Park electricity-hydrogen coupling intelligent energy scheduling method oriented to source load uncertainty

Abstract

The invention, which relates to the energy scheduling field, discloses a source-load-uncertainty-oriented park electro-hydrogen coupling intelligent energy scheduling method comprising the following steps: constructing a park electro-hydrogen coupling energy management and control model based on two-stage stochastic programming, the model being a scene-aggregated two-stage neural network structure; fitting the second-stage rescheduling cost of the park electro-hydrogen coupled energy system based on the neural network model; equivalently embedding the trained neural network model into the optimization problem; and solving the optimization problem, obtaining an electricity-hydrogen coupling scheduling instruction, and scheduling the park electricity-hydrogen coupling energy system according to the scheduling instruction. According to the invention, the two-stage scheduling model based on neural network fitting can give play to the advantages of data driving, considers the geographic factor characteristics of renewable energy and the characteristics of a production energy consumption mode, achieves the targeted precise modeling of a target park, and can guarantee the absolute optimization of a scheduling decision based on an optimization conversion neural network embedding method. And the safety and economy of scheduling operation are improved.

Description

Technical Field

[0001] This invention relates to the field of energy dispatching, and in particular to a smart energy dispatching method for park-based electricity-hydrogen coupling in the face of source-load uncertainty. Background Technology

[0002] The energy system of industrial parks is a core unit of modern industrial energy consumption, and its energy utilization efficiency and management level are directly related to the realization of the national "dual-carbon" strategic goals. Promoting the green and low-carbon transformation of industrial parks and building an efficient, clean, and low-carbon energy system has become an inevitable trend in current industrial development. Against this backdrop, green innovation in production processes is particularly crucial for the energy-intensive steel industry. Traditional blast furnace-converter long-process steelmaking technology heavily relies on fossil fuels such as coke, making it a major source of carbon and pollutant emissions within the parks. In recent years, hydrogen metallurgy technology has been considered a disruptive direction for the deep decarbonization of the steel industry. Hydrogen, as an excellent clean reducing agent, only produces water during the reduction ironmaking process, theoretically eliminating carbon dioxide emissions, thus greatly reducing environmental pollution and potentially improving reduction efficiency. With the rapid development of renewable energy (such as wind power and photovoltaics), the "green electricity to green hydrogen" technology route (i.e., water electrolysis to produce hydrogen) provides a clean hydrogen source guarantee for hydrogen-powered steelmaking. Therefore, integrated industrial park electro-hydrogen coupling systems combining multiple energy forms such as electricity and hydrogen have emerged. Such systems aim to significantly improve overall energy efficiency and reduce losses and waste of "green electricity" and "green hydrogen" in all stages of production, storage, transportation and use through the coupling and complementarity of multiple energy sources.

[0003] However, due to fluctuations in renewable energy output and temporary adjustments to industrial production plans within the park, the source and load of the park's electro-hydrogen coupling system exhibit high uncertainty, threatening the economic efficiency and reliability of the park's energy system operation. The uncertainty on the source side is mainly reflected in the output fluctuations of renewable energy sources (such as wind and solar power), which are affected by weather, seasons, and other factors, with prediction errors reaching 10% or more, making the traditional "demand-driven supply" dispatching model unsuitable. The uncertainty on the load side stems from the diverse energy consumption behaviors of various users within the industrial park, such as sudden load changes in electric arc furnaces in steel production and the start-up and shutdown impacts of hydrogen energy equipment, further exacerbating the difficulty of supply and demand matching. These factors necessitate that the park's energy management incorporate these uncertainties.

[0004] In recent years, uncertainty scheduling theory has been increasingly incorporated into energy management problems. Considering the operational scenarios of park energy management, the scheduling problem is often modeled as a two-stage optimization problem. Before uncertainty occurs, the first stage solves for energy pre-scheduling decisions to obtain the energy system's operational baseline. After source-load fluctuations occur, the second stage solves for system rescheduling decisions, i.e., the readjustment of each link in the system to ensure supply-demand balance. Based on the modeling method of uncertainty, existing technologies can be mainly divided into robust optimization and stochastic programming methods. Robust optimization uses the worst-case scenario as the boundary, which can ensure system safety, but the scheduling scheme is too conservative, leading to a significant decrease in economic efficiency. Stochastic programming describes uncertainty through probability distributions or scenario sets, but its performance is limited by two points: if uncertainty is assumed to be based on a specific probability distribution (such as a normal distribution), complex uncertainties in actual production conditions may lead to model mismatch, thus reducing the accuracy of park energy scheduling; if the system's operational performance under uncertainty is described based on scenarios, a massive number of scenarios are required to ensure representativeness, resulting in low computational efficiency and difficulty in meeting scheduling requirements in terms of solution time.

[0005] Therefore, those skilled in the art are dedicated to developing a smart energy dispatching method for electricity-hydrogen coupling in industrial parks that addresses source-load uncertainty. Summary of the Invention

[0006] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is that the energy management system for parks based on stochastic programming has low accuracy in energy scheduling under actual production conditions, low computational efficiency in massive scenarios, and difficulty in meeting scheduling requirements in terms of solution time.

