KR20210060558A - Driving method, support device, learning device, and oil station driving condition setting support system - Google Patents

Abstract

An operation method for operating an apparatus for distilling crude oil to produce a plurality of oil components includes: a design process of estimating the amount of water contained in the crude oil after the change or the flow rate of each of the plurality of oil components when the oil type is changed; A preliminary preparation step of preparing in advance for receiving the changed crude oil based on the moisture content or flow rate of the crude oil after the change; A change process that initiates the acceptance of crude oil after the change; And an adjustment step of adjusting operating conditions for distilling the crude oil after the change. In the adjustment step, the target set value of the control amount for controlling the device is adjusted according to the state value indicating the state of the device for distilling crude oil after the change.

Description

Driving method, support device, learning device, and oil station driving condition setting support system

The present invention relates to a driving method for operating an apparatus for manufacturing petroleum products, a support device that can be used for the driving method, a learning device, and a system for setting operation conditions of an oil refinery.

In oil refineries for refining crude oil to produce petroleum products, crude oil extracted from various oil wells is accommodated as a raw material depending on market price, production volume in each oil well, transportation situation of crude oil from each oil well, and the like. The crude oil contained in the crude oil tank is introduced into an atmospheric distillation column and separated into a plurality of fractions having different boiling points. A plurality of fractions are further processed and upgraded in a downstream device as necessary to produce petroleum products.

When the oil type of the crude oil to be treated is changed (hereinafter referred to as "change of oil type"), the operating condition of equipment such as an atmospheric distillation column or a heating furnace rapidly changes due to changes in the composition of hydrocarbon oil or water contained in the crude oil. It is subject to change. Conventionally, an experienced operator has set appropriate operating conditions by adjusting set values of various control amounts.

Japanese Unexamined Patent Application Publication No. H05-189062

However, whether or not the optimum driving condition can be realized depends on the experience and skill of the operator, and therefore, there is no choice but to rely on a small number of experienced and highly skilled operators, which is a tremendous burden. In addition, since it was difficult to stipulate a large number of driving patterns in operation to change the oil type of crude oil, complex interrelationships or key points of the operation sequence, etc., an experienced driver could adjust the driving conditions while paying attention to what points and in what order. It was difficult to educate other drivers about whether or not.

In order to solve these problems, a technique has been proposed that statistically processes the operation method of an experienced operator and expresses it as a logistic function, etc. (for example, see Patent Document 1), but in oil refineries, a number of devices Since the operating state of is set by a large number of control amounts, and they complexly interfere with each other, this approach has limitations.

The present invention has been devised in view of such a situation, and its object is to provide a technology that supports setting of driving conditions capable of realizing desirable operation of an oil refinery.

In order to solve the above problem, the operation method according to an embodiment of the present invention is an operation method for operating an apparatus for distilling crude oil to produce a plurality of oils, and when the oil type is changed, it is included in the crude oil after the change. A design process of estimating the amount of water or the flow rate of each of the plurality of oils; A preliminary preparation step of preparing in advance for receiving the changed crude oil based on the moisture content or flow rate of the crude oil after the change; A change process that initiates the acceptance of crude oil after the change; And an adjustment step of adjusting operating conditions for distilling the crude oil after the change. In the adjustment step, the target set value of the control amount for controlling the device is adjusted according to the state value indicating the state of the device for distilling crude oil after the change.

Another embodiment of the present invention is a support device. This apparatus is an apparatus for distilling crude oil to produce a plurality of fractions, comprising: a design step of estimating the amount of water contained in the crude oil after the change or the flow rate of each of the plurality of fractions when the oil type is changed; A preliminary preparation step of preparing in advance for receiving the changed crude oil based on the moisture content or flow rate of the crude oil after the change; A change process that initiates the acceptance of crude oil after the change; And an acquisition unit for acquiring information necessary to proceed with each step included in the operating method when an operating method including an adjustment step of adjusting an operating condition for distilling the changed crude oil is executed; And a presentation unit for presenting the information acquired by the acquisition unit.

Another embodiment of the present invention is a learning device. The device includes: a state value acquisition unit that acquires a state value indicating a state of the device for distilling crude oil; And a learning unit that learns, by machine learning, a method for calculating a recommended value of a control amount for controlling the device when the oil type is changed, based on the state value.

Another embodiment of the present invention is a system for supporting setting operating conditions of an oil refinery. The oil refinery operation condition setting support system includes: a support device that supports setting operation conditions of an apparatus for distilling crude oil to produce a plurality of oils; And a learning device that learns a policy used in the support device by machine learning, wherein the learning device includes: an acquisition unit that acquires a state value indicating a state of the device; And a learning unit that learns, by machine learning, a method of calculating a recommended value of a control amount for controlling the device when the oil type is changed, based on the state value, wherein the supporting device indicates the state of the device when the oil type is changed. A state value acquisition unit that acquires a state value; A calculation unit that calculates a recommended value of a control amount for controlling the device using the measures learned by the learning device, based on the state value; And an output unit that presents the calculated recommended value to the operator, or sets the calculated recommended value as a target set value of the control amount to the device.

Here, any combination of the above constituent elements and conversion of the expression of the present invention among methods, apparatuses, systems, recording media, computer programs, etc. are also effective as embodiments of the present invention.

Advantageous Effects of Invention According to the present invention, it is possible to provide a technology that supports setting of operating conditions capable of realizing desirable operation of an oil refinery.

1 is a diagram schematically showing the configuration of a refinery.
Fig. 2 is a diagram showing an overall configuration of a system for supporting setting operating conditions of an oil refinery according to an embodiment.
3 is a diagram showing the procedure of an oil type change operation.
Fig. 4 is a diagram showing details of pre-accommodation preparation (S16) in oil type change operation.
Fig. 5 is a diagram showing details of a change in the charge pump flow rate control setting (S18) in the oil type change operation.
6 is a diagram showing details of a crude oil tank change (S20) in an oil type change operation.
Fig. 7 is a diagram showing details of flow rate adjustment (S22) toward a heating furnace in an oil type change operation.
Fig. 8 is a diagram showing the details of fine adjustment (S24) of the flow rate of an intermediate oil in the oil type change operation.
9 is a diagram showing details of overhead temperature adjustment and other points to be noted in the oil type change operation (S28).
Fig. 10 is a diagram showing details of flow rate adjustment (S30) of each reflux in oil type change operation.
11 is a diagram showing a configuration of a learning device according to an embodiment.
12 is a diagram showing a configuration of a driving condition setting support device and a control device according to an embodiment.

The oil refinery operation condition setting support system according to the embodiment supports the setting of the operation conditions of the oil refinery. In the present embodiment, in particular, a case where the oil refinery supports setting of operating conditions at the time of oil type change will be described. The oil refinery operation condition setting support system manually manages the sequence of oil type change operations that have been performed by skilled operators in the past, and information necessary to properly proceed with the oil type change operation, the amount of control to be adjusted, and the amount of attention to be paid attention to. By presenting points and the like to the driver, the setting of driving conditions by the driver is supported. In addition, the oil station operation condition setting support system, according to a plurality of state values indicating the states of the plurality of devices provided in the oil station, the recommended values of the plurality of control amounts for controlling the plurality of devices, the measures learned by machine learning. By calculating using (function) and presenting the calculated recommended value to the operator, the setting of driving conditions by the operator is supported.

1 schematically shows the configuration of a refinery 3. The crude oil stored in the crude oil tank 10a and the crude oil tank 10b is pumped from the crude oil tank 10a and the crude oil tank 10b by the filling pump 11, and is then exchanged with the reflux oil of the distillation column 15. It is preheated, mixed with the injected water, and introduced into a desalting device (desalter) 12. On the other hand, in FIG. 1, a charge pump 11 is provided at each outlet of the crude oil tank 10a and the crude oil tank 10b, but in another example, crude oil may be pumped from a plurality of crude oil tanks by a common charge pump. . In the desalting device 12, impurities such as moisture, salt, iron and soil contained in crude oil are removed as wastewater. Crude oil that has passed through the desalination unit 12 is further heated by heat exchange with each oil extracted from the distillation column 15 and bottom oil, and introduced into a pre-flash drum 13 do. Of the crude oil introduced into the pre-flash drum 13, the evaporated low boiling point fraction is directly introduced into the distillation column 15, and the high boiling point fraction of the liquid is heated in the heating furnace 14 and then introduced into the distillation column 15. The load on the heating furnace 14 can be reduced by introducing the low-boiling-point fraction contained in the preheated crude oil into the distillation column 15 in advance. On the other hand, the preflash drum 13 may not be provided, and in that case, all crude oil may be heated in the heating furnace 14 and then introduced into the distillation column 15. In addition, a sub-distillation column may be provided instead of the pre-flash drum 13, and in that case, the fraction classified by the sub-distillation column may not be introduced into the distillation column 15 and may be separated as a product.

In the distillation column 15, crude oil is separated into a plurality of fractions having different boiling points. Each fraction extracted from the distillation column 15 is introduced into the stripper 16. In the stripper 16, each fraction is brought into contact with superheated steam to adjust the flash point, and the low-boiling fraction is refluxed to the distillation column 15. Each fraction that has passed through the stripper 16 is cooled by the crude oil before distillation in a heat exchanger, and becomes each fraction of kerosene, light gas oil, and heavy gas oil. The low-boiling fraction extracted from the overhead of the distillation column 15 is temporarily stored in an overhead accumulator 17, and the gas component becomes a liquefied petroleum gas raw material, or introduced into a gas recovery device, and a liquid component Becomes gasoline. The bottom oil extracted from the bottom of the distillation column 15 is cooled by the crude oil before distillation in a heat exchanger, and becomes atmospheric residual oil.

When the crude oil processed by the distillation column 15 is changed from crude oil stored in the crude oil tank 10a to other crude oil from the crude oil stored in the crude oil tank 10b, the desalination device 12, The operating conditions of each device such as the pre-flash drum 13, the heating furnace 14, the distillation column 15, the stripper 16, and the overhead accumulator 17 can be rapidly changed, as well as from the distillation column 15. The composition and flow rate of each oil extracted may fluctuate. In a conventional oil refinery, in order to ensure that each oil component produced during the oil type change operation does not deviate from the required specifications, each oil component is extracted from the distillation column 15 until the oil type change operation is completed and the operation transitions to a stable normal operation. The flow rate was suppressed to be lower than the normal operation. Therefore, during the oil type change operation, more crude oil is extracted from the distillation column 15 as the bottom oil, and the yield of atmospheric residual oil, which is a lower oil fraction, increases, and the yield of gasoline or kerosene, which are more valuable oils. It was degraded. Therefore, in order to improve the production efficiency in the refinery (3), even during the oil type change operation, the yield of more valuable oil is increased, the amount of tower oil is reduced, and the transition from the oil type change operation to the normal operation is required. There was a need for a technology that could shorten the time it took.

The oil refinery operation condition setting support system of this embodiment properly manages each process of the oil type change operation according to the sequence of oil type change operation that has been executed by an experienced operator, and supports the setting of operating conditions by the operator. . Thereby, since the oil type change operation can be optimized and leveled to a high level without depending on the operator's experience or skill, it is possible to improve the yield of oil having a high value during the oil type change operation, and in the oil type change operation. The time it takes to transition to normal operation can be shortened. Thereby, it is possible to improve the production efficiency and profits in the oil refinery.

