Multi-objective collaborative optimization method and system for integrated energy system

Abstract

The invention belongs to the technical field of energy system optimization, and particularly provides a multi-objective collaborative optimization method and system for an integrated energy system. Comprising the steps of generating a multi-dimensional uncertainty time sequence scene of the integrated energy system; establishing a high-fidelity dynamic behavior model library; defining decision variables, and constructing a multi-target two-stage stochastic programming optimization model by taking the economy, environmental protection and toughness of the comprehensive energy system as a three-dimensional optimization target; solving the multi-objective two-stage stochastic programming optimization model by adopting a self-adaptive multi-objective evolution algorithm assisted by an agent model; and according to a solving result, realizing a mixed multi-attribute decision based on a fuzzy analytic hierarchy process and a multi-criterion compromise solution sorting method. According to the method, the park-level IES global optimal, high-robustness and high-efficiency planning design is realized from uncertainty modeling, equipment characteristic description and large-scale solution to multi-attribute decision full-chain innovation.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary fields of energy system engineering, artificial intelligence and operations research, and particularly relates to a multi-objective collaborative optimization method and system for integrated energy systems. Background Technology

[0002] Under global climate change, the construction of new power systems with a high proportion of renewable energy is accelerating, and the transformation of energy systems towards cleaner, electrified, and intelligent systems has become an inevitable trend. As the core concentrated unit of urban energy consumption, the low-carbon transformation of industrial parks is of great importance. Integrated Energy Systems (IES) integrate multiple energy subsystems such as cooling, heating, electricity, gas, and transportation within a region through physical or information coupling technologies, covering the entire process of source, grid, load, and storage. This enables cascaded energy utilization, multi-energy complementarity, and supply-demand coordination, and has become a core technological path for improving overall energy efficiency and promoting the consumption of renewable energy.

[0003] IES planning needs to balance economic efficiency, environmental friendliness, and reliability. However, its operating environment is subject to multiple uncertainties—the output of renewable energy (photovoltaics and wind power) fluctuates intermittently, user electricity / heat / cooling loads change randomly, the behavior of flexible resources such as electric vehicles (EVs) is unpredictable, and equipment performance evolves dynamically with operating conditions and life cycle. This places high demands on the accuracy and robustness of the planning scheme. An ideal IES plan needs to simultaneously achieve: accurate characterization of multi-dimensional uncertainties in time series, dynamic reflection of equipment performance throughout its entire life cycle, coordinated optimization of long-term capacity configuration and short-term operating strategies, efficient solution of large-scale optimization problems, and scientific integration of multiple decision preferences.

[0004] Existing IES planning techniques have significant bottlenecks: Firstly, the scenario generation often adopts the "typical day" method (using one typical day from each of the four seasons to represent the whole year), which breaks the continuity of time and fails to capture the impact of extreme events such as continuous rain and heat waves on energy storage capacity and system resilience. Furthermore, it ignores the nonlinear coupling relationship between ambient temperature and photovoltaic output and air conditioning load, resulting in insufficient robustness of capacity configuration schemes and large deviations in economic evaluation. Secondly, equipment modeling is generally simplified to static models (such as setting the COP of heat pumps and the efficiency of fuel cells as fixed constants). For destructive equipment such as energy storage batteries, only the lifespan is converted into annualized cost, ignoring the strong nonlinear coupling relationship between their degradation rate and depth of charge and discharge (DOD), current rate (C-rate), and operating temperature. This can easily lead to short-sighted behavior of "abusing" energy storage, resulting in the actual total life cycle cost far exceeding expectations.

[0005] The equipment capacity configuration (long-term planning) and daily operation scheduling (short-term strategy) of IES are strongly coupled. Unreasonable capacity configuration will compress the operation optimization space, and capacity planning that is detached from refined operation will easily lead to investment waste or insufficient capacity. The ideal solution is to build an integrated "capacity-operation" optimization model. However, this model contains continuous capacity variables and massive 0-1 integer operation variables (covering multiple scenarios and 8760 hours a year), forming a super-large-scale mixed integer nonlinear programming (MINLP) problem. In practice, it faces the "curse of dimensionality": traditional deterministic solvers (such as CPLEX and Gurobi) have difficulty finding the global optimal solution, and metaheuristic algorithms (such as genetic algorithms) need to complete the full-scenario operation simulation to evaluate a single candidate solution. A single evaluation takes several hours, and the overall optimization cycle takes several weeks or even months, resulting in a significant efficiency bottleneck.

