Control device, server, management system, computer program, learning model, and control method

Abstract

To provide the control device, the server, the control server, the computer program, the learning model and the control method, that can reuse combustible refuse as an industrial raw material with high efficiency.SOLUTION: The control device comprises: a gas information acquisition part that acquires gas information of gas converted by a gasification furnace that converts collected refuse into the gas; a control information acquisition part that acquires control information for controlling a gas purification device that purifies the gas converted by the gasification furnace; a characteristic information acquisition part that acquires characteristic information including information on purified gas purified by the gas purification device; and a generation part that generates a learning model by machine learning based on the gas information, the control information and the characteristic information.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a control device, a server, management System , a computer program, a learning model, and a control method.

Background Art

[0002] Although the energy equivalent amount of combustible waste discharged in Japan is larger than that of fossil resources used to produce plastic materials, many combustible wastes are incinerated or landfilled.

[0003] Patent Document 1 discloses a waste treatment facility that can reduce the amount of landfill by subjecting magnetic substances separated from incineration ash discharged from a waste incinerator to magnetic separation and subjecting the separated magnetic substances to a reduction metallization process to reduce the iron oxide that has conventionally been landfilled.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in a waste treatment facility such as that of Patent Document 1, the reuse of combustible waste remains partial. In addition, combustible waste is miscellaneous and heterogeneous, and the variation in the components and composition contained therein is large, so it has been difficult to reuse it as an industrial raw material.

[0006] The present invention has been made in view of such circumstances, and an object thereof is to provide a control device, a server, management System , a computer program, a learning model, and a control method capable of efficiently reusing combustible waste as an industrial raw material.

Means for Solving the Problems

[0007] The control device according to an embodiment of the present invention is a control device that controls a gas purification device, and includes a gas information acquisition unit that acquires gas information of gas converted by a gasification furnace that converts collected garbage into gas, a control information acquisition unit that acquires control information for controlling the gas purification device that purifies the gas converted by the gasification furnace, a characteristic information acquisition unit that acquires characteristic information including information on purified gas purified by the gas purification device, and a generation unit that generates a learning model by machine learning based on the gas information, control information, and characteristic information.

[0008] The control device according to an embodiment of the present invention is a control device that controls a gas purification device, and includes a learning model learned based on gas information of gas converted by a gasification furnace that converts collected garbage into gas, control information for controlling the gas purification device that purifies the gas converted by the gasification furnace, and characteristic information including information on purified gas purified by the gas purification device, a gas information acquisition unit that acquires gas information of gas converted by the gasification furnace, and an output unit that inputs the gas information acquired by the gas information acquisition unit into the learning model and outputs control information for controlling the gas purification device.

[0009] The server according to an embodiment of the present invention includes a collection unit that collects identification information for identifying a plant, gas information of gas converted by a gasification furnace that converts collected garbage into gas, control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, and characteristic information including information on purified gas purified by the gas purification device from each of a plurality of garbage treatment plants, and a storage unit that stores the gas information, control information, and characteristic information collected by the collection unit in association with the identification information.

[0010] The management server according to an embodiment of the present invention includes a collection unit that collects the degree of deterioration of an adsorption device in a gas purification device that purifies gas converted by a gasification furnace that converts collected garbage into gas from each of a plurality of garbage treatment plants, and a storage unit that stores the degree of deterioration collected by the collection unit in association with the identification information.

[0011] A computer program according to an embodiment of the present invention causes a computer to execute a process of acquiring gas information of gas converted by a gasifier that converts collected garbage into gas, a process of acquiring control information for controlling a gas purification device that purifies the gas converted by the gasifier, a process of acquiring characteristic information including information on purified gas purified by the gas purification device, and a process of generating a learning model by machine learning based on the gas information, control information, and characteristic information.

[0012] A computer program according to an embodiment of the present invention causes a computer to execute a process of inputting acquired gas information into a learning model learned based on gas information of gas converted by a gasifier that converts collected garbage into gas, control information for controlling a gas purification device that purifies the gas converted by the gasifier, and characteristic information including information on purified gas purified by the gas purification device, and outputting control information for controlling the gas purification device.

[0013] A learning model according to an embodiment of the present invention is learned based on gas information of gas converted by a gasifier that converts collected garbage into gas, control information for controlling a gas purification device that purifies the gas converted by the gasifier, and characteristic information including information on purified gas purified by the gas purification device.

[0014] A control method according to an embodiment of the present invention is a control method for controlling a gas purification device, and includes acquiring gas information of gas converted by a gasifier that converts collected garbage into gas, acquiring control information for controlling the gas purification device that purifies the gas converted by the gasifier, acquiring characteristic information including information on purified gas purified by the gas purification device, and generating a learning model by machine learning based on the gas information, control information, and characteristic information.

[0015] The control method according to an embodiment of the present invention is a control method for controlling a gas purification device, which acquires gas information of the gas converted by a gasifier that converts collected garbage into gas, and based on characteristic information including the gas information of the gas converted by the gasifier, control information for controlling the gas purification device that purifies the gas converted by the gasifier, and information on the purified gas purified by the gas purification device, inputs the acquired gas information into a learned learning model, and outputs control information for controlling the gas purification device.

Effect of the Invention

[0016] According to the present invention, combustible garbage can be reused efficiently as an industrial raw material.

Brief Description of the Drawings

[0017]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Mode for Carrying Out the Invention

[0018] Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a schematic diagram showing an example of the configuration of an ethanol production system 100 according to the present embodiment. The ethanol production system is installed, for example, in a waste treatment facility, and includes a gasification furnace 10, a gas purification device 20, an ethanol production device 30, and a control device 50.

[0019] The gasification furnace 10 is a furnace that can steam-roast waste (combustible waste) in a low-oxygen state and decompose it to the molecular level (for example, including carbon monoxide gas and hydrogen gas). The gas purification device 20 can remove and purify the impurity gas contained in the gas converted by the gasification furnace 10 and extract the required gas (for example, carbon monoxide gas and hydrogen gas). The ethanol production device 30 can produce ethanol by using the required gas extracted by the gas purification device 20 with a catalyst (for example, a metal catalyst, a microorganism catalyst, etc.). Ethanol has the same C2 structure as ethylene, which accounts for about 60% of petrochemical products, and can be induced into induced chemical materials such as plastics by converting it into ethylene monomer or butadiene monomer by an existing chemical process. Note that the combustible waste only needs to be combustible, and industrial waste, general waste, agricultural waste, etc. are given as examples, but it is not particularly limited thereto. Also, different from the combustible waste, for example, an organic compound and / or an inorganic compound may be used as a substitute for the waste of the present embodiment, and the organic compound and / or the inorganic compound is converted into a gas containing one or more of CO, CO2, and H2 as the main component of the converted gas. If so, the present invention is applicable.

[0020] The control device 50 includes a control unit 51 that controls the entire device, a gas information acquisition unit 52, a characteristic information acquisition unit 53, a communication unit 54, a storage unit 55, a sensor information acquisition unit 56, a recording medium reading unit 57, a determination unit 58, and a processing unit 60. The processing unit 60 includes a reward calculation unit 61, an action selection unit 62, and an action evaluation unit 63.

[0021] The control unit 51 can be configured by a CPU, a ROM, a RAM, etc.

[0022] The gas information acquisition unit 52 acquires the gas information of the gas converted by the gasification furnace 10. The gas information includes, for example, the concentration of impurity gas (contaminant) taken out from the gasification furnace 10. The impurity gas includes, for example, gases such as hydrogen cyanide, benzene, toluene, ethylbenzene, xylene, and dioxin, but is not limited thereto. Note that the concentration of the impurity gas varies depending on the components and composition of miscellaneous waste.

[0023] The processing unit 60 has a function as a control information acquisition unit and acquires control information for controlling the operation of the gas purification device 20. The details of the control information will be described later.

[0024] The characteristic information acquisition unit 53 acquires characteristic information including information on the purified gas purified by the gas purification device 20. The information on the purified gas includes, for example, the purity of carbon monoxide gas and hydrogen gas. Further, the information on the purified gas may include the purity of carbon dioxide gas, or may include the concentration of impurity gas that could not be removed. Further, the characteristic information includes various activities of a catalyst (which may be a known catalyst, for example, a metal catalyst or a microorganism) that generates ethanol from carbon monoxide gas and hydrogen gas, the purity or amount of ethylene generated by the ethanol generation device 30, etc.

[0025] The communication unit 54 has a function of communicating with the management server 200 and the learning server 300 via the network 1 described later, and can transmit and receive required information. The management server 200 and the learning server 300 will be described later.

[0026] The storage unit 55 is composed of, for example, a hard disk or a flash memory, and can store information acquired from outside the control device 50, information such as the processing results inside the control device 50, and the like.

[0027] The sensor information acquisition unit 56 acquires sensor information from the gas purification device 20. The details of the sensor information will be described later.

[0028] The recording medium reading unit 57 can read a computer program recorded on a recording medium (not shown) that records a computer program defining the processing of the control device 50.

[0029] In the ethanol production device 30, when ethanol is produced using microorganisms as an example of a catalyst, the determination unit 58 determines the activity of the microorganisms based on the state of the microorganisms. The activity includes, for example, the reaction rate and survival rate of the microorganisms. The state of the microorganisms can be monitored in real time from outside the culture solution layer to determine the activity of the microorganisms. Note that the state of the microorganisms may be monitored offline to determine the activity of the microorganisms. Thereby, for example, when the activity of the microorganisms decreases, nutrients can be added to reactivate them, and the ethanol production rate can be maintained at a high level. Note that when a metal catalyst is used as an example of the catalyst, the determination unit 58 may not be provided.

[0030] The processing unit 60 can be configured by combining hardware such as, for example, a CPU (e.g., a multi-processor equipped with a plurality of processor cores), a GPU (Graphics Processing Units), a DSP (Digital Signal Processors), an FPGA (Field-Programmable Gate Arrays), and the like. Also, the processing unit 60 may be configured by a virtual machine or a quantum computer, etc. The agent described later is a virtual machine existing on a computer, and the state of the agent is changed by parameters or the like. Also, the processing unit 60 may learn on a computer other than another (control device 50).

[0031] The processing unit 60 has a function as a generation unit and can generate a learning model by machine learning based on gas information, control information, and characteristic information. For machine learning, for example, deep learning, reinforcement learning, deep reinforcement learning, etc. can be used. For example, when using reinforcement learning, the gas information is regarded as the "state", the control information is regarded as the "action", and the "reward" is calculated based on the characteristic information, and the value of the Q-value or Q-function (action value function) can be learned.

[0032] That is, the action selection unit 62 has a function as an action output unit and outputs control information based on the gas information acquired by the gas information acquisition unit 52 and the Q-value or the value of the Q-function (action evaluation information) of the action evaluation unit 63. The action evaluation unit 63 includes the evaluation value of the action in reinforcement learning, specifically, includes the Q-value or the value of the Q-function (action value function). That is, the action selection unit 62 selects and outputs an action from the actions that can be taken in the acquired state based on the evaluation value of the action in the acquired state (gas information).

[0033] The reward calculation unit 61 calculates a reward based on the acquired characteristic information. The calculation of the reward can be such that when the characteristic information is within the required value or range, it is positive (with a reward), and when the characteristic information does not reach the required value or is not within the range, it is 0 (without a reward) or negative (penalty).

[0034] The action selection unit 62 has a function as an update unit and updates the Q-value or the value of the Q-function of the action evaluation unit 63 so that the reward calculated by the reward calculation unit 61 increases. Thereby, even when the concentration of the gas (impurity gas) input to the gas purification device 20 exceeds the threshold due to large fluctuations in the components and composition of the garbage, the learning model can be learned so that the control information can be output such that the characteristic information on the output side of the gas purification device 20 is within the required value or range.

[0035] By using the learned learning model, even when the components and composition of the waste vary, the operation of the gas purification device 20 can be optimized to generate the desired ethanol, so that the combustible waste can be reused as an industrial raw material with high efficiency.

[0036] The control unit 51 can store the Q value or the value of the Q function (action evaluation information) of the updated action evaluation unit 63 in the storage unit 55. By reading out the action evaluation information stored in the storage unit 55, the learned learning model can be reproduced.

[0037] FIG. 2 is a schematic diagram showing an example of the main configuration of the gas purification device 20. The gas purification device 20 includes a gas pipeline that communicates with the output side of the gasification furnace 10 and also communicates with the input side of the ethanol generation device 30. In the middle of the gas pipeline, a buffer tank 21, two adsorption devices 22, 23, a compressor 25, and a buffer tank 24 are interposed from the gasification furnace 10 side. The buffer tanks 21 and 24 are for temporarily storing gas. Electromagnetic valves are provided in the gas pipelines on the inlet side and the outlet side of the adsorption devices 22 and 23.

[0038] The adsorption device 22 houses a gas adsorption member 221, and the adsorption device 23 houses a gas adsorption member 231. Sensor units 27 are provided at required locations of each of the adsorption device 22 and the adsorption device 23. In FIG. 2, for the sake of convenience, the sensor units 27 are shown outside the adsorption devices 22 and 23, but the installation positions of the sensor units 27 are not limited to the example of FIG. 2.

[0039] The adsorption devices 22 and 23 are used alternately one by one. For example, in one cycle time, the adsorption device 22 is used, and in the next cycle time, the adsorption device 23 is used instead of the adsorption device 22. Thereafter, the same switching is repeated. In one cycle time, operations such as the increase and decrease of the pressure in the adsorption device, the desorption and cleaning of the adsorption device (for example, the gas adsorption member) are performed.

[0040] Figure 3 is an explanatory diagram showing the principle of the pressure swing adsorption method. In the figure, the vertical axis represents the adsorption capacity, and the horizontal axis represents the gas pressure. In Figure 3, the adsorption isotherm of the impurity gas and the adsorption isotherm of carbon monoxide gas or hydrogen gas are schematically illustrated. The principle of the pressure swing adsorption (PSA) method is as follows. That is, when the pressure in the adsorption device is raised and lowered, the difference in the adsorption capacity of the impurity gas (the difference between symbols A1 and A2) is larger than the difference in the adsorption capacity of carbon monoxide gas or hydrogen gas (the difference between symbols B1 and B2). As a result, more impurity gas is adsorbed and removed by the gas adsorption member than carbon monoxide gas or hydrogen gas. The carbon monoxide gas and hydrogen gas that were not adsorbed by the gas adsorption member are sent to the ethanol production device 30.

[0041] The operation control unit 26 includes a gas flow rate control unit 261, a temperature adjustment unit 262, a humidity adjustment unit 263, an adsorption device switching unit 264 that switches the operation of the adsorption devices 22 and 23, and a communication unit 265. The communication unit 265 has a communication function and can transmit and receive predetermined information to and from the control device 50.

[0042] The operation control unit 26 has a function as a usage history acquisition unit and acquires the usage history of the adsorption devices 22 and 23 (for example, the gas adsorption members 221 and 231). The usage history includes, for example, the cumulative usage time, the number of cleaning times, and the like.

[0043] The sensor unit 27 is composed of a plurality of sensors of different types and can detect the degree of deterioration of the adsorption devices 22 and 23 (for example, the gas adsorption members 221 and 231). The degree of deterioration can be determined, for example, by the color or dirt on the surfaces of the gas adsorption members 221 and 223 after cleaning the gas adsorption members 221 and 231, the amount of impurities adsorbed at a predetermined cycle time, and the like.

[0044] The sensor unit 27 can detect the presence or absence of a desorption operation of the adsorption devices 22 and 23 (for example, the gas adsorption members 221 and 231). The desorption operation of the adsorption devices 22 and 23 can be, for example, an unintentional desorption operation.

[0045] The communication unit 265 can transmit information such as the usage history, degradation degree, and presence or absence of a desorption operation of the adsorption devices 22 and 23 to the control device 50.

[0046] The gas flow rate control unit 261 controls the gas flow rate based on the control information output by the control device 50.

[0047] The temperature adjustment unit 262 adjusts the gas temperature based on the control information output by the control device 50.

[0048] The humidity adjustment unit 263 adjusts the gas humidity based on the control information output by the control device 50.

[0049] The adsorption device switching unit 264 adjusts the cycle time of the operation switching of the adsorption devices 22 and 23 based on the control information output by the control device 50.

[0050] Next, the learning mode of the processing unit 60 of the control device 50 will be described.

[0051] FIG. 4 is a schematic diagram showing an example of reinforcement learning according to the present embodiment. Reinforcement learning is a machine learning algorithm in which an agent placed in a certain environment takes actions against the environment and seeks a policy (a rule serving as an index when the agent takes actions) that maximizes the obtained reward. In reinforcement learning, the agent is like a learner who takes actions against the environment and is the object of learning. The environment updates the state and gives a reward in response to the actions of the agent. An action is an action that the agent can take against a certain state of the environment. A state is the state of the environment held by the environment. A reward is given to the agent when the agent causes a desirable result to act on the environment. The reward can be, for example, a positive, negative, or 0 value. In the case of a positive value, it is the reward itself; in the case of a negative value, it is a penalty; and in the case of a 0 value, there is no reward. Also, an action evaluation function is a function that determines the evaluation value of an action in a certain state, and can also be represented in a table format like a table. In Q-learning, it is called a Q-function, Q-value, evaluation value, etc. Q-learning is one of the frequently used methods in reinforcement learning. Hereinafter, Q-learning will be described, but the reinforcement learning may alternatively be different from Q-learning.

[0052] In the present embodiment, the gasifier 10, the gas purification device 20, the ethanol production device 30, and the reward calculation unit 61 in the processing unit 60 correspond to the "environment", and the action selection unit 62 and the action evaluation unit 63 correspond to the "agent". The action evaluation unit 63 corresponds to the above-described Q-function and Q-value and corresponds to an action evaluation function (action evaluation information).

[0053] First, when the action selection unit 62 acquires the state s t , based on the action evaluation unit 63, from among the actions that can be taken in the state s t , the action a t with the highest evaluation (for example, the largest value of the Q-function) is selected and output to the gas purification device 20 as control information. The gas purification device 20 performs operation control based on the control information.

[0054] Next, when the action selection unit 62 acquires the state s t+1 , it acquires the state s and at the same time receives the reward r from the reward calculation unit 61t+1 Obtain state s t The time t and state s at which t is obtained t+1 The time (interval) between the time t+1 at which t+1 is obtained can be set as appropriate, for example, it can be 1 second, 10 seconds, 30 seconds, 1 minute, 2 minutes, etc., but is not limited thereto.

[0055] When the gas purification device 20 performs operation control based on action a t (control information), the characteristic information changes. The reward calculation unit 61 calculates the reward r t based on the characteristic information that has changed based on action a t+1 (control information). When the action selection unit 62 acts on the gas purification device 20 to obtain a desirable result, a high-value (positive value) reward is calculated. When the reward is 0, there is no reward, and when the reward is a negative value, it is a penalty. The reward calculation unit 61 can calculate the reward based on at least one of the purity of the carbon monoxide gas and hydrogen gas purified by the gas purification device 20, the purity or amount of ethanol purified by the ethanol production device 30, and the activity of the catalyst in the ethanol production device 30. Thereby, the operation control of the gas purification device 20 can be performed so that the characteristic information falls within the required value or range. Note that the reward may be calculated using the concentration of the impurity gas output by the gas purification device 20. In this case, the higher the concentration of the impurity gas, the greater the penalty can be set.

[0056] The action selection unit 62 updates, based on the obtained state s t+1 and reward r t+1 the value of, for example, the Q-function or the Q-value of the action evaluation unit 63. More specifically, the action selection unit 62 updates the value of the Q-function or the Q-value in the direction of maximizing the reward for the action. Thereby, the action that is expected to have the maximum value in a certain state of the environment can be learned.

[0057] By repeating the above-described processing and repeating the update of the action evaluation unit 63, the action evaluation unit 63 that can maximize the reward can be learned.

[0058] In Q-learning, a table (also referred to as a Q-table) of size (the number of states s × the number of actions a) can be updated. However, when the number of states becomes large as in this embodiment, a method of expressing the Q-function using a neural network can be adopted.

[0059] FIG. 5 is a schematic diagram showing an example of the configuration of the neural network model section of this embodiment. The neural network model section represents the processing section 60 (specifically, the action selection section 62 and the action evaluation section 63). The neural network model section has an input layer 601, an intermediate layer 602, and an output layer 603. The number of input neurons in the input layer 601 can be the number of types of impurity gases. The concentrations of impurity gas G1, impurity gas G2,..., impurity gas Gn are input to the input neurons in the input layer 601. The number of types of impurity gases is, for example, about 400, but is not limited thereto.

[0060] The number of output neurons in the output layer 603 can be the number of action options. In FIG. 5, for the sake of convenience, the number of output neurons in the output layer 603 is set to 2. One output neuron outputs the value of the Q-function when the cycle time is lengthened, and the other output neuron outputs the value of the Q-function when the cycle time is shortened.

[0061] Machine learning (deep reinforcement learning) using the neural network model section can be performed as follows. That is, when the state s is input to the input neurons of the neural network model section t the output neurons output Q(s t , a t ). Here, Q is a function that stores the evaluation of the action a in the state s. The update of the Q-function can be performed by Equation (1).

[0062]

Equation

[0063] In Equation (1), s trepresents the state at time t, and a t represents the state s t represents the actions that can be taken in state s. α represents the learning rate (where 0 < α < 1), and γ represents the discount rate (where 0 < γ < 1). The learning rate α is also referred to as the learning coefficient and is a parameter that determines the learning speed (step size). That is, the learning rate α is a parameter that adjusts the update amount of the Q-value or the value of the Q-function. The discount rate γ is a parameter that determines how much to discount the evaluation of future states (rewards or penalties) when updating the Q-function. That is, when the evaluation in a certain state is related to the evaluation in past states, it is a parameter that determines how much to discount rewards and penalties.

[0064] In Equation (1), r t+1 is the reward obtained as a result of the action. When no reward is obtained, it becomes 0, and in the case of a penalty, it becomes a negative value. In Q-learning, the second term of Equation (1), {r t+1 +γ·maxQ(s t+1 ,a t+1 ) - Q(s t ,a t )}, is made to be 0. That is, the Q(s t ,a t ) of the Q-function is such that the neural network model part's parameters are learned so that it becomes the sum of the reward (r t+1 ) and the maximum value among the actions possible in the next state s t+1 (γ·maxQ(s t+1 ,a t+1 ). The parameters of the neural network model part are updated so that the error between the expected value of the reward and the current action evaluation approaches 0. In other words, the value of (γ·maxQ(s t+1 ,a t+1 )) is corrected based on the value of the current Q(s t ,a t ) and the maximum evaluation value obtained among the actions executable in the state s t after executing the action a t+1 .

[0065] When an action is executed in a certain state, a reward is not necessarily obtained. For example, a reward may be obtained after repeating the action several times. Equation (2) represents the update equation of the Q function when a reward is obtained in Equation (1), avoiding the problem of divergence. Equation (3) represents the update equation of the Q function when no reward is obtained in Equation (1).

[0066] In the example of FIG. 5, the number of output neurons was 2, but it is not limited to this.

[0067] FIG. 6 shows an example of action a t is an explanatory diagram. As shown in FIG. 6, when action a t is the control of the cycle time (the cycle time of switching between the suction device 22 and the suction device 23), specifically, actions such as increasing the cycle time, decreasing the cycle time, or not changing the cycle time can be used. Here, how much to increase or decrease the cycle time can be set as appropriate. When action a t is the control of the gas temperature, specifically, actions such as increasing the temperature, decreasing the temperature, or not changing the temperature can be used. Here, how much to increase or decrease the temperature can be set as appropriate. When action a t is the control of the gas amount, specifically, actions such as increasing the gas amount, decreasing the gas amount, or not changing the gas amount can be used. Here, how much to increase or decrease the gas amount can be set as appropriate. Also, when action a t is the control of the gas humidity, specifically, actions such as increasing the humidity, decreasing the humidity, or not changing the humidity can be used. Here, how much to increase or decrease the humidity can be set as appropriate. The output neuron can be configured to output the Q function by combining all or part of the actions illustrated in FIG. 6. Although not shown, the gas pressure may be included in action a t , for example, actions such as increasing the gas pressure, decreasing the gas pressure, or not changing the gas pressure can be included.

[0068] FIG. 7 is a schematic diagram showing another example of the configuration of the neural network model unit of the present embodiment. The difference from the neural network model unit illustrated in FIG. 5 is that the number of output neurons is increased instead of being 2. In the example of FIG. 7, different types of actions are combined. For example, as shown in FIG. 7, the output neurons are the value of the Q-function when doing nothing, the value of the Q-function when increasing the cycle time and increasing the gas amount, the value of the Q-function when increasing the cycle time and decreasing the gas amount,..., the value of the Q-function when decreasing the cycle time, increasing the gas amount, and raising the gas temperature,..., the value of the Q-function when increasing the cycle time and adding a nutrient to the microorganism, and so on. Note that the number of output neurons and the types of outputs are not limited to the example of FIG. 7.

[0069] Note that, as the neural network model unit shown in FIGS. 5 and 7, a so-called convolutional neural network (CNN) may be used.

[0070] Next, the operation control mode of the gas purification apparatus 20 by the control apparatus 50 will be described.

[0071] The processing unit 60 (specifically, the action selection unit 62 and the action evaluation unit 63 as learning models) is learned based on the characteristic information including the gas information of the gas converted by the gasifier 10, the control information for controlling the gas purification apparatus 20, and the information of the purified gas purified by the gas purification apparatus 20.

[0072] The processing unit 60 acquires the gas information of the gas converted by the gasifier 10.

[0073] The processing unit 60 inputs the gas information into the learning model (action selection unit 62 and action evaluation unit 63) and outputs the control information for controlling the gas purification apparatus 20.

[0074] The control unit 51 can control the gas purification device 20 based on the control information output by the processing unit 60. As a result, even when the components and composition of the waste vary, the operation of the gas purification device 20 can be optimized to produce the desired ethanol, enabling the efficient reuse of combustible waste as an industrial raw material.

[0075] FIG. 8 is a schematic diagram showing an example of the concentration of impurity gas when the operation of the gas purification device 20 is controlled by the control device 50. The left figure shows the impurity gas input to the gas purification device 20, and the right figure shows the impurity gas output from the gas purification device 20. In the figure, the vertical axis represents the gas concentration, and the horizontal axis represents time. As shown in the left figure, each time the collected combustible waste is put into the gasification furnace 10 (for example, about once every 1 minute to 30 minutes), the concentration of the impurity gas fluctuates due to the variation in the components and composition of the waste, and may exceed the threshold value. When the impurity gas with a concentration exceeding the threshold value is taken out from the gas purification device 20 and input to the ethanol production device 30, for example, the purity of the produced ethanol decreases.

[0076] In this embodiment, since the operation of the gas purification device 20 is controlled using a learned learning model, as shown in the right figure, the concentration of the impurity gas becomes less than the threshold value, and it is possible to prevent the impurity gas with a concentration exceeding the threshold value from being input to the ethanol production device 30.

[0077] The control unit 51 can store, in the storage unit 55, characteristic information including information on the purified gas purified by the gas purification device 20 in the operation control mode of the gas purification device 20. As a result, when the components and composition of the waste vary, the characteristic information obtained as a result of optimizing the operation of the gas purification device 20 can be collected.

[0078] The control unit 51 can transmit the acquired gas information, the output control information, and the acquired characteristic information to the learning server 300 described later via the communication unit 54 in the operation control mode of the gas purification device 20.

[0079] In the operation control mode of the gas purification device 20, the processing unit 60 can re-train the learning model based on the acquired gas information, the output control information, and the acquired characteristic information. Thereby, the operation of the gas purification device 20 can be further optimized.

[0080] The control unit 51 can transmit information such as the usage history, the degree of deterioration, and the presence or absence of a desorption operation of the adsorption devices 22 and 23 acquired from the gas purification device 20 to the management server 200 described later via the communication unit 54.

[0081] Next, the processing in the learning mode of the present embodiment will be described.

[0082] FIG. 9 is a flowchart showing an example of the processing procedure of machine learning according to the present embodiment. For convenience, the main body of the processing will be described as the processing unit 60. The processing unit 60 sets the parameters of the neural network model unit to initial values (S11). The processing unit 60 acquires the state s t (S12). The state s t is the gas information of the gas converted by the gasification furnace 10, and specifically, is the concentration of impurity gas.

[0083] The processing unit 60 selects and executes an action a t that can be taken in the state s (S13). The action a t is control information for the operation control of the gas purification device 20, and specifically, can combine all or part of the cycle time, the gas amount, the gas temperature, and the gas humidity. Further, the action a t may include an action related to the nutrient given to the microorganism (catalyst). t The processing unit 60 acquires the state s

[0084] resulting from the action a t (S14), and the reward r t+1 (S15), and the reward r t+1Obtain it (S15). The reward can be calculated based on the characteristic information. Here, the characteristic information can include at least one of the information of the purified gas purified by the gas purification device 20 (for example, the purity of carbon monoxide gas and hydrogen gas), the purity or amount of ethanol purified by the ethanol production device 30, and the activity of microorganisms when using microorganisms as an example of the catalyst. Note that the reward may be 0 (no reward).

[0085] The processing unit 60 uses the above formula (1) to make the value of the Q function at the current time (s t , a t ) be the maximum reward obtained among the actions executable in the state s t+1 and learn (update) the parameters of the neural network model unit (S16).

[0086] The processing unit 60 determines whether to end the processing (S17). Here, whether to end the processing may be determined by whether the parameters of the neural network model unit have been updated a predetermined number of times, or it can be determined by whether the characteristic information has reached the allowable value or within the allowable range, etc.

[0087] When the processing is not ended (NO in S17), the processing unit 60 sets the state s t+1 as the state s t (S18) and continues the processing after step S13. When the processing is ended (YES in S17), the processing unit 60 stores the parameters of the neural network model unit in the storage unit 55 (S19) and ends the processing. Note that the processing shown in FIG. 9 can be repeated. Also, the processing shown in FIG. 9 can be repeatedly executed for each different learning model.

[0088] Instead of the learning for updating the parameters of the neural network model as described above, in the learning using the Q-table, in the initial state of Q-learning, the Q-values in the Q-table can be initialized with, for example, random numbers. Once a difference occurs in the expected value of the reward at the initial stage of Q-learning, a situation may occur where it is impossible to transition to a state that has not yet been experienced and impossible to reach the goal. Therefore, when determining an action for a certain state, the probability ε can be used. Specifically, with a certain probability ε, an action can be randomly selected and executed from all actions, and with a probability (1 - ε), an action with the maximum Q-value can be selected and executed. Thereby, learning can be appropriately advanced regardless of the initial state of the Q-values.

[0089] FIG. 10 is a flowchart showing an example of the processing procedure in the operation control mode of the control device 50 according to the present embodiment. For convenience, the main body of the processing will be described as the control unit 51. The control unit 51 reads the parameters of the neural network model (S31), acquires the state s t (S32), and stores the acquired state s t in the storage unit 55 (S33). The state s t is the gas information of the gas converted by the gasification furnace 10, and specifically, it is the concentration of the impurity gas.

[0090] The control unit 51 outputs an action a t for the state s t based on the learning model (S34), and stores the output action a t in the storage unit 55 (S35). The action a t is control information for the operation control of the gas purification device 20, and specifically, it can be a combination of all or part of the cycle time, gas amount, gas temperature, and gas humidity. Further, the action a t may include an action regarding the nutrient given to the microorganism (catalyst).

[0091] The control unit 51 performs the operation control of the gas purification device 20 based on the output action a t (S36), and acquires the characteristic information (S37). The control unit 51 stores the acquired characteristic information in the storage unit 55 (S38).

[0092] The control unit 51 determines whether or not the operation of the gas purification apparatus 20 has ended (S39). If the operation has not ended (NO in S39), it acquires the state s t+1 (S40), sets the state s t+1 as the state s t and continues the processing after step S34. If the operation of the gas purification apparatus 20 has ended (YES in S39), the control unit 51 transmits the state, action, and characteristic information stored in the storage unit to the server (learning server 300) (S42) and ends the processing.

[0093] Further, the processing unit 60 can re-learn the learning model (action selection unit 62 and action evaluation unit 63) based on the gas information acquired by the gas information acquisition unit 52, the control information output by the processing unit 60, and the characteristic information acquired by the characteristic information acquisition unit 53. Thereby, the operation of the gas purification apparatus 20 can be further optimized.

[0094] The control unit 51 and the processing unit 60 of the present embodiment can also be realized using a computer equipped with a CPU (processor), GPU, RAM (memory), etc. For example, a computer program and data (e.g., a learned Q-function or Q-value, etc.) recorded on a recording medium (e.g., an optical readable disk recording medium such as a CD-ROM) can be read by a recording medium reading unit 57 (e.g., an optical disk drive) and stored in the RAM. It may be stored in a hard disk (not shown) and stored in the RAM when the computer program is executed. By loading a computer program defining the procedure of each process as shown in FIGS. 9 and 10 into the RAM (memory) provided in the computer and executing the computer program with the CPU (processor), the control unit 51 and the processing unit 60 can be realized on the computer.

[0095] In the above-described embodiment, Q-learning has been described as an example of machine learning. Alternatively, other learning algorithms such as another TD learning (Temporal Difference Learning) may be used. For example, a learning method that updates the value of a state instead of updating the value of an action, like Q-learning, may be used. In this method, the value V(s t ) of the current state St is updated by the formula V(s t ) <- V(s t ) + α·δt. Here, δt = r t+1 + γ·V(s t+1 ) - V(s t ), where α is the learning rate and δt is the TD error.

[0096] As described above, according to this embodiment, the collected combustible waste can be converted into ethanol with extremely high production efficiency, and the combustible waste can be reused as an industrial raw material with high efficiency.

[0097] In the above-described embodiment, one waste treatment facility (also referred to as a plant) has been described. However, this embodiment can also be applied to a plurality of plants installed in a plurality of locations (regions).

[0098] FIG. 11 is a schematic diagram showing an example of the configuration of a management system for managing a plurality of plants. As shown in FIG. 11, each control device 50 provided in a plurality of plants is connected to a network 1 such as the Internet. A management server 200 and a learning server 300 are connected to the network 1. Information can be transmitted and received between each control device 50, the management server 200, and the learning server 300 via the network 1. The management server 200 includes a CPU 201, a RAM 202, a ROM 203, and a plant DB 204, and a display device 210 is connected thereto. Note that the management server 200 (CPU 201) can control the processing of the display device 210. The learning server 300 includes a processing unit 301 and a plant DB 302. The processing unit 301 can have the same configuration as the processing unit 60 of the control device 50.

[0099] In the operation control mode of the gas purification device 20, each control device 50 can transmit the acquired gas information, the output control information, the acquired characteristic information, and the identification information for identifying the plant to the learning server 300. The learning server 300 can collect the identification information for identifying the plant, the gas information of the gas converted by the gasifier 10, the control information for controlling the gas purification device 20, and the characteristic information including the information of the purified gas purified by the gas purification device 20 from each control device 50. The learning server 300 can store the collected gas information, control information, and characteristic information in association with the identification information in the plant DB 302. Thereby, for each plant, the information necessary for optimizing the operation of the gas purification device 20 can be collected and recorded.

[0100] The learning server 300 can collect information on how to control the operation of the gas purification device 20 in order to obtain desired characteristic information when the components and composition of the waste vary. Also, by transmitting the same information from the control devices 50 of each of a plurality of waste treatment facilities (plants), the learning server 300 can collect information on how to control the operation of the gas purification device 20 in each plant in order to obtain desired characteristic information.

[0101] The processing unit 301 can learn a learning model based on the collected gas information, control information, and characteristic information. Thereby, the learning server 300 can generate a customized learning model for each waste treatment facility (plant) installed in various regions. When a control device 50 is newly installed in an existing waste treatment facility or when a new plant is constructed, a learning model suitable for each plant can be distributed. When distributing the learning model to a plant (specifically, within the control device 50), the learning model (algorithm and parameters, etc.) can be encrypted and distributed using a secret key or the like. Each control device 50 can decrypt it using its own secret key.

[0102] In addition, each control device 50 can transmit information such as the usage history, degree of deterioration, presence or absence of a desorption operation, and activity of a catalyst (for example, a microorganism) of the adsorption devices 22 and 23, which is acquired from the gas purification device 20, to the management server 200.

[0103] In the management server 200, based on the usage history, the remaining number of usage times, remaining usage time, etc. until the adsorption devices 22 and 23 are replaced can be calculated, and thus the replacement timing of the adsorption devices 22 and 23 can be estimated. Also, by transmitting similar information from the control devices of each of a plurality of waste treatment facilities (plants), the management server 200 can estimate the replacement timing of the adsorption devices 22 and 23 in the gas purification device 20 in each plant.

[0104] FIG. 12 is a schematic diagram showing an example of a plant list screen 211 displayed on the display device 210. As shown in FIG. 12, the plant list screen 211 has a plant ID display area 212, a degree of deterioration display area 213 of the adsorption device, an alert display area 214, and an activity display area 215 of a catalyst (for example, a microorganism). In the management server 200, that is, an operator who monitors the display screen of the display device 210 can determine whether maintenance, inspection, or replacement of the adsorption devices 22 and 23 is necessary in each plant based on the degree of deterioration of each adsorption device of each plant displayed in the degree of deterioration display area 213 of the adsorption device. In the example of FIG. 12, in any plant, the degree of deterioration of the adsorption device has not reached the value to be replaced.

[0105] In addition, in the management server 200, that is, an operator who monitors the display screen of the display device 210 can recognize that there has been an unintentional desorption operation of the adsorption devices 22 and 23 when the alert in the alert display area 214 lights up or blinks. Thereby, for example, it is possible to detect the installation of an adsorption device that is not a genuine product and prevent the installation of non-genuine products. Also, by transmitting similar information from the control devices 50 of each of a plurality of waste treatment facilities (plants), the management server 200 can detect the installation of an adsorption device that is not a genuine product in each plant and prevent the installation of non-genuine products.

[0106] Also, in the management server 200, that is, the operator who monitors the display screen of the display device 210 can identify whether the activity in the activity display area 215 of the catalyst is OK or NG. In FIG. 12, the activity is assumed to be OK. Thereby, for example, when the activity of the microorganism decreases, an instruction to remotely inject a nutrient can be given, the microorganism can be reactivated, and the ethanol production rate can be maintained at a high level.

[0107] FIG. 13 is a flowchart showing an example of the procedure of the process of the management server 200. Hereinafter, for convenience, the main body of the process will be described with the CPU 201. The CPU 201 acquires the deterioration information of the gas adsorption devices 22 and 23 in the gas purification device 20 of each of the plurality of plants (S101), and displays the degree of deterioration of the gas adsorption devices 22 and 23 for each plant (S102).

[0108] The CPU 201 determines whether or not it has acquired the desorption operation information of the gas adsorption devices 22 and 23 (S103). Here, the desorption operation information is information indicating that there has been an unintentional desorption operation of the gas adsorption devices 22 and 23, and does not include the desorption operation when cleaning the gas adsorption devices 22 and 23.

[0109] When the desorption operation information of the gas adsorption devices 22 and 23 is acquired (YES in S103), the CPU 201 outputs an alert for the corresponding plant (for example, the alert in the alert display area 214 illustrated in FIG. 12) (S104). The output of the alert may be the lighting or blinking of a display lamp, or may be output as sound. Also, it may be notified to the portable terminal device (not shown) of the operator. When the desorption operation information of the gas adsorption devices 22 and 23 has not been acquired (NO in S103), the CPU 201 performs the process of step S105 described later.

[0110] The CPU 201 determines whether it has acquired the activity information of the catalyst (e.g., microorganism) for ethanol production (S105). If it has acquired the activity information (YES in S105), it displays the activity of the catalyst (e.g., microorganism) for each plant (S106), and determines whether to end the process (S107). If it has not acquired the activity information (NO in S105), the CPU 201 performs the process of step S107. If the process is not ended (NO in S107), the CPU 201 continues the processes after step S101. If the process is ended (YES in S107), the process is ended.

[0111] In the above-described embodiment, each of the management server 200 and the learning server 300 may be composed of a plurality of servers, or the management server 200 and the learning server 300 may be integrated into one server.

[0112] The control device according to the present embodiment is a control device that controls a gas purification device, and includes a gas information acquisition unit that acquires gas information of gas converted by a gasification furnace that converts collected garbage into gas, a control information acquisition unit that acquires control information for controlling the gas purification device that purifies the gas converted by the gasification furnace, a characteristic information acquisition unit that acquires characteristic information including information of purified gas purified by the gas purification device, and a generation unit that generates a learning model by machine learning based on the gas information, control information, and characteristic information.

[0113] The computer program according to the present embodiment causes a computer to execute a process of acquiring gas information of gas converted by a gasification furnace that converts collected garbage into gas, a process of acquiring control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, a process of acquiring characteristic information including information of purified gas purified by the gas purification device, and a process of generating a learning model by machine learning based on the gas information, control information, and characteristic information.

[0114] The control method according to this embodiment is a control method for controlling a gas purification apparatus, which acquires gas information of gas converted by a gasifier that converts collected garbage into gas, acquires control information for controlling the gas purification apparatus that purifies the gas converted by the gasifier, acquires characteristic information including information on purified gas purified by the gas purification apparatus, and generates a learning model by machine learning based on the gas information, control information, and characteristic information.

[0115] The gas information acquisition unit acquires gas information of gas converted by a gasifier that converts collected garbage into gas. The gasifier is a furnace that can steam-roast garbage in a low-oxygen state and decompose it to the molecular level (for example, including carbon monoxide gas and hydrogen gas). The gas information includes, for example, the concentration of impurity gas (contaminant) generated by the gasifier. Note that the concentration of the impurity gas varies depending on the components and composition of miscellaneous garbage.

[0116] The control information acquisition unit acquires control information for controlling a gas purification apparatus that purifies the gas converted by the gasifier. The gas purification apparatus can remove and purify the impurity gas contained in the gas converted by the gasifier and extract the required gas (for example, carbon monoxide gas and hydrogen gas). The control information is information for operation control of the gas purification apparatus.

[0117] The characteristic information acquisition unit acquires characteristic information including information on purified gas purified by the gas purification apparatus. The information on the purified gas includes, for example, the purity of carbon monoxide gas and hydrogen gas. Also, the information on the purified gas can include the concentration of impurity gas that could not be removed. The purified gas can be converted into ethanol using a catalyst (for example, a metal catalyst, a microbial catalyst, etc.). Ethanol has the same C2 structure as ethylene, which accounts for about 60% of petrochemical products, and can be induced into induced chemical materials such as plastics by converting it into ethylene monomer or butadiene monomer by an existing chemical process. The characteristic information includes, for example, the activity of a catalyst (for example, a microorganism) that generates ethanol from carbon monoxide gas and hydrogen gas, the purity or amount of the generated ethylene, and the like.

[0118] The generation unit generates a learning model by machine learning based on gas information, control information, and characteristic information. For machine learning, for example, deep learning, reinforcement learning, deep reinforcement learning, etc. can be used. For example, when using reinforcement learning, the gas information is regarded as the "state", the control information is regarded as the "action", and the "reward" is calculated based on the characteristic information, and the value of the Q-value or Q-function (action value function) may be learned.

[0119] With the above configuration, since the variation in the components and composition of the garbage is large, even when the concentration of the gas (impurity gas) input to the gas purification device exceeds the threshold value, the learning model can be learned so that the characteristic information on the output side of the gas purification device can output control information that falls within the required value or range. By using the learned learning model, even when the components and composition of the garbage vary, the operation of the gas purification device can be optimized to generate the desired ethanol, so that the combustible garbage can be reused as an industrial raw material with high efficiency.

[0120] In the control device according to the present embodiment, the generation unit includes an action output unit that outputs the control information based on the gas information acquired by the gas information acquisition unit and the action evaluation information, a reward calculation unit that calculates a reward based on the characteristic information acquired by the characteristic information acquisition unit, and an update unit that updates the action evaluation information so that the reward calculated by the reward calculation unit increases.

[0121] The action output unit outputs control information based on the gas information acquired by the gas information acquisition unit and the action evaluation information. The action evaluation information is the evaluation value of the action in reinforcement learning and is the same as the Q-value or Q-function (action value function). That is, the action output unit selects and outputs an action from the actions that can be taken in the acquired state based on the evaluation value of the action in the acquired state.

[0122] The reward calculation unit calculates a reward based on the characteristic information acquired by the characteristic information acquisition unit. When calculating the reward, if the characteristic information is within the required value or range, it can be set to positive (with reward), and if the characteristic information does not reach the required value or is not within the range, it can be set to 0 (without reward) or negative (penalty).

[0123] The update unit updates the action evaluation information so that the reward calculated by the reward calculation unit increases. Thereby, even when the concentration of the gas (impurity gas) input to the gas purification device exceeds the threshold value due to large fluctuations in the components and composition of the waste, the learning model can be made to learn so that the characteristic information on the output side of the gas purification device is within the required value or range, and control information can be output.

[0124] The control device according to the present embodiment includes a storage unit that stores the action evaluation information updated by the update unit.

[0125] The storage unit stores the action evaluation information updated by the update unit. By reading out the action evaluation information stored in the storage unit, a learned learning model can be reproduced.

[0126] In the control device according to the present embodiment, the gas information acquisition unit acquires gas information including the concentration of the impurity gas.

[0127] The gas information acquisition unit acquires gas information including the concentration of the impurity gas. The impurity gas includes, for example, gases such as hydrogen cyanide, benzene, toluene, ethylbenzene, xylene, and dioxin, but is not limited thereto. Thereby, the impurity gas can be removed, and the contaminants contained in the gas from the gasification furnace can be thoroughly removed.

[0128] In the control device according to the present embodiment, the characteristic information acquisition unit acquires characteristic information including at least one of the purity of carbon monoxide gas and hydrogen gas, the purity or amount of ethanol, and the activity of a catalyst for generating ethanol from carbon monoxide gas and hydrogen gas.

[0129] The characteristic information acquisition unit acquires characteristic information including at least one of the purity of carbon monoxide gas and hydrogen gas, the purity or amount of ethanol, and the activity of a catalyst (e.g., a microorganism) that generates ethanol from carbon monoxide gas and hydrogen gas. Thereby, the characteristic information can be made to be within a required value or range.

[0130] In the control device according to the present embodiment, the control information acquisition unit acquires control information including at least one of the gas amount, gas temperature, and gas humidity of the gas purification device, and the switching cycle time of the adsorption device in the gas purification device.

[0131] The control information acquisition unit acquires control information including at least one of the gas amount, gas temperature, and gas humidity of the gas purification device, and the switching cycle time of the adsorption device in the gas purification device. The gas adsorption member is a member provided in the adsorption device that adsorbs and captures impurities. The switching cycle time is, for example, the usage time of one of the adsorption devices when two adsorption devices are alternately switched and used. During the switching cycle time, the gas adsorption member of the unused adsorption device can be desorbed, and the impurities attached to the gas adsorption member can be washed. Thereby, the operation control of the gas purification device can be performed so that the characteristic information is within a required value or range.

[0132] The control device according to the present embodiment is a control device that controls a gas purification device, and includes a learning model learned based on characteristic information including gas information of the gas converted by a gasification furnace that converts collected garbage into gas, control information for controlling the gas purification device that purifies the gas converted by the gasification furnace, and information of the purified gas purified by the gas purification device, a gas information acquisition unit that acquires the gas information of the gas converted by the gasification furnace, and an output unit that inputs the gas information acquired by the gas information acquisition unit into the learning model and outputs control information for controlling the gas purification device.

[0133] The computer program according to this embodiment causes a computer to execute a process of acquiring gas information of gas converted by a gasification furnace that converts collected garbage into gas, and a process of inputting the acquired gas information into a learning model learned based on characteristic information including the gas information of the gas converted by the gasification furnace, control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, and information of purified gas purified by the gas purification device, and outputting control information for controlling the gas purification device.

[0134] The learning model according to an embodiment of the present invention is learned based on characteristic information including gas information of gas converted by a gasification furnace that converts collected garbage into gas, control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, and information of purified gas purified by the gas purification device.

[0135] The control method according to this embodiment is a control method for controlling a gas purification device, which acquires gas information of gas converted by a gasification furnace that converts collected garbage into gas, and inputs the acquired gas information into a learning model learned based on characteristic information including the gas information of the gas converted by the gasification furnace, control information for controlling the gas purification device that purifies the gas converted by the gasification furnace, and information of purified gas purified by the gas purification device, and outputs control information for controlling the gas purification device.

[0136] The learning model is learned based on characteristic information including gas information of gas converted by a gasification furnace that converts collected garbage into gas, control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, and information of purified gas purified by the gas purification device.

[0137] The gas information includes, for example, the concentration of impurity gas (contaminants) generated by a gasification furnace. Note that the concentration of the impurity gas varies depending on the components and composition of miscellaneous waste. The control information is information for controlling the operation of the gas purification device. The characteristic information includes, for example, the purity of carbon monoxide gas and hydrogen gas, the activity of a catalyst (e.g., a microorganism) for generating ethanol from carbon monoxide gas and hydrogen gas, the purity or amount of the generated ethylene, etc.

[0138] The learning model is learned using, for example, deep learning, reinforcement learning, deep reinforcement learning, etc.

[0139] The gas information acquisition unit acquires the gas information of the gas converted by the gasification furnace. The gasification furnace is a furnace that can roast garbage in a low-oxygen state and decompose it to the molecular level (including, for example, carbon monoxide gas and hydrogen gas). The gas information includes, for example, the concentration of impurity gas (contaminants) generated by the gasification furnace. Note that the concentration of the impurity gas varies depending on the components and composition of miscellaneous waste.

[0140] The output unit inputs the gas information acquired by the gas information acquisition unit into the learning model and outputs control information for controlling the gas purification device. Thereby, even when the components and composition of the garbage vary, the operation of the gas purification device can be optimized to generate the desired ethanol, so that the combustible garbage can be reused as an industrial raw material with high efficiency.

[0141] The control device according to the present embodiment controls the gas purification device based on the control information output by the output unit.

[0142] The gas purification device is controlled based on the control information output by the output unit. Thereby, even when the components and composition of the garbage vary, the operation of the gas purification device can be optimized to generate the desired ethanol, so that the combustible garbage can be reused as an industrial raw material with high efficiency.

[0143] The control device according to the present embodiment includes a characteristic information acquisition unit that acquires characteristic information including information on the purified gas purified by the gas purification device, and a storage unit that stores the characteristic information acquired by the characteristic information acquisition unit.

[0144] The characteristic information acquisition unit acquires characteristic information including information on the purified gas purified by the gas purification device, and the storage unit stores the acquired characteristic information. As a result, when the components and composition of the waste vary, it is possible to collect the characteristic information obtained as a result of optimizing the operation of the gas purification device.

[0145] The control device according to the present embodiment includes a transmission unit that transmits the gas information acquired by the gas information acquisition unit, the control information output by the output unit, and the characteristic information acquired by the characteristic information acquisition unit to the server.

[0146] The transmission unit transmits the gas information acquired by the gas information acquisition unit, the control information output by the output unit, and the characteristic information acquired by the characteristic information acquisition unit to the server. As a result, the server can collect information on how to control the operation of the gas purification device in order to obtain the desired characteristic information when the components and composition of the waste vary. Further, by transmitting similar information from the control devices of a plurality of waste treatment facilities (plants), the server can collect information on how to control the operation of the gas purification device in each plant in order to obtain the desired characteristic information.

[0147] The control device according to the present embodiment includes a learning processing unit that re-learns the learning model based on the gas information acquired by the gas information acquisition unit, the control information output by the output unit, and the characteristic information acquired by the characteristic information acquisition unit.

[0148] The learning processing unit re-learns the learning model based on the gas information acquired by the gas information acquisition unit, the control information output by the output unit, and the characteristic information acquired by the characteristic information acquisition unit. As a result, the operation of the gas purification device can be further optimized.

[0149] The control device according to this embodiment includes a usage history acquisition unit that acquires the usage history of the adsorption device in the gas purification device, and a transmission unit that transmits the usage history acquired by the usage history acquisition unit to the management server.

[0150] The usage history acquisition unit acquires the usage history of the adsorption device in the gas purification device. The usage history includes, for example, the cumulative usage time, the number of cleaning times, and the like.

[0151] The transmission unit transmits the usage history acquired by the usage history acquisition unit to the management server. Based on the usage history, the management server can estimate the remaining usage times, remaining usage time, etc. until the adsorption device is replaced, thereby estimating the replacement timing of the adsorption device. Also, by transmitting similar information from the control devices of each of a plurality of waste treatment facilities (plants), the management server can estimate the replacement timing of the adsorption device in the gas purification device at each plant.

[0152] The control device according to this embodiment includes a degradation degree acquisition unit that acquires the degradation degree of the adsorption device in the gas purification device, and the transmission unit transmits the degradation degree acquired by the degradation degree acquisition unit to the management server.

[0153] The degradation degree acquisition unit acquires the degradation degree of the adsorption device in the gas purification device. The degradation degree can be determined, for example, by the color or dirt on the surface of the gas adsorption member after cleaning the gas adsorption member, the amount of impurities adsorbed at a predetermined cycle time, and the like.

[0154] The transmission unit transmits the degradation degree acquired by the degradation degree acquisition unit to the management server. Based on the degradation degree, the management server can determine whether maintenance, inspection, or replacement of the adsorption device is necessary. Also, by transmitting similar information from the control devices of each of a plurality of waste treatment facilities (plants), the management server can determine whether maintenance, inspection, or replacement of the adsorption device in the gas purification device at each plant is necessary.

[0155] The control device according to this embodiment includes a desorption operation acquisition unit that acquires the presence or absence of a desorption operation of the adsorption device in the gas purification device. When the desorption operation acquisition unit acquires that there is a desorption operation, the transmission unit transmits the fact that there was a desorption operation to the management server.

[0156] The desorption operation acquisition unit acquires the presence or absence of a desorption operation of the adsorption device in the gas purification device. The desorption operation of the adsorption device can be, for example, an unintentional desorption operation.

[0157] When the desorption operation acquisition unit acquires that there is a desorption operation, the transmission unit transmits the fact that there was a desorption operation to the management server. In the management server, since it is possible to determine the presence or absence of an unintentional desorption operation of the adsorption device, for example, it is possible to detect the installation of an adsorption device that is not a genuine product and prevent the installation of non-genuine products. Also, by transmitting similar information from the control devices of each of a plurality of waste treatment facilities (plants), the management server can detect the installation of an adsorption device that is not a genuine product in each plant and prevent the installation of non-genuine products.

[0158] The control device according to this embodiment includes a determination unit that determines the activity level of microorganisms based on the state of microorganisms that produce ethanol from carbon monoxide gas and hydrogen gas purified by the gas purification device.

[0159] The determination unit determines the activity level of microorganisms based on the state of microorganisms that produce ethanol from carbon monoxide gas and hydrogen gas purified by the gas purification device. The activity level includes, for example, the reaction rate and survival rate of the microorganisms. The activity level of the microorganisms can be determined by monitoring the state of the microorganisms in real time from outside the culture solution layer. Note that the activity level of the microorganisms may also be determined offline. Thereby, for example, when the activity level of the microorganisms decreases, nutrients can be added to reactivate them, and the production rate of ethanol can be maintained at a high level.

[0160] The server according to this embodiment collects, from each of a plurality of garbage treatment plants, identification information for identifying the plant, gas information of the gas converted by a gasification furnace that converts the collected garbage into gas, control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, and characteristic information including information on the purified gas purified by the gas purification device, and includes a storage unit that stores the gas information, control information, and characteristic information collected by the collection unit in association with the identification information.

[0161] The collection unit collects, from each of a plurality of garbage treatment plants, identification information for identifying the plant, gas information of the gas converted by a gasification furnace that converts the collected garbage into gas, control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, and characteristic information including information on the purified gas purified by the gas purification device.

[0162] The storage unit stores the gas information, control information, and characteristic information collected by the collection unit in association with the identification information. Thereby, for each plant, information necessary for optimizing the operation of the gas purification device can be collected and recorded.

[0163] The management server according to this embodiment includes a collection unit that collects, from each of a plurality of garbage treatment plants, identification information for identifying the plant and the degree of deterioration of an adsorption device in a gas purification device that purifies the gas converted by a gasification furnace that converts the collected garbage into gas, and a storage unit that stores the degree of deterioration collected by the collection unit in association with the identification information.

[0164] The collection unit collects, from each of a plurality of garbage treatment plants, identification information for identifying the plant and the degree of deterioration of an adsorption device in a gas purification device that purifies the gas converted by a gasification furnace that converts the collected garbage into gas.

[0165] The storage unit stores the degree of deterioration collected by the collection unit in association with the identification information. Thereby, for each plant, the degree of deterioration of the adsorption device in the gas purification device can be grasped.

[0166] In addition, in the present embodiment, organic compounds and / or inorganic compounds can be used as substitutes for garbage. In this case, the control device includes a gas information acquisition unit that acquires gas information of the gas converted by a gasification furnace that converts organic compounds and / or inorganic compounds into gas, a control information acquisition unit that acquires control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, a characteristic information acquisition unit that acquires characteristic information including information on the purified gas purified by the gas purification device, and a generation unit that generates a learning model by machine learning based on the gas information, the control information, and the characteristic information. Further, the control device includes a learning model learned based on the gas information of the gas converted by a gasification furnace that converts organic compounds and / or inorganic compounds into gas, the control information for controlling a gas purification device that purifies the gas converted by the gasification furnace, and the characteristic information including information on the purified gas purified by the gas purification device, a gas information acquisition unit that acquires the gas information of the gas converted by the gasification furnace, and an output unit that inputs the gas information acquired by the gas information acquisition unit into the learning model and outputs control information for controlling the gas purification device.

Explanation of Signs

[0167] 1 Network 10 Gasification furnace 20 Gas purification device 22, 23 Adsorption device 221, 231 Gas adsorption member 26 Operation control unit 261 Gas flow rate control unit 262 Temperature adjustment unit 263 Humidity adjustment unit 264 Adsorption device switching unit 265 Communication unit 30 Ethanol generation device 50 Control device 51 Control unit 52 Gas information acquisition unit 53 Characteristic information acquisition unit 54 Communication unit 55 Storage unit 56 Sensor information acquisition unit 57 Recording medium reading unit 58 Judgment unit 60 Processing unit 61 Reward calculation unit 62 Action selection unit 63 Action evaluation unit 200 Management server 201 CPU 202 RAM 203 ROM 204 Plant DB 210 Display device 300 Learning server 301 Processing unit 302 Plant DB

Claims

1. A control device for controlling a gas purification device, comprising: a gas information acquisition unit that acquires, as gas information, the concentration of impurity gas contained in the gas converted by a gasification furnace that converts collected combustible waste into gas; a control information acquisition unit that acquires, as control information, the gas amount of the gas purification device that is output to the gas purification device that purifies the gas converted by the gasification furnace and controls the operation of the gas purification device; a characteristic information acquisition unit that acquires, as characteristic information, the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas; a generation unit that generates a learning model in which action evaluation information for selecting an action that maximizes the reward obtained when one action is selected from among the actions selectable for a certain state is learned by reinforcement learning using, as parameters, the state represented by the gas information, the action represented by the control information, and the reward based on the characteristic information; A control device comprising the above.

2. The generation unit includes: an action output unit that outputs, to the gas purification device, the control information selected from among the control information selectable for the gas information acquired by the gas information acquisition unit based on the action evaluation information; a reward calculation unit that calculates the reward so that the reward becomes a large value when the characteristic information acquired by the characteristic information acquisition unit falls within a required value or range as a result of the operation control of the gas purification device based on the control information output by the action output unit; an update unit that updates the action evaluation information so that control information that increases the reward calculated by the reward calculation unit can be selected; The control device according to claim 1, further comprising: The update unit updates the action evaluation information to generate the learning model.

3. The control device according to claim 2, further comprising a storage unit that stores the action evaluation information updated by the update unit.

4. The characteristic information acquisition unit: acquires characteristic information including the activity of a catalyst that produces ethanol from carbon monoxide gas and hydrogen gas purified by the gas purification device. The control device according to any one of claims 1 to 3.

5. A control device for controlling a gas purification device, comprising: A state represented by gas information which is the concentration of impurity gas contained in the gas converted by a gasifier that converts collected combustible waste into gas, an action represented by control information which is the gas amount of a gas purification device that is output to the gas purification device that purifies the gas converted by the gasifier and controls the operation of the gas purification device, and a reward based on characteristic information which is the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas are used as parameters respectively in reinforcement learning, and a learning model is learned in which action evaluation information for selecting an action that increases the reward obtained when selecting one action from among the actions selectable for a certain state is learned, and a gas information acquisition unit that acquires the concentration of impurity gas contained in the gas converted by the gasifier as gas information, an output unit that inputs the gas information acquired by the gas information acquisition unit into the learning model and outputs, as control information, the gas amount for controlling the gas purification device A control device comprising:

6. The control device according to claim 5, wherein the gas purification device is controlled based on the control information output by the output unit.

7. a characteristic information acquisition unit that acquires, as characteristic information, the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas, a storage unit that stores the characteristic information acquired by the characteristic information acquisition unit The control device according to claim 5 or claim 6, comprising:

8. The control device according to claim 7, further comprising a transmission unit that transmits the gas information acquired by the gas information acquisition unit, the control information output by the output unit, and the characteristic information acquired by the characteristic information acquisition unit to a server.

9. The control device according to claim 7 or claim 8, further comprising a learning processing unit that relearns the learning model based on the gas information acquired by the gas information acquisition unit, the control information output by the output unit, and the characteristic information acquired by the characteristic information acquisition unit.

10. a usage history acquisition unit that acquires the usage history of the adsorption device in the gas purification device, a transmission unit that transmits the usage history acquired by the usage history acquisition unit to a management server The control device according to any one of claims 1 to 9, comprising:

11. comprising a degradation degree acquisition unit that acquires the degradation degree of the adsorption device in the gas purification device, The transmission unit The control device according to claim 10, which transmits the degree of deterioration acquired by the degree-of-deterioration acquisition unit to the management server.

12. comprising a desorption-operation acquisition unit that acquires the presence or absence of a desorption operation of an adsorption device in the gas purification device, the transmission unit transmits the fact that there is a desorption operation to the management server when the desorption-operation acquisition unit acquires that there is a desorption operation. The control device according to claim 10 or claim 11.

13. The control device according to any one of claims 1 to 12, comprising a determination unit that determines the activity of the microorganism based on the state of the microorganism that generates ethanol from the carbon monoxide gas and hydrogen gas purified by the gas purification device.

14. A collection unit that collects, from each of a plurality of plants for garbage treatment, identification information for identifying the plant, gas information that is the concentration of impurity gas contained in the gas converted by a gasification furnace that converts collected combustible garbage into gas, control information that is the gas amount of a gas purification device that is output to a gas purification device that purifies the gas converted by the gasification furnace and controls the operation of the gas purification device, and the purity of the carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol generated from the carbon monoxide gas and hydrogen gas; A learning model in which action evaluation information for selecting an action that increases the reward obtained when selecting one action from among the actions selectable for a certain state is learned by reinforcement learning using, as parameters, the state represented by the gas information collected by the collection unit, the action represented by the control information, and the reward based on the characteristic information; A storage unit that stores the gas information, control information, and characteristic information collected by the collection unit in association with the identification information A server comprising.

15. A management system comprising the control device according to any one of claims 1 to 13 and a management server, the management server A collection unit that collects, from each of a plurality of plants for garbage treatment, identification information for identifying the plant and the degree of deterioration of an adsorption device in a gas purification device that purifies the gas converted by a gasification furnace that converts collected combustible garbage into gas; A storage unit that stores the degree of deterioration collected by the collection unit in association with the identification information A management system comprising.

16. In a computer, A process of acquiring, as gas information, the concentration of impurity gas contained in the gas converted by a gasification furnace that converts collected combustible garbage into gas; A process of outputting the gas converted by the gasifier to a gas purification device that purifies the gas and acquiring, as control information, the gas amount of the gas purification device for controlling the operation of the gas purification device; A process of acquiring, as characteristic information, the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas; A process of generating a learning model in which action evaluation information for selecting an action that increases the reward obtained when selecting one action from among the actions selectable for a certain state is learned by reinforcement learning using, as parameters, the state represented by the gas information, the action represented by the control information, and the reward based on the characteristic information; A computer program for causing the above to be executed.

17. On a computer, A process of acquiring, as gas information, the concentration of impurity gas contained in the gas converted by a gasifier that converts collected combustible waste into gas; The state represented by gas information that is the concentration of impurity gas contained in the gas converted by the gasifier; The action represented by control information that is the gas amount of the gas purification device for outputting the gas converted by the gasifier to a gas purification device that purifies the gas and controlling the operation of the gas purification device, and A process of inputting the acquired gas information into a learning model in which action evaluation information for selecting an action that increases the reward obtained when selecting one action from among the actions selectable for a certain state is learned by reinforcement learning using, as parameters, the state represented by the gas information, the action represented by the control information, and the reward based on the characteristic information that is the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas, and outputting, as control information, the gas amount for controlling the gas purification device; A computer program for causing the above to be executed.

18. The state represented by gas information that is the concentration of impurity gas contained in the gas converted by a gasifier that converts collected combustible waste into gas; The action represented by control information that is the gas amount of the gas purification device for outputting the gas converted by the gasifier to a gas purification device that purifies the gas and controlling the operation of the gas purification device, and A learning model in which reward information based on the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas, respectively, is used as a parameter, and action evaluation information for selecting an action that increases the reward obtained when selecting one action from among the actions selectable for a certain state is learned by reinforcement learning.

19. A control method for controlling a gas purification device, comprising: acquiring, as gas information, the concentration of impurity gas contained in the gas converted by a gasifier that converts collected combustible waste into gas; acquiring, as control information, the gas amount of the gas purification device for outputting to the gas purification device that purifies the gas converted by the gasifier and for controlling the operation of the gas purification device; acquiring, as characteristic information, the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas; A control method for generating a learning model in which, by reinforcement learning using, as parameters, the state represented by the gas information, the action represented by the control information, and the reward based on the characteristic information, action evaluation information for selecting an action that increases the reward obtained when selecting one action from among the actions selectable for a certain state is learned.

20. A control method for controlling a gas purification device, comprising: acquiring, as gas information, the concentration of impurity gas contained in the gas converted by a gasifier that converts collected combustible waste into gas; a state represented by gas information that is the concentration of impurity gas contained in the gas converted by the gasifier; an action represented by control information that is the gas amount of the gas purification device for outputting to the gas purification device that purifies the gas converted by the gasifier and for controlling the operation of the gas purification device, and A control method for inputting the acquired gas information into a learning model in which reward information based on the purity of carbon monoxide gas and hydrogen gas purified by the gas purification device, and the purity or amount of ethanol produced from carbon monoxide gas and hydrogen gas, respectively, is used as a parameter, and action evaluation information for selecting an action that increases the reward obtained when selecting one action from among the actions selectable for a certain state is learned by reinforcement learning, and outputting, as control information, the gas amount for controlling the gas purification device.

Images