[0007] To achieve the above objectives, this invention provides a smart energy dispatching method for electricity-hydrogen coupling in industrial parks, accommodating source-load uncertainty. The method includes the following steps: S101: Construct a park-based electric-hydrogen coupled energy management model based on two-stage stochastic programming. The model is set as a two-stage neural network structure of scenario aggregation, including a first-stage electric-hydrogen coupled energy scheduling problem model and a second-stage energy scheduling model that takes into account the uncertainty of production energy demand. S103: Fitting the second-stage rescheduling cost of the park's electric-hydrogen coupled energy system based on a neural network model; S105: Equivalently embed the trained neural network model into the optimization problem; S107: Solve the optimization problem using an optimization solver, obtain the electric-hydrogen coupling scheduling instructions, and schedule the park's electric-hydrogen coupling energy system according to the scheduling instructions.

[0008] Further, step S101 includes the following sub-steps: S1011: Determine the composition and equipment parameters of the park's electro-hydrogen coupling system to provide a parameter basis for subsequent optimization problems; S1012: Construct a deterministic first-stage energy scheduling model for the electro-hydrogen coupling problem, providing a model foundation for constructing the second-stage energy scheduling model; S1013: Construct the second-stage energy dispatch model for the aforementioned uncertainty, and determine the objective function and constraints of the second-stage model.

[0009] Furthermore, the park's electro-hydrogen coupling system includes an electrical subsystem and a hydrogen energy subsystem, wherein, In the power subsystem, non-gas turbine units, the power grid, and renewable energy sources work together as the power supply source. The power load includes the park's power load and the electrolysis hydrogen production load. The power subsystem is equipped with energy storage. In the hydrogen energy subsystem, the hydrogen source is jointly provided by the hydrogen pipeline network and the electrolytic hydrogen production in the park. The hydrogen load includes the hydrogen load in the park and the hydrogen-to-electricity load. The equipment parameters of the park's electro-hydrogen coupling system include the capacity parameters and operating efficiency parameters of the equipment in the power subsystem and the hydrogen energy subsystem.

[0010] Further, step S1012 includes the following sub-steps: S10121: Construct the objective function for the first-stage electro-hydrogen coupled energy dispatch problem model, where the objective function is the operating cost of the electro-hydrogen coupled energy system.

[0011] S10122: Construct the operating constraints of the power subsystem in the industrial park, including power balance constraints, generation limitation constraints, energy storage constraints, and power constraints; The power balance constraint is:

[0012] S10123: Construct the operational constraints of the hydrogen energy subsystem in the industrial park, including hydrogen energy load balance constraints, electricity-to-hydrogen energy conversion efficiency constraints, gas storage limitation constraints, and power constraints. The hydrogen energy load balance constraint is:

[0013] in, For the first phase of scheduling decisions, For unit electricity purchase cost, To purchase electricity, Unit gas purchase cost For the amount of gas purchased; for Non-gas power generation at any time for Hydrogen-to-electricity generation at any time for The park purchases electricity from the power grid at any time. for Real-time energy storage charging capacity, for Energy storage discharge rate at any time for Renewable energy generation at all times for Real-time park power load, for The amount of electricity used for the constant conversion of electricity to hydrogen; for Purchase hydrogen at any time for Hydrogen storage gas release rate at all times for Hydrogen production rate at any time during electro-hydrogen conversion for Hydrogen load consumption at any time for Hydrogen consumption during hydrogen-to-electricity conversion at any time for Constant hydrogen intake volume.

[0014] Furthermore, under the operational constraints of the power subsystem and the hydrogen energy subsystem, the second-stage energy dispatch model is as follows:

[0015] The objective function of the second-stage energy dispatch model is:

[0016]

[0017] in, For variables in a two-stage decision based on a one-stage decision, This is a factor affecting the second-stage uncertainty. For standby capacity costs, This represents the total cost of the second phase. To purchase electricity again cost, For the cost of repurchasing gas, To reduce costs by reducing workload.

[0018] Further, in step S103, the neural network model includes a multi-layer neural network, with the last layer being a fully connected neural network. The neural network model maps the input first-stage decision and the scene set, and outputs the expected target function value for the second stage. The specific mapping method is as follows:

[0019] The loss function of the neural network model is the mean squared error function:

[0020] in, For scene collection, For the number of embedding vectors, For embedding vector identifiers, Predict the probability of occurrence for the k-th scenario. For neural networks, For the scene.

[0021] Furthermore, the neural network model completes the mapping operation using the following steps: S1031: Input each scene in the scene set independently into the neural network. In this process, the scene set is embedded into a hidden layer; S1032: Sum and average the embedding vectors, and process the result before inputting it into the neural network. ; S1033: The neural network The final embedding vector of the scene set is obtained after processing.

[0022] Furthermore, in training the neural network model, the dataset is generated using the following method: S1034: Sample a set of scenarios with a random cardinality from an uncertain distribution to generate a random feasible first-stage decision; S1035: Solve the objective function of the second-stage energy dispatch model using a solver; S1036: Calculation The calculation results are then used as labels for the samples in the dataset.

[0023] Further, in step S105, after training, the neural network model is equivalently transformed into mixed-integer linear constraints and embedded into the optimization problem. The objective function of the second-stage energy scheduling model is fitted by mixed-integer linear programming, i.e.:

[0024] in, Neural networks with equivalent embeddings.

[0025] Further, step S107 includes the following sub-steps: S1071: Obtain K typical samples and infer the embedding vectors using the trained neural network model. ; S1072: Insert the embedding vector Substitute the solution into the embedded neural network model and solve it using an optimized solver. S1073: The optimization solver outputs a scheduling command for the park's hydrogen energy equipment; S1074: Send dispatch instructions to the energy equipment controller to achieve energy management.

