CN114169711A - Accident emergency disposal scheme generation method, device and system for distillation device - Google Patents

Accident emergency disposal scheme generation method, device and system for distillation device Download PDF

Info

Publication number
CN114169711A
CN114169711A CN202111406804.8A CN202111406804A CN114169711A CN 114169711 A CN114169711 A CN 114169711A CN 202111406804 A CN202111406804 A CN 202111406804A CN 114169711 A CN114169711 A CN 114169711A
Authority
CN
China
Prior art keywords
emergency
emergency disposal
agent
state
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111406804.8A
Other languages
Chinese (zh)
Inventor
陈曦
谭南鑫
曾俊逸
宋安驰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong University of Science and Technology
Original Assignee
Huazhong University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Priority to CN202111406804.8A priority Critical patent/CN114169711A/en
Publication of CN114169711A publication Critical patent/CN114169711A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D3/00—Distillation or related exchange processes in which liquids are contacted with gaseous media, e.g. stripping
    • B01D3/42—Regulation; Control
    • C—CHEMISTRY; METALLURGY
    • C10—PETROLEUM, GAS OR COKE INDUSTRIES; TECHNICAL GASES CONTAINING CARBON MONOXIDE; FUELS; LUBRICANTS; PEAT
    • C10G—CRACKING HYDROCARBON OILS; PRODUCTION OF LIQUID HYDROCARBON MIXTURES, e.g. BY DESTRUCTIVE HYDROGENATION, OLIGOMERISATION, POLYMERISATION; RECOVERY OF HYDROCARBON OILS FROM OIL-SHALE, OIL-SAND, OR GASES; REFINING MIXTURES MAINLY CONSISTING OF HYDROCARBONS; REFORMING OF NAPHTHA; MINERAL WAXES
    • C10G7/00—Distillation of hydrocarbon oils
    • C10G7/12—Controlling or regulating
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Chemical & Material Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Educational Administration (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Oil, Petroleum & Natural Gas (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Operations Research (AREA)
  • Marketing (AREA)
  • General Chemical & Material Sciences (AREA)
  • Development Economics (AREA)
  • Organic Chemistry (AREA)
  • Feedback Control In General (AREA)
  • Vaporization, Distillation, Condensation, Sublimation, And Cold Traps (AREA)

Abstract

The invention discloses a method, a device and a system for generating an accident emergency disposal scheme of a distillation device, belonging to the field of accident treatment of distillation devices, wherein the method comprises the following steps: s1: constructing an original mechanism environment model corresponding to the distillation device; s2: constructing an intelligent agent corresponding to emergency disposal personnel; s3: controlling the intelligent agent in the current intelligent agent state to select and execute the current emergency disposal behavior, and inputting the current emergency disposal behavior into the original mechanism environment model so as to obtain experience data; s4: randomly extracting the empirical data and inputting the empirical data to the deep neural network for learning training; s5: obtaining a target mechanism environment model according to an accident scene to be processed; acquiring a plurality of corresponding emergency disposal behaviors based on the target mechanism model and the trained deep neural network model; and splicing the plurality of emergency disposal behaviors to obtain an emergency disposal scheme of the distillation plant accident. The invention can improve the rationality, effectiveness and accuracy of the emergency disposal scheme generation of the accident.

