Method and system for optimally controlling power of short-process electric arc furnace of iron and steel enterprise

Abstract

The invention provides an iron and steel enterprise short-process electric arc furnace power optimization control method and system. The method comprises the following steps that the production process of an iron and steel enterprise is analyzed; according to an analysis result, establishing a steel enterprise short-process production scheduling resource task model; establishing an iron and steel enterprise short-process electric arc furnace power optimization control model; and based on the established electric arc furnace power optimization control model. On the premise of considering the power demand response demand of the power grid, the power demand of the iron and steel enterprise and the incidence relation of the production process, the power utilization response behavior of the enterprise is analyzed, the response potential of the production process is evaluated and excavated, a feasible scheduling strategy with supply and demand interaction between the power grid and the enterprise as the characteristic is provided, the power grid is promoted to stabilize peak and valley, and the enterprise is guided to reduce cost and increase efficiency.

Description

Technical Field

[0001] The present invention relates to the field of coordinated optimization and scheduling of energy systems, and in particular to a power optimization control method and system for a short-process electric arc furnace in a steel enterprise. Background Art

[0002] Steel enterprises have many problems, such as large electricity consumption, high electrification level, many types of adjustable loads, and less constraints on production safety conditions. They are the key targets for exploring the potential of load-side demand response. Especially for short-process steel enterprises that use scrap steel and other metal waste as input raw materials and the arc of the electrode of the electric arc furnace as the smelting heat source, compared with long-process steel enterprises that use iron ore as smelting raw materials and fossil energy such as coke and coal gas as the main source of melting energy, they are more in line with the requirements of deepening energy conservation and low-carbon transformation in the industrial field and promoting green transformation of the steel industry in terms of steelmaking energy consumption, production efficiency, carbon emissions, etc.

[0003] The short process of steel production is essentially a mixed production process with large-scale, multi-stage, multi-parallel equipment, multi-intermediate products, batch and continuous production. Scrap steel is processed into finished or semi-finished products such as steel and ingots after a series of physical changes and chemical reactions in multiple processes and multiple production modes. At the same time, power resources are continuously consumed in the process of executing production tasks to maintain the normal operation of processing equipment, which makes the material conversion, equipment operation and power demand of the entire manufacturing process present a complex relationship of coupling, coexistence and interdependence. As a process-based production industry, the power load regulation of the steel industry in its production process generally follows strict process restrictions. The start-up and operation time of the equipment is determined in advance according to the production schedule, and power cannot be cut off at will once it is working. At present, traditional load regulation technology cannot analyze the power response behavior of enterprises and optimize the power control of enterprises under the premise of considering the constraints associated with production resources such as materials, power, and equipment in various links of steel production, thereby affecting the mining of demand response potential of steel enterprises.

[0004] Wang Haibo, & Zhang Li. (2021). Supply and demand interactive scheduling model considering the production process of short-process steel enterprises. Automation of Electric Power Systems, 45(15), 13. Aiming at short-process steel enterprises whose adjustable potential is closely related to the production process, a supply and demand interactive scheduling model considering the production process of such users is proposed to mobilize their enthusiasm for participating in grid scheduling and promote the consumption of renewable energy. The resource task network method is used to model the coupling relationship between each link of steel production and production resources, and the processing method of the connection relationship between the processes is improved to make the formulated production operation plan more realistic and feasible; the different interests of the two parties, grid scheduling and enterprise users, are comprehensively considered to establish a two-level planning scheduling model to truly reflect the interactive game process between power supply and demand; in order to match the model solution with the interactive process, a new heterogeneous decomposition algorithm is adopted based on the hierarchical iteration idea to accelerate the convergence speed of the solution while ensuring that only real-time electricity prices and purchased power are interacted between layers. This method does not convert certain specific process operation requirements in the production process into constraints on the available quantity of resources and the start time of the task according to the characteristics of each process and process of the short-process steel production process. It also does not analyze the electricity response behavior of the enterprise under the premise of taking into account the relationship between the power demand response needs of the power grid, the electricity demand of the steel enterprise and the production process, and solve the power optimization control problem of the short-process electric arc furnace of the steel enterprise under different demand response invitations.

[0005] Therefore, there is an urgent need for a power optimization control method for steel enterprises to solve the power optimization control problem of short-process electric arc furnaces in steel enterprises under different demand response invitations. Summary of the invention

[0006] In order to solve the deficiencies in the prior art, the present invention provides a power optimization control method for a short-process electric arc furnace in a steel enterprise. First, the production process of the steel enterprise is analyzed, and the interactive relationship between the production and transportation processes and various production resources is clarified according to the process characteristics of each process in the short process of steel production and the power consumption characteristics of the main production equipment; then, the resource and timing constraints in the steel production process are abstracted into a mathematical model using the Resource-Task Network (RTN) method, and an improvement plan is proposed for the deficiencies in the way the connection relationship between processes is handled, so as to express the power demand of the short-process steel enterprise under the premise of ensuring the uninterrupted operation of the connected processes, thereby laying a model foundation for the optimization of its power consumption behavior.

[0007] The present invention adopts the following technical solution.