[0026] In a preferred embodiment of the present invention, compared with the prior art, the present invention has the following beneficial effects: 1. The second-stage scheduling model based on neural network fitting in this invention can leverage the advantages of data-driven approaches, taking into account the geographical characteristics of renewable energy and the characteristics of production and energy consumption patterns, to achieve targeted and accurate modeling of target industrial parks.

[0027] 2. The neural network of the present invention can compress the scene and realize the structure of the number of scenes and the complexity of energy scheduling. Based on the neural network embedding method of optimization transformation, with the help of the optimization solver, the absolute optimality of scheduling decision can be guaranteed.

[0028] 3. This invention utilizes the concept of hierarchical neural networks to separate parameter inputs from scheduling decision inputs, thereby achieving effective control over the size of the embedded network, effectively reducing the size of the neural network that needs to be optimized, effectively improving the solution efficiency of energy scheduling problems, and enhancing the performance and applicability of the proposed framework.

[0029] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of an energy dispatching method according to a preferred embodiment of the present invention; Figure 2 This is a schematic diagram of a scenario illustrating a preferred embodiment of the energy dispatching method of the present invention; Figure 3 This is a schematic diagram of the neural network structure of an energy dispatching method according to a preferred embodiment of the present invention. Detailed Implementation

[0031] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0032] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0033] To address the shortcomings of existing technologies, this invention proposes a smart energy scheduling method for park-based electric-hydrogen coupling systems accommodating source-load uncertainty. This method is a stochastic programming approach based on neural network embedding, overcoming the limitations of traditional methods. The method aims to leverage the powerful adaptive capabilities of neural networks to fit the final operating cost of the park's electric-hydrogen coupling energy system under different first-stage scheduling decisions and source-load uncertainty disturbances. This method avoids energy scheduling errors caused by inaccurate mechanistic modeling, compresses various source-load fluctuation scenarios, and then equivalently embeds the neural network into the optimization problem through equivalent transformations, ultimately achieving precise energy scheduling by solving the optimization problem.

[0034] like Figure 1 As shown, this invention proposes a smart energy dispatching method for electricity-hydrogen coupling in industrial parks, addressing source-load uncertainty, specifically including the following steps: S101: Construct a park-based electric-hydrogen coupling energy management model based on two-stage stochastic programming.

[0035] In this embodiment, the model adopts a two-stage neural network structure for scenario aggregation, including a first-stage model of the electric-hydrogen coupling energy scheduling problem and a second-stage energy scheduling model that takes into account the uncertainty of production energy demand. The construction of the park electric-hydrogen coupling energy management model based on two-stage stochastic programming includes the following sub-steps: S1011: Determine the composition and equipment parameters of the park's electro-hydrogen coupling system to provide a parameter basis for subsequent optimization problems; S1012: Construct a deterministic first-stage model of the electro-hydrogen coupling energy dispatch problem, providing a model foundation for constructing a second-stage energy dispatch model; S1013: Construct a second-stage energy dispatch model with uncertainty, and determine the objective function and constraints of the second-stage model.

[0036] In this embodiment, the park's electro-hydrogen coupling system includes an electrical subsystem and a hydrogen energy subsystem, wherein, In the power subsystem, non-gas turbines, the power grid, and renewable energy serve as the power supply sources. The power load includes the park's power load and the electrolysis hydrogen production load. The power subsystem is equipped with energy storage. In the hydrogen energy subsystem, the hydrogen pipeline network and the electrolytic hydrogen production in the park serve as the hydrogen source, and the hydrogen load includes the hydrogen load in the park and the hydrogen-to-electricity load. The equipment parameters of the park's electro-hydrogen coupling system include the capacity parameters and operating efficiency parameters of the equipment in the power subsystem and the hydrogen energy subsystem.

[0037] The construction of a deterministic first-stage model for the electro-hydrogen coupled energy dispatch problem includes the following sub-steps: S10121: Construct the objective function for the first-stage model of the electro-hydrogen coupled energy dispatch problem. The objective function is the operating cost of the electro-hydrogen coupled energy system.

[0038] S10122: Construct operational constraints for the power subsystem in the industrial park, including power balance constraints, generation limitation constraints, energy storage constraints, and power constraints; The power balance constraints are:

[0039] S10123: Construct operational constraints for hydrogen energy subsystems in industrial parks, including hydrogen energy load balance constraints, electricity-to-hydrogen energy conversion efficiency constraints, gas storage limitation constraints, and power constraints. The hydrogen energy charge balance constraint is:

[0040] in, For the first phase of scheduling decisions, For unit electricity purchase cost, To purchase electricity, Unit gas purchase cost For the amount of gas purchased; for Non-gas power generation at any time for Hydrogen-to-electricity generation at any time for The park purchases electricity from the power grid at any time. for Real-time energy storage charging capacity, for Energy storage discharge rate at any time for Renewable energy generation at all times for Real-time park power load, for The amount of electricity used for the constant conversion of electricity to hydrogen; for Purchase hydrogen at any time for Hydrogen storage gas release rate at all times for Hydrogen production rate at any time during electro-hydrogen conversion for Hydrogen load consumption at any time for Hydrogen consumption during hydrogen-to-electricity conversion at any time for Constant hydrogen intake volume.

[0041] In this embodiment, under the operating constraints of the power subsystem and the hydrogen energy subsystem, the second-stage energy dispatch model is as follows:

[0042] The objective function of the second-stage energy dispatch model is:

[0043]

[0044] in, For variables in a two-stage decision based on a one-stage decision, This is a factor affecting the second-stage uncertainty. For standby capacity costs, This represents the total cost of the second phase. To purchase electricity again cost, For the cost of repurchasing gas, To reduce costs by reducing workload.