2 shows the overall configuration of a system for supporting setting operating conditions of an oil refinery according to an embodiment. The oil refinery operation condition setting support system 1 is for learning the oil refinery 3 for refining crude oil to produce petroleum products, and the measures used to support the setting of operating conditions in the oil refinery 3 A learning device 40 is provided. The milking station 3 and the learning device 40 are connected by an arbitrary communication network 2 such as the Internet or an in-house connection system, and are operated in an arbitrary form of operation such as on-premises, cloud, and edge computing.

The oil refinery 3 includes a control target device 5 such as an atmospheric distillation column or a heating furnace installed in the oil refinery 3, and a control device 20 that sets a control amount for controlling the operating conditions of the control target device 5. ), and a driving condition setting support device 30 that supports setting of the driving conditions of the oil production station 3 using the measures learned by the learning device 40. The driving condition setting support device 30 manages the sequence of the oil type change operation, and informs the operator of the information necessary to properly proceed the oil type change operation, the state value to be stopped, the amount of control to be adjusted, points to be noted, etc. present. In addition, the driving condition setting support device 30 determines the recommended target set values of the plurality of control amounts according to the plurality of state values representing the states of the plurality of control target devices 5, and the measures learned by machine learning. And present it to the operator.

3 shows the procedure of the oil type change operation. The oil type change operation includes a design process (S1), a preliminary preparation process (S2), a change process (S3), and an adjustment process (S4). In the design process (S1), first, the product yield obtained from the changed crude oil is set based on information such as where the crude oil is collected, etc. stored in the crude oil tank 10, and the moisture content is confirmed by a sample test or the like ( S10), each flow rate of a plurality of oils extracted from the distillation column 15 is set (S12). The moisture content and the flow rate of oil may be estimated using any known technique such as a linear programming (LP) model.

Next, in the preliminary preparation step (S2), the time step of the oil type change operation is set (S14), and based on the water content of the crude oil after the change or the flow rate of the oil, the desalination device 12 and the preflash drum 13 , The heating furnace 14, the distillation column 15, the stripper 16, the overhead accumulator 17, etc., while preparing in advance for receiving the crude oil after the change (S16), the charge pump 11 The setting of the flow rate control of is changed from automatic to manual (S18).

Next, in the change process (S3), the crude oil tank 10 from which crude oil is extracted by the filling pump 11 is changed, and the receiving of the crude oil after the change is started (S20).

Next, in the adjustment process (S4), the operating conditions for distilling the crude oil after the change are adjusted. In the adjustment process (S4), while prioritizing preventing the off-spec of the intermediate oil, the flow rate adjustment toward the heating furnace 14 (S22), fine adjustment of the flow rate of the intermediate oil (kerosene, diesel, heavy diesel) ( S24), quality adjustment of intermediate oil (S26), overhead temperature adjustment and confirmation of other points of interest (S28), flow adjustment of each reflux (S30), crude oil preheating flow balance adjustment (S32), each setting is an operation indicator It includes the confirmation of whether or not it is internal (S34), and the quality of the intermediate oil (S36) (the order is not the same). When the intermediate oil satisfies the predetermined quality (Y in S36), the oil type change operation is completed and the normal operation proceeds. When the intermediate fraction does not meet the predetermined quality (N in S36), the process returns to S24, and the quality adjustment of the intermediate fraction is continued. These steps are executed in the time step set in S14.

Fig. 4 shows the details of pre-accommodation preparation (S16) in the oil type change operation. In the pre-accommodation preparation (S16), the flow rate of water injected into the crude oil before introducing the crude oil into the desalination device 12, based on the water content of the crude oil after the change or the flow rate of the oil, the liquid level of the pre-flash drum 13 , The flow rate of crude oil introduced from the heating furnace 14 to the distillation column 15, the liquid level of the overhead accumulator 17, and the flow rate of each oil component of kerosene, light diesel, and heavy light oil are adjusted. For example, when it is estimated that the amount of water contained in crude oil after change is greater than before change, the flow rate of water injected into crude oil is made smaller in advance. In addition, when it is estimated that the composition ratio of the low-boiling-point oil contained in the crude oil after the change is higher than that of the crude oil before the change, the liquid level is expected to decrease by evaporating more low-boiling-point oils than before the change in the pre-flash drum 13. Put it. On the other hand, after the crude oil is unloaded from the tanker into the crude oil tank, it is allowed to stand in the crude oil tank for a sufficient period of time to separate the distilled water. Unlike the composition ratio of each oil component, it can be adjusted in advance. When the moisture content of crude oil is previously reduced to a level that does not affect the operation, it is not necessary to take into account the moisture content of the crude oil after the change in pre-accommodation preparation (S16).

5 shows the details of the charge pump flow rate control setting change (S18) in the oil type change operation. When the filling pump 11 to be operated is changed and the crude oil tank 10 for extracting crude oil is changed, the pump discharge pressure rapidly rises due to the change of the liquid level in the crude oil tank 10 from a low to a high place, resulting in a flow rate. Since this may fluctuate, the function of automatically controlling the flow rate of the charge pump 11 is temporarily changed to off, so that subtle adjustments can be made manually.