[0008] The present invention provides a power optimization control method for a short-process electric arc furnace in a steel enterprise, the method comprising the following steps:

[0009] Step 1: Analyze the production process of steel enterprises and obtain the interactive relationship between equipment, materials and processes of steel enterprises;

[0010] Step 2: Based on the analysis results of step 1, establish a resource task model for short-process production scheduling of steel enterprises;

[0011] Step 3: Based on the short-process production scheduling resource task model of the steel enterprise established in step 2, establish the power optimization control model of the short-process electric arc furnace of the steel enterprise; solve the optimization control model to obtain the power optimization control strategy of the short-process electric arc furnace of the steel enterprise.

[0012] The present invention further includes the following preferred embodiments.

[0013] In step 1, analyzing the production process of the steel enterprise includes: obtaining the short-process production process of the steel enterprise; and analyzing the electricity consumption characteristics of the steel enterprise's production equipment.

[0014] The short-process production process of steel enterprises includes primary smelting, decarburization, refining, continuous casting and rolling.

[0015] The electricity consumption characteristics of steel enterprise production equipment are as follows:

[0016]

[0017] Where: P i t is the power consumption of the equipment at time t; is the startup time of device i; is the operating time of the equipment, P i rated is the rated power of device i.

[0018] In step 1, the interactive relationship between the steel enterprise equipment, materials and processes includes: at any time, a steel enterprise equipment can only process one heat; all heats in each casting are continuously cast in a set order; the equipment and heats only interact at the start and end of the production process; and the heat is transported immediately after the production process is completed.

[0019] In step 2, establishing a resource task model for short-process production scheduling of steel enterprises includes: dividing the components of the production process into resources and tasks, and obtaining the interaction process between resources and tasks; drawing a resource task framework diagram for short-process production scheduling of steel enterprises based on the obtained interaction process between resources and tasks; and establishing a mathematical expression for the available quantity of resources based on the drawn resource task framework diagram for short-process production scheduling of steel enterprises.

[0020] The mathematical expression for the available quantity of a resource is as follows:

[0021]

[0022] in, is the available quantity of resource r in time period t; is the start variable of task k at the beginning of period t, a value of 1 indicates start, and a value of 0 indicates not start; is the relative time point θ during the execution of task k k The quantity of output resource r, where negative values ​​represent input; It represents the number of resources r provided by the outside world in time period t; K is the task set; R is the resource set.

[0023] In step 3, establishing a power optimization control model for a short-process electric arc furnace in a steel enterprise includes: modeling the power optimization sequence decision process of a short-process electric arc furnace in a steel enterprise as a Markov decision process with T discrete time steps; and constructing an objective function of the arc furnace power optimization control model.

[0024] Modeling the power optimization sequence decision process of the short process electric arc furnace of the steel enterprise as a Markov decision process with T discrete time steps includes: constructing the state-value function of the steel enterprise; and constructing the state-behavior value function of the steel enterprise.

[0025] The formula of the state-value function of the steel enterprise is as follows:

[0026]

[0027] Among them, π indicates the probability of an action, r represents the immediate reward obtained by the agent after the state transition at a certain moment, γ is the discount factor, which indicates the importance of the agent to future rewards, and V π (st) represents the expected value of the cumulative return of the agent at a certain moment under the strategy π.

[0028] The formula of the steel enterprise state-behavior value function is as follows:

[0029]

[0030] Among them, q π (st,at) is the agent in state S t Next select action a t , r represents the immediate reward obtained by the agent after the state transfer at a certain moment, γ is the discount factor, which indicates the importance of the agent to future rewards, P represents the probability that the agent transfers to state s' after selecting action a in state s, and V(s) represents the expected value of the cumulative reward obtained by the agent after the state transfers to s' at a certain moment.

[0031] The objective function of the arc furnace power optimization control model is constructed according to the following formula:

[0032]

[0033] Where j represents the task running time, J represents the maximum task time, V(s j ) represents the state value of the steel enterprise at time t, and the agent is in the state transfer to s j 'The expected value of the cumulative return after , r j,t It represents the immediate reward obtained by the agent after the state transition at time j, and γ is the discount factor, which indicates the importance of the agent to future rewards.

[0034] At the same time, the present invention also provides a power optimization control system for a short-process electric arc furnace in a steel enterprise, the system comprising a production process analysis module, a resource task model construction module and a power optimization control model construction module;

[0035] The production process analysis module is used to analyze the production process of steel enterprises, including primary refining, decarburization, refining, continuous casting and rolling;

[0036] The resource task model building module builds a resource task model for short-process production scheduling of steel enterprises based on the analysis results of the production process analysis module;

[0037] The power optimization control model building module is used to establish the power optimization control model of the short-process electric arc furnace in steel enterprises.

[0038] A terminal comprises a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of any one of the methods.

[0039] A computer-readable storage medium stores a computer program, which implements the steps of any one of the methods when executed by a processor.

[0040] The beneficial effect of the present invention is that, compared with the prior art, it analyzes the electricity response behavior of enterprises under the premise of taking into account the relationship between the power demand response needs of the power grid and the electricity demand of steel enterprises and the production process, evaluates and taps the response potential of the production process, and proposes a feasible scheduling strategy characterized by the interaction between the power grid and the enterprise supply and demand, which promotes the power grid to smooth out peaks and valleys and guides enterprises to reduce costs and increase efficiency.