[0045] S103: Fitting the second-stage rescheduling cost of the park's electric-hydrogen coupled energy system based on a neural network model.

[0046] In this embodiment, the neural network model constructed by the present invention includes a multi-layer neural network, with the last layer being a fully connected neural network. The neural network model maps the input first-stage decision and the scene set, and outputs the expected target function value for the second stage. The specific mapping method is as follows:

[0047] The loss function of the neural network model is the mean squared error function:

[0048] in, For scene collection, For the number of embedding vectors, For embedding vector identifiers, Predict the probability of occurrence for the k-th scenario. For neural networks, For the scene.

[0049] The neural network model completes the mapping operation using the following steps: S1031: Input each scene in the scene set independently into the neural network. In this process, the scene set is embedded into a hidden layer; S1032: Sum and average the embedding vectors, and process the result before inputting it into the neural network. ; S1033: Neural Network The final embedding vector of the scene set is obtained after processing.

[0050] To train the neural network model, this embodiment uses the following method to generate the dataset: S1034: Sample a set of scenarios with a random cardinality from an uncertain distribution to generate a random feasible first-stage decision; S1035: Solve the objective function of the second-stage energy dispatch model using a solver; S1036: Calculation The calculation results are then used as labels for the dataset samples.

[0051] S105: Equivalent embedding of a trained neural network model into an optimization problem.

[0052] In this embodiment, after the neural network model is trained, it is equivalently transformed into mixed-integer linear constraints and embedded into the optimization problem. The objective function of the second-stage energy scheduling model is fitted by mixed-integer linear programming, that is:

[0053] in, Neural networks with equivalent embeddings.

[0054] S107: Solve the optimization problem using an optimization solver, obtain the electric-hydrogen coupling scheduling instructions, and schedule the electric-hydrogen coupling energy system in the park according to the scheduling instructions.

[0055] In this embodiment, the following sub-steps are included: S1071: Obtain K typical samples and infer the embedding vectors using a trained neural network model. ; S1072: Embed the vector Substitute the solution into the embedded neural network model and solve it using an optimized solver; S1073: Optimize the solver output to schedule the park's hydrogen energy equipment; S1074: Send dispatch instructions to the energy equipment controller to achieve energy management.

[0056] Compared with existing technologies, the intelligent energy scheduling method for park-based electricity-hydrogen coupling oriented to source-load uncertainty provided by the embodiments of the present invention has the following beneficial technical effects.

[0057] 1. Existing technologies for park energy management based on stochastic programming often rely on assumed probability distributions or sample approximations. However, assuming uncertainty is based on a specific probability distribution (such as a normal distribution) can lead to model mismatch in complex uncertainties under actual production conditions, thus reducing the accuracy of park energy scheduling. If the model is based on describing the system's performance under uncertainty in specific scenarios, a massive number of scenarios are required to ensure representativeness, resulting in low computational efficiency and difficulty in meeting scheduling requirements in terms of solution time. This invention proposes a data-driven two-stage stochastic programming method to solve the park energy management problem under source-load uncertainty. First, the proposed method leverages the powerful fitting ability of neural networks to fit the first-stage energy scheduling and source-load uncertainty to the second-stage readjustment cost. Then, the trained neural network is transformed into a set of mixed-integer linear constraints through optimization techniques and equivalently embedded into the optimization problem. An optimization solver is then used to achieve accurate energy scheduling solutions. The two-stage scheduling model based on neural network fitting can leverage the advantages of data-driven approaches, considering the geographical characteristics of renewable energy and the characteristics of production energy consumption patterns, to achieve targeted and accurate modeling of the target park. Simultaneously, the neural network can compress scenarios, achieving a structured approach to the number of scenarios and the complexity of energy scheduling. Subsequently, the neural network embedding method based on optimization transformation, with the help of an optimization solver, can guarantee the absolute optimality of scheduling decisions.

[0058] 2. Existing technologies use general neural network structures to fit the two-stage rescheduling cost under different pre-scheduling decisions and random fluctuations in source load scenarios. This requires directly inputting a large number of scenarios and a large-scale neural network, leading to the introduction of too many 0-1 variables in the neural network embedding optimization problem, resulting in excessive computational overhead and failing to meet scheduling requirements. This invention introduces a two-stage network structure for scenario aggregation. First, the first neural network structure extracts and aggregates scene features. The second stage fits the aggregated features with the two-stage results under the pre-scheduling decision. In the neural network embedding optimization problem, since there are no scheduling decision variables in the first stage, the scene can be input into the first-stage network, and only the second-stage network is used for optimization embedding, greatly reducing the complexity of the optimization problem.

[0059] 3. This invention utilizes the concept of hierarchical neural networks to separate parameter inputs from scheduling decision inputs, thereby achieving effective control over the size of the embedded network. This invention effectively reduces the size of the neural network requiring embedding optimization, significantly improves the efficiency of solving energy scheduling problems, and enhances the performance and applicability of the proposed framework.

[0060] The present invention will now be described in detail with reference to preferred embodiments.

[0061] The integrated operation and management method for the park's electro-hydrogen coupling energy system proposed in this invention is as follows: Figure 1 As shown, it mainly includes four steps: ① Construction of an energy management model for the park based on two-stage stochastic programming; ② Fitting the two-stage rescheduling cost of an electric-hydrogen coupled energy system in a park based on neural networks; ③ Equivalently embed the trained neural network into the optimization problem; ④ Solve the optimization problem to obtain the electro-hydrogen coupling scheduling instructions.