[0006] Furthermore, IES planning is a multi-objective optimization problem. Its Pareto optimal solution set needs to be selected by combining the preferences of investors, operators, users, and the government. However, existing linear weighting methods, subjective weight setting, and traditional methods such as TOPSIS are difficult to handle the ambiguity of preferences and the conflict of objectives. The decision-making process lacks scientificity and transparency and cannot meet the needs of engineering practice. Summary of the Invention

[0007] To overcome the shortcomings of the existing technologies, this invention provides a multi-objective collaborative optimization method and system for integrated energy systems. It innovates the entire chain from uncertainty modeling, equipment characteristic characterization, large-scale solution to multi-attribute decision-making, overcomes existing bottlenecks, and achieves global optimality, high robustness, and high efficiency planning and design of park-level IES.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a multi-objective collaborative optimization method for integrated energy systems.

[0009] A multi-objective collaborative optimization method for integrated energy systems includes the following steps: Based on conditional generative adversarial networks, multidimensional uncertain time-series scenarios of integrated energy systems are generated. Establish a high-fidelity dynamic behavior model library that includes a BESS health state dual-scale decay model; Define decision variables and construct a multi-objective two-stage stochastic programming optimization model with the economic efficiency, environmental protection and resilience of the integrated energy system as three-dimensional optimization objectives. An adaptive multi-objective evolution algorithm assisted by a surrogate model is used to solve a multi-objective two-stage stochastic programming optimization model. Based on the solution results, a hybrid multi-attribute decision-making method is realized using fuzzy hierarchical analysis and multi-criteria compromise solution ranking method. This provides alternative schemes and performance interpretation reports for multi-objective collaborative optimization of integrated energy systems, which are then used by decision-makers for final decision-making.

[0010] The second aspect of this invention provides a multi-objective collaborative optimization system for integrated energy systems.

[0011] A multi-objective collaborative optimization system for integrated energy systems, comprising: The scenario generation module is configured to generate multi-dimensional uncertain time-series scenarios of a comprehensive energy system based on a conditional generative adversarial network. The model building module is configured to: build a high-fidelity dynamic behavior model library including a BESS health state dual-scale decay model; The optimization model building module is configured to: define decision variables, take the economic efficiency, environmental protection and resilience of the integrated energy system as three-dimensional optimization objectives, and build a multi-objective two-stage stochastic programming optimization model. The solution module is configured to use an adaptive multi-objective evolution algorithm assisted by a surrogate model to solve a multi-objective two-stage stochastic programming optimization model. The decision-making module is configured to: based on the solution results, implement hybrid multi-attribute decision-making using fuzzy hierarchical analysis and multi-criteria compromise solution ranking method, and provide alternative schemes and performance interpretation reports for multi-objective collaborative optimization of the integrated energy system for decision-makers to make final decisions. A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the multi-objective collaborative optimization method for integrated energy systems as described in the first aspect of the present invention.

[0012] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the multi-objective collaborative optimization method for integrated energy systems as described in the first aspect of the present invention.

[0013] The above one or more technical solutions have the following beneficial effects: This invention addresses the technical problems in existing integrated energy system optimization, such as distorted uncertainties, simplified equipment models, disconnect between planning and operation, low solution efficiency, and strong decision-making subjectivity. It proposes a multi-objective collaborative optimization method for integrated energy systems based on conditional generative adversarial networks (CGANs) and surrogate model-assisted evolutionary algorithms. The method involves: acquiring historical system data; using a CGAN model to learn and generate long-term uncertain operating scenarios with high-fidelity spatiotemporal correlations; combining scenario reduction techniques to obtain typical scenarios and probabilities; constructing a high-fidelity dynamic equipment model library including heat pump nonlinear performance, dual-scale decay of storage SOH, V2G aggregation response, and electricity-hydrogen-electricity energy conversion models; establishing a multi-objective two-stage stochastic programming (MINLP) model with three dimensions: life-cycle cost, carbon emissions, and system resilience, using equipment capacity configuration as the first-stage decision variable and scenario operation scheduling as the second-stage decision variable; and using a surrogate model-assisted adaptive multi-objective evolutionary algorithm for solving the problem. Initial samples are generated using LHS to train the GPR surrogate model, and key samples are selected for real-world evaluation using NSGA-III global search and EHVI criteria. The model is dynamically updated to achieve efficient iteration and outputs the Pareto optimal frontier. The method is based on Fuzzy... AHP integrates the weights of multiple expert preferences calculation criteria, combines the VIKOR method to calculate the group benefits, individual regrets, and trade-offs of the proposed solutions, and then selects and outputs a final recommended solution list that takes into account the needs of all parties. This achieves synergistic optimization of integrated energy systems under the objectives of economy, environmental protection, and resilience, significantly improving the rationality and solution efficiency of the optimized solutions.

[0014] This invention, through full-chain technological innovation, not only ensures the robustness, life-cycle optimization, and comprehensiveness of the planning scheme, but also significantly improves the solution efficiency. At the same time, it realizes the scientification and transparency of the decision-making process. It is applicable to the planning and design of various park-level integrated energy systems and has extremely strong engineering application value and promotion potential.