Description

Accident emergency disposal scheme generation method, device and system for distillation device
Technical Field
The invention belongs to the technical field of accident handling of a distillation device, and particularly relates to an accident emergency disposal scheme generation method, device and system of the distillation device.
Background
The emergency treatment of the accident of the product oil mixing distillation device relates to the characteristics of a complex chemical system, the development trend of the accident is difficult to predict, secondary derivative accidents are easy to cause, and the destructiveness is serious. The damage caused by the accidents is huge for families, enterprises and even the whole society, and the emergency management work of the accidents gradually becomes the focus of the attention of the enterprises and the countries. At present to the reply of distillation plant production incident, mainly for emergency treatment personnel carry out subjective emergency treatment or refer to set emergency treatment scheme through self experience and carry out corresponding processing, current distillation plant accident emergency treatment scheme is all to single accident, and the emergency treatment who relates to moreover is comparatively general, has increased the degree of difficulty of emergency treatment work.
Patent CN104268710A discloses a method for generating an emergency disposal scheme for a rail transit road network, which performs disposal main point extraction by analyzing a text plan, gives disposal main point attribute information to the disposal main points according to event feature dimensions, time feature dimensions, location feature dimensions and the like, stores the disposal main point attribute information as a disposal main point library, matches emergency information with disposal main points, extracts disposal with a high matching degree, and generates a disposal scheme. The thesis "fast generation method of emergency disposal scheme for unconventional emergency based on scenario reconstruction" analyzes the relationship between scenarios, splits the occurred scenarios, recombines them according to a certain rule, generates the scenario most similar to the current scenario, and generates the emergency disposal scheme by referring to the emergency disposal measures corresponding to the occurred scenarios. The two emergency disposal scheme generation methods have high dependence on historical cases, a perfect disposal key point database needs to be established for disposal key point matching, a perfect case database needs to be established for scene reconstruction, and the establishment of the databases is a long-term process and needs to be continuously updated and maintained, so that the emergency disposal scheme generation methods are difficult to be applied to practice.
In addition, although there are a few historical cases for the distillation apparatus, the number of types of apparatuses is large, and the number of auxiliary devices involved is very large, and in addition to the influence of environmental factors, the accident situation of the distillation apparatus is very complicated, and the accident development also involves a complicated chemical mechanism, so that it is difficult to obtain an effective emergency accident disposal plan through the historical cases.
Disclosure of Invention
Aiming at the defects or improvement requirements of the prior art, the invention provides a method, a device and a system for generating an accident emergency treatment scheme of a distillation device, aiming at obtaining empirical data based on dynamic simulation and deep neural network learning and generating the accident emergency treatment scheme of the distillation device without depending on historical data, so that the technical problem that the generated accident emergency treatment scheme is low in accuracy due to high dependence on historical cases is solved.
To achieve the above object, according to one aspect of the present invention, there is provided a method for generating an emergency treatment plan for a distillation apparatus, including:
s1: carrying out full-flow dynamic simulation on a distillation device to construct an original mechanism environment model corresponding to the distillation device;
s2: constructing an agent corresponding to emergency disposal personnel, and setting a deep learning framework of the agent, wherein the deep learning framework comprises defining an agent state, an emergency task reward function and a deep neural network model;
s3: controlling the intelligent agent in the current intelligent agent state to select and execute the current emergency disposal behavior, inputting the current emergency disposal behavior into the original mechanism environment model, calculating reward information by using the emergency task reward function, and updating the intelligent agent state; storing the current agent state, the current emergency disposition behavior, the reward information, and the updated agent state as experience data into an experience pool;
s4: randomly extracting the experience data from the experience pool and inputting the experience data to the deep neural network for learning training until a superposition value corresponding to the emergency task reward of the agent is converged;
s5: updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model; acquiring a plurality of corresponding emergency disposal behaviors based on the target mechanism model and the trained deep neural network model; and splicing the plurality of emergency disposal behaviors to obtain an emergency disposal scheme of the distillation plant accident.
In one embodiment, the original mechanistic environment model includes: a distillation plant dynamic model and a fault handling model; the S1 includes:
s11: establishing a full-process steady-state model of the distillation device through process simulation; constructing a distillation plant dynamic model based on the full process steady state model, the equipment dimensional data of the distillation plant and the controller parameters;
s12: and establishing a fault handling model around common emergency handling behaviors, and performing emergency handling simulation by using the fault handling model so as to obtain a mapping relation between the current emergency handling behaviors and variables in the distillation plant dynamic model.
In one embodiment, the state of the intelligent object defined in S2 includes: accident information, running state information, operating state information and target information;
the accident information comprises an accident type, an accident occurrence position and equipment fault information; the running state information comprises the change condition of each process parameter; the operating state information includes: the location of the agent and the fire equipment it carries; the target information refers to a current region to which the agent is going.
In one embodiment, the emergency mission reward function defined in S2 is: r ═ rtime+rpara+rresult(ii) a Wherein the content of the first and second substances,
rtimethe reward corresponding to the execution time for the emergency treatment,
Figure BDA0003373081360000031
k is a positive constant; [ t ] ofi,1,ti,2]Time window for item i emergency treatment, tiTime to perform the emergency disposition;
rparathe reward, r, corresponding to the maximum deviation of the process parameters of the distillation apparatusparaThe definition is as follows:
Figure BDA0003373081360000032
c is the penalty of the maximum deviation of the parameters being more than 10 percent;
rresultreward for harm caused by improper disposal of agent, rresultIncluding accident hazard indexes corresponding to tower rushing, tower flooding, leakage, fire and explosion.
In one embodiment, the deep neural network model includes: a predicted network and a target network; the S3 includes:
s31: controlling the intelligent agent in the current intelligent agent state s by using a greedy algorithm to randomly select and execute an emergency disposal behavior according to the probability of epsilon, and selecting and executing an emergency disposal behavior with the maximum Q value according to the probability of 1-epsilon; the Q value is a result obtained by inputting the current agent state s into the prediction network;
s32: inputting the executed emergency disposal behavior a into the original mechanism environment model to obtain consequence data, calculating reward information r corresponding to the emergency disposal behavior a by using the consequence data and the emergency task reward function, and updating the state s' of the agent after the emergency disposal behavior a is executed;
s33: and storing experience data (s, a, r, s ') formed by the current agent state s, the emergency handling behavior a, the reward information r and the updated agent state s' into an experience pool.
In one embodiment, the S4 includes:
s41: randomly extracting a plurality of groups of the empirical data (s, a, r, s') from the empirical pool, gamma being an attenuation coefficient, st+1For the state of the agent at the next moment, θ-A parameter of the target network;
s42: a target value y is calculated from the target network,
Figure BDA0003373081360000041
stis as follows
The state of the agent at the previous moment, theta is a parameter of the prediction network;
s43: updating the parameters of the prediction network by using a gradient descent method, wherein the calculation formula of the loss function is as follows: loss ═ y-Q(s)t,a；θ))2；
S44: synchronizing the updated parameter theta of the prediction network to the parameter theta of the target network after a preset number of time steps-；
S45: and repeating the training of the deep neural network from S41 to S44 until the deep neural network converges, and saving the parameters of the prediction network at the moment.
In one embodiment, the S5 includes:
s51: updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model;
s52: inputting the current state of the agent into the trained deep neural network to obtain a Q value; executing a target emergency disposal behavior corresponding to the maximum Q value, and inputting the target emergency disposal behavior into the target mechanism environment model to obtain a corresponding intelligent agent state after the emergency disposal behavior is executed; inputting the updated state of the agent into the trained deep neural network to update the Q value, executing a target emergency disposal behavior corresponding to the updated maximum Q value, and so on; until the emergency disposal task is completed;
s53: and splicing the executed target emergency disposal behaviors to obtain an accident emergency disposal scheme of the distillation device.
According to another aspect of the present invention, there is provided an accident emergency plan generating apparatus of a distillation apparatus, including:
the construction module is used for carrying out full-flow dynamic simulation on the distillation device so as to construct an original mechanism environment model corresponding to the distillation device;
the system comprises a setting module, a data processing module and a data processing module, wherein the setting module is used for constructing an intelligent agent corresponding to emergency disposal personnel and setting a deep learning framework of the intelligent agent, and the deep learning framework comprises the definition of an intelligent agent state, an emergency task reward function and a deep neural network model;
the acquisition module is used for controlling the intelligent agent in the current intelligent agent state to select and execute the current emergency disposal behavior, inputting the current emergency disposal behavior into the original mechanism environment model, calculating reward information by using the emergency task reward function and updating the intelligent agent state; storing the current agent state, the current emergency disposition behavior, the reward information, and the updated agent state as experience data into an experience pool;
the training module is used for randomly extracting the experience data from the experience pool and inputting the experience data into the deep neural network for learning training until a superposition value corresponding to the emergency task reward of the intelligent agent is converged;
the generating module is used for updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model; acquiring a plurality of corresponding emergency disposal behaviors based on the target mechanism model and the trained deep neural network model; and splicing the plurality of emergency disposal behaviors to obtain an emergency disposal scheme of the distillation plant accident.
According to another aspect of the present invention, there is provided an emergency treatment plan generation system for a distillation plant, comprising a memory storing a computer program and a processor implementing the steps of the method when the computer program is executed by the processor.
According to another aspect of the invention, a computer-readable storage medium is provided, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
the method provided by the invention constructs an original mechanism environment model corresponding to the distillation device by performing full-flow dynamic simulation on the distillation device; and then establishing an intelligent agent corresponding to emergency disposal personnel for simulation, and performing iterative simulation, evaluation and training learning based on an environmental mechanism model of the distillation device, the intelligent agent and a neural network established based on a time sequence difference reinforcement learning method to obtain an optimized emergency disposal scheme of the distillation device, so that the rationality, effectiveness and accuracy of the emergency disposal scheme generation are improved. In addition, the operation state of the distillation device under the accident is simulated in the full-flow simulation mode, and the change of the operation state information under the influence of the accident is simulated at the mechanism level, so that the space-time evolution characteristic of the accident scene is reflected, the effectiveness of simulation data can be improved, and the accuracy of the accident emergency disposal scheme of the distillation device is improved.
Drawings
FIG. 1 is a flow chart of a method for emergency incident solution generation for a distillation plant provided in an embodiment of the present invention;
fig. 2 is a schematic diagram of an accident emergency scenario generation method of the distillation apparatus provided in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
As shown in fig. 1, the present invention provides a method for generating an emergency treatment plan for an accident of a distillation apparatus, including:
s1: carrying out full-flow dynamic simulation on the distillation device to construct an original mechanism environment model corresponding to the distillation device;
s2: constructing an agent corresponding to emergency disposal personnel, and setting a deep learning framework of the agent, wherein the deep learning framework comprises defining an agent state, an emergency task reward function and a deep neural network model;
s3: controlling the intelligent agent in the current intelligent agent state to select and execute the current emergency disposal behavior, inputting the current emergency disposal behavior into the original mechanism environment model, calculating reward information by using an emergency task reward function, and updating the intelligent agent state; storing the current agent state, the current emergency disposal behavior, the reward information and the updated agent state as experience data into an experience pool;
s4: randomly extracting experience data from the experience pool, inputting the experience data into a deep neural network for learning and training until a superposition value corresponding to the emergency task reward of the intelligent agent is converged;
s5: updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model; acquiring a plurality of corresponding emergency disposal behaviors based on a target mechanism model and a trained deep neural network model; and splicing the plurality of emergency disposal behaviors to obtain an emergency disposal scheme of the distillation plant accident.
Specifically, S1: carrying out full-flow dynamic simulation on the distillation device, and constructing a distillation device mechanism environment model; s2: constructing an intelligent agent of emergency disposal personnel, and setting a deep reinforcement learning framework for the intelligent agent to perform an emergency disposal task of the distillation plant accident; s3: under a simulation environment, an intelligent agent of emergency disposal personnel continuously executes emergency disposal behaviors, inputs the behaviors into a distillation device mechanism environment model, explores accident situations, and stores obtained data into an experience pool; s4: randomly extracting a certain amount of data from the experience pool as a sample, and performing learning training on the deep neural network by using a deep reinforcement learning method to achieve convergence of reward superposition values acquired by the intelligent agent; s5: setting an accident scene of the distillation device, obtaining the optimal emergency disposal behavior corresponding to the accident scene according to the trained deep neural network model, splicing the emergency disposal behaviors selected by the intelligent agent, and generating a disposal flow.
In one embodiment, the original mechanistic environment model includes: a distillation plant dynamic model and a fault handling model; s1 includes:
s11: establishing a full-process steady-state model of the distillation device through process simulation; constructing a dynamic model of the distillation device based on the full-process steady-state model, the equipment size data of the distillation device and the controller parameters;
s12: and establishing a fault handling model around common emergency handling behaviors, and performing emergency handling simulation by using the fault handling model so as to obtain a mapping relation between the current emergency handling behaviors and variables in the dynamic model of the distillation device.
Specifically, the original mechanism environment model of the distillation device comprises a distillation device dynamic model and a fault handling model, wherein the distillation device dynamic model is established based on process simulation software and can simulate the real-time change condition of relevant process parameters of the distillation device; the fault handling model is established for emergency handling to the distillation plant. S11 includes: establishing a full-process steady-state model of the distillation device through process simulation software, setting equipment size data such as the number of tower plates, the plate spacing, the tower diameter and the like of the distillation device and the position and parameters of a controller on the basis, and establishing a dynamic model of the distillation device; s12 includes: the method comprises the steps of establishing a model capable of supporting simulation fault handling to perform emergency handling simulation around common emergency handling behaviors such as pump starting, valve opening, fire extinguishing and the like, and establishing a corresponding relation between the emergency handling behaviors and dynamic model variables of the distillation device. Wherein, the dynamic model of the middle tray of the rectifying tower in the distillation device comprises:
equation of material balance
The total material balance equation of the tower plate is as follows:
Figure BDA0003373081360000081
the material balance equation of each component of the tower plate is as follows:
Figure BDA0003373081360000082
combining formula (1) and formula (2) gives:
Figure BDA0003373081360000083
equation of phase equilibrium
yn,i＝kn,ixn,i (4)
Equation for normalizing mole fraction
Figure BDA0003373081360000091
Figure BDA0003373081360000092
Equation of energy balance
Figure BDA0003373081360000093
Because of the fact that
Figure BDA0003373081360000094
The value of (c) is very small, so to simplify the equation, neglecting it here, the tray gas phase flow calculation formula is obtained from equation (7):
Figure BDA0003373081360000095
the symbols in the formulae (1) to (8) have the meanings shown in Table 1.
Figure BDA0003373081360000096
TABLE 1 symbols and their meanings
In one embodiment, the states of the agent defined in S2 include: accident information, running state information, operating state information and target information;
the accident information comprises an accident type, an accident occurrence position and equipment fault information; the running state information comprises the change condition of each process parameter; the operation state information includes: the location of the agent and the fire equipment it carries; the target information refers to the current region to which the agent is going.