[0062] The specific implementation steps of this invention are described below: Step 1: Construction of a Park Electricity-Hydrogen Coupling Energy Management Model Based on Two-Stage Stochastic Programming: This step constructs the basic decision-making framework of a two-stage stochastic programming model for the park's electro-hydrogen coupling system, clarifying the system's operational objectives, the objects of operational control, and operational constraints—that is, the optimization objective, decision variables, and constraints of the optimization problem. It mainly includes the following steps: S1 clarifies the composition and equipment parameters of the park's electro-hydrogen coupling system: In this embodiment, the park's electro-hydrogen coupling system is used as the scheduling object, such as... Figure 2 As shown, the system consists of an electrical subsystem and a hydrogen energy subsystem. In the electrical subsystem, non-gas turbines, the power grid, and renewable energy sources jointly serve as the power supply sources. The electrical loads include the park's electrical load and the hydrogen electrolysis production load. The subsystem is also equipped with energy storage. In the hydrogen energy subsystem, the hydrogen pipeline network and the park's on-site hydrogen electrolysis production serve as the hydrogen source. The hydrogen loads include the park's hydrogen consumption and hydrogen-to-electricity loads. During the implementation of this invention, it is necessary to first clarify the capacity, operating efficiency, and other parameters of each device in the above subsystems to provide a parameter basis for subsequent optimization.

[0063] S2 One-Stage Deterministic Model Construction: To construct a scheduling model for an electric-hydrogen coupled energy system in a park that considers source-load uncertainty, it is necessary to first build a deterministic one-stage electric-hydrogen coupled energy scheduling problem model. This provides a model foundation for subsequently constructing a two-stage stochastic programming scheduling model that considers source-load uncertainty.

[0064] S2.1 Determination of the objective function: In a deterministic scenario, the operating cost of an electro-hydrogen coupled energy system can be expressed as: (1) in, For unit electricity purchase cost, To purchase electricity Unit gas purchase cost This refers to the amount of gas purchased.

[0065] S2.2 Constraints on the Operation of Hydrogen Energy in Industrial Parks: 1) Construction of power subsystem operation constraints: The operation of the power subsystem is subject to the following physical constraints: (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) in, Constraint (2) is for power balance, where They are respectively The data includes non-gas-fired power generation, hydrogen-to-electricity generation, electricity purchased from the grid by the park, energy storage charging and discharging, renewable energy generation, park power load, and electricity consumption for hydrogen-to-electricity conversion.

[0066] Constraints (3)-(5) are power generation restrictions, where They represent gas units respectively. exist The lower limit of power generation, power generation capacity, and upper limit of power generation capacity at any given time. , , These represent the lower limit, upper limit, and lower limit of the power generation capacity of hydrogen-to-electricity conversion, respectively.

[0067] Constraint (6) represents the energy conversion efficiency of hydrogen to electricity. express The amount of hydrogen consumed in the instantaneous conversion of electricity. The efficiency coefficient for hydrogen-to-electricity conversion. for The amount of electricity generated by hydrogen conversion at any given moment.

[0068] Constraints (7)-(9) are energy storage constraints, where for The amount of energy stored in a time-sensitive energy storage system. and These are the charging and discharging efficiencies of the energy storage, respectively. Constraint (9) requires that the energy storage capacity at the end of the scheduling process be no less than that at the beginning of the scheduling process, in order to ensure the sustainable operation of the system.

[0069] Constraints (10)-(13) limit the maximum / minimum power of energy storage, electricity purchase, renewable energy generation and electrical load.

[0070] 2) Operational constraints of the hydrogen energy subsystem

[0071] The operation of the hydrogen energy subsystem is subject to the following physical constraints: (14) (15) (16) (17) (18) (19) (20) (twenty one) (twenty two) (twenty three) Condition (14) represents the hydrogen energy load balance, where They are respectively The data includes the amount of hydrogen purchased at any given time, the amount of hydrogen released from storage, the amount of hydrogen produced by electricity-to-hydrogen conversion, the amount of hydrogen consumed by hydrogen load, the amount of hydrogen consumed by hydrogen-to-electricity conversion, and the amount of hydrogen introduced into storage.

[0072] Condition (15) represents the energy conversion efficiency from electro-to-hydrogen, where For the amount of gas produced by electro-hydrogen conversion, The electro-hydrogen conversion efficiency coefficient. for The power of electro-hydrogen conversion at all times.

[0073] Condition (16) is The upper and lower limits of power.

[0074] Constraints (17)-(23) are gas storage restrictions. for Hydrogen storage capacity at all times They represent The efficiency coefficients of electro-hydrogen conversion and hydrogen-to-electricity conversion are given at any time. Constraint (18) indicates that the hydrogen produced by electro-hydrogen conversion will flow to the hydrogen storage station or be produced directly. Constraints (19)-(20) limit the upper and lower limits of the amount of hydrogen stored. Constraints (21)-(23) limit the upper and lower limits of the amount of gas produced by electro-hydrogen conversion and the gas flow destination of the two gas production destinations.

[0075] S3 constructs a two-stage energy dispatch model that takes into account the uncertainty of energy demand for production.

[0076] A deterministic model was established in stage S2. However, in the actual operation of the energy system, uncertainties exist on both the supply and demand sides. On the supply side, renewable energy sources such as photovoltaics and wind turbines are affected by weather conditions, making it difficult to accurately predict their output, thus introducing uncertainty. On the demand side, considering uncertainties such as (mainly load) orders, operating conditions, and human factors, actual energy demand and plans will fluctuate to some extent. These fluctuations in energy load uncertainty may lead to insufficient energy supply, thereby reducing the economic efficiency of the energy system operation. Therefore, it is necessary to construct a two-stage stochastic programming model that considers uncertainties on both the supply and demand sides. That is, in the first stage, in addition to solving the operating baseline for deterministic scheduling, it is also necessary to calculate the reserve capacity of each energy component. Thus, in the second stage, when uncertainties occur, corresponding adjustments to energy can be made to ensure real-time energy balance.

[0077] S3.1 Two-stage stochastic programming model

[0078] The two-stage stochastic programming model of the park's point-coupled system can be expressed as follows:

[0079] (twenty four) in, For the first stage of scheduling decision, , These are the uncertainties for the two-stage electrical load, hydrogen load, and renewable energy, respectively. (2)-(13) are the operating constraints of the power subsystem, and (14)-(23) are the operating constraints of the hydrogen energy subsystem.

[0080] Among them, the cost of spare capacity It consists of the following formula: (25) in, For non-gas unit standby capacity cost, Cost of hydrogen-to-electricity backup capacity, Cost of backup capacity for electro-hydrogen conversion, Cost of backup gas supply capacity, Cost of backup power capacity Cost of backup capacity for electrical energy storage Cost of backup capacity for hydrogen energy storage.

[0081] Non-gas unit standby capacity cost: (26) Hydrogen-to-electricity backup capacity cost: (27) Cost of spare capacity for electro-hydrogen conversion: (28) Gas source backup capacity cost: (29) Cost of backup power capacity: (30) Cost of backup capacity for electrical energy storage: (31) Hydrogen energy storage backup capacity cost (32) in, The unit cost of the spare capacity for each section, These are the unit prices for non-gas units, hydrogen-to-electricity units, electricity-to-hydrogen units, gas-source units, and positive and negative standby power capacity, respectively. These are the unit prices for positive and negative standby capacity of renewable energy storage / energy storage / discharge, respectively. These represent the unit price for positive and negative standby capacity of renewable energy hydrogen storage / discharge.

[0082] These refer to the positive and negative backup capacities of non-gas units, hydrogen-to-electricity units, electricity-to-hydrogen units, gas sources, and power sources. These represent the positive and negative backup capacities of renewable energy storage / discharge. These represent the positive and negative backup capacities for renewable energy hydrogen storage / discharge.

[0083] S3.2 Determination of the Objective Function of the Two-Stage Model (33) (34) (35) (36) , This represents the total cost of the second phase.

[0084] Equation (33) is the total cost = cost of repurchasing electricity + cost of repurchasing gas + cost of load reduction, where For the cost of repurchasing electricity, For the cost of repurchasing gas, To reduce costs by reducing workload. These are the unit prices for repurchased electricity and resold electricity, respectively. These are the unit prices for repurchased gas and resold gas, respectively. These represent the unit costs of electricity load shedding and hydrogen load shedding, respectively. Two-stage decision variables. These are respectively the amounts of electricity repurchase, electricity resale, gas repurchase, gas resale, electricity load reduction, and hydrogen load reduction.

[0085] S3.3 Operational Constraints of the Two-Phase Model

[0086] 1) Power System Model

[0087] In the second phase of the problem, the energy operation readjustment of the power system is affected by capacity constraints.

[0088] Specifically, it includes: Non-gas turbine standby capacity constraints: (37) (38) Reserve capacity constraints for hydrogen-fired power units: (39) (40) Electro-hydrogen conversion reserve capacity constraints: (41) (42) Energy storage backup capacity constraints: (43) (44) (45) (46) Repurchase and resale of electricity: (47) (48) Power load shedding constraints: (49) in,

[0089] These represent the reserve capacity for increases / decreases in non-gas turbine power, hydrogen turbine power, and electricity-to-hydrogen power, respectively. Second-stage decision variables.

[0090] These represent the following: electricity purchased / sold, increase / decrease in power of non-gas turbine units, increase / decrease in power of hydrogen turbine units, increase / decrease in electrical energy release, electricity load shedding, increase / decrease in electricity-to-hydrogen conversion power, increase / decrease in electrical energy storage, and change in electricity load.

[0091] Furthermore, considering the energy supply and demand balance of the system, the readjustment amount must satisfy the balance constraint: (50) Furthermore, for actual equipment such as gas turbine units, non-gas turbine units, and electricity-to-hydrogen conversion units, at any given moment, at most one of the positive and negative adjustments is non-zero, which can be represented as complementary relaxation constraints: (51) (52) (53) (54) (55) (56) Complementary relaxation constraints can be transformed into linear constraints using the Big M method.

[0092] 2) Hydrogen model

[0093] In the second phase of the problem, the energy operation readjustment of the power system is affected by capacity constraints.

[0094] Specifically, this includes: generator standby capacity constraints: (57) (58) (59) (60) Electro-hydrogen conversion reserve capacity constraints: (61) (62) (63) (64) Gas storage backup capacity constraints: (65) (66) (67) (68) Repurchase and resale of gas: (69) (70) Gas load shedding constraints: (71) Among them, the two-stage decision variables of the power subsystem These are the amounts of gas purchased / sold, the increase / decrease in electricity-to-hydrogen gas production, the increase / decrease in hydrogen storage gas release, and the amount of hydrogen load reduction. These represent the increase / decrease in gas consumption for hydrogen-fired power generation, the increase / decrease in gas charging for hydrogen-fired energy storage, and the change in hydrogen load.

[0095] Overall equilibrium constraints: (72) Similarly, the complementary relaxation constraints of the hydrogen system are as follows: (73) (74) (75) (76) (77) Similarly, it can be transformed into a linear constraint using the Big M method.

[0096] Thus, the complete form of the two-stage problem is as follows:

[0097] (78) Step 2: Fitting the two-stage rescheduling cost of the park's electro-hydrogen coupled energy system based on neural networks. Existing two-stage stochastic optimization methods have significant limitations in handling electric-hydrogen coupled energy scheduling. They either rely excessively on specific probability distribution models, leading to poor decision-making performance, or their expectation-based methods, based on scenario modeling, result in excessively high problem complexity, making them difficult to solve. To address these issues, this invention proposes an energy scheduling method for industrial parks based on constraint learning and two-stage stochastic programming. This method eliminates the dependence on probability distribution functions and avoids the model complexity problem in massive scenarios. The specific steps are as follows: S2.1 Network Structure like Figure 3 As shown, a diagram is presented from ( → The mapping. The neural network model is input with a one-stage decision. Given a finite set of any number of scenes from the entire scene set, the neural network outputs a two-stage expected objective function value. First, each scene in the scene set is independently input into the neural network. In the middle, the scene set Embedded into a hidden layer, and then... The embedded vectors are summed and averaged, then input into the neural network. The processing ultimately yields the final embedding vectors of the scene set. .

[0098] The advantage of this customized network structure lies in its high sensitivity to network size during subsequent neural network embedding processes, such as... Figure 3 The neural network structure shown can be achieved by simply using a fully connected neural network. Embedded, this network structure separates scene compression from scheduling decisions, which can effectively reduce the computational burden on the solver.

[0099] S2.2 Dataset Generation Method

[0100] Training supervised second-stage value approximation models for neural networks requires a diverse dataset of input-output pairs. To generate such datasets for given two-stage stochastic programming problems, we employ an iterative procedure. First, we generate a stochastic feasible first-stage decision. We sample a random cardinality from an uncertainty distribution. The scene set. Then, the complete form of the two-stage problem (78) given in this paper is substituted into the solver to solve it, thereby calculating The calculated values ​​are then used as the labels for the samples in this dataset. The choice should balance the following trade-offs: the time required to generate a sample of second-stage values ​​for a given first-stage solution, and the time required to estimate the expected second-stage values ​​of a set of first-stage decisions within a given time budget. Specifically, if If the value is larger, on average more time will be spent determining the expected value across a large number of scenarios; while if If the space is smaller, the decision space in the first stage will be explored more fully because the expected value estimate will be based on fewer scenarios.

[0101] S2.3 Network Training

[0102] Based on general neural network methods, the training set and test set are divided in a 4:1 ratio, using scene sets as the basis. For input, For output, The loss function is the mean squared error function: , During the training phase, and All parameters are updated. The trained network can achieve a good estimate of the two-stage objective function.

[0103] Step 3: Equivalent embedding optimization problem of the trained neural network

[0104] After training, the neural network can be equivalently transformed into mixed integer linear constraints and embedded into the optimization problem. The specific process is described in detail below.

[0105] The single-layer inference of the above neural network can be represented as: (79) in, , and These are the neural network's first... The neuron values, weight matrix, and bias vector of a layer. It is the activation function, which in this case is the ReLU function.

[0106] To facilitate the introduction of subsequent transformations, let: (80) yes The Row element.

[0107] Then, the ReLU function The following linear constraints can be encoded using the Big-M method: (81) (82) (83) (84) (85) From the above, the objective function of the two-stage stochastic programming can be fitted by mixed-integer linear programming, that is:

[0108] (86) When optimizing the solution of the problem, Through neural networks The inference operations are directly obtained, only by neural networks. The transformations (81)-(85) are embedded into the optimization problem.

[0109] Step 4: Solve the optimization problem to obtain the electro-hydrogen coupling scheduling instructions.

[0110] After the model is built, we first obtain A typical sample is obtained, and then the above neural network inference is used to obtain... Then The embedded model can be solved using commercial or open-source optimization solvers to obtain scheduling instructions for the park's electric and hydrogen energy equipment, which can then be sent to the energy equipment controller to achieve energy management.

[0111] Compared with existing technologies, this invention can improve the quality of operation and scheduling decisions for park-based electric-hydrogen coupled energy systems under source-load uncertainty interference, thereby improving operational safety and economy. Specifically, the scheme adopted in this invention breaks through the limitations of traditional stochastic programming methods, refines the characterization of energy pre-scheduling decisions and energy rescheduling costs under different source-load fluctuations, and thus obtains reliable and high-performance scheduling and operation management decisions for park-based electric-hydrogen coupled energy systems.

[0112] In terms of application implementation, the technical solution of this invention can achieve low-cost, large-scale deployment and migration. Benefiting from the advantages of the neural network embedding framework, the neural network model is continuously updated and integrated into the optimization problem and operation management problem in real time. This means that as the operational data of the park's electric-hydrogen coupling energy system becomes increasingly abundant, the system model can be adaptively updated, thereby continuously improving the decision-making performance of the park's electric-hydrogen coupling system operation management.