[0041] The present invention completes the common production process of long-process and short-process steel enterprises, the electricity consumption characteristics of production equipment, and the interactive analysis of equipment, material and process, uses the resource task network method to establish a short-process steel production scheduling RTN model, takes the steel enterprise as the intelligent body, and models the power optimization sequence decision process of the short-process electric arc furnace of the steel enterprise as a Markov decision process with T discrete time steps. The deep Q learning algorithm is used to train the electric arc furnace power optimization control model, and the power optimization control problem of the short-process electric arc furnace of the steel enterprise under different demand response invitations is solved, thereby forming a power optimization control method for the short-process electric arc furnace of the steel enterprise under demand response invitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a power optimization control method for a short-process electric arc furnace in a steel enterprise according to an embodiment of the present invention;

[0043] Figure 2 It is a short-process production flow chart of a steel enterprise in an embodiment of the present invention;

[0044] Figure 3 It is a RTN block diagram of the short-process refining process of a steel enterprise in an embodiment of the present invention;

[0045] Figure 4 A discrete time representation diagram in an embodiment of the present invention;

[0046] Figure 5 The figure is a graphical description of the available quantity of resources in the embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0048] The present invention provides a power optimization control method for a short-process electric arc furnace in a steel enterprise. Figure 1 The figure shows a flow chart of a power optimization control method for a short-process electric arc furnace in a steel enterprise, and the method comprises the following steps:

[0049] Step 1: Analyze the production process of the steel enterprise and obtain the interactive relationship between the equipment, materials and processes of the steel enterprise.

[0050] In one embodiment of the present invention, in step 1, analyzing the production process of the steel enterprise includes: obtaining the short-process production process of the steel enterprise; and analyzing the electricity consumption characteristics of the production equipment of the steel enterprise.

[0051] In one embodiment of the present invention, the production process of a short-process steel enterprise is specifically as follows: metal waste such as scrap steel is loaded into an electric arc furnace (EAF), electrified to melt it into molten steel and poured into a ladle, and then transported to an argon oxygen decarburization furnace (AOD) by transportation equipment such as an overhead crane and a crane for decarburization, and then transported to a refining furnace (LF) for deoxidation, desulfurization, and removal of inclusions and other refining operations. The refined molten steel is injected into a tundish above a continuous caster (CC), and is continuously cast into ingots through operations such as crystallizer cooling, ingot drawing by a straightening machine, and cutting by a flame cutting machine. Finally, different types of rolling mills are selected according to order requirements to roll the ingots into steel products that meet specifications, as shown in the following example. Figure 2 shown.

[0052] From the above production process, it can be seen that the main production equipment of short-process steel enterprises includes EAF, AOD, LF, CC and rolling mills, and the processes completed on them can be called primary refining, decarburization, refining, continuous casting and rolling in sequence. Considering that the order contracts received by steel enterprises do not have exactly the same requirements for steel shape specifications, and the working principles of different types of rolling mills usually have obvious differences, it is difficult to uniformly describe their electricity consumption rules. Therefore, this invention does not discuss the adjustable potential of the rolling process for the time being, and only takes EAF, AOD, LF, CC as key power-consuming equipment to analyze the influencing factors of their power consumption characteristics.

[0053] In one embodiment of the present invention, the analysis of the power consumption characteristics of the production equipment of the steel enterprise is as follows: Since the arcing and shutdown processes of EAF and LF are usually completed within 5-10 seconds, and the start and stop time of the two types of motor-driven equipment AOD and CC are also calculated in seconds, which can be ignored compared with their working hours, and the actual operating power of the above equipment has a small random fluctuation range due to changes in factors such as the temperature in the furnace and the state of the charge. From the scheduling level, its power consumption characteristics can be simplified as follows:

[0054]

[0055] Where: P i t is the power consumption of the equipment at time t; is the startup time of device i; is the operating time of the equipment, P i rated is the rated power of device i.

[0056] Since the rated power of the production equipment is known, the operating time of the production equipment (i.e. the duration of the corresponding production task) can be obtained through furnace and casting design based on the order information of downstream manufacturers. Therefore, the electricity consumption of short-process steel companies is only related to the start-up time of various production equipment. The latter is not only limited by the power supply capacity, but also depends on the interactive process of input and output, occupation and release of materials and equipment in the production and transportation process.

[0057] In one embodiment of the present invention, the interactive relationship between the equipment, materials and processes of a steel enterprise is as follows: According to the analysis of the steel production process, EAF, AOD and LF all process molten steel in batches in units of furnaces (molten steel contained in the same ladle), while CC uses the buffering effect of the tundish to combine multiple furnaces of similar steel grades into castings to achieve continuous casting of ingots. The continuous casting process cannot be interrupted to prevent the "interruption" of molten steel from causing product quality degradation or even equipment damage, but sufficient time must be left between adjacent castings to replace the tundish and crystallizer (referred to as the casting replacement operation). The former is to prevent the occurrence of piercing accidents caused by long-term high-temperature erosion, and the latter is to meet the requirements of the order contract for the size of the ingot. Therefore, steel manufacturing can be regarded as a large-scale, multi-stage, multi-intermediate product batch production process as a whole, in which the interactive relationship between the equipment and materials in the form of furnaces and castings and each production and transportation process can be summarized as follows:

[0058] 1) The interaction between equipment, materials and production processes in steel enterprises

[0059] ① Each batch is processed according to the established process path, but no specific production equipment is specified for each production process;

[0060] ②At any time, one device can only process one heat, and one heat can only be processed on one device;

[0061] ③ Each furnace is allowed to wait for a certain period of time between two adjacent production processes, but there is an upper limit to this time to prevent the molten steel from cooling too long and causing damage to product quality;

[0062] ④ All the batches in each pouring must be cast continuously in strict accordance with the given order without interruption in the middle;

[0063] ⑤ A certain amount of time is required between adjacent pours on the same continuous casting machine to facilitate the replacement of the tundish, crystallizer, and equipment adjustment;

[0064] ⑥ Equipment and furnaces only interact with the corresponding production process at the start and end times, and the available quantity of equipment and furnaces remains unchanged during the execution of the production process.

[0065] 2) The interaction between equipment, materials and transportation processes in steel enterprises

[0066] ① Since the available number of freight equipment such as cranes is generally not limited, the interaction between the transportation process and freight equipment can be temporarily ignored;

[0067] ② In order to speed up the production process, the furnace is transported immediately after the production process is completed;

[0068] ③ Since similar equipment is usually concentrated in the same area of ​​the factory, the furnace transportation time between adjacent production links has nothing to do with the specific production equipment, that is, the transportation process does not interact with the production equipment. Based on the interaction between the above equipment, materials, electricity and each process, the production constraints that must be met for steel production scheduling and even optimizing the electricity consumption behavior of the enterprise can be extracted. Considering that steel production scheduling is essentially a scheduling problem of batch processes, and there may be several parallel devices in each production stage, the present invention uses RTN, a modeling method that can effectively solve batch scheduling problems involving parallel operations of multiple similar devices, to model the short process of steel production.

[0069] Step 2: Based on the analysis results of step 1, establish a resource task model for short-process production scheduling of steel enterprises.

[0070] In one embodiment of the present invention, in step 2, establishing a short-process production scheduling resource task model for a steel enterprise includes: dividing the constituent elements of the production process into resources and tasks, and obtaining the interaction process between resources and tasks; drawing a short-process production scheduling resource task framework diagram for the steel enterprise according to the obtained interaction process between resources and tasks; and establishing an expression of the short-process production scheduling resource task for the steel enterprise based on the drawn short-process production scheduling resource task framework diagram for the steel enterprise.

[0071] In this implementation, the Resource-task Network (RTN) is a modeling tool that combines intuitive graphical representation with rigorous mathematical expression. Its graphical representation and mathematical model are composed of two basic elements: resources and tasks. After dividing the constituent elements of a production process into two categories according to the definitions of resources and tasks, the interaction process between the two is intuitively described and abstractly represented. The specific definitions and expressions are shown in Table 1.

[0072] Table 1 Basic elements of resource tasks

[0073]

[0074] When performing the refining task of short-process steelmaking, the input materials are the furnaces waiting for LF processing, the output materials are the furnaces after refining, the equipment used is LF, and the energy resources consumed are power resources. Directed arcs are used to represent the interaction between the above resources and the refining task, and the RTN block diagram can be drawn as follows: Figure 3As shown, the arrow pointing from resource to task indicates the input of resource, and vice versa indicates the output of resource. Since LF is occupied and released at the start and end of the refining task, its interaction with the refining task can be represented by a bidirectional arrow. In addition, considering that power resources need to be continuously consumed during the execution of the refining task to maintain the normal operation of LF (referred to as continuous interaction), while the interaction between materials and LF and the refining task only occurs at the start and end of the task (referred to as discrete interaction), the present invention uses solid arcs and dotted arcs to distinguish the two.

[0075] After using the interactive relationship between resources and tasks to draw the RTN block diagram of the production process, a more general mathematical abstract expression can be formed based on this intuitive graphical representation. Since the available number of resources only changes during the interaction with the corresponding tasks, the task start time is used as the decision variable, and the interaction between the task and the resources during execution is quantitatively described to establish a mathematical expression for the available number of resources. Then, by analyzing the interactive relationship between resources and tasks, relevant constraints are set on the available number of resources and the task start time, so as to achieve accurate modeling of the production process while ensuring that the order tasks are completed on time. Taking the refining process as an example, the mathematical modeling process of RTN can be summarized as follows:

[0076] ①Time display

[0077] Before establishing the mathematical model of RTN, we must first establish a unified model time representation method to describe the interaction process between various resources and tasks. At present, the commonly used time representation methods are mainly divided into two types: continuous time representation and discrete time representation. The former regards the length of the time interval as a variable parameter, and accurately arranges the start and stop times of each task in the scheduling time domain by constructing complex timing constraints that determine the execution order of each task: the latter divides the time range under study into equally spaced time periods, and assumes that the task is only started at the endpoints of the time period, thereby greatly narrowing the search range of the model solution space. Although the continuous time representation improves the solution accuracy of the model compared with the discrete time representation, it is easy to introduce nonlinear constraints into the model, which not only increases the difficulty of solving the model, but may even lead to unsolvable situations, and it is difficult to achieve coordination with the power grid dispatching using discrete time period decision-making examples. The RTN models based on the two time representation equations were compared and tested. The results show that the discrete time model is more suitable for handling large-scale industrial scheduling problems than the continuous time model, and the calculation time of the latter is 1-2 orders of magnitude higher than the former. For this reason, this project chooses to establish the mathematical model of RTN based on discrete time representation. The definitions of various time parameters used are as follows Figure 4 shown.