[0015] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0016] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0017] Figure 1 This is a schematic diagram of the physical structure of a typical park-level integrated energy system in Embodiment 1 of the present invention.

[0018] Figure 2 This is a flowchart illustrating the overall technical process of the multi-objective collaborative optimization method according to Embodiment 1 of the present invention.

[0019] Figure 3 This is a detailed flowchart of the "Adaptive Multi-Objective Evolution Algorithm Assisted by Agent Model" in Embodiment 1 of the present invention.

[0020] Figure 4 This is a schematic diagram of the framework of the "Fuzzy AHP-VIKOR hybrid multi-attribute decision" method used for final decision support in Embodiment 1 of the present invention. Detailed Implementation

[0021] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0023] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0024] Example 1 To address the aforementioned technical challenges in the planning and optimization of integrated energy systems at the park level, this invention discloses a multi-objective collaborative optimization method for integrated energy systems based on conditional generative adversarial networks and agent model-assisted evolutionary algorithms. This method constructs a complete technology chain encompassing "data-driven scenario generation—mechanism and data-integrated dynamic modeling—multi-dimensional objective integrated optimization—algorithm-level computational acceleration—scientific multi-attribute decision-making."

[0025] This invention addresses the multi-objective collaborative optimization needs of integrated energy systems by formally defining the multi-objective optimization problem. It utilizes conditional generative adversarial networks (CGAN) to generate data-driven uncertain temporal scenarios, and combines mechanisms and data to construct high-fidelity dynamic behavior models of key system equipment. This results in a multi-dimensional integrated optimization model encompassing equipment capacity planning and multi-scenario operation scheduling. Furthermore, it accelerates the computation of large-scale optimization problems by relying on surrogate models to assist evolutionary algorithms, and finally completes the selection of Pareto optimal solutions through scientific multi-attribute decision-making.

[0026] This invention achieves synergistic optimization of integrated energy systems under the objectives of economy, environmental protection and resilience, and significantly improves the rationality of optimization schemes and solution efficiency.

[0027] like Figure 2 As shown, a multi-objective collaborative optimization method for an integrated energy system includes the following steps: Based on conditional generative adversarial networks, multidimensional uncertain time-series scenarios of integrated energy systems are generated. Establish a high-fidelity dynamic behavior model library that includes a BESS health state dual-scale decay model; Define decision variables and construct a multi-objective two-stage stochastic programming optimization model with the economic efficiency, environmental protection and resilience of the integrated energy system as three-dimensional optimization objectives. An adaptive multi-objective evolution algorithm assisted by a surrogate model is used to solve a multi-objective two-stage stochastic programming optimization model. Based on the solution results, a hybrid multi-attribute decision-making method is realized using fuzzy hierarchical analysis and multi-criteria compromise solution ranking method. This provides alternative schemes and performance interpretation reports for multi-objective collaborative optimization of integrated energy systems, which are then used by decision-makers for final decision-making.

[0028] The method of this embodiment will now be explained in detail. The technical implementation steps of this embodiment include: Generation of multidimensional uncertain temporal scenes based on conditional generative adversarial networks (CGAN); Establish a high-fidelity dynamic behavior model library; Construct a multi-objective two-stage stochastic programming optimization model; An adaptive multi-objective evolution algorithm assisted by a surrogate model is used to solve the problem; Hybrid multi-attribute decision-making based on fuzzy hierarchical analysis (Fuzzy AHP) and VIKOR.

[0029] In this embodiment, VIKOR means the multi-criteria compromise solution ranking method described above.

[0030] Furthermore, the generation of multidimensional uncertain temporal scenes based on conditional generative adversarial networks specifically includes model structure and principles, adversarial training process, scene generation and reduction.

[0031] Furthermore, the model structure, principle, and method include: CGAN consists of a generator (G) and a discriminator (D), both of which are deep neural networks. The generator G typically employs a Long Short-Term Memory (LSTM) network or a Transformer architecture, which is adept at processing temporal data, and receives random noise vectors. z (Used to introduce randomness) and a conditional vector containing meteorological data, load data, and time stamps from the past k time periods. c The goal is to generate one or more future moments and conditions. c A multidimensional data sequence that is logically and statistically matched This sequence covers photovoltaic output, various load demands, EV availability, etc.; the discriminator D typically employs a convolutional neural network (CNN) architecture, receiving data-condition pairs. And output data x Under given conditions c The following are the probabilities of real data. The core task is "authentication".

