Electro-hydrogen energy scheduling strategy solving method and device and storage medium

Abstract

The invention discloses an electric hydrogen energy scheduling strategy solving method and device and a storage medium, and belongs to the technical field of energy scheduling, and the method comprises the steps: building an RIES day-ahead global scheduling model and an sIES intra-day local correction model according to the architecture that a RIES in an electric hydrogen energy network is divided into a plurality of sIES; solving the RIES day-ahead global scheduling model through an improved primal-dual gradient method according to the historical data of the electric hydrogen energy network, and obtaining a global scheduling strategy comprising an energy circulation coefficient and an energy interaction amount; the energy circulation coefficient is used for adjusting the interaction priority among the sIES; and the global scheduling strategy is used for real-time scheduling and operation data acquisition, rolling optimization is performed on the sIES intraday local correction model through multi-agent reinforcement learning, the energy circulation coefficient and the energy interaction amount are periodically and dynamically corrected, and a real-time scheduling strategy is generated. According to the method, the global optimal scheduling strategy is adjusted in real time, so that the problem that the existing scheduling strategy is easy to cause resource waste is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy scheduling, and specifically relates to a method, device and storage medium for solving an electric-hydrogen energy scheduling strategy. Background Art

[0002] With the rapid development of high-proportion renewable energy integration and electricity-hydrogen coupling technology, integrated energy systems are accelerating their transformation toward multi-energy complementarity, low carbon, and high efficiency. Electricity-hydrogen coupling systems, which achieve bidirectional conversion between electricity and hydrogen through core equipment such as hydrogen electrolysis and fuel cells, have become a key path to improving renewable energy absorption capacity and addressing the challenges of wind and solar power curtailment.

[0003] Extensive research has been conducted on the optimal scheduling of electric-hydrogen systems, primarily focusing on traditional methods such as two-stage robust optimization and model predictive control. However, when addressing complex scenarios such as the high volatility of wind and solar power output and the coupled operating schedules of multiple devices, bottlenecks such as insufficient model accuracy and delayed real-time decision-making persist. In particular, existing research on multi-timescale collaborative optimization often employs a single-scale optimization framework with a fixed time resolution, making it difficult to balance the economic efficiency of day-ahead planning with the flexibility of real-time adjustments, resulting in high system operating costs.

[0004] Among existing technologies for optimizing electric-hydrogen energy systems, scheduling methods based on deep reinforcement learning face significant bottlenecks in dynamic environmental perception and decision-making mechanisms. Traditional deep reinforcement learning frameworks lack detailed modeling of physical constraints in electric-hydrogen coupled systems, such as electrolyzer start-up and shutdown hysteresis, hydrogen tank pressure dynamic response, and fuel cell ramp rate. Furthermore, their single-layer state space design struggles to effectively characterize the multi-timescale correlations between hydrogen storage's intertemporal inertia and electrical energy fluctuations.

[0005] What is more serious is that the existing methods ignore the temporal coupling relationship between the electricity spot market and the hydrogen energy futures market. They neither consider the traction effect of the two-stage day-ahead and real-time transactions on the energy storage scheduling strategy, nor establish a two-way value mapping mechanism between hydrogen energy backup capacity and grid frequency regulation requirements, which will lead to multiple wastes of energy resources.

[0006] Specifically, in the day-ahead planning stage, conservative energy storage scheduling strategies lead to insufficient renewable energy absorption, such as insufficient reserved production capacity for electrolyzers, resulting in excessively high wind / solar power curtailment rates. Secondly, in the real-time operation stage, the hydrogen energy storage system lacks a value linkage mechanism with the grid frequency regulation, and cannot respond in a timely manner even if the grid experiences frequency regulation needs due to frequency fluctuations. At the same time, due to the imbalance of energy transfer across time periods, the utilization rate of wind power hydrogen production capacity is insufficient during the night when electricity consumption is low, resulting in a shortage of hydrogen energy supply during the daytime peak, and the need to start gas-fired units to make up for it, resulting in additional gas resource consumption and additional peak-shaving costs. Even more seriously, the hydrogen storage system is in a sub-full load state for a long time due to its failure to participate in frequency regulation services. When the tank pressure reaches the safety threshold, it is forced to release hydrogen, resulting in direct energy loss, and ultimately forming a chain of resource waste: "day-ahead forecast deviation - real-time response lag - loss of frequency regulation benefits - low energy storage utilization - forced energy discharge." Summary of the Invention

[0007] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, device and storage medium for solving the electric hydrogen energy scheduling strategy, so as to solve the problem that the current electric hydrogen energy scheduling strategy easily causes waste of resources by adjusting the global optimal scheduling strategy in real time.

[0008] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions:

[0009] In a first aspect, the present invention provides a method for solving an electric hydrogen energy scheduling strategy, comprising:

[0010] S1: Based on the architecture of RIES in the electric-hydrogen energy network, which is responsible for multiple SIES, a RIES day-ahead global dispatch model and a SIES intraday local correction model are established;

[0011] S2: Based on the historical data of the electric hydrogen energy network, the RIES day-ahead global scheduling model of step S1 is solved by the improved primal-dual gradient method to obtain a global scheduling strategy including the energy flow coefficient and energy interaction amount; the energy flow coefficient adjusts the interaction priority between each SIES;

[0012] S3: Use the global scheduling strategy of step S2 for real-time scheduling and collect operating data. Through multi-agent reinforcement learning, perform rolling optimization on the sIES intraday local correction model of step S1, periodically and dynamically correct the energy circulation coefficient and energy interaction amount, and generate a real-time scheduling strategy.

[0013] In the aforementioned method for solving the electric hydrogen energy scheduling strategy, in step S1, the RIES day-ahead global scheduling model includes: an objective function and constraints;

[0014] The goal of the RIES day-ahead global scheduling model is to maximize the RIES operational benefits. The objective function The calculation formula is:

[0015] ,

[0016] Where, represents the income of the kth IERO; represents the cost of the k-th IERO; represents the benefit of the k-th IERO delivering electricity to the outside; represents the revenue of the k-th IERO from delivering hydrogen to the outside; represents the external grid electricity purchase cost of the k-th IERO; represents the external hydrogen source cost of the kth IERO; represents the internal operation and maintenance cost of the k-th IERO;

[0017] The constraints of the RIES day-ahead global scheduling model include energy balance constraints;

[0018] The energy balance constraints include electric energy load balance constraints, thermal energy load balance constraints and hydrogen energy load balance constraints;

[0019] The power load balance constraint formula is:

[0020] ,

[0021] Where, represents the amount of electricity purchased from the external power grid by the kth IERO during period t; represents the amount of electricity sold by the k-th IERO to the external grid during period t; represents the wind power output of the kth IERO in period t; represents the power output of the CHP unit of the k-th IERO in period t; represents the energy consumption of the k-th IERO during period t of the electric-hydrogen hybrid energy storage; Represents the set of IERO; represents the amount of electric energy interaction transmitted from the kth IERO to the lth IERO in period t; represents the electrical load of the kth IERO in period t;

[0022] The heat load balance constraint formula is:

[0023] ,

[0024] Where, represents the thermal energy output of the k-th IERO during period t of electric-hydrogen hybrid energy storage; represents the heat output of the CHP unit of the k-th IERO in period t; It represents the amount of heat energy interaction transferred from the kth IERO to the lth IERO in time period t; represents the heat energy load of the kth IERO in period t;

[0025] The hydrogen energy load balance constraint formula is:

[0026] ,

[0027] Where, represents the external hydrogen source output of the kth IERO in period t, represents the hydrogen energy output of the k-th IERO during period t; represents the hydrogen energy output of the k-th IERO during period t; represents the hydrogen energy delivered from the kth IERO to the lth IERO during period t; represents the hydrogen energy load of the kth IERO in period t.

[0028] In the aforementioned method for solving the electric hydrogen energy scheduling strategy, in step S1, the sIES intraday local correction model includes: an objective function and constraints;

[0029] The goal of the sIES intraday local correction model is to maximize the energy interaction benefits of each sIES under non-cooperative game. The calculation formula of the objective function is:

[0030]

[0031] Where, represents the benefit obtained by the mth sIES through energy interaction under the kth IERO; Intraday phase defined for the day-ahead phase The flow coefficient of energy i from the m-th sIES under the k-th IERO to the n-th sIES under the l-th IERO in the time period; is the time period of the sIES intraday local correction model, is the period of the sIES intraday local correction model; Represents the set of sIES under a single IERO; Indicates the intraday phase defined in the day-ahead phase The flow coefficient of energy i from the nth sIES under the lth IERO to the mth sIES under the kth IERO in the time period; Intraday stage The interaction amount of energy i delivered from the m-th sIES under the k-th IERO to the n-th sIES under the l-th IERO in the time period; Intraday stage The interaction amount of energy i transmitted from the nth sIES under the lth IERO to the mth sIES under the kth IERO in the time period; Represents the set of sIES under a single IERO;

[0032] The constraints of the sIES intraday local correction model include: benefit fairness constraint, energy circulation coefficient balance constraint and energy interaction balance constraint;

[0033] The benefit fairness constraint formula is:

[0034] ,

[0035] Where, represents the operating benefit of the kth IERO considering energy interaction; Indicates that the day-ahead phase is formulated in the intraday phase The flow coefficient of energy i delivered by the kth IERO in the time period; represents the benefit obtained by the k-th IERO through energy interaction; It represents the operating benefit of the kth IERO without considering energy interaction;

[0036] The energy flow coefficient balance constraint formula is:

[0037] ,

[0038] Where, Indicates the intraday period set in the day-ahead period The kth IERO in the period transmits the circulation coefficient of energy i to the lth agent; Indicates the intraday period set in the day-ahead period The flow coefficient of energy i delivered by the l-th IERO to the k-th agent in the period; represents the set of IERO, Represents the energy type set, i takes e / H / c;

[0039] The energy flow coefficient acts as a weight factor and directly affects the calculation of energy interaction. When the flow coefficient of a sIES is higher than that of other nodes, its energy interaction ratio in energy interaction increases, thus obtaining a higher scheduling priority.

[0040] The energy interaction balance constraint formula is:

[0041] ,

[0042] Where, In the intraday stage The amount of interaction between the kth IERO and the lth IERO in a period of time, In the intraday stage The amount of interaction in which the l-th IERO transmits i energy to the k-th IERO in the period.

[0043] In the aforementioned method for solving the electric-hydrogen energy scheduling strategy, step S2 includes:

[0044] S21: The energy interaction quantity is used as the primal variable of the primal-dual gradient method, and the energy circulation coefficient is used as the dual variable of the primal-dual gradient method;

[0045] Auxiliary variables are introduced to decouple the global energy interaction constraints and reconstructed into distributed solutions for each IERO;

[0046] S22: Constructing the augmented Lagrangian function with penalty term;

[0047] S23: Set the initial values ​​of energy interaction, auxiliary variables and energy circulation coefficients among each IERO;

[0048] S24: Each IERO updates its energy interaction according to the constructed augmented Lagrangian function;

[0049] S25: Coordinate and update auxiliary variables according to the updated energy interaction amount;

[0050] S26: updating the energy circulation coefficient according to the updated energy interaction amount and auxiliary variables;

[0051] S27: Calculate the original residual and dual residual based on the updated energy interaction, auxiliary variables and energy circulation coefficient;

[0052] S28: Determine the convergence status based on the calculated primal residual and dual residual;

[0053] S29: Loop through steps S24 to S28, and output the energy interaction amount and energy flow coefficient of each IERO at convergence as the global scheduling strategy.

[0054] In the aforementioned method for solving the electric-hydrogen energy scheduling strategy, in step S21, the introduction of auxiliary variables to decouple the global energy interaction constraint includes:

[0055] Introducing auxiliary variables and , constrain the original energy interaction of each IERO:

[0056] ,

[0057] Translates to:

[0058] , to achieve distributed solution.

[0059] The aforementioned solution method for the electric-hydrogen energy scheduling strategy, after introducing auxiliary variables to decouple the global energy interaction constraints, includes:

[0060] For electricity, heat and hydrogen energy, energy interaction benefit optimization models with auxiliary variables are constructed respectively, and augmented Lagrangian functions with penalty terms are constructed;

[0061] Use KKT optimality conditions to derive auxiliary variables and set consistency constraints , solve for the auxiliary variable and The analytical decoupling expression of is used to transform the global coupling problem into a local optimization problem in which each IERO is calculated independently.

[0062] The aforementioned method for solving the electric-hydrogen energy scheduling strategy, step S3 includes: constructing a multi-agent system;

[0063] The agent design includes: each sIES acts as an independent agent, responsible for its own interactive decision-making of various types of energy;

[0064] The state space includes: energy flow coefficient, energy interaction, sIES load, and the output or energy consumption of wind power, combined heat and power (CHP), electric-hydrogen hybrid energy storage, and external hydrogen sources;

[0065] The action space includes: discretizing and correcting the energy flow coefficient and energy interaction quantity output by the RIES day-ahead global scheduling model;

[0066] The reward function is calculated as follows:

[0067] ,

[0068] Where, satisfying the constraints refers to meeting the constraints of the sIES intraday local correction model.