[0113] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A smart energy dispatching method for electricity-hydrogen coupling in industrial parks, accommodating source-load uncertainty, characterized in that: The method includes the following steps: S101: Construct a park-based electric-hydrogen coupled energy management model based on two-stage stochastic programming. The model is set as a two-stage neural network structure of scenario aggregation, including a first-stage electric-hydrogen coupled energy scheduling problem model and a second-stage energy scheduling model that takes into account the uncertainty of production energy demand. S103: Fitting the second-stage rescheduling cost of the park's electric-hydrogen coupled energy system based on a neural network model; S105: Equivalently embed the trained neural network model into the optimization problem; S107: Solve the optimization problem using an optimization solver, obtain the electric-hydrogen coupling scheduling instructions, and schedule the park's electric-hydrogen coupling energy system according to the scheduling instructions.

2. The method as described in claim 1, characterized in that, Step S101 includes the following sub-steps: S1011: Determine the composition and equipment parameters of the park's electro-hydrogen coupling system to provide a parameter basis for subsequent optimization problems; S1012: Construct a deterministic first-stage energy scheduling model for the electro-hydrogen coupling problem, providing a model foundation for constructing the second-stage energy scheduling model; S1013: Construct the second-stage energy dispatch model for the aforementioned uncertainty, and determine the objective function and constraints of the second-stage model.

3. The method as described in claim 2, characterized in that, The park's electro-hydrogen coupling system includes an electrical subsystem and a hydrogen energy subsystem, wherein... In the power subsystem, non-gas turbine units, the power grid, and renewable energy sources work together as the power supply source. The power load includes the park's power load and the electrolysis hydrogen production load. The power subsystem is equipped with energy storage. In the hydrogen energy subsystem, the hydrogen source is jointly provided by the hydrogen pipeline network and the electrolytic hydrogen production in the park. The hydrogen load includes the hydrogen load in the park and the hydrogen-to-electricity load. The equipment parameters of the park's electro-hydrogen coupling system include the capacity parameters and operating efficiency parameters of the equipment in the power subsystem and the hydrogen energy subsystem.

4. The method as described in claim 3, characterized in that, Step S1012 includes the following sub-steps: S10121: Construct the objective function for the first-stage electro-hydrogen coupled energy dispatch problem model, where the objective function is the operating cost of the electro-hydrogen coupled energy system. S10122: Construct the operating constraints of the power subsystem in the industrial park, including power balance constraints, generation limitation constraints, energy storage constraints, and power constraints; The power balance constraint is: S10123: Construct the operational constraints of the hydrogen energy subsystem in the industrial park, including hydrogen energy load balance constraints, electricity-to-hydrogen energy conversion efficiency constraints, gas storage limitation constraints, and power constraints. The hydrogen energy load balance constraint is: in, For the first phase of scheduling decisions, For unit electricity purchase cost, To purchase electricity, Unit gas purchase cost For the amount of gas purchased; for Non-gas power generation at any time for Hydrogen-to-electricity generation at any time for The park purchases electricity from the power grid at any time. for Real-time energy storage charging capacity, for Energy storage discharge rate at any time for Renewable energy generation at all times for Real-time park power load, for The amount of electricity used for the constant conversion of electricity to hydrogen; for Purchase hydrogen at any time for Hydrogen storage gas release rate at all times for Hydrogen production rate at any time during electro-hydrogen conversion for Hydrogen load consumption at any time for Hydrogen consumption during hydrogen-to-electricity conversion at any time for Constant hydrogen intake volume.

5. The method as described in claim 4, characterized in that, Under the operating constraints of the power subsystem and the hydrogen energy subsystem, the second-stage energy dispatch model is as follows: The objective function of the second-stage energy dispatch model is: in, For two-stage decision variables based on one-stage decision-making, This is a factor affecting the second-stage uncertainty. For standby capacity costs, This represents the total cost of the second phase. To purchase electricity again cost, For the cost of repurchasing gas, To reduce costs by reducing workload.

6. The method as described in claim 5, characterized in that, In step S103, the neural network model includes a multi-layer neural network, with the last layer being a fully connected neural network. The neural network model maps the input first-stage decision and the scene set, and outputs the expected objective function value for the second stage. The specific mapping method is as follows: The loss function of the neural network model is the mean squared error function: in, For scene collection, For the number of embedding vectors, For embedding vector identifiers, Predict the probability of occurrence for the k-th scenario. For neural networks, For the scene.

7. The method as described in claim 6, characterized in that, The neural network model completes the mapping operation using the following steps: S1031: Input each scene in the scene set independently into the neural network. In this process, the scene set is embedded into a hidden layer; S1032: Sum and average the embedding vectors, and process the result before inputting it into the neural network. ; S1033: The neural network The final embedding vector of the scene set is obtained after processing.

8. The method as described in claim 7, characterized in that, The dataset was generated using the following method to train the neural network model: S1034: Sample a set of scenarios with a random cardinality from an uncertain distribution to generate a random feasible first-stage decision; S1035: Solve the objective function of the second-stage energy dispatch model using a solver; S1036: Calculation The calculation results are then used as labels for the samples in the dataset.

9. The method as described in claim 8, characterized in that, In step S105, after training, the neural network model is equivalently transformed into mixed-integer linear constraints and embedded into the optimization problem. The objective function of the second-stage energy scheduling model is fitted by mixed-integer linear programming, i.e.: in, Neural networks with equivalent embeddings.

10. The method as described in claim 9, characterized in that, Step S107 includes the following sub-steps: S1071: Obtain K typical samples and infer the embedding vectors using the trained neural network model. ; S1072: Insert the embedding vector Substitute the solution into the embedded neural network model and solve it using an optimized solver. S1073: The optimization solver outputs a scheduling command for the park's hydrogen energy equipment; S1074: Send dispatch instructions to the energy equipment controller to achieve energy management.

Images