6 shows the details of the crude oil tank change S20 in the oil type change operation. When the charge pump 11 to be operated is changed and the extraction of crude oil from the crude oil tank 10 storing the changed crude oil is started, the flow rate of the charge pump 11 and the desalination device 12 from the crude oil tank 10 After the lapse of a time that is roughly determined based on the length of the pipe up to) and the like, the modified crude oil reaches the desalting device 12. The flow rate of water injected into the crude oil introduced into the desalination device 12 is adjusted in advance in S16 based on the moisture amount estimated in S10, but the moisture remaining in the crude oil tank 10 or mixed into crude oil during transportation of crude oil. Since there is a possibility that the crude oil contains a different amount of water than the estimated amount of water, etc., the amount of water in the desalting device 12 is checked, and adjustment is made as necessary. When the amount of water contained in crude oil is large, for example, an overcurrent occurs in the desalination device 12, the interface is not clearly visible, the liquid level and the temperature of the preflash drum 13 fluctuate together, or the heating furnace 14 Since the inlet temperature decreases or the pressure of the distillation column 15 increases, these state values are checked. When it is confirmed that the amount of water contained in the crude oil is large, the amount of water to be injected into the crude oil may be decreased, the amount of the antifoaming agent injected may be increased, or the amount of superheated steam introduced into the distillation column 15 may be decreased. As described above, when it is confirmed in advance that the moisture content of crude oil has been reduced to a level that does not affect the operation, there are cases where it is not necessary to perform adjustment regarding the moisture content.

7 shows the details of flow rate adjustment (S22) toward the heating furnace in the oil type change operation. After the changed crude oil reaches the desalination device 12, the passage of time approximately determined based on the length or flow rate of the pipe from the desalination device 12 to the preflash drum 13 and the heating furnace 14, etc. Later, since the changed crude oil reaches the preflash drum 13 and the heating furnace 14, the liquid level of the preflash drum 13 is checked, and the flow rate toward the heating furnace 14 is adjusted as necessary. .

Fig. 8 shows the details of fine adjustment (S24) of the flow rate of the intermediate oil in the oil type change operation. By the process up to S22, preparation for introducing the modified crude oil into the distillation column 15 is prepared, and the apparatus upstream of the distillation column 15 can be expected to be basically in a state in which automatic operation is possible. , Fine adjustment is made so that the withdrawal amount of the oil produced from the crude oil after the change is optimal within the specifications required for each oil. First, the pump flow rate for extracting each intermediate oil from the stripper 16 is adjusted. The flow rate of naphtha extracted from the overhead is adjusted according to the liquid level of the overhead accumulator 17. The flow rates of kerosene, light diesel, and heavy diesel extracted from the stripper 16 are gradually adjusted to the optimum flow rate while checking the quality of each intermediate oil. The quality of each intermediate fraction is analyzed by an online analysis system or the like.

9 shows details of overhead temperature adjustment and other points to be noted in the oil type change operation (S28). When the overhead temperature falls below a predetermined value, acidic substances are condensed in the overhead, and corrosion of the device material may occur due to the acidic substances, so that the overhead temperature is not lower than the predetermined value, it is removed from the overhead. The flow rate of the oil inside, the flow rate of the reflux to the overhead, and the like are adjusted. In addition, it checks whether conditions such as design temperature, pressure limit, and flow rate limit of each device are satisfied, and adjusts as necessary.

10 shows the details of the flow rate adjustment (S30) of each reflux in the oil type change operation. It is introduced from the distillation column 15 to the heat exchanger to heat the crude oil before distillation, and the flow rate of a plurality of reflux that is cooled and refluxed to the distillation column 15 is calculated as the temperature of each tray in the distillation column 15, and the temperature of the preheated crude oil. It adjusts according to etc., and the temperature distribution of the distillation column 15 is made an optimum, and energy saving is also aimed at.

In the oil type change operation according to the above procedure, the setting of the control amount for adjusting the flow rate or the liquid level may be performed manually by the operator, but in the oil station operation condition setting support system 1 of the present embodiment, the oil station ( In order to further improve the production efficiency in 3), the recommended value of each control amount in the oil type change operation is calculated by using the method learned by machine learning.

11 shows a configuration of a learning device according to an embodiment. The learning device 40 includes a state value acquisition unit 41, an action determination unit 42, a reward value acquisition unit 43, an action value function update unit 44, a neural network 45, and a learning control unit 46. , A simulator 47, a driving data acquisition unit 48, a driving data holding unit 49, and a simulator learning unit 50. These configurations are realized by a CPU, a memory, a program loaded in the memory, etc. of an arbitrary computer, speaking of hardware components, but functional blocks realized by their association are shown here. Therefore, it is understood by those skilled in the art that these functional blocks can be realized in various forms by only hardware, software, or a combination thereof.

The operation data acquisition unit 48 includes a state value indicating the state of each control target device 5 when the oil station 3 is operated, a target set value of the control amount set by each control device 20, and the oil station (3) Measured values indicating the environment, conditions, and the like are acquired from the oil production station 3 as operation data, and stored in the operation data holding unit 49.

The simulator learning unit 50 learns a simulator 47 that simulates the behavior of the milking station 3 by machine learning. The simulator learning unit 50 refers to the driving data stored in the driving data holding unit 49 as teacher data, and learns the difference from the simulator 47. The simulator 47 may simulate the operation behavior of the entire oil refinery 3, and the desalting device 12, the pre-flash drum 13, the heating furnace 14, the distillation column 15, and the stripper 16 , The overhead accumulator 17 or the like may be a combination of simulating the respective driving behavior of the device. When the simulator 47 is configured by a combination of a plurality of simulators that simulate each control target device 5, the simulator learning unit 50 first learns each of the plurality of simulators, and the accuracy of each simulator You may learn the whole simulator 47 which combines a plurality of simulators after individually improving it. By learning the simulator 47 using the operation data of the oil refinery 3 in the past, it is possible to adjust the universally created simulator according to the environment and configuration of the oil refinery 3, so that it can be used in the simulator. It is possible to improve the accuracy of the estimation.