[0069] In the aforementioned method for solving the electric-hydrogen energy scheduling strategy, step S3 further includes using the constructed multi-agent system rolling optimization:

[0070] D1: Using the energy flow coefficient and energy interaction output by the RIES day-ahead global scheduling model as the benchmark scheduling strategy, initialize the Q function of each agent;

[0071] D2: Collect operating data, including actual wind power output, photovoltaic output and load measurement values;

[0072] D3: Adoption -greedy strategy for action selection;

[0073] D4: Output the corrected energy flow coefficient and energy interaction amount based on the action selection;

[0074] D5: Each agent exchanges state and action information through communication, and updates the Q value until the Q value change of all agents is less than the preset Q value change threshold;

[0075] D6: Loop through steps D2 to D5 according to the preset cycle, continuously tracking wind and solar power output fluctuations and load changes, and generating a real-time scheduling strategy.

[0076] In a second aspect, the present invention provides an electric hydrogen energy scheduling strategy solving device, comprising a model building module, a global solving module and a real-time correction module;

[0077] The model building module is used to establish a RIES day-ahead global scheduling model and a SIES intraday local correction model based on the architecture of RIES managing multiple SIES in the electric hydrogen energy network;

[0078] The global solution module is used to solve the RIES day-ahead global scheduling model of the model construction module using an improved primal-dual gradient method based on historical data of the electric hydrogen energy network to obtain a global scheduling strategy including energy circulation coefficients and energy interaction quantities; the energy circulation coefficients adjust the interaction priorities between each SIES;

[0079] The real-time correction module is used to apply the global scheduling strategy of the global scheduling strategy solving module to real-time scheduling and collect operating data, perform rolling optimization on the SIES intraday local correction model of the model construction module through multi-agent reinforcement learning, periodically and dynamically correct the energy circulation coefficient and energy interaction amount, and generate a real-time scheduling strategy.

[0080] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect.

[0081] Compared with the prior art, the present invention has the following beneficial effects:

[0082] The present invention solves the problem that the current electric hydrogen energy scheduling strategy easily causes waste of resources by adjusting the global optimal scheduling strategy in real time.

[0083] The present invention adopts a two-stage hybrid optimization method:

[0084] In the recent stage: the primal-dual gradient method solves global coupling constraints, avoiding the privacy and computational bottlenecks of centralized optimization; the primal-dual method is improved to decouple interaction constraints, achieve distributed solution, and reduce computational complexity;

[0085] Intraday stage: MARL handles high-dimensional uncertainty and enables distributed real-time decision making; -greedy balances exploration and exploitation to avoid local optimality; multi-agent parallelism improves response speed.

[0086] The present invention applies game theory: sIES competes for energy interaction benefits through non-cooperative games within a day, and RIES ensures global convergence through penalty factors.

[0087] The present invention achieves maximum efficiency in energy resource utilization through electricity-hydrogen synergy: hydrogen energy is used as a cross-time energy storage medium to balance the intermittency of renewable energy, global optimization is performed on the day before to ensure economy, and MARL correction is performed within the day to deal with uncertainty.

[0088] Through two-stage collaboration, the electric-hydrogen energy scheduling strategy solution method of the present invention improves the renewable energy absorption rate and supply-demand matching elasticity, improves economic efficiency and saves costs, and solves the problem that the current electric-hydrogen energy scheduling strategy is prone to waste of resources.

[0089] The present invention constructs a two-stage collaborative optimization architecture of day-ahead and intra-day. By decoupling day-ahead capacity pre-planning and real-time power adjustment decisions, it integrates the cross-period value transfer model of hydrogen energy storage and the deep strategy migration mechanism to break through the adaptability barriers of traditional deep reinforcement learning in the electric-hydrogen coupling system. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a framework diagram of the method for solving the electric hydrogen energy scheduling strategy of Example 1 of the present invention. DETAILED DESCRIPTION

[0091] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. Unless there is a conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0092] The term "and / or" in this document simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. Additionally, the character " / " in this document generally indicates an "or" relationship between the related objects.

[0093] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0094] Example 1:

[0095] Figure 1This is a framework diagram of a method for solving an electric hydrogen energy scheduling strategy in Example 1 of the present invention. This embodiment introduces a method for solving an electric hydrogen energy scheduling strategy, including:

[0096] S1: Based on the architecture of RIES in the electric-hydrogen energy network, which is responsible for multiple SIES, a RIES day-ahead global dispatch model and a SIES intraday local correction model are established;

[0097] S2: Based on the historical data of the electric hydrogen energy network, the RIES day-ahead global scheduling model of step S1 is solved by the improved primal-dual gradient method to obtain a global scheduling strategy including the energy flow coefficient and energy interaction amount; the energy flow coefficient adjusts the interaction priority between each SIES;

[0098] S3: Apply the global scheduling strategy from step S2 to real-time scheduling and collect operational data. Using multi-agent reinforcement learning, perform rolling optimization on the SIES intraday local correction model from step S1. Dynamically modify the energy flow coefficient and energy interaction periodically to generate a real-time scheduling strategy. In this example, the periodicity is 15 minutes, and the energy flow coefficient and energy interaction are dynamically modified to generate a real-time scheduling strategy.

[0099] The specific implementation of the present invention is described below:

[0100] In step S1, RIES day-ahead global scheduling model

[0101] When multiple secondary integrated energy systems (SIESs) accept unified dispatch from a regional integrated energy system (RIES), achieving interconnected and collaborative energy management, energy interactions occur between the SIESs. In this scenario, each integrated energy agency (IERO) should consider the operational benefits of energy interactions when making dispatch decisions. For RIES managers and dispatching organizations, the goal of implementing multi-SIES collaborative energy management should be to maximize the overall operational benefits of the RIESs.

[0102] The day-ahead global dispatch decision for RIES aims to maximize operational efficiency. Operating costs include the operation and maintenance expenses of the units in each sIES, while operating revenue comes from the energy supplied by each IERO to the sIES it manages. For any sIES, the energy types include electricity e, thermal energy c, and hydrogen energy H.

[0103] The constraints of the RIES day-ahead global scheduling model include energy balance constraints;

[0104] The energy balance constraints include electric energy load balance constraints, thermal energy load balance constraints, and hydrogen energy load balance constraints. A period T of the RIES day-ahead global scheduling model is 24 hours, including 24 time points. At any given time point t, where t ranges from 1 to 24, the following energy balance constraints must be followed during the operation and management of each IERO:

[0105] Electric energy load balancing constraints:

[0106] ,

[0107] Where, represents the amount of electricity purchased from the external power grid by the kth IERO during period t; represents the amount of electricity sold by the k-th IERO to the external grid during period t; represents the wind power output of the kth IERO in period t; represents the power output of the CHP unit of the k-th IERO in period t; represents the energy consumption of the k-th IERO during period t of the electric-hydrogen hybrid energy storage; Represents the set of IERO; represents the amount of electric energy interaction transmitted from the kth IERO to the lth IERO in period t; represents the electrical load of the kth IERO in period t;

[0108] Thermal load balance constraints:

[0109] ,

[0110] Where, represents the thermal energy output of the k-th IERO during period t of electric-hydrogen hybrid energy storage; represents the heat output of the CHP unit of the k-th IERO in period t; It represents the amount of heat energy interaction transferred from the kth IERO to the lth IERO in time period t; represents the heat energy load of the kth IERO in period t;

[0111] Hydrogen energy load balance constraints:

[0112] ,

[0113] Where, represents the external hydrogen source output of the kth IERO in period t, represents the hydrogen energy output of the k-th IERO during period t; represents the hydrogen energy output of the k-th IERO during period t; represents the amount of hydrogen energy interaction delivered from the kth IERO to the lth IERO; represents the hydrogen energy load of the kth IERO in period t.

[0114] In the case of energy interaction, the goal of the RIES day-ahead global scheduling model is to maximize the RIES operational benefits, which is achieved by maximizing the operational benefits of energy management decisions by maximizing each IERO. The objective function The formula is as follows:

[0115] ,

[0116] Where, represents the income of the kth IERO; represents the cost of the k-th IERO; represents the benefit of the k-th IERO delivering electricity to the outside; represents the revenue of the k-th IERO from delivering hydrogen to the outside; represents the external grid electricity purchase cost of the k-th IERO; represents the external hydrogen source cost of the kth IERO; represents the internal cogeneration and energy storage unit operation and maintenance cost of the kth IERO;

[0117] The external grid purchase cost of the kth IERO in a period T (24 hours) of the RIES day-ahead global dispatch model is The calculation formula is:

[0118] ,

[0119] Where, The coefficient of purchasing electricity from the external power grid; represents the electric energy purchased by the k-th IERO from the external grid during period t; represents the time step of the RIES day-ahead global scheduling model (the duration of a period). In this embodiment, The value is 1h;

[0120] The external hydrogen source cost of the kth IERO in a period T (24 hours) of the RIES day-ahead global scheduling model The calculation formula is:

[0121] ,

[0122] Where, The coefficient of purchasing hydrogen from outside; represents the hydrogen energy purchased from the outside by the k-th IERO in period t;

[0123] The internal operation and maintenance cost of the kth IERO in a period T (24 hours) of the RIES day-ahead global scheduling model The calculation formula is:

[0124] ,

[0125] Where, represents the cost coefficient of electric-hydrogen hybrid energy storage; represents the energy consumption of the k-th IERO during period t of the electric-hydrogen hybrid energy storage; The coefficient of purchasing hydrogen from outside; represents the hydrogen consumption of the k-th IERO in the electric-hydrogen hybrid energy storage at time t; represents the hydrogen consumption of the k-th IERO cogeneration unit in period t; represents the sinking coefficient caused by wind curtailment; represents the wind power forecast of the k-th IERO in period t. In this embodiment, The data comes from wind power generation in a typical area; represents the wind power output of the kth IERO in period t;

[0126] The revenue of the k-th IERO from transmitting electricity to the outside in a period T (24 hours) of the RIES day-ahead global dispatch model is The calculation formula is:

[0127] ,

[0128] Where, The coefficient representing the electricity sold to the external grid; represents the electric energy delivered by the k-th IERO to the outside during period t (including the electric energy delivered by the k-th IERO to the external energy distribution network outside the RIES, and the electric energy delivered by the k-th IERO within the RIES to other IEROs);

[0129] The revenue of the k-th IERO from delivering hydrogen to the outside in a period T (24 hours) of the RIES day-ahead global scheduling model is The calculation formula is:

[0130] ,

[0131] Where, The coefficient of selling hydrogen to the outside world; represents the hydrogen energy delivered to the outside by the k-th IERO during period t (including the hydrogen energy delivered by the k-th IERO to the external energy distribution network outside RIES, and the hydrogen energy delivered by the k-th IERO within RIES to other IEROs);

[0132] In step S1, the sIES intraday local correction model:

[0133] During daily operation, due to fluctuations in renewable energy or changes in load, the actual operation may deviate from the day-ahead plan. Based on non-cooperative game theory, each SIES dynamically adjusts its energy interaction scheduling strategy with the goal of maximizing its own energy interaction benefits.

[0134] During the overall RIES scheduling process, the interactive energy flows between SIESs will offset each other. The SIES intraday local correction model has a cycle T' of 8 hours, including 32 time periods. In any given time period t', t' ranges from 1 to 32. In the SIES system, it is necessary to establish an optimization objective function for the energy interaction benefits of each SIES, as shown in the following formula:

[0135] (10),

[0136] Where, represents the benefit obtained by the mth sIES through energy interaction under the kth IERO; Intraday phase defined for the day-ahead phase The flow coefficient of energy i from the m-th sIES under the k-th IERO to the n-th sIES under the l-th IERO in the time period; is the time period of the sIES intraday local correction model, is the period of the sIES intraday local correction model; Represents the set of sIES under a single IERO; Indicates the intraday phase defined in the day-ahead phase The flow coefficient of energy i from the nth sIES under the lth IERO to the mth sIES under the kth IERO in the time period; Intraday stage The interaction amount of energy i delivered from the m-th sIES under the k-th IERO to the n-th sIES under the l-th IERO in the time period; Intraday stage The interaction amount of energy i transmitted from the nth sIES under the lth IERO to the mth sIES under the kth IERO in the time period; Represents the set of sIES under a single IERO; Represents a sum operation.

[0137] In the non-cooperative game process, for each sIES, the following operational constraints must be met:

[0138] The benefit fairness constraint formula is:

[0139] ,

[0140] Where, represents the operating benefit of the kth IERO considering energy interaction; Indicates that the day-ahead phase is formulated in the intraday phase The flow coefficient of energy i delivered by the kth IERO in the time period; represents the benefit obtained by the k-th IERO through energy interaction; It represents the operating benefit of the kth IERO without considering energy interaction;

[0141] The energy flow coefficient balance constraint formula is:

[0142] (12),

[0143] Where, Indicates the intraday period set in the day-ahead period The kth IERO in the period transmits the circulation coefficient of energy i to the lth agent; Indicates the intraday period set in the day-ahead period The flow coefficient of energy i delivered by the l-th IERO to the k-th agent in the period; represents the set of IERO, Represents a set of energy types, i takes the value of e / H / c; the energy flow coefficient adjusts the interaction priority between each sIES;

[0144] The energy flow coefficient acts as a weight factor and directly affects the calculation of energy interaction. When the energy flow coefficient of a sIES is higher than that of other nodes, its energy interaction ratio in energy interaction increases, thus obtaining a higher scheduling priority.

[0145] Energy interaction balance constraints:

[0146] ,

[0147] Where, In the intraday stage The amount of interaction between the kth IERO and the lth IERO in a period of time, In the intraday stage The amount of interaction in which the l-th IERO transmits i energy to the k-th IERO in the period.

[0148] In step S2, the improved primal-dual gradient method scheduling solution includes:

[0149] Y1 Basic Principles of Primal-Dual Gradient Method

[0150] The principle of the primal-dual gradient method is explained as follows. Consider the following equality constrained optimization problem:

[0151]

[0152] Where, is a convex function, .

[0153]

[0154] Where, is the Lagrange multiplier (also called dual variable). Assuming that there is an optimal solution, it is easy to see that strong duality holds, then There must be a saddle point. yes A saddle point of and They must be the optimal solutions of Equation 13 and its dual problem respectively. Therefore, the saddle point problem corresponding to Equation (13) can be solved by , we get the optimal solution of formula (13), and the primal-dual gradient method is an efficient method for solving saddle point problems. Its core idea is to perform gradient descent and gradient ascent on the primal variable and the dual variable respectively. Lagrangian function The gradient of is:

[0155] ,

[0156] Then the discrete-time form of the primal-dual gradient method, the primal-dual gradient algorithm, is:

[0157] ,

[0158] ,

[0159] Its continuous-time form, the primal-dual gradient dynamics, is:

[0160] ,

[0161] ,

[0162] Y2 Distributed scheduling model solution based on primal-dual gradient method

[0163] First, it is important to recognize that since the interaction between agents is a global coupling variable, all constraints, except for the energy interaction constraint, are only related to the energy interaction of each sIES itself. To facilitate the processing of energy interactions between IERO agents, auxiliary variables are introduced to reconstruct the interaction constraint model.

[0164] Step S2 includes:

[0165] S21: The energy interaction quantity is used as the primal variable of the primal-dual gradient method, and the energy circulation coefficient is used as the dual variable of the primal-dual gradient method;

[0166] Auxiliary variables are introduced to decouple the global energy interaction constraints and reconstructed into distributed solutions for each IERO;

[0167] S22: Constructing the augmented Lagrangian function with penalty term;

[0168] S23: Set the initial values ​​of energy interaction, auxiliary variables and energy circulation coefficients among each IERO;

[0169] S24: Each IERO updates its energy interaction according to the constructed augmented Lagrangian function;

[0170] S25: Coordinate and update auxiliary variables according to the updated energy interaction amount;

[0171] S26: updating the energy circulation coefficient according to the updated energy interaction amount and auxiliary variables;

[0172] S27: Calculate the original residual and dual residual based on the updated energy interaction, auxiliary variables and energy circulation coefficient;

[0173] S28: Determine the convergence status based on the calculated primal residual and dual residual;

[0174] S29: Loop through steps S24 to S28, and output the energy interaction amount and energy flow coefficient of each IERO at convergence as the global scheduling strategy.

[0175] In this embodiment, the specific implementation process of step S21 is as follows:

[0176] Before the introduction of auxiliary variables, the energy interaction must satisfy the two-way balance constraints. The original interaction constraints are as follows:

[0177] ,

[0178] Where, represents the interaction amount of energy i delivered from the kth IERO to the lth IERO in period t, It represents the interaction amount of energy i delivered from the l-th IERO to the k-th IERO in period t;

[0179] Formula (20) indicates that the energy i delivered by the k-th IERO to the l-th IERO must be equal to the energy i delivered by the l-th IERO from the k-th IERO. This constraint leads to the direct coupling of the interaction quantities of all IEROs and requires a centralized solution.

[0180] Introducing auxiliary variables and , transform formula (20) into:

[0181]

[0182] Each SIES only needs local updates and , no global information is required.

[0183] In this embodiment, the specific implementation process of step S22 is as follows:

[0184] Construct the augmented Lagrangian form:

[0185] ,

[0186] Where, The auxiliary variable represents the energy i delivered from the k-th IERO to the l-th IERO in period t; Represents the preset penalty factor; represents the square operation of the Euclidean norm;

[0187] The specific process of each IERO distributed solution is as follows:

[0188] In this embodiment, step S23 specifically includes:

[0189] Set the initial energy interaction and the corresponding dual variables.

[0190] ,

[0191] Where, is the interaction amount of energy i delivered from the kth IERO to the lth IERO at the initial moment; is the historical average of the amount of interaction between the k-th IERO and the l-th IERO that transmits energy i at the initial moment; is the auxiliary variable of energy i delivered from the k-th IERO to the l-th IERO at the initial moment; is the circulation coefficient of energy i of the initial k-th IERO.

[0192] In this embodiment, steps S24 to S26 specifically include:

[0193] The energy interaction of the j+1th iteration is obtained by variable splitting based on the alternating direction multiplier method. , auxiliary variables and the dual variable (energy flow coefficient) The calculation formulas are:

[0194] ,

[0195] In this embodiment, steps S27 and S28 specifically include:

[0196] Convergence residual judgment. The original residual of the j+1th iteration and the dual residual The calculation formulas are:

[0197] ,

[0198] when and When , the scheduling problem converges and the iteration terminates.

[0199] Where, is the preset original residual convergence threshold, is the preset dual residual convergence threshold.

[0200] It should be noted that during the auxiliary variable update process, the auxiliary variable and Not completely decoupled.

[0201] In this embodiment, the auxiliary variables are conditioned by KKT conditions. and Decoupling, including:

[0202] For electricity, heat and hydrogen energy, energy interaction benefit optimization models with auxiliary variables are constructed respectively, and augmented Lagrangian functions with penalty terms are constructed;

[0203] Use KKT optimality conditions to derive auxiliary variables and combine them with consistency constraints , solve for the auxiliary variable and The analytical decoupling expression of is used to transform the global coupling problem into a local optimization problem in which each IERO is calculated independently.

[0204] The global coupling problem refers to an optimization problem that needs to consider the interrelated constraints between all sIES (sub-integrated energy systems) in the entire RIES (regional integrated energy system). Its core feature is that variables and constraints have direct dependencies across nodes.

[0205] A local optimization problem refers to converting a global problem into an optimization problem that can be solved independently by each sIES through mathematical decomposition (such as auxiliary variable method and KKT condition), relying only on local variables and limited neighborhood information.

[0206] Taking the power interaction between SIES as an example, the following describes the specific decoupling process of the power interaction variables in the SIES power interaction benefit optimization model using the KKT conditions (Karush-Kuhn-Tucker conditions).

[0207] By global coupling constraints, introducing auxiliary variables, constructing Lagrangian functions and using the alternating direction multiplier method, the optimal model for the electric energy interaction benefit is obtained as follows:

[0208] ,

[0209] Where, is the energy flow coefficient that only considers the transmission of electric energy from the kth IERO to the lth IERO; Represents the transpose operation; is the auxiliary variable that only considers the energy transmitted from the lth IERO to the kth IERO; After iteration j+1, only the interaction of the l-th IERO transmitting power to the k-th IERO is considered; It is the energy flow coefficient that only considers the energy transmitted from the lth IERO to the kth IERO; is the auxiliary variable that only considers the energy transmission from the kth IERO to the lth IERO; is the interaction amount of electric energy transmitted from the kth IERO to the lth IERO after iteration j+1. The core function of formula (26) is to decouple the bidirectional energy interaction constraint by optimizing auxiliary variables. Its principle is based on the ideas of augmented Lagrangian method and alternating direction multiplier method.

[0210] Apply KKT conditions to deduce the above problem and get The analytical expression of :

[0211]

[0212] The derivation process of formula (27) is as follows:

[0213] Construct the Lagrangian function for formula (26):

[0214] ,

[0215] Formula 28 is derived based on energy interaction constraints and the augmented Lagrangian method. Starting from the two-way equilibrium constraint, auxiliary variable decoupling is introduced. Combined with the objective function of maximizing the energy interaction benefit of each SIES in the SIES intraday local correction model, the constraints are incorporated in the form of dual variable penalty terms and quadratic penalty terms containing penalty factors. The formula is finally derived by taking partial derivatives of the auxiliary variables and using the KKT optimality condition.

[0216] According to the KKT condition, for each sIES, the partial derivative is 0:

[0217]

[0218] Formula (28) is respectively and The partial derivative of can be obtained:

[0219] ,

[0220] Right now:

[0221] ,

[0222] And because of the consistency constraint: ,

[0223] Substituting formula (32) into formula (31), we get:

[0224] ,

[0225] The analytical decoupling expression decomposes the global coupling constraint into a distributed computation form that only depends on local and adjacent node variables, enabling each IERO to be solved independently based on local information while ensuring global consistency by synchronizing auxiliary variables through the coordinator.

[0226] So far, in the j+1th iteration process, and Decoupling.

[0227] This decoupling process allows each SIES sub-run scheduling solution to be solved independently, thus achieving the decentralization goal of the primal-dual gradient method.

[0228] Step S3 includes:

[0229] Building multi-agent systems;

[0230] The agent design includes: each sIES acts as an independent agent, responsible for its own interactive decision-making of various types of energy;

[0231] The state space includes: energy flow coefficient, energy interaction, SIES load, and the output or energy consumption of wind power, CHP, electric-hydrogen hybrid energy storage and external hydrogen source;

[0232] The action space includes: discretizing and correcting the energy flow coefficient and energy interaction quantity output by the RIES day-ahead global scheduling model;

[0233] The reward function is calculated as follows:

[0234] ,

[0235] Where, satisfying the constraints refers to meeting the constraints of the sIES intraday local correction model.

[0236] Step S3 also includes using the constructed multi-agent system to perform rolling optimization on the sIES intraday local correction model:

[0237] D1: Using the energy flow coefficient and energy interaction output by the RIES day-ahead global scheduling model as the benchmark scheduling strategy, initialize the Q function of each agent;

[0238] D2: Collect operating data, including actual wind power output, photovoltaic output and load measurement values;

[0239] D3: Adoption -greedy strategy for action selection;

[0240] D4: Output the corrected energy flow coefficient and energy interaction amount based on the action selection;

[0241] D5: Each agent exchanges state and action information through communication, and updates the Q value until the Q value change of all agents is less than the preset Q value change threshold;

[0242] D6: Loop through steps D2 to D5 according to the preset cycle, continuously tracking wind and solar power output fluctuations and load changes, and generating a real-time scheduling strategy.

[0243] In this embodiment, the specific implementation process of step S3 is as follows:

[0244] Considering the randomness of renewable energy generation and the volatility of demand-side energy load, the output and energy load of each SIES unit often fluctuate significantly within a day. Therefore, intraday rolling corrections are performed based on the day-ahead dispatch results.

[0245] Taking into account the real-time volatility of distributed power output during the day, a multi-agent reinforcement learning intelligent algorithm is introduced to achieve rapid tracking of the daily output curve and dynamic adjustment of the scheduling strategy. Multi-agent reinforcement learning algorithm Multi-agent is an important field of distributed artificial intelligence (DAI) research. It consists of multiple freely computing deep reinforcement learning single agents, which communicate and coordinate with other agents. Multi-agent reinforcement learning can decompose complex and large systems into distributed systems that are easy to manage and control. It is one of the important methods to solve complex system optimization and multi-agent game problems. This application sets different sIES of a single IERO as different agents to construct a multi-agent system, where the Q value calculation formula of agent z in the c+1th step is:

[0246] ,

[0247] Where s is the state at the current moment; is the action of agent z at the current moment; is the preset learning rate; is the preset discount rate; is the reward of agent z; is the Q value of agent z at step c; is the action space of agent z; The state of the next moment; is the action of agent z at the next moment;

[0248] In the exploration of reinforcement learning, the reward depends on the energy interaction benefit of the sIES and whether it satisfies the operation constraints, thereby obtaining feedback signals from the environment.

[0249] ,

[0250] The state of the agent includes the following elements: energy flow coefficient and power, the load size of each SIES, the output or energy consumption of wind power, cogeneration, electric-hydrogen hybrid energy storage, and external hydrogen source. Considering that a day is divided into T time sections, in this embodiment, T represents 24 hours; within the day, it is divided into three T's, each T' represents 8 hours. Then, the state s of the agent at any time is:

[0251]

[0252] Where, The energy flow coefficient in the intraday stage, including the electricity flow coefficient, heat energy flow coefficient and hydrogen energy flow coefficient; It represents the energy interaction amount in the daily stage, including the electric energy interaction amount, thermal energy interaction amount and hydrogen energy interaction amount.

[0253] Actions are the decision-making control outputs of each agent. Based on the action requirements of reinforcement learning, the basic circulation coefficient and transaction power upper and lower limits are discretized into 2D fixed values ​​to obtain the action space.

[0254] ,

[0255] ,

[0256] Where, is the correction value of the energy circulation coefficient; is the correction amount of energy interaction; The upper limit of the flow coefficient correction; The upper limit of the transaction volume correction.

[0257] In summary, the energy flow coefficient of the day before can be obtained based on the primal-dual gradient method Interaction with energy , and based on multi-agent reinforcement learning, real-time sensing of intraday power fluctuations can output decision corrections and Therefore, the energy circulation coefficient corrected within the day can be obtained Energy interaction with the correction As shown in the following formula:

[0258] ,

[0259] ,

[0260] In the process of action selection of intelligent agents, if each intelligent agent unanimously selects the action with the largest Q value, reinforcement learning can easily fall into the local optimum. In order to strike a balance between exploration and utilization, it is usually adopted -greedy strategy to update the strategy. The -greedy strategy is to choose the action that is currently considered optimal in most cases, but randomly select other actions with a certain probability to avoid falling into the local optimum. The calculation formula for action selection of the -greedy strategy is:

[0261] ,

[0262] Wherein, arg max() is a function that finds the index of the maximum value in the array. In this embodiment, the arg max function is used to select the current optimal action, that is, the action with the largest current Q value; Represents the actions of the 1st to nth agents; is a random number in the interval [0,1]; To preset the greed coefficient, ; Indicates that the agent randomly chooses actions.

[0263] Example 2:

[0264] Based on the same inventive concept as the first embodiment, this embodiment introduces a device for solving an electric hydrogen energy scheduling strategy, including a model building module, a global solution module, and a real-time correction module;

[0265] The model building module is used to establish a RIES day-ahead global scheduling model and a SIES intraday local correction model based on the architecture of RIES managing multiple SIES in the electric hydrogen energy network;

[0266] The global solution module is used to solve the RIES day-ahead global scheduling model of the model construction module using an improved primal-dual gradient method based on historical data of the electric hydrogen energy network to obtain a global scheduling strategy including energy circulation coefficients and energy interaction quantities; the energy circulation coefficients adjust the interaction priorities between each SIES;

[0267] The real-time correction module is used to apply the global scheduling strategy of the global scheduling strategy solving module to real-time scheduling and collect operating data, perform rolling optimization on the SIES intraday local correction model of the model construction module through multi-agent reinforcement learning, periodically and dynamically correct the energy circulation coefficient and energy interaction amount, and generate a real-time scheduling strategy.

[0268] Example 3:

[0269] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method described in Example 1 are implemented.

[0270] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0271] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0272] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0273] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0274] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for solving electric hydrogen energy scheduling strategy, characterized in that: include: Based on the architecture of RIES in charge of multiple SIES in the electric-hydrogen energy network, a RIES day-ahead global dispatch model and a SIES intraday local correction model are established; Based on the historical data of the electric hydrogen energy network, the RIES day-ahead global dispatch model is solved using an improved primal-dual gradient method to obtain a global dispatch strategy that includes energy circulation coefficients and energy interaction quantities. The energy flow coefficient adjusts the interaction priority between sIES; The global scheduling strategy is used for real-time scheduling and operation data collection. The SIES intraday local correction model is optimized through multi-agent reinforcement learning. The energy circulation coefficient and energy interaction amount are periodically and dynamically corrected to generate a real-time scheduling strategy.

2. The method for solving the electric hydrogen energy scheduling strategy according to claim 1 is characterized in that: The RIES day-ahead global scheduling model includes: an objective function and constraints; The goal of the RIES day-ahead global scheduling model is to maximize the RIES operational benefits. The objective function The calculation formula is: ， Where, represents the income of the kth IERO; represents the cost of the k-th IERO; represents the benefit of the k-th IERO delivering electricity to the outside; represents the revenue of the k-th IERO from delivering hydrogen to the outside; represents the external grid electricity purchase cost of the k-th IERO; represents the external hydrogen source cost of the kth IERO; represents the internal operation and maintenance cost of the k-th IERO; The constraints of the RIES day-ahead global scheduling model include energy balance constraints; The energy balance constraints include electric energy load balance constraints, thermal energy load balance constraints and hydrogen energy load balance constraints; The power load balance constraint formula is: ， Where, represents the amount of electricity purchased from the external power grid by the kth IERO during period t; represents the amount of electricity sold by the k-th IERO to the external grid during period t; represents the wind power output of the kth IERO in period t; represents the power output of the CHP unit of the k-th IERO in period t; represents the energy consumption of the k-th IERO during period t of the electric-hydrogen hybrid energy storage; Represents the set of IERO; represents the amount of electric energy interaction transmitted from the kth IERO to the lth IERO in period t; represents the electrical load of the kth IERO in period t; The heat load balance constraint formula is: ， Where, represents the thermal energy output of the k-th IERO during period t of electric-hydrogen hybrid energy storage; represents the heat output of the CHP unit of the k-th IERO in period t; It represents the amount of heat energy interaction transferred from the kth IERO to the lth IERO in time period t; represents the heat energy load of the kth IERO in period t; The hydrogen energy load balance constraint formula is: ， Where, represents the external hydrogen source output of the kth IERO in period t, represents the hydrogen energy output of the k-th IERO during period t; represents the hydrogen energy output of the k-th IERO during period t; represents the hydrogen energy delivered from the kth IERO to the lth IERO during period t; represents the hydrogen energy load of the kth IERO in period t.

3. The method for solving the electric hydrogen energy scheduling strategy according to claim 2 is characterized in that: The sIES intraday local correction model includes: an objective function and constraints; The goal of the sIES intraday local correction model is to maximize the energy interaction benefits of each sIES under non-cooperative game. The calculation formula of the objective function is: ， Where, represents the benefit obtained by the mth sIES through energy interaction under the kth IERO; Intraday phase defined for the day-ahead phase The flow coefficient of energy i from the m-th sIES under the k-th IERO to the n-th sIES under the l-th IERO in the time period; is the time period of the sIES intraday local correction model, is the period of the sIES intraday local correction model; Represents the set of sIES under a single IERO; Indicates the intraday phase defined in the day-ahead phase The flow coefficient of energy i from the nth sIES under the lth IERO to the mth sIES under the kth IERO in the time period; Intraday stage The interaction amount of energy i delivered from the m-th sIES under the k-th IERO to the n-th sIES under the l-th IERO in the time period; Intraday stage The interaction amount of energy i transmitted from the nth sIES under the lth IERO to the mth sIES under the kth IERO in the time period; Represents the set of sIES under a single IERO; The constraints of the sIES intraday local correction model include: benefit fairness constraint, energy circulation coefficient balance constraint and energy interaction balance constraint; The benefit fairness constraint formula is: ， Where, represents the operating benefit of the kth IERO considering energy interaction; Indicates that the day-ahead phase is formulated in the intraday phase The flow coefficient of energy i delivered by the kth IERO in the time period; represents the benefit obtained by the k-th IERO through energy interaction; It represents the operating benefit of the kth IERO without considering energy interaction; The energy flow coefficient balance constraint formula is: ， Where, Indicates the intraday period set in the day-ahead period The kth IERO in the period transmits the circulation coefficient of energy i to the lth agent; Indicates the intraday period set in the day-ahead period The flow coefficient of energy i delivered by the l-th IERO to the k-th agent in the period; represents the set of IERO, Represents the energy type set, i takes e / H / c; The energy interaction balance constraint formula is: ， Where, In the intraday stage The amount of interaction between the kth IERO and the lth IERO in a period of time, In the intraday stage The amount of interaction in which the l-th IERO transmits i energy to the k-th IERO in the period.

4. The method for solving the electric hydrogen energy scheduling strategy according to claim 3 is characterized in that: Based on the historical data of the electric hydrogen energy network, the RIES day-ahead global dispatch model is solved using the improved primal-dual gradient method. The global dispatch strategy including energy flow coefficient and energy interaction is obtained, including: The energy interaction quantity is used as the primal variable of the primal-dual gradient method, and the energy circulation coefficient is used as the dual variable of the primal-dual gradient method; Auxiliary variables are introduced to decouple the global energy interaction constraints and reconstructed into distributed solutions for each IERO; Construct an augmented Lagrangian function with penalty terms; Set the initial values ​​of energy interaction, auxiliary variables and energy circulation coefficients among each IERO; The following steps are executed cyclically to output the energy interaction amount and energy flow coefficient of each IERO at convergence as the global scheduling strategy: Each IERO updates its energy interaction amount according to the constructed augmented Lagrangian function; Coordinate and update auxiliary variables according to the updated energy interaction amount; Update the energy circulation coefficient according to the updated energy interaction amount and auxiliary variables; Calculate the original residual and dual residual based on the updated energy interaction, auxiliary variables and energy circulation coefficient; The convergence is determined based on the calculated primal and dual residuals.

5. The method for solving the electric hydrogen energy scheduling strategy according to claim 4 is characterized in that: The introduction of auxiliary variables to decouple global energy interaction constraints includes: Introducing auxiliary variables and , constrain the original energy interaction of each IERO: ， Translates to: , to achieve distributed solution.

6. The method for solving the electric hydrogen energy scheduling strategy according to claim 5 is characterized in that: After introducing auxiliary variables to decouple the global energy interaction constraints, it includes: For electricity, heat and hydrogen energy, energy interaction benefit optimization models with auxiliary variables are constructed respectively, and augmented Lagrangian functions with penalty terms are constructed; Use KKT optimality conditions to derive auxiliary variables and set consistency constraints , solve for the auxiliary variable and The analytical decoupling expression of is used to transform the global coupling problem into a local optimization problem in which each IERO is calculated independently.

7. The method for solving the electric hydrogen energy scheduling strategy according to claim 1 is characterized in that: The global scheduling strategy is used for real-time scheduling and operation data collection. The SIES intraday local correction model is optimized through multi-agent reinforcement learning. The energy flow coefficient and energy interaction amount are periodically and dynamically corrected to generate a real-time scheduling strategy, including: building a multi-agent system; The agent design includes: each sIES acts as an independent agent, responsible for its own interactive decision-making of various types of energy; The state space includes: energy flow coefficient, energy interaction, sIES load, and the output or energy consumption of wind power, combined heat and power (CHP), electric-hydrogen hybrid energy storage, and external hydrogen sources; The action space includes: discretizing and correcting the energy flow coefficient and energy interaction quantity output by the RIES day-ahead global scheduling model; The reward function is calculated as follows: ， Where, satisfying the constraints refers to meeting the constraints of the sIES intraday local correction model.

8. The method for solving the electric hydrogen energy scheduling strategy according to claim 7 is characterized in that: The global scheduling strategy is used for real-time scheduling and operation data collection. The SIES intraday local correction model is optimized through multi-agent reinforcement learning. The energy flow coefficient and energy interaction amount are periodically and dynamically corrected to generate a real-time scheduling strategy. The following is also included: the SIES intraday local correction model is optimized through the constructed multi-agent system: The energy flow coefficient and energy interaction output by the RIES day-ahead global scheduling model are used as the benchmark scheduling strategy to initialize the Q function of each agent. The following steps are executed cyclically according to the preset period to continuously track the fluctuation of wind and solar power output and load changes and generate a real-time dispatch strategy: Collect operating data, including actual wind power output, photovoltaic output and load measurement values; use -greedy strategy for action selection; According to the action selection, the corrected energy flow coefficient and energy interaction amount are output; Each agent exchanges state and action information through communication and updates the Q value until the Q value change of all agents is less than the preset Q value change threshold.

9. A device for solving electric hydrogen energy scheduling strategy, characterized in that: Includes model building module, global solution module and real-time correction module; The model building module is used to establish a RIES day-ahead global scheduling model and a SIES intraday local correction model based on the architecture of RIES managing multiple SIES in the electric hydrogen energy network; The global solution module is used to solve the RIES day-ahead global scheduling model of the model construction module using an improved primal-dual gradient method based on historical data of the electric hydrogen energy network to obtain a global scheduling strategy including energy circulation coefficients and energy interaction quantities; the energy circulation coefficients adjust the interaction priorities between each SIES; The real-time correction module is used to apply the global scheduling strategy of the global scheduling strategy solving module to real-time scheduling and collect operating data, perform rolling optimization on the SIES intraday local correction model of the model construction module through multi-agent reinforcement learning, periodically and dynamically correct the energy circulation coefficient and energy interaction amount, and generate a real-time scheduling strategy.

10. 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 8 are implemented.

Images