Specifically, the agent status S is defined as S (event, State, Pos, Goal), and includes Accident information, operation status information, and target information. The accident information comprises an accident type, an accident occurrence position and equipment fault information; the running state information refers to the change condition of each process parameter; the operation state information comprises the position of the intelligent agent and fire fighting equipment carried by the intelligent agent; the target information refers to the current region to which the agent is going. The location of the accident and the location of the agent are both represented by two-dimensional coordinates. Defining the action set of the intelligent agent, including various emergency treatment behaviors such as moving, opening an inlet valve of a pump, opening an outlet valve of the pump, closing the inlet valve of the pump, closing the outlet valve of the pump, starting the pump, taking out fire-fighting equipment, treating leaked oil, extinguishing fire, pressing an alarm and the like.
In one embodiment, the emergency mission reward function defined in S2 is: r ═ rtime+rpara+rresult(ii) a Wherein the content of the first and second substances,
rtimethe reward corresponding to the execution time for the emergency treatment,
Figure BDA0003373081360000101
k is a positive constant; [ t ] ofi,1,ti,2]Time window for item i emergency treatment, tiTime to perform the emergency disposition;
rparathe reward, r, corresponding to the maximum deviation of the process parameters of the distillation apparatusparaThe definition is as follows:
Figure BDA0003373081360000102
c is the penalty of the maximum deviation of the parameters being more than 10 percent;
rresultreward corresponding to harm caused by improper disposal of intelligent agent, and harm r caused by improper disposal of emergency disposal personnelresultIs shown as rresult＝{d1,d2,d3,…,dn}，d1,d2,d3,…,dnMark punchTower flooding, leakage, fire and explosion.
In particular, the reward r is determined according to the time at which the emergency treatment is performedtimeThe reward is used for ensuring the timeliness of emergency treatment, rtimeThe definition is as follows:
Figure BDA0003373081360000111
where k is a positive constant, the weight of the portion of the reward; [ t ] ofi,1,ti,2]Time window for item i emergency treatment, tiTime to perform the emergency treatment.
Determining the reward r according to the maximum deviation of the process parameters of the distillation plantparaThe reward is used to ensure the correctness of emergency treatment,
Figure BDA0003373081360000112
where c is a positive constant, a penalty of greater than 10% of the maximum deviation of the parameter.
Determining a reward r from a hazard caused by improper disposal of an agentresultWhen the temperature at the top of the tower drops, the liquid level of the reflux tank drops, the liquid level and the pressure at the bottom of the tower rise and the like, the tower is flooded; when the temperature and pressure at the top of the tower rise, the liquid level of the tower kettle falls and the like, the tower is flushed; when the temperature of oil in the pipeline of the reboiling furnace rises to 500 ℃, explosion occurs; when the leakage accident occurs and the disposal is not in time, the leakage quantity is increased to cause fire. When the accidents happen, part of process parameters of the distillation device exceed the threshold values, and the parameters are difficult to adjust to normal states through other treatment measures, the whole device must stop running, and r isresultThe definition is as follows: r isresult＝{d1,d2,d3,…,dn}。
Finally, the total reward function r of the intelligent agent emergency disposal task is the sum of all rewards, and the calculation formula is as follows: r ═ rtime+rpara+rresult。
In one embodiment, the deep neural network model includes: a predicted network and a target network; s3 includes:
s31: controlling the intelligent agent in the current intelligent agent state s by using a greedy algorithm to randomly select and execute an emergency disposal behavior according to the probability of epsilon, and selecting and executing an emergency disposal behavior with the maximum Q value according to the probability of 1-epsilon; the Q value is a result obtained by inputting the current agent state s into the prediction network;
s32: inputting the executed emergency disposal behavior a into an original mechanism environment model to obtain consequence data, calculating reward information r corresponding to the emergency disposal behavior a by using the consequence data and an emergency task reward function, and updating the state s' of the intelligent agent after the emergency disposal behavior a is executed;
s33: experience data (s, a, r, s ') formed by the current agent state s, the emergency disposition action a, the reward information r and the updated agent state s' are stored in an experience pool.
Wherein, the deep reinforcement learning algorithm is defined as DQN algorithm, the deep neural network structure comprises a prediction network and a target network, and the parameters are respectively defined as theta and theta-The two networks have the same structure, and the network layers are sequentially as follows: the intelligent agent state s input layer, two full connection layers and an action value output layer, wherein the activation function of the full connection layer is a ReLU function.
Specifically, after the deep reinforcement learning framework is defined, sample data for training the neural network needs to be acquired, and it is obviously unrealistic to acquire the sample data from the reality, so that under the simulation condition established in S1, the agent continuously takes emergency treatment behaviors through a greedy algorithm and inputs the emergency treatment behaviors into the distillation device environment mechanism model, calculates rewards of the emergency treatment behaviors according to consequence data acquired by the mechanism environment model, and stores state information, the taken emergency treatment behaviors and corresponding rewards in an experience pool. In this embodiment, the specific implementation steps are as follows:
under a simulation environment, an intelligent agent of emergency disposal personnel continuously executes an emergency disposal action, and the distillation device environment and a mechanism model calculate the result of the emergency disposal action to obtain the reward corresponding to the action; and the intelligent agent continuously acquires the current intelligent agent state s of the intelligent agent, inputs the s into the prediction network and obtains output Q(s), namely a Q value sequence of all emergency treatment behaviors.
Selecting and executing emergency disposal behaviors a randomly according to the probability of epsilon, and selecting and executing the emergency disposal behavior a with the maximum Q value according to the probability of 1-epsilon; in order for the agent to explore the unknown state more, the initial value of epsilon is larger, e.g. set to 0.9, after which the agent reduces the value of epsilon a little for each emergency treatment action performed, i.e. reduces the probability of randomly selecting an emergency treatment action, e.g. reduces epsilon by e for each emergency treatment action performed-6(ii) a Under the action of the emergency disposal behavior a, obtaining the state s' of the next moment, and calculating the obtained reward r; store (s, a, r, s') in the experience pool D.
In one embodiment, S4 includes:
s41: randomly extracting a plurality of groups of experience data (s, a, r, s') from an experience pool;
s42: a target value y is calculated from the target network,
Figure BDA0003373081360000131
s43: updating parameters of the prediction network by using a gradient descent method, wherein the calculation formula of the loss function is as follows: loss ═ y-Q(s)t,a；θ))2；
S44: synchronizing the updated parameter theta of the prediction network to the parameter theta of the target network after a preset number of time steps-；
S45: and repeating the training of the deep neural network from S41 to S44 until the deep neural network converges, and saving the parameters of the prediction network at the moment.
Specifically, a certain amount of data is randomly extracted from an experience pool to serve as a sample, and a deep neural network is learned and trained by using a deep reinforcement learning method. In this embodiment, the specific implementation steps are as follows:
firstly, only storing data in an experience pool, and extracting the data for training when the data amount reaches a certain degree, such as 200 pieces of data;
randomly extracting a batch of data (s, a, r, s') from the experience pool, wherein the data amount extracted each time can be adjusted according to needs, such as randomly extracting 128 groups of data each time;
calculating a target value y according to the target network, wherein the formula is as follows:
Figure BDA0003373081360000132
where gamma is the attenuation coefficient, st+1Is the state of the next moment, theta-Parameters of the target network;
updating the parameters of the prediction network according to a gradient descent method, wherein an optimizer is Adam, and the calculation formula of Loss is as follows: loss ═ y-Q(s)t,a；θ))2，stThe state is the current time;
after a certain time, synchronizing theta of the parameters of the prediction network to theta of the target network-I.e. theta-＝θ；
Sixthly, continuously circulating the processes from the second step to the fifth step, training the deep neural network until the deep neural network converges, and storing the parameters of the prediction network.
In one embodiment, S5 includes:
s51: updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model;
s52: inputting the current state of the agent into a trained deep neural network to obtain a Q value; executing a target emergency disposal behavior corresponding to the maximum Q value, and inputting the target emergency disposal behavior into a target mechanism environment model to obtain a corresponding intelligent agent state after the emergency disposal behavior is executed; inputting the updated state of the agent into the trained deep neural network to update the Q value, executing the target emergency disposal behavior corresponding to the updated maximum Q value, and so on; until the emergency disposal task is completed;
s53: and splicing the executed multiple target emergency disposal behaviors to obtain an accident emergency disposal scheme of the distillation device.
Specifically, after the deep neural network training is completed, the accident scene of the distillation device is set, and the emergency disposal behaviors selected by the intelligent agent are spliced according to the trained deep neural network model to generate a disposal flow. As shown in fig. 2, the method provided by the present invention specifically includes the following steps: enabling the dynamic model of the distillation device to run to a stable state, changing relevant process parameters in the dynamic model according to the set accident scene, initializing other parameters, loading stored parameters of a prediction network, and starting emergency disposal simulation; the method comprises the steps that an intelligent agent of emergency disposal personnel obtains a current intelligent agent state s as input of a prediction network, output Q(s) of the prediction network is obtained, emergency disposal behaviors with the maximum Q values are executed, the behaviors are input into a distillation device mechanism environment model, state information is updated according to the behaviors, and the steps are repeated until the intelligent agent completes an accident emergency disposal task of the distillation device; and splicing the emergency disposal behaviors executed by the intelligent agent to generate an accident emergency disposal scheme of the distillation device.
According to another aspect of the present invention, there is provided an accident emergency plan generating apparatus of a distillation apparatus, including: the device comprises a construction module, a setting module, an acquisition module, a training module and a generation module, wherein the construction module is used for carrying out full-flow dynamic simulation on the distillation device so as to construct an original mechanism environment model corresponding to the distillation device; the setting module is used for constructing an agent corresponding to the emergency disposal personnel and setting a deep learning framework of the agent, and the deep learning framework comprises the definition of an agent state, an emergency task reward function and a deep neural network model; the acquisition module is used for controlling the intelligent agent in the current intelligent agent state to select and execute the current emergency disposal behavior, inputting the current emergency disposal behavior into the original mechanism environment model, calculating reward information by using an emergency task reward function, and updating the intelligent agent state; storing the current agent state, the current emergency disposal behavior, the reward information and the updated agent state as experience data into an experience pool; the training module is used for randomly extracting experience data from the experience pool and inputting the experience data into the deep neural network for learning training until the superposition value corresponding to the emergency task reward of the intelligent agent is converged; the generating module is used for updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model; acquiring a plurality of corresponding emergency disposal behaviors based on a target mechanism model and a trained deep neural network model; and splicing the plurality of emergency disposal behaviors to obtain an emergency disposal scheme of the distillation plant accident.
The invention also provides an accident emergency disposal scheme generation system of the distillation device, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor realizes the steps of the accident emergency disposal method when executing the computer program.
The present invention also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, carries out the steps of the emergency handling method of an accident.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (10)