[0078] In the figure, δ represents the time step; d k represents the duration of task k, when the number of time periods τ is used k When making allowancek =[d k / δ], [·] is the rounding up operator; θ k represents the relative time points during the execution of task k with the start time as the reference time point, and its value range is {0, 1, …, τ k}; T is the total number of scheduling periods.

[0079] Taking the refining task of a certain batch h as an example, assuming that δ is 15 minutes and the duration of the refining task is 40 minutes, the number of time periods occupied by the refining task is [40 / 15]=3, and the relative time points during the execution of the refining task with the start time as the reference time point are {0, 1, 2, 3}.

[0080] ② Resource balance constraints

[0081] In discrete time representation, tasks are only started at the end of each time period. Therefore, the available quantity of resources in time period / is composed of the following factors: first, its available quantity in time period t-1; second, its interaction with all tasks at the beginning of time period / ; third, its quantity provided by the outside world in time period / . Based on this balance relationship, the mathematical expression of the available quantity of resources can be established as:

[0082]

[0083] Where: is the available quantity of resource r in time period t; is the start variable of task k at the beginning of period t, a value of 1 indicates start, and a value of 0 indicates not start; is the relative time point θ during the execution of task k k The amount of output resource r at (negative value indicates input); It represents the number of resources r provided by the outside world in time period t; K is the task set; R is the resource set.

[0084] For energy resources EN , if the energy storage equipment is not considered, its consumption in each period is equal to the supply, and there is no available quantity that can be transferred to the next period, so formula (2) can be simplified to:

[0085]

[0086] Among them, Π′ r represents the number of resources r provided by the outside world in time period t. The parameters representing the interaction between resources and tasks in formula (2) and formula (3) are Taken from the interaction matrix u between task k and resources k , which is defined as:

[0087] θ k =0 1 … τk

[0088]

[0089] Where: u k is s×(τ k +1)-dimensional matrix, s is the number of resource types required for production, {r 1 , r 2 ,…,r s} is a set of various resources required for production. If task k has no interaction with a resource, then it can be k Delete the corresponding all-zero row vector. Taking the refining task of heat h as an example, based on the previous assumptions, its interaction relationship matrix with resources can be expressed as:

[0090]

[0091] Where: is the refining task of heat h; Z is the heat h after refining; waste is the heat h to be refined; e is the electricity resource.

[0092] In order to illustrate the construction principle of formula (2) vividly, based on the interactive relationship matrix 4 between the refining task and resources of the heat h shown in formula (5), a graphical description is given, as follows: Figure 2-4 As shown in the figure, when refining task 2, is started in period t-8 and A2+0, the available quantity of furnace and LF will change at the starting point of period / . This is because the starting point of period / is 4, and the recorded L, is the relative time point where the furnace and LF interact. Even if the furnace output and LF release actually occur within the period, moving this interaction event to the starting point of the next period in discrete time representation does not affect the start of subsequent tasks. In addition, although the available quantity of power resources remains at 0, 2, interacts with power resources at every moment during execution. This process is expressed in equation (2) and Figure 5 The average power consumption at the beginning of the time period is used to represent the power demand, as shown in formula (6). The resulting power demand is fully provided by external power sources (such as purchasing power from the power grid).

[0093]

[0094] in, is the available quantity of resource r in time period t; is the available quantity of resource r in time period t-1; Represents the quantity of resources r provided by the outside world in time period t.

[0095] In summary, formula (2) closely links the available number of resources with the task launch capability. This can be used as a constraint to accurately express the availability of resources by using the balance relationship. At the same time, it also proposes a simpler control method for the difficult-to-arrange task execution sequence from the perspective of managing the available number of resources. This is called the resource balance constraint in the RTN model.

[0096] ③ Resource quantity constraints

[0097] In the actual production process, the available quantity of all resources must be within the given upper and lower non-negative boundary range. This resource quantity constraint can be expressed in mathematical language as:

[0098]

[0099] Where: is the lower bound of the available quantity of resource r: is the upper bound of the available quantity of resource r. h , the minimum available quantity of resources to interact with is 0, and the number of furnaces invested is L h in The output furnace layer L h out The maximum available quantity is 1. The available quantity of LF cannot exceed its initial quantity, and the available quantity of power resources is always kept at 0, that is:

[0100]

[0101] in, It represents the resources required to produce the furnace layer Lhout at time t. It represents the resources required to invest in furnace Lhin at time t, represents the resources required by LF at time t, Indicates the resources required by LF at the start time, Indicates the available amount of power resources at the start time.

[0102] ④Process constraints

[0103] In addition to ensuring the order of task execution and limiting the number of available resources, in specific process operations, it is also necessary to set relevant constraints on factors that affect the task start time, such as the number of task starts and the task connection relationship. Taking the refining task L as an example, it is required to be started only once within the scheduling time range. This constraint can be expressed as:

[0104]

[0105] in, It represents the number of times the refining task L is started at time t, and T represents the scheduling time.

[0106] ⑤Completion constraints

[0107] The final product quantity should meet the order production requirements at the end of the scheduling cycle, that is,

[0108]

[0109] Where: R final T is the available quantity of the final product in period T; order The order output.

[0110] Based on the above RTN block diagram drawing and mathematical modeling process, the RTN model can be constructed according to the interactive relationship between resources and tasks in the production process under study. From the entire modeling process, it can be seen that the RTN method has the following three advantages in solving batch scheduling problems: 1) RTN regards all production resources including materials, energy, and equipment as resources, and uniformly describes the input and output process when interacting with tasks representing processing operations. In particular, it is considered that there is no difference between the equipment that executes the task and other resources involved in production, so that the complexity of the model can be greatly reduced when dealing with production scheduling problems containing parallel equipment and the same processing time for similar equipment; 2) The RTN model is universal and universal. Although the present invention only analyzes the refining process of short-process steelmaking as an example when introducing RTN modeling, it can express the production scheduling problem of a large range of batch processes instead of modeling only a single process: 3) The RTN model contains only a few basic constraints, the model is simple and extensible, and based on this framework, it can easily convert certain specific process operation requirements in the production process into constraints on the available number of resources and the start time of the task. Therefore, RTN can provide an effective modeling tool for the present invention to analyze the production scheduling problem of a short process of steel production with large-scale, multi-stage, and multi-parallel equipment characteristics.

[0111] Step 3: Based on the short-process production scheduling resource task model of the steel enterprise established in step 2, establish the power optimization control model of the short-process electric arc furnace of the steel enterprise; solve the optimization control model to obtain the power optimization control strategy of the short-process electric arc furnace of the steel enterprise.

[0112] In one embodiment of the present invention, in step 3, establishing a power optimization control model for a short-process electric arc furnace of a steel enterprise includes: modeling the power optimization sequence decision process of a short-process electric arc furnace of a steel enterprise as a Markov decision process with T discrete time steps; and constructing the objective function of the arc furnace power optimization control model.

[0113] In one embodiment of the present invention, modeling the power optimization sequence decision process of a short-process electric arc furnace of a steel enterprise as a Markov decision process with T discrete time steps includes: constructing a state-value function of the steel enterprise; and constructing a state-behavior value function of the steel enterprise.

[0114] In this embodiment, according to the actions of the arc furnace at T moments, the trajectory sequence of the agent can be obtained, including the state, action, transfer state and immediate reward at each moment. The cumulative reward G of the agent during the training period is the cumulative discounted reward obtained by the agent from the beginning to T moments, as shown below:

[0115] G=r 0 +γr 1 +…+γ T-1-t r T-1 (11)

[0116] Among them, r 0 represents the instant reward obtained by the agent at time 0, r T-1 represents the instantaneous reward obtained by the agent at time T-1. In each state, the strategy π specifies the probability of an action, that is, π(a|s) = p(a|s). Under the strategy π, the expected value of the agent's cumulative return at time t is defined as the state-value function v π (s t ), as follows:

[0117]

[0118] Correspondingly, the agent is in state s t Next select action a t The expected value of the cumulative return obtained is defined as the state-behavior value function q π (s t ,a t ), as follows:

[0119]

[0120] The goal of reinforcement learning is to improve the action strategy π and find the optimal strategy π* so that the agent can obtain the maximum cumulative discounted reward by following the strategy π*. Therefore, in order to find the optimal action strategy π*, it is necessary to maximize qπ(s t ,a t ). Calculate qπ(s t ,a t ) is called strategy evaluation, according to qπ(s t ,a t ) is called a policy improvement. When taking actions based on the maximum qπ, the policy π is called a greedy policy, as shown below:

[0121]

[0122] In the process of interaction between the agent and the environment, in order to ensure the agent's utilization of strategy π and action exploration, the e-greedy algorithm is usually used.

[0123] In one embodiment of the present invention, the state space of the Markov decision process not only reflects the environmental information that the agent can observe, but also includes the changes in state after different actions are executed. The agent guides the action selection according to the state to obtain more rewards. The effective design of the state space is conducive to the correlation between the neural network learning state and action and the correlation between the state and reward.

[0124] Taking the steel enterprise as an intelligent agent, the observed state includes not only its own state, but also the state in the environment that is beneficial to decision-making. Considering the production scenario of demand response, the state space of the intelligent agent is defined as follows:

[0125]

[0126] in, is the time index of T discrete time steps, indicating the time information of the current agent; DR t is the demand response load regulation index starting at time t; p sub t is the demand response load regulation subsidy price starting at time t, in yuan / kWh; c t The unfinished indicators of the production plan for the day; is the time-of-use electricity price at time t; The status of the electric arc furnace, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch; The status of the refining Aod furnace, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch; The status of the refining LF furnace, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch; It is the status of the continuous casting process, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch.

[0127] In one embodiment of the present invention, the action space includes all actions involved in the intelligent agent's ability to change the state of the environment. For an arc furnace, the action space is the equipment power.

[0128] In another embodiment of the present invention, regardless of whether it is a value function-based or a policy-based reinforcement learning algorithm, the ultimate goal is to maximize the cumulative discount reward. By designing a reward mechanism consistent with the optimization goal, that is, the sum of the electricity cost of the arc furnace operation and the benefits of participating in the demand response, and the penalty for the model recording the unfinished production index at time T-1 as the reward points, the intelligent agent is guided to complete the predetermined production plan and guide the intelligent agent to select the optimal action in multiple states.

[0129] Reward mechanism design method: In order to find the optimal action strategy π*, it is necessary to maximize qπ(st,at). The process of calculating qπ(st,at) is called strategy evaluation. Improving strategy π according to the value of qπ(st,at) is called strategy improvement. When taking action based on the maximum qπ, strategy π is called a greedy strategy.

[0130] Multiple states:

[0131] in, is the time index of T discrete time steps, indicating the time information of the current agent; DR t is the demand response load regulation index starting at time t; p sub t is the demand response load regulation subsidy price starting at time t, in yuan / kWh; c t The unfinished indicators of the production plan for the day; is the time-of-use electricity price at time t; The status of the electric arc furnace, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch; The status of the refining Aod furnace, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch; The status of the refining LF furnace, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch; It is the status of the continuous casting process, including whether the equipment is occupied at time t, whether the equipment power is adjustable, and the processing progress of the current equipment processing batch.

[0132] In one embodiment of the present invention, in order to solve the problem of power control of short-process electric arc furnace in steel enterprises under demand response, the power optimization control method of electric arc furnace based on deep Q learning is trained. The state value and action-state value of the agent at time t are respectively expressed as V(s j ) and Q(s j ,a j ). Since the reward functions are both cost and benefit, the state value V(s j ) can be regarded as the accumulated income of steel enterprises.

[0133] According to the Bellman equation, the agent objective function can be regarded as maximizing the state value, as shown below:

[0134]

[0135] In the Markov decision process MDP model in the technical solution of the present invention, the state change value after the agent selects the matching action a in state s is determined, so The objective function of the arc furnace power optimization control model is as follows:

[0136]

[0137] Among them, j represents the task running time, J represents the maximum task time, V(s j ) represents the state value of the steel enterprise at time t, and the agent is in state transfer to s j 'The expected value of the cumulative return after , r j,t It represents the immediate reward obtained by the agent after the state transition at time j, and γ is the discount factor, which indicates the importance of the agent to future rewards.

[0138] In order to avoid the curse of dimensionality, the present invention uses a neural network to fit the state function V(s j ,θ). Among them, θ is the set of neural network model parameters.

[0139] V(s j ,θ)←V(s j ,θ)+α[r j，t +γV(s j ′,θ)-V(s j ,θ)] (19)

[0140] Among them, r j,t +γV(s j ',θ) is the TD target value, and α is the incremental step size. In order to stabilize the training process of the neural network, the present invention refers to the target network V'(s j ,θ') and the experience return pool M. V'(s j ,θ') is used to fit the TD target value, and the update process of the state value function of the agent is as follows:

[0141] V(s j ,θ)→V(s j ,θ)+α[r j,t +γV′(s j 'θ′)-V(s j ,θ)] (20)

[0142] For a sample m of size m sampled from the experience replay pool M, the update process of the parameter θ is as follows:

[0143]

[0144] Where lr is the learning rate of the neural network training. After completing multiple parameter updates of the network V(sj,θ), the weight parameters of V(sj,θ) are copied and updated to the target network V'(sj,θ').

[0145] In order to maintain a certain exploration rate during training, the training algorithm uses the e-greedy algorithm. During the training phase, the cloud scheduling platform samples random numbers from the interval [0,1]. The platform randomly determines the arc furnace power with a probability of e, and selects actions using the arc furnace power control strategy based on deep Q learning with a probability of (1-e). During the testing phase, e is set to 0.

[0146] In one embodiment of the present invention, a steel enterprise is selected for simulation. The steel enterprise has 2 electric arc furnaces, 2 AOD refining furnaces, and 2 LF refining furnaces. The rated power of the electric arc furnace is 4500kW. The time step δ=15min, and the energy consumption of one batch of the electric arc furnace is 6000kWh. The operation time of one batch of AOD furnace is 5δ=75min, the operation time of one batch of LF furnace is 75min, and the continuous casting batch is 45min.

[0147] Demand response invitation 1: start time 9:00; end time 11:00; response subsidy price 2.4 yuan / kWh; Demand response invitation 2: start time 14:00; end time 16:00; response subsidy price 4.8 yuan / kWh.

[0148] According to the simulation of this model, the enterprise load control is carried out. During the two time periods, the total response of the two arc furnace equipment is calculated to be 9400kWh, with a profit of 375,600 yuan. It is recommended to carry out load response.