The learning control unit 46 provides a method for calculating the recommended value of the control amount to be set in each control target device 5 in the driving condition setting support device 30 in the oil type change operation, for in-depth reinforcement learning. Obtained by

In reinforcement learning, an agent placed in an environment acts on the environment and seeks a measure to maximize the reward earned by that action. The action value function and measures to maximize the expectation of the sum of rewards obtained by repeating the steps of the agent inducing an action against the environment, the environment updating the state and evaluating the action, and informing the agent of the state and reward. Optimize. More specifically, the action determination unit 42 determines a target set value of the control amount for controlling the oil type change operation in the oil refinery 3, and the state value acquisition unit 41 sets the determined target. A plurality of state values indicating a state after a predetermined period of time of the oil refinery 3 in which the value is set and operated is acquired, the step value acquisition unit 43 acquires the step value for the state, and the action value function update unit (44) optimizes the action value function and measures based on the obtained payoff value.

In this embodiment, the target of the control amount to the plurality of control target devices 5 in the state (s) of the oil production station 3 defined by the state values of the plurality of control target devices 5 and the state (s). Since the combination of the action (a) options for inputting a set value becomes a vast number, deep reinforcement learning is performed in which the action value function is approximated by the neural network 45. The algorithm for deep reinforcement learning may be DQN (Deep Q-Learning Network), DDQN (Double DQN), or any other algorithm. The neural network 45 may be a feedforward neural network (FFNN) such as a multilayer perceptron neural network, a simple perceptron neural network, and a convolutional neural network (CNN), or may be a neural network of any other form. In the input layer of the neural network 45, all state values representing the states of all control target devices 5 are input, and from the output layer, the values of target set values of all control quantities input to all control target devices 5 Is displayed.

The learning control unit 46 determines the policy and content of learning, and performs deep reinforcement learning. In this embodiment, the learning control unit 46 uses the past operation data in the milking station 3 stored in the operation data holding part 49 to change the oil type performed in the past in the milking station 3 A driving performance learning mode that learns measures from driving behavior and a virtual driving learning mode that learns measures from the behavior of oil type change driving simulated under unknown driving conditions are controlled using the simulator 47.

The learning control unit 46 sets initial conditions such as oil type and moisture content before and after the change to start the trial, and determines the target set value of the control amount, and the refinery controlled by using the target set value of the determined control amount. Acquisition of a plurality of status values indicating the state after a predetermined time of the oil type change operation is performed according to the sequence of the oil type change operation described above, and when the order of the oil type change operation is completed, one trial is terminated. Start. The learning control unit 46 executes before completing the sequence of the oil type change operation when a predetermined condition is satisfied as it is clear that the execution during execution does not result in a good result, such as that the obtained reward value is less than a predetermined value. May be terminated and the next trial may be started.

In the driving performance learning mode, the learning control unit 46 sets a target set value actually set by the operator in the past and the target set value according to the past driving data stored in the driving data holding unit 49 It is repeated to acquire a plurality of state values after being actually operated. That is, the action determination unit 42 determines the setting of the target set value actually set by the driver in the past as the next action, according to the driving data stored in the driving data holding unit 49, and the state value acquisition unit 41 ) Acquires a plurality of state values stored in the operation data holding unit 49 as state values indicating the state of each control target device 5 after the target set value is set. Since the enforcement proceeds according to the driving data stored in the driving data holding unit 49, the learning may proceed without going through the action determining unit 42. The reward value acquisition unit 43 acquires a reward value for the state of the oil refinery 3 presented by the past operation data, and the action value function update unit 44 sends the reward value acquisition unit 43 to the The action value function expressed by the neural network 45 is updated on the basis of the reward value obtained thereby. Thereby, whether or not the control by the operator in the oil type change operation actually executed in the past is good or not can be reflected in the action value function expressed by the neural network 45. Details of the calculation of the payoff value and the update of the action value function will be described later.

In the virtual driving learning mode, the learning control unit 46 sets a target set value by the action determination unit 42 and a plurality of state values after a predetermined time simulated by the simulator 47 in which the target set value is set. Repeat the acquisition of. The action determination unit 42 determines target set values of a plurality of control amounts input to the simulator 47. The action determination unit 42 determines target set values of a plurality of control amounts at random or based on the action value function expressed by the neural network 45. The action determining unit 42 determines a target set value of the control amount at random according to a known arbitrary algorithm such as the ε-greedy method, or sets a target of the control amount at which the expected value is maximum based on the action value function. You can choose whether to determine the value. Thereby, while implementing a wide variety of options, learning can be carried out efficiently, and the time until learning can be completed can be shortened. Further, the behavior determination unit 42 may select an action that is not selected from the past driving data stored in the driving data holding unit 49. In this way, it is possible to search for behaviors that have not been selected by the operator in the past oil type change operation but can produce good results. The learning control unit 46 may set a state value reflecting the influence of the disturbance at random timing in the simulator 47, and learn an appropriate coping method for the disturbance.

The state value acquisition unit 41 acquires a plurality of state values representing states of the plurality of control target devices 5 from the simulator 47. The step value acquisition unit 43 acquires a step value for the state of the oil manufacturing station 3 presented by a plurality of state values acquired by the state value acquisition unit 41. This maintenance value is a numerical value of whether or not the oil type change operation performed in the oil refinery 3 is satisfactory. More specifically, the maintenance value is (1) the time it takes to reach a predetermined operating state after changing the oil type of the crude oil to be processed, (2) the yield of a plurality of oils, and (3) the amount consumed in the adjustment process. Numericalization is performed based on at least one of the amount of energy, (4) the degree of satisfaction of the operating conditions required in the adjustment step, and (5) the evaluation by the operator for the driving situation in the adjustment step, or a combination thereof. The weight of each of these factors for numerically converting the reward value may be determined according to the operating policy of the oil refinery 3. The reward value may be digitized based on another evaluation factor in place of any one of the above evaluation factors or in addition to the above evaluation factors.

When the step value is numerically converted based on the evaluation factor 5, the evaluation by the operator used in the evaluation factor 5 may be provided from the operator terminal 60 to the learning device 40. The driver terminal 60 includes an evaluation acquisition unit 61 and an evaluation transmission unit 62. The evaluation acquisition unit 61 displays the status of the oil type change operation executed in the oil refinery 3 or the condition of the oil type change operation virtually executed in the simulator 47 of the learning device 40, etc. It is presented to the driver through an input device, and an evaluation of the driving situation is obtained from the driver through an input device. The evaluation transmission unit 62 transmits the evaluation by the operator acquired by the evaluation acquisition unit 61 to the learning device 40 via a communication device or the like. The operator terminal 60 may be realized by the learning device 40, may be realized by the operation condition setting support device 30 or the control device 20 of the oil station 3, and as a device different from them. May be realized.

The action value function update unit 44 updates the action value function expressed by the neural network 45 based on the pay value acquired by the pay value acquisition unit 43. In the action value function update unit 44, the output of the action value function of the set of actions taken by the action determination unit 42 in a certain state (s) is obtained by the action determination unit 42 in a certain state (s). As a result of the action taken, the weight of the neural network 45 is learned so as to approximate the expected value of the sum of the reward value obtained by the reward value acquisition unit 43 and the reward value that will be obtained when the optimal action is continued thereafter. Let it. That is, the action value function update unit 44 includes the sum of the reward value actually obtained by the reward value acquisition section 43, the expected value of the reward value to be obtained after that, multiplied by the time discount, and the output value of the action value function. The weight of each combination of each layer of the neural network 45 is adjusted so as to reduce the error therebetween. Accordingly, the weight is updated so that the action value calculated by the neural network 45 approaches the true value, and learning proceeds.

The learning by the driving performance learning mode and the learning by the virtual driving learning mode may be performed at an arbitrary number of times, in order, or in combination. For example, first, in the driving performance learning mode, learning is carried out using past driving data, and to some extent, whether the setting of the target set value in the past driving is good or not is reflected in the behavior value function. , With the virtual driving learning mode, learning may proceed with a wide range of options in more various driving conditions.

It is also possible to verify whether or not an accurate behavior decision can be made using the learned neural network 45 using past driving data. For example, as in the driving performance learning mode, the behavior determination unit 42 uses the learned neural network 45 in parallel while performing the type change driving according to the driving data stored in the driving data holding unit 49. To decide what to do next. When the behavior determined by the behavior determination unit 42 is different from the past driving performance stored in the driving data holding unit 49, the behavior determination unit 42 determines whether the behavior determined is good or not, based on a reward value obtained thereafter. So that the behavior is not selected by the behavior determination unit 42, or the same behavior as the past driving performance is determined by the behavior determination unit 42 when it is evaluated as being not good behavior, Adjust the neural network 45. Whether the behavior determined by the behavior determination unit 42 is good or not may be evaluated, for example, based on the accumulated value of the reward value for the past driving performance progressed according to the driving data after a predetermined time, and the behavior Subsequent driving behavior in the case of taking the action determined by the determination unit 42 may be estimated by the simulator 47, and further evaluated based on the integrated value of the reward value for the estimated driving behavior until a predetermined time later.

In this figure, for simplicity of explanation, the learning device 40 is shown as a single device, but the learning device 40 may be realized by a plurality of servers using cloud computing technology, distributed processing technology, or the like. . Thereby, the time taken to improve the accuracy of learning can be drastically shortened.

12 shows the configuration of a driving condition setting support device and a control device according to an embodiment. The control device 20 includes a control unit 21 and an operation panel 22.

The operation panel 22 displays various state values indicating the operation state of the oil refinery 3 and target set values of various control amounts set by the control device 20 on the display device, while various control amounts The input of the target set value of is received from the operator.

The control unit 21 includes a state value acquisition unit 23, a state value transmission unit 24, and a set value input unit 25. These functional blocks can also be realized in various forms by using only hardware, only software, or a combination thereof.

The state value acquisition unit 23 acquires various state values representing the operation state and operation result of the oil refinery 3 from various sensors or measuring devices provided in the control target device 5 and the like, and the operation panel 22 ) On the display device. The state value transmission unit 24 transmits the state value acquired by the state value acquisition unit 23 to the driving condition setting support device 30 and the learning device 40. The set value input unit 25 inputs target set values of various control amounts received from the operator through the operation panel 22 to the control target device 5 and displays them on the display device of the operation panel 22. The set value input unit 25 may automatically input the recommended value of the control amount acquired from the driving condition setting support device 30 to the control target device 5.

The driving condition setting support device 30 includes a control unit 31.

The control unit 31 includes a state value receiving unit 32, a recommended value calculation unit 33, a recommended value output unit 34, a policy update unit 35, an information presentation unit 36, and a sequence management unit 37. do. These functional blocks can also be realized in various forms by using only hardware, only software, or a combination thereof.

The state value receiving unit 32 acquires a plurality of state values from the state value transmitting unit 24 of the control device 20. The recommended value calculation unit 33 calculates recommended values of a plurality of control amounts from a plurality of state values received by the state value receiving unit 32 using the measures learned by the learning device 40. The recommended value output unit 34 outputs the recommended values of the plurality of control amounts calculated by the recommended value calculation unit 33 to the operation panel 22 or the set value input unit 25 of the control device 20. The policy update unit 35 acquires the policy relearned by the learning device 40 and updates the recommended value calculation unit 33.

The sequence management unit 37 maintains the sequence of the oil type change operation described above, and during the oil type change operation, the order of the oil type change operation, information necessary to properly proceed each process of the oil type change operation, the amount of control to be adjusted, and attention Points to be done are presented to the information presentation unit 36. The information presentation unit 36 presents the above-described information on the operation panel 22 of the control device 20.

Thereby, it is possible to optimize the setting of the operating conditions in the oil type change operation, which has been operated based on the experience of an experienced operator in the past, and level it to a high level, so that the production in the oil refinery (3) Efficiency can be improved. Further, since the load on the heating furnace 14 can be reduced and the efficiency in the heat exchanger can be improved, energy consumed in the oil refinery 3 can be reduced. In addition, since adjustment or maintenance of the target set value becomes unnecessary, the load of system management and maintenance can be reduced.

In the above, the present invention has been described based on Examples. This embodiment is an illustration, and it is understood by those skilled in the art that various modifications are possible for each combination of constituent elements and treatment processes, and that such modifications are also within the scope of the present invention.

An operation method according to an embodiment of the present invention is an operation method for operating an apparatus for distilling crude oil to produce a plurality of oils, and when the oil type is changed, the amount of water contained in the crude oil after the change or each of the plurality of oils The design process of estimating the flow rate of; A preliminary preparation step of preparing in advance for receiving the changed crude oil based on the moisture content or flow rate of the crude oil after the change; A change process that initiates the acceptance of crude oil after the change; And an adjustment step of adjusting operating conditions for distilling the crude oil after the change. In the adjustment step, the target set value of the control amount for controlling the device is adjusted according to the state value indicating the state of the device for distilling crude oil after the change.

In the preliminary preparation process, based on the oil type, moisture content, or flow rate after the change, the flow rate of water injected into the crude oil before introducing the crude oil into the desalination device, the device for temporarily storing the crude oil, the liquid level, and the heating to heat the crude oil. The flow rate of the crude oil introduced into the distillation column for distilling crude oil from the furnace, the liquid level level of the device for temporarily storing the oil that has flowed out from the distillation column, or the flow rate of a plurality of oil components may be adjusted.

The apparatus for temporarily storing crude oil may include a free flash drum.

The adjustment process includes a process of acquiring a state value; A calculation process of calculating a recommended value of the control amount based on the state value; And presenting the calculated recommended value to the operator, or setting the calculated recommended value as a target set value in the apparatus.

In the calculation process, a recommended value may be calculated using a method learned by machine learning.

The policy may be learned by reinforcement learning.

The measure is the time it takes from changing the oil type of the crude oil to be processed to reach a predetermined operating state, the yield of a plurality of oils, the amount of energy consumed in the adjustment process, the degree of satisfaction of the operating conditions required in the adjustment process, and The learning may be performed by reinforcement learning using a reward value based at least on either one of the evaluations by the operator for the driving situation in the adjustment step or a combination thereof.

The policy may be learned by reinforcement learning using a step value based on a state value when the device was operated in the past and a target set value.

The countermeasure may be learned by reinforcement learning using a step value based on a state value when a target set value is set in a simulator simulating the driving state of the device.

Another embodiment of the present invention is a support device. This apparatus is an apparatus for distilling crude oil to produce a plurality of fractions, comprising: a design step of estimating the amount of water contained in the crude oil after the change or the flow rate of each of the plurality of fractions when the oil type is changed; A preliminary preparation step of preparing in advance for receiving the changed crude oil based on the moisture content or flow rate of the crude oil after the change; A change process that initiates the acceptance of crude oil after the change; And an acquisition unit for acquiring information necessary to proceed with each step included in the operating method when an operating method including an adjustment step of adjusting an operating condition for distilling the changed crude oil is executed; And a presentation unit for presenting the information acquired by the acquisition unit.

In the adjustment step, the state value acquisition unit for acquiring a state value indicating a state of the device; A calculation unit that calculates a recommended value of a control amount for controlling the device using the measures learned by machine learning, based on the state value; And an output unit that presents the calculated recommended value to the operator, or sets the calculated recommended value to the device as a target set value of the control amount.

The measures include the time it takes from changing the oil type of the crude oil to be processed to reach a predetermined operating state, the yield of a plurality of oils obtained by distilling the crude oil, the amount of energy consumed in the adjustment process, and the required in the adjustment process. The learning may be performed by reinforcement learning using a reward value at least based on either the degree of satisfaction of the driving condition and the evaluation by the operator for the driving situation in the adjustment step, or a combination thereof.

Another embodiment of the present invention is a learning device. The device includes: a state value acquisition unit that acquires a state value indicating a state of the device for distilling crude oil; And a learning unit that learns, by machine learning, a method for calculating a recommended value of a control amount for controlling the device when the oil type is changed, based on the state value.

The learning unit includes the time required to reach a predetermined operating state after changing the oil type of the crude oil to be processed, the yield of a plurality of oil fractions obtained by distilling the crude oil, the amount of energy consumed, and the degree of satisfaction of required operating conditions, And reinforcement learning using a reward value based at least on either one of the evaluation by the driver for the driving situation or a combination thereof.

The present invention can be used in an oil refinery operating condition setting support system that supports setting of operating conditions capable of realizing desirable operation of an oil refinery.

1: Oil refinery operation condition setting support system
3: refinery
5: Control target device
10: crude oil tank
11: filling pump
12: desalting device
13: pre-flash drum
14: heating furnace
15: distillation column
16: stripper
17: overhead accumulator
20: control device
21: control unit
22: operation panel
23: status value acquisition unit
24: status value transmitter
25: setting value input section
26: evaluation acquisition unit
27: evaluation transmitter
30: Driving condition setting support device
31: control unit
32: status value receiver
33: recommended value calculation unit
34: recommended value output
35: policy update unit
36: information presentation unit
37: process control unit
40: learning device
41: status value acquisition unit
42: action decision section
43: Reward value acquisition unit
44: action value function update unit
45: neural network
46: learning control
47: simulator
48: operation data acquisition unit
49: operation data holding unit
50: simulator learning department
60: operator terminal
61: evaluation acquisition unit
62: evaluation transmitter

Claims (14)

It is an operating method for operating an apparatus for distilling crude oil to produce a plurality of oils,
A design step of estimating the amount of water contained in the crude oil after the change of oil type or the flow rate of each of the plurality of oils;
A preliminary preparation step of preliminary preparation for accommodating the changed crude oil based on the moisture content of the crude oil after the change or the flow rate;
A change process that initiates the acceptance of crude oil after the change; And
Including an adjustment step of adjusting the operating conditions for distilling the crude oil after the change,
In the adjustment step, a target set value of a control amount for controlling the device is adjusted according to a state value indicating a state of the device for distilling crude oil after the change. The method of claim 1,
In the preliminary preparation step, based on the moisture content of the crude oil after the change or the flow rate, the flow rate of water injected into the crude oil before introducing the crude oil into the desalination device, the liquid level of the device for temporarily storing crude oil, and heating the crude oil The operating method, characterized in that the flow rate of crude oil introduced into the distillation column for distilling crude oil from the heating furnace for, the liquid level of the device for temporarily storing the oils flowing out from the distillation column, or the flow rates of the plurality of oils are adjusted. The method according to claim 1 or 2,
The adjustment process,
Acquiring the state value;
A calculation process of calculating a recommended value of the control amount based on the state value; And
And presenting the calculated recommended value to an operator, or setting the calculated recommended value as a target set value in the device. The method of claim 3,
In the calculation process, the driving method, characterized in that the recommended value is calculated by using a method learned by machine learning. The method of claim 4,
The driving method, characterized in that the measures are learned by reinforcement learning. The method of claim 5,
The measures include the time taken from changing the oil type of the crude oil to be processed until reaching a predetermined operating state, the yield of the plurality of oils, the amount of energy consumed in the adjustment process, and the operating conditions required in the adjustment process. A driving method characterized in that the learning is performed by reinforcement learning using a reward value based at least on a degree of satisfaction and an evaluation by an operator for a driving situation in the adjustment step, or a combination thereof. The method of claim 6,
The driving method, characterized in that the measures are learned by reinforcement learning using a step value based on the state value and the target set value when the device has been operated in the past. The method according to claim 6 or 7,
The driving method, characterized in that the method is learned by reinforcement learning using a step value based on the state value when the target set value is set in a simulator that simulates a driving state of the device. An apparatus for distilling crude oil to produce a plurality of fractions, comprising: a design step of estimating an amount of water contained in the crude oil after the change or flow rates of each of the plurality of fractions when the oil type is changed; A preliminary preparation step of preliminary preparation for accommodating the changed crude oil based on the moisture content of the crude oil after the change or the flow rate; A change process that initiates the acceptance of crude oil after the change; And an acquisition unit for acquiring information necessary to proceed with each step included in the operating method when an operating method including an adjustment step of adjusting an operating condition for distilling the changed crude oil is executed; And
And a presentation unit for presenting the information acquired by the acquisition unit. The method of claim 9,
In the above adjustment process,
A state value acquisition unit that acquires a state value indicating a state of the device;
A calculation unit that calculates a recommended value of a control amount for controlling the device using the measures learned by machine learning, based on the state value; And
And an output unit that presents the calculated recommended value to an operator or sets the calculated recommended value to the device as a target set value of the control amount. The method of claim 10,
The measures include the time taken from changing the oil type of the crude oil to be processed until reaching a predetermined operating state, the yield of a plurality of oils obtained by distilling the crude oil, the amount of energy consumed in the adjustment step, and the adjustment step. A support device, characterized in that the learning is performed by reinforcement learning using a reward value based at least on one or a combination of the degree of satisfaction of the driving conditions required by the operator and the evaluation by the operator for the driving situation in the adjustment step. . A state value acquisition unit that acquires a state value indicating a state of an apparatus for distilling crude oil; And
And a learning unit that learns, by machine learning, a method for calculating a recommended value of a control amount for controlling the device when the oil type is changed, based on the state value. The method of claim 12,
The learning unit includes the time required to reach a predetermined operating state after changing the oil type of the crude oil to be processed, the yield of a plurality of oils obtained by distilling the crude oil, the amount of energy consumed, and the degree of satisfaction of required operating conditions. And a reinforcement learning using a reward value based at least on any one of an evaluation by a driver for a driving situation, or a combination thereof, to learn the method. A support device that supports setting operating conditions of an apparatus for distilling crude oil to manufacture a plurality of oils; And
A learning device for learning the measures used in the support device by machine learning,
The learning device,
An acquisition unit that acquires a state value indicating a state of the device; And
A learning unit that learns, by machine learning, a method for calculating a recommended value of a control amount for controlling the device at the time of oil type change, based on the state value,
The support device,
A state value acquisition unit that acquires a state value indicating a state of the device when the oil type is changed;
A calculation unit that calculates a recommended value of a control amount for controlling the device using the measures learned by the learning device, based on the state value; And
And an output unit configured to present the calculated recommended value to an operator or to set the calculated recommended value to the device as a target set value of the control amount.

Images