1. A method for generating an emergency disposal plan for a distillation apparatus, comprising:
s1: carrying out full-flow dynamic simulation on a distillation device to construct an original mechanism environment model corresponding to the distillation device;
s2: constructing an agent corresponding to emergency disposal personnel, and setting a deep learning framework of the agent, wherein the deep learning framework comprises defining an agent state, an emergency task reward function and a deep neural network model;
s3: controlling the intelligent agent in the current intelligent agent state to select and execute the current emergency disposal behavior, inputting the current emergency disposal behavior into the original mechanism environment model, calculating reward information by using the emergency task reward function, and updating the intelligent agent state; storing the current agent state, the current emergency disposition behavior, the reward information, and the updated agent state as experience data into an experience pool;
s4: randomly extracting the experience data from the experience pool and inputting the experience data to the deep neural network for learning training until a superposition value corresponding to the emergency task reward of the agent is converged;
s5: updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model; acquiring a plurality of corresponding emergency disposal behaviors based on the target mechanism model and the trained deep neural network model; and splicing the plurality of emergency disposal behaviors to obtain an emergency disposal scheme of the distillation plant accident.
2. The emergency incident disposition scheme generating method of a distillation apparatus according to claim 1, wherein the primitive mechanistic environment model includes: a distillation plant dynamic model and a fault handling model; the S1 includes:
s11: establishing a full-process steady-state model of the distillation device through process simulation; constructing a distillation plant dynamic model based on the full process steady state model, the equipment dimensional data of the distillation plant and the controller parameters;
s12: and establishing a fault handling model around common emergency handling behaviors, and performing emergency handling simulation by using the fault handling model so as to obtain a mapping relation between the current emergency handling behaviors and variables in the distillation plant dynamic model.
3. The emergency incident disposition scheme generating method of a distillation apparatus according to claim 1, wherein the intelligent state defined in S2 includes: accident information, running state information, operating state information and target information;
the accident information comprises an accident type, an accident occurrence position and equipment fault information; the running state information comprises the change condition of each process parameter; the operating state information includes: the location of the agent and the fire equipment it carries; the target information refers to a current region to which the agent is going.
4. The emergency treatment plan generation method for accidents of distillation equipment according to claim 1, wherein the emergency mission reward function defined in S2 is: r ═ rtime+rpara+rresult(ii) a Wherein the content of the first and second substances,
rtimethe reward corresponding to the execution time for the emergency treatment,
Figure FDA0003373081350000021
k is a positive constant; [ t ] ofi,1,ti,2]Time window for item i emergency treatment, tiTime to perform the emergency disposition;
rparathe reward, r, corresponding to the maximum deviation of the process parameters of the distillation apparatusparaThe definition is as follows:
Figure FDA0003373081350000022
c is the penalty of the maximum deviation of the parameters being more than 10 percent;
rresultreward for harm caused by improper disposal of agent, rresultIncluding accident hazard indexes corresponding to tower rushing, tower flooding, leakage, fire and explosion.
5. The emergency incident treatment plan generation method of a distillation apparatus according to claim 1, wherein the deep neural network model comprises: a predicted network and a target network; the S3 includes:
s31: controlling the intelligent agent in the current intelligent agent state s by using a greedy algorithm to randomly select and execute an emergency disposal behavior according to the probability of epsilon, and selecting and executing an emergency disposal behavior with the maximum Q value according to the probability of 1-epsilon; the Q value is a result obtained by inputting the current agent state s into the prediction network;
s32: inputting the executed emergency disposal behavior a into the original mechanism environment model to obtain consequence data, calculating reward information r corresponding to the emergency disposal behavior a by using the consequence data and the emergency task reward function, and updating the state s' of the agent after the emergency disposal behavior a is executed;
s33: and storing experience data (s, a, r, s ') formed by the current agent state s, the emergency handling behavior a, the reward information r and the updated agent state s' into an experience pool.
6. The emergency incident disposition scheme generating method of a distillation apparatus according to claim 5, wherein the S4 comprises:
s41: randomly extracting a plurality of sets of said experience data (s, a, r, s') from said experience pool;
s42: a target value y is calculated from the target network,
Figure FDA0003373081350000031
gamma is the attenuation coefficient, st+1For the state of the agent at the next moment, θ-A parameter of the target network;
s43: updating the parameters of the prediction network by using a gradient descent method, wherein the calculation formula of the loss function is as follows: loss ═ y-Q(s)t,a；θ))2，stThe state of the agent at the current moment is theta, and theta is a parameter of the prediction network;
s44: synchronizing the updated parameter theta of the prediction network to the parameter theta of the target network after a preset number of time steps-；
S45: and repeating the training of the deep neural network from S41 to S44 until the deep neural network converges, and saving the parameters of the prediction network at the moment.
7. The emergency incident disposition scheme generating method of a distillation apparatus according to any one of claims 1 to 6, wherein the S5 includes:
s51: updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model;
s52: inputting the current state of the agent into the trained deep neural network to obtain a Q value; executing a target emergency disposal behavior corresponding to the maximum Q value, and inputting the target emergency disposal behavior into the target mechanism environment model to obtain a corresponding intelligent agent state after the emergency disposal behavior is executed; inputting the updated state of the agent into the trained deep neural network to update the Q value, executing a target emergency disposal behavior corresponding to the updated maximum Q value, and so on; until the emergency disposal task is completed;
s53: and splicing the executed target emergency disposal behaviors to obtain an accident emergency disposal scheme of the distillation device.
8. An emergency incident disposition scheme generating apparatus for a distillation apparatus, comprising:
the construction module is used for carrying out full-flow dynamic simulation on the distillation device so as to construct an original mechanism environment model corresponding to the distillation device;
the system comprises a setting module, a data processing module and a data processing module, wherein the setting module is used for constructing an intelligent agent corresponding to emergency disposal personnel and setting a deep learning framework of the intelligent agent, and the deep learning framework comprises the definition of an intelligent agent state, an emergency task reward function and a deep neural network model;
the acquisition module is used for controlling the intelligent agent in the current intelligent agent state to select and execute the current emergency disposal behavior, inputting the current emergency disposal behavior into the original mechanism environment model, calculating reward information by using the emergency task reward function and updating the intelligent agent state; storing the current agent state, the current emergency disposition behavior, the reward information, and the updated agent state as experience data into an experience pool;
the training module is used for randomly extracting the experience data from the experience pool and inputting the experience data into the deep neural network for learning training until a superposition value corresponding to the emergency task reward of the intelligent agent is converged;
the generating module is used for updating parameters in the original mechanism environment model according to the accident scene to be processed to obtain a target mechanism environment model; acquiring a plurality of corresponding emergency disposal behaviors based on the target mechanism model and the trained deep neural network model; and splicing the plurality of emergency disposal behaviors to obtain an emergency disposal scheme of the distillation plant accident.
9. An emergency disposal plan generating system for a distillation plant, comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.
CN202111406804.8A 2021-11-24 2021-11-24 Accident emergency disposal scheme generation method, device and system for distillation device Pending CN114169711A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111406804.8A CN114169711A (en) 2021-11-24 2021-11-24 Accident emergency disposal scheme generation method, device and system for distillation device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111406804.8A CN114169711A (en) 2021-11-24 2021-11-24 Accident emergency disposal scheme generation method, device and system for distillation device