[0149] At the same time, the present invention also realizes a power optimization control system for a short-process electric arc furnace in a steel enterprise, the system comprising a production process analysis module, a resource task model construction module and a power optimization control model construction module;

[0150] The production process analysis module is used to analyze the production process of steel enterprises, including primary refining, decarburization, refining, continuous casting and rolling;

[0151] The resource task model building module builds a resource task model for short-process production scheduling of steel enterprises based on the analysis results of the production process analysis module;

[0152] The power optimization control model building module is used to establish the power optimization control model of the short-process electric arc furnace in steel enterprises.

[0153] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0154] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0155] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0156] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A power optimization control method for a short-process electric arc furnace in a steel enterprise, characterized in that: The method comprises the following steps: Step 1: Analyze the production process of steel enterprises and obtain the interactive relationship between equipment, materials and processes of steel enterprises; Step 2: Based on the analysis results of step 1, establish a resource task model for short-process production scheduling of steel enterprises; Step 3: Based on the short-process production scheduling resource task model of the steel enterprise established in step 2, establish the power optimization control model of the short-process electric arc furnace of the steel enterprise; solve the optimization control model to obtain the power optimization control strategy of the short-process electric arc furnace of the steel enterprise.

2. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 1, characterized in that: In step 1, the analysis of the steel enterprise production process includes: Obtain the short-process production procedures of steel enterprises; Analyze the electricity consumption characteristics of steel enterprise production equipment.

3. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 2, characterized in that: The short-process production process of steel enterprises includes primary smelting, decarburization, refining, continuous casting and rolling.

4. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 2, characterized in that: The electricity consumption characteristics of steel enterprise production equipment are as follows: Where: P i t is the power consumption of the equipment at time t; is the startup time of device i; is the operating time of the equipment, P i rated is the rated power of device i.

5. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 1, characterized in that: In step 1, the interactions among equipment, materials, and processes of a steel enterprise include: At any given time, a piece of steel company equipment can only process one heat; All the heats in each casting batch are cast continuously in the set order; Equipment and furnaces only interact at the start and end of the production process; And the heat is transported immediately after the production process is completed.

6. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 1, characterized in that: In step 2, the resource task model for short-process production scheduling of steel enterprises is established, including: Divide the components of the production process into resources and tasks, and obtain the interaction process between resources and tasks; Draw a resource task diagram for short-process production scheduling of steel enterprises based on the interaction process between resources and tasks; Based on the drawn resource task diagram of short-process production scheduling of steel enterprises, a mathematical expression for the available quantity of resources is established.

7. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 6, characterized in that: The mathematical expression for the available quantity of a resource is as follows: in, is the available quantity of resource r in time period t; is the start variable of task k at the beginning of period t, a value of 1 indicates start, and a value of 0 indicates not start; is the relative time point θ during the execution of task k k The quantity of output resource r, where negative values ​​represent input; It represents the number of resources r provided by the outside world in time period t; K is the task set; R is the resource set.

8. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 1, characterized in that: In step 3, the power optimization control model of the short-process electric arc furnace of the steel enterprise is established, including: The power optimization sequence decision process of the short-process electric arc furnace in the steel enterprise is modeled as a Markov decision process with T discrete time steps. Construct the objective function of the arc furnace power optimization control model.

9. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 8, characterized in that: The power optimization sequence decision process of the short-process electric arc furnace in a steel enterprise is modeled as a Markov decision process with T discrete time steps, including: Construct the state-value function of steel enterprises; And construct the state-behavior value function of steel enterprises.

10. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 9, characterized in that: The formula of the state-value function of the steel enterprise is as follows: Among them, π indicates the probability of an action, r represents the immediate reward obtained by the agent after the state transition at a certain moment, γ is the discount factor, which indicates the importance of the agent to future rewards, and V π (st) represents the expected value of the cumulative return of the agent at a certain moment under the strategy π.

11. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 9, characterized in that: The formula of the steel enterprise state-behavior value function is as follows: Among them, q π (st,at) is the agent in state S t Next select action a t , r represents the immediate reward obtained by the agent after the state transfer at a certain moment, γ is the discount factor, which indicates the importance of the agent to future rewards, P represents the probability that the agent transfers to state s' after selecting action a in state s, and V(s) represents the expected value of the cumulative reward obtained by the agent after the state transfers to s' at a certain moment.

12. The power optimization control method for a short-process electric arc furnace in a steel enterprise according to claim 8, characterized in that: The objective function of the arc furnace power optimization control model is constructed according to the following formula: Where j represents the task running time, J represents the maximum task time, V(s j ) represents the state value of the steel enterprise at time t, and the agent is in the state transfer to s j 'The expected value of the cumulative return after , r j,t It represents the immediate reward obtained by the agent after the state transition at time j, and γ is the discount factor, which indicates the importance of the agent to future rewards.

13. A power optimization control system for a short-process electric arc furnace in a steel enterprise using the method according to any one of claims 1 to 12, the system comprising a production process analysis module, a resource task model construction module and a power optimization control model construction module, characterized in that: The production process analysis module is used to analyze the production process of steel enterprises, including primary refining, decarburization, refining, continuous casting and rolling; The resource task model building module builds a resource task model for short-process production scheduling of steel enterprises based on the analysis results of the production process analysis module; The power optimization control model building module is used to establish the power optimization control model of the short-process electric arc furnace in steel enterprises.

14. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images