[0032] Furthermore, the adversarial training process includes: G and D are optimized through adversarial training in a "zero-sum game" manner, with the training objective being a minimax problem. G needs to generate data that approximates reality to "deceive" D (make it...). As close to 1 as possible), D needs to accurately distinguish between real data and generated data (so that...) For real data x As close as possible to 1, for the generated data To minimize the impact of GAN training, this invention preferably employs the Wasserstein GAN with Gradient Penalty (WGAN-GP) loss function to avoid the instability problem of traditional GAN ​​training. By alternately optimizing G and D to Nash equilibrium, G can learn the intrinsic joint probability distribution of real data under given conditions, capturing the nonlinear temporal correlation, higher-order statistical moments, and extreme event patterns among variables.

[0033] Specifically, the objective function is:

[0034] In the formula, The loss function value for WGAN-GP. The distribution learned by the generator Distribution of generated data The expectation operator, For discriminator D, the "generated data-condition pair" is... The output, For the true data distribution, These are random interpolation points between real and generated data. The gradient penalty coefficient is... For discriminator D to interpolate data gradient L 2-norm, This is the gradient penalty term.

[0035] Furthermore, the scene generation and reduction method includes: after training, inputting different random noises into G. z and historical conditions c This process generates thousands to tens of thousands of annual (8760-hour) operational scenarios with statistical characteristics highly consistent with real data, preserving temporal correlation and variable coupling. Subsequently, a scenario reduction technique based on Dynamic Time Warping (DTW) distance and K-medoids clustering is employed to cluster these massive scenarios into... N s A representative typical scenario is selected, and calculations are performed for each scenario. s probability of occurrence This is to balance the need to retain uncertain information with the need to control computational complexity.

[0036] Furthermore, the establishment of a high-fidelity dynamic behavior model library specifically includes the BESS health state (SOH) dual-scale decay model, the multi-agent (ABM) V2G aggregation model, the heat pump COP model, and the electric-hydrogen-electric chain model.

[0037] Furthermore, the BESS health state dual-scale decay model includes: distinguishing and coupling the "microscopic damage" caused to the battery by the operating strategy and the "macroscopic aging" within the planning cycle, at each scheduling moment on the hourly (operating scale). t The State of Health (SOH) of a Battery Safety System (BESS) is defined as the ratio of the current maximum available capacity to the initial capacity and the ratio of the current internal resistance to the initial internal resistance, both considered constant. However, the irreversible damage caused by its charging and discharging behavior is quantified using a comprehensive empirical model fitted based on extensive battery aging experimental data. At the end of a planning iteration cycle (one year), the battery's SOH is updated. The updated SOH directly affects the battery's maximum available capacity, charging and discharging efficiency, and upper and lower power limits in the scheduling model for the following year, forming a closed-loop feedback between planning, operation, and equipment health evolution.

[0038] Specifically, the comprehensive empirical model fitted based on a large amount of battery aging experimental data can be expressed by the following formula:

[0039]

[0040] In the formula, For scheduling time t The amount of irreversible capacity decay caused by a single charge-discharge cycle of the internal BESS. This is a function of the capacity decay coefficient. Depth of battery discharge, The charge / discharge current ratio of the battery. The battery body temperature during BESS operation. The charging and discharging power of BESS within the scheduling time t. To schedule the time step, For scheduling time t The irreversible increase in internal resistance caused by the single charge-discharge behavior of the internal BESS. The coefficient function is added to the internal resistance.

[0041] Specifically, after a planning iteration cycle (one year) ends, the State of Harmony (SOH) of the battery is updated, and the updated values ​​include: The capacity SOH update formula can be expressed as:

[0042] In the formula, For the first Annual BESS capacity health status For the first Annual BESS capacity health status For the probability-weighted expectation operator of all typical scenarios, Index for typical scenarios, Typical scenario Next All scheduling times within the year The sum of capacity decay, For the first Annual capacity decay due to "calendar aging" This is the initial rated capacity of BESS.

[0043] The formula for updating the internal resistance SOH can be expressed as:

[0044] In the formula, For the first Annual BESS internal resistance health status, For the first Annual BESS internal resistance health status, Typical scenario Next All scheduling times within the year The total increase in internal resistance, For the first The annual increase in internal resistance due to "calendar aging". Let be the initial internal resistance of BESS.

[0045] Calendar aging and It is a function of the annual average state of charge and the annual average temperature, and usually follows the Arrhenius equation:

[0046] In the formula, This is the amount of time the calendar ages. They are directly proportional. The activation energy for the battery aging reaction. Let be the ideal gas constant. This refers to the battery's average annual operating temperature.

[0047] Furthermore, the method of the multi-agent (ABM) V2G aggregation model includes: constructing an EV vehicle-to-grid (V2G) aggregation model using multi-agent (ABM), where each EV user is defined as an agent with independent attributes (vehicle type, commