation index |di'_sim-di'_opt| described in the section [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method]. The first to third terms on the right side of equation (17), Tr_ope(t) and Tc_ope(t e ), Q_ope(t) is the value that an experienced operator uses during non-steady operation (start time t s ~End time t e ) are the actual values ​​of the oven battery temperature, coke temperature, and input heat quantity when coke of the desired quality is produced by manually operating the opening of the control valve 5. The actual value of the oven battery temperature Tr_ope(t) and the actual value of the coke temperature Tc_ope(t e ), and the actual value Q_ope(t) of the input heat quantity corresponds to the optimal value di'_opt of the variable in the evaluation function explained in the section [Weighting coefficient determination device and weighting coefficient determination method].

[0159] The first to third terms on the right side of equation (17) are Tr_sim(t) and Tc_sim(t e ), Q(t)_sim(t) are the calculated values ​​of the oven battery temperature, coke temperature, and input heat amount when the value of the evaluation function J in equation (15) is minimized within the range that satisfies the constraint equation in equation (16). The calculated value di'_opt of the variables in the evaluation function explained in the section [Weighting coefficient determination device and weighting coefficient determination method] is calculated based on the oven battery temperature, coke temperature, and input heat amount calculated values ​​Tr_sim(t), Tc_sim(t e ), Q(t)_sim(t). In equation (17), the variables di(=Tr(t), Tr_ref, Tc(t) in the evaluation function included in equation (15) correspond to e ), Q(t)), the target furnace temperature trajectory Tr_ref is not included in the calculation of the difference evaluation index. e ), Q(t)_sim(t) may be calculated together with the target reactor temperature trajectory Tr_ref used for control by the control unit 640 as described in [Application Example].

[0160] When determining the weighting coefficient wi in [Application Example], the functional configuration of the weighting coefficient determination device 200 is, for example, as shown in Fig. 12. In Fig. 12, the weighting coefficient determination device 200 includes a calculation unit 210, a weighting coefficient determination unit 220, an evaluation index determination unit 230, a storage unit 240, and an output unit 250. The calculation unit 210 further includes an acquisition unit 610, a predicted value calculation unit 620, and a target furnace temperature determination unit 630, which are described in the [Application Example] section. As explained in the section [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method], the storage unit 240 stores the actual value of the physical quantity, which is obtained when determining the optimal value di'_opt of the variable in the evaluation function and is used to calculate the evaluation index ei of the evaluation function J, in association with the optimal value di'_opt of the variable in the evaluation function.

[0161] An example of the actual values ​​of such physical quantities in the [Application Example] will be described with reference to Fig. 13. Fig. 13 is a diagram illustrating an example of information stored in the storage unit 240. As mentioned above, in the [Application Example], the optimal value di'_opt of the variables in the evaluation function is determined by the value obtained by a skilled operator during non-steady operation (t s ~t e ) are the actual values ​​of the oven battery temperature, coke temperature, and input heat amount Tr_ope(t), Tc_ope(t) when coke of the desired quality is produced by manually operating the opening of the control valve 5. e ), Q_ope(t). In the following explanation, the actual values ​​of the furnace battery temperature, coke temperature, and input heat amount, Tr_ope(t), Tc_ope(t e ), Q_ope(t) are also referred to as operator operation history. Figure 13(a) is a diagram showing an example of operator operation history. As explained in the section [Application Example], the furnace battery temperature and input heat amount change from the start time t s From the end time t e The coke temperature is the temperature at the end of the unsteady operation.

[0162] The actual values ​​of the physical quantities obtained when determining the optimal values ​​di'_opt of the variables in the evaluation function, as explained in the section [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method], and used to calculate the evaluation index ei of the evaluation function J, correspond to the operating conditions when the operator operation history was obtained. As shown in FIG. 13(b), the actual values ​​of the physical quantities obtained when the operator operation history was obtained correspond to the operating conditions when the operator operation history was obtained. s The start time is the same as the end time t e The operating conditions for the period ending at the same time as the time t are stored in the storage unit 240. The operating conditions are information acquired by the acquisition unit 610 in step S1001 of FIG. 10, and include the pause period, the carbonization time, and the target coke temperature. The operator operation history shown in FIG. 13(a) and the operating conditions shown in FIG. 13(b) are stored in the storage unit 240. s ~t e are related to each other via

[0163] 10 is executed by the calculation unit 210 (the acquisition unit 610, the predicted value calculation unit 620, and the target furnace temperature determination unit 630) in accordance with the above-described correspondence relationships. In step S1001, the acquisition unit 610 acquires the operating conditions stored in the storage unit 240.

[0164] In step S1013, the target reactor temperature determination unit 630 calculates variables in the evaluation function, including the design variables when the value of the evaluation function J is the smallest among the values ​​of the evaluation function J for the candidates of the target reactor temperature trajectory Tr_ref calculated when the convergence condition is satisfied. Specifically ... e ), Q(t). The target furnace temperature determination unit 630 calculates the variables Tr(t), Tr_ref, Tc(t) in the evaluation function. e ), Q(t), Tr_ref is determined as the optimal solution for the target reactor temperature trajectory Tr_ref, and Tr(t), Tc(t e ), Q(t) are the calculated values ​​of the variables in the evaluation function Tr_sim(t), Tc_sim(t e), Q_sim(t) are stored in the storage unit 240. In the following description, the calculated values ​​Tr_sim(t) and Tc_sim(t) of the variables in the evaluation function stored in the storage unit 240 are used. e ), Q_sim(t) are also referred to as control history. In FIG. 13(c), the control history for the same period as the unsteady operation shown in FIG. 13(a) is stored in the storage unit 240. The operator operation history shown in FIG. 13(a), the operating conditions shown in FIG. 13(b), and the control history shown in FIG. 13(c) are stored in the storage unit 240 at time t s ~t e are related to each other via

[0165] Then, as explained in the section [Weighting Coefficient Determination Device and Weighting Coefficient Determination Method], the weighting coefficient determination unit 220 determines the optimal solutions w1_opt, w2_opt, and w3_opt of the weighting coefficients in equation (15). In addition, the evaluation index determination unit 230 determines the evaluation indexes (|Tr_sim(t)-Tr_ope(t)|, |Tc_sim(t)) included in the evaluation function J in equation (15). e )-Tc_ope(t e)|, |Q(t)_sim(t)-Q_ope(t)|), an evaluation index to be included in the evaluation function J of equation (15) is determined. As described above, for example, when the absolute values ​​of the optimal solutions w1_opt, w2_opt, and w3_opt of the weighting coefficients of equation (15) determined by the weighting coefficient determination unit 220 are equal to or less than a positive threshold, the evaluation index determination unit 230 sets (changes) the weighting coefficients to 0 (zero). In doing so, the output unit 250 outputs information indicating 0 (zero) as the value of the weighting coefficient for the weighting coefficients changed to 0 (zero) by the evaluation index determination unit 220 out of the optimal solutions w1_opt, w2_opt, and w3_opt of the weighting coefficients determined by the weighting coefficient determination unit 220, and outputs information indicating the optimal solutions of the weighting coefficients determined by the weighting coefficient determination unit 220 for other weighting coefficients. In this case, the processing device 600 sets the values ​​of the weighting coefficients w1 to w3 of the evaluation function J in equation (15) to values ​​indicated in the information output by the output unit 250. Furthermore, for example, when the output unit 250 outputs information on the optimal solution wi_opt of the weighting coefficients determined by the weighting coefficient determination unit 220, and information on the evaluation indexes ei determined by the evaluation index determination unit 230 as the evaluation indexes ei to be included in the evaluation function J and the evaluation indexes ei determined as the evaluation indexes ei not to be included in the evaluation function J, the processing device 600 may discard the optimal solution wi_opt of the weighting coefficients and set it to 0 (zero) for the weighting coefficients for the evaluation indexes ei determined as the evaluation indexes ei not to be included in the evaluation function J. For example, the output unit 250 may transmit this information to the processing device 600, display it on a computer display, or store it in a storage medium readable by the processing device 600.

[0166] FIG. 14 shows examples of the target furnace temperature trajectory Tr_ref, the estimated furnace temperature trajectory Tr_sim, and the operator-intervention furnace temperature trajectory Tr_ope. The operator-intervention furnace temperature trajectory Tr_ope shown in FIG. 14 is the actual value of the furnace battery temperature Tr_ope(t) included in the operator operation history. As described above, the actual furnace battery temperature value included in the operator operation history is the furnace battery temperature when a skilled operator manually operates the control valve 5 during unsteady operation to produce coke of the desired quality in actual operation. The target furnace temperature trajectory Tr_ref is the target furnace temperature trajectory determined by the target furnace temperature determination unit 630 using the weighting coefficients w1 to w3 determined by the weighting coefficient determination device 200 using the operator operation history. The estimated furnace temperature trajectory Tr_sim is the predicted value of the furnace battery temperature Tr(t) calculated by the predicted value calculation unit 620 when the target furnace temperature trajectory is determined by the target furnace temperature determination unit 630 using the weighting coefficients w1 to w3 determined by the weighting coefficient determination device 200 using the operator operation history. As shown in FIG. 14, it can be seen that by determining the weighting coefficients w1 to w3 as explained in this section, an estimated furnace temperature trajectory Tr_sim close to the operator-intervention furnace temperature trajectory Tr_ope is calculated.

[0167] As described above, this section has illustrated a case where the weighting coefficients w1 to w3 in the evaluation function J (Equation (15)) are determined when solving the target furnace temperature trajectory Tr_ref, which is an index for the operation of the control valve 5 when producing a product such as coke, using an optimization problem. Therefore, it is possible to produce a product (coke) so that the evaluation index in the evaluation function J is high.

[0168] The storage unit 240 also stores the operator operation history (FIG. 13(a)) and the operating conditions (FIG. 13(b)) in association with each other. The calculation unit 210 then calculates the values ​​Tr(t), Tr_ref, and Tc(t) of the variables in the evaluation function based on the operating conditions stored in association with the operator operation history. e ), Q(t). Therefore, the calculation unit 210 calculates the calculated values ​​Tr(t), Tr_ref, Tc(t) of the variables in the evaluation function under the same operating conditions as when the optimal value di_opt of the variables in the evaluation function was obtained. e), Q(t) can be calculated.

[0169] [Other variations] The above-described embodiments of the present invention can be realized by a computer executing a program. A computer-readable recording medium on which the program is recorded and a computer program product such as the program can also be applied as embodiments of the present invention. Examples of recording media that can be used include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, magnetic tapes, non-volatile memory cards, and ROMs. Furthermore, the above-described embodiments of the present invention are merely examples of specific embodiments for carrying out the present invention, and the technical scope of the present invention should not be construed as being limited by these. In other words, the present invention can be embodied in various forms without departing from its technical concept or main features. [Explanation of symbols]

[0170] 1. Coke oven 2. Carbonization chamber 3 Combustion chamber 4 Furnace wall 5. Regulating valve 6 thermometer 7 Extrusion Ram 8 Thermometer 9 Guide car 200 Weighting coefficient determination device 210 Calculation Unit 220 Weighting coefficient determination unit 240 Storage section 250 Output section 600 Processing Equipment 610 Acquisition Department 620 Prediction Value Calculation Unit 621 Reactor State Calculation Unit 622 Input heat amount calculation section 630 Target furnace temperature determination unit 640 Control Unit t k Dry distillation time t tStreet time t s Start time of non-routine operation t e End time of non-routine operation t m The time when the operating conditions changed T c Coke Temperature T r Furnace temperature Tr_ref Target furnace temperature trajectory Tr_mes Actual furnace temperature trajectory Tr_sim Estimated furnace temperature trajectory Tr_ope Operator intervention furnace temperature trajectory time0 Unsteady operation period time1 Time to reach minimum furnace temperature time2 Minimum furnace temperature maintenance time ΔTr1 Change in furnace temperature at the start of unsteady state ΔTr2 Change in furnace temperature at the end of unsteady state

Claims

1. A weight coefficient determination device that determines weight coefficients for a plurality of evaluation indexes included in an evaluation function used to solve an optimization problem by inverse reinforcement learning, a calculation means for calculating a calculated value when the value of the evaluation function is maximum or minimum for a variable in the evaluation function that is a variable that determines the evaluation index; a weight coefficient determination means for calculating a value of a difference evaluation index that evaluates a difference between a calculated value of a variable in the evaluation function calculated by the calculation means and a preset optimal value of the variable in the evaluation function, and for determining a value of the weight coefficient by solving an optimization problem using a difference evaluation function that is an evaluation function including the difference evaluation index.

2. The calculation means Calculating a calculated value of a variable in the evaluation function using the candidate values ​​of the weighting coefficients; The weighting coefficient determination means determining, as the value of the weighting coefficient, a candidate value of the weighting coefficient used in calculating the calculated value of the variable in the evaluation function when the value of the difference evaluation function satisfies a predetermined condition; 2. The weighting coefficient determination device according to claim 1, wherein the candidate values ​​of the weighting coefficients are repeatedly changed until the value of the difference evaluation function satisfies a predetermined condition.

3. the optimal value of the variable in the evaluation function and the actual value of the physical quantity obtained when the optimal value of the variable in the evaluation function was determined, and used to calculate the evaluation index, are stored in a storage medium in a mutually associated manner; 3. The weighting coefficient determination device according to claim 1, wherein the calculation means calculates the calculated value of the variable in the evaluation function using an actual value of the physical quantity stored in association with the optimal value of the variable in the evaluation function.

4. The variables in the evaluation function include:

4. The weighting coefficient determination device according to claim 1, wherein the weighting coefficient determination device includes a variable indicating a physical quantity whose value changes as a result of processing the article.

5. The calculation means 5. The weighting coefficient determination device according to claim 4, wherein a prediction calculation is performed for a value of at least one of the variables in the evaluation function other than a design variable.

6. The weighting coefficient determination device according to any one of claims 1 to 5, further comprising evaluation index determination means that determines whether or not to include the evaluation index for the weighting coefficient in the evaluation function based on the value of the weighting coefficient determined by the weighting coefficient determination means, and determines the evaluation index to be included in the evaluation function based on the result of the determination.

7. The evaluation index is: a first evaluation index relating to a difference between a predicted value and a target value of a physical quantity representing a carbonization state of coke during unsteady operation; a second evaluation index relating to a difference between a predicted value and a target value of a furnace temperature, which is the temperature of a combustion chamber in a coke oven, during unsteady operation; Including, The calculation means a predicted value calculation means for calculating a predicted value of the furnace temperature during the unsteady operation based on a first influencing factor which is a factor that influences the predicted value of the furnace temperature during the unsteady operation, and for calculating a predicted value of the physical quantity during the unsteady operation based on a second influencing factor which is a factor that influences the predicted value of the physical quantity during the unsteady operation; a target furnace temperature determination means for calculating calculated values ​​of variables in the evaluation function and values ​​of the evaluation function based on the predicted value of the furnace temperature during the unsteady operation and the predicted value of the physical quantity during the unsteady operation, and for determining values ​​of variables in the evaluation function including a target furnace temperature trajectory, which is a time change of a target value of the furnace temperature during the unsteady operation, based on the value of the evaluation function; The weighting coefficient determination device according to any one of claims 1 to 6, comprising:

8. The evaluation index is: The weighting coefficient determination device according to claim 7 , further comprising a third evaluation index related to the amount of heat input to the combustion chamber during the unsteady operation.

9. A weight coefficient determination method for determining weight coefficients for a plurality of evaluation indexes included in an evaluation function used to solve an optimization problem by inverse reinforcement learning, comprising: a calculation step of calculating a calculated value when the value of the evaluation function is maximum or minimum for a variable in the evaluation function that is a variable that determines the evaluation index; a weight coefficient determination step of calculating the value of a difference evaluation index that evaluates the difference between the calculated value of the variable in the evaluation function calculated in the calculation step and a preset optimal value of the variable in the evaluation function, and determining the value of the weight coefficient by solving an optimization problem using a difference evaluation function that is an evaluation function including the difference evaluation index.

10. A program for causing a computer to function as each of the means of the weighting coefficient determination device according to any one of claims 1 to 8.

Images