Publications (1)

Publication Number Publication Date
CN114169711A true CN114169711A (en) 2022-03-11

Family

ID=80480743

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111406804.8A Pending CN114169711A (en) 2021-11-24 2021-11-24 Accident emergency disposal scheme generation method, device and system for distillation device

Country Status (1)

Country Link
CN (1) CN114169711A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115569395A (en) * 2022-10-13 2023-01-06 四川大学 Rectifying tower intelligent safety monitoring method based on neural network
CN115645972A (en) * 2022-09-13 2023-01-31 安徽理工大学 Extraction equipment for chemical engineering pharmacy

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115645972A (en) * 2022-09-13 2023-01-31 安徽理工大学 Extraction equipment for chemical engineering pharmacy
CN115645972B (en) * 2022-09-13 2024-04-16 安徽理工大学 Extraction equipment for chemical pharmacy
CN115569395A (en) * 2022-10-13 2023-01-06 四川大学 Rectifying tower intelligent safety monitoring method based on neural network

Similar Documents

Publication Publication Date Title
Li et al. Dam monitoring data analysis methods: A literature review
CN114169711A (en) Accident emergency disposal scheme generation method, device and system for distillation device
Su et al. Smart performance-based design for building fire safety: Prediction of smoke motion via AI
CN100412871C (en) System and method to generate domain knowledge for automated system management
Abdolhamidzadeh et al. A new method for assessing domino effect in chemical process industry
Martorell et al. Constrained optimization of test intervals using a steady-state genetic algorithm
Ferreira et al. Comparing evolutionary methods for reservoir computing pre-training
Intisar et al. Classification of online judge programmers based on rule extraction from self organizing feature map
Cheng et al. A decision-making framework for load rating planning of aging bridges using deep reinforcement learning
Vavilova et al. Application of genetic algorithm for adjusting the structure of multilayered neural network for prediction of investment processes
Ren et al. A new interval prediction method for displacement behavior of concrete dams based on gradient boosted quantile regression
Paredes-Frigolett et al. A novel method for rule extraction in a knowledge-based innovation tutoring system
CN113537614A (en) Construction method, system, equipment and medium of power grid engineering cost prediction model
CN117333011A (en) Fire risk assessment method, device, equipment and medium
Brajnik et al. Qualitative modeling and simulation of socio-economic phenomena
Espinosa-Paredes et al. Modeling of the High Pressure Core Spray Systems with fuzzy cognitive maps for operational transient analysis in nuclear power reactors
CN113300884B (en) GWO-SVR-based step-by-step network flow prediction method
Mateou et al. Tree-structured multi-layer fuzzy cognitive maps for modelling large scale, complex problems
CN112700005B (en) Abnormal event processing method and device based on Monte Carlo tree search
CN111221741B (en) Method for automatically generating abnormal unit test based on genetic algorithm and log analysis
CN111160662A (en) Risk prediction method, electronic equipment and storage medium
Ajdani et al. Providing a consistent method to model the behavior and modelling intrusion detection using a hybrid particle swarm optimization-logistic regression algorithm
CN111179070A (en) Loan risk timeliness prediction system and method based on LSTM
Núñez-Carrera et al. Failure analysis of the standby liquid control system for a boiling water reactor with fuzzy cognitive maps
CN110942259B (en) Community gas equipment risk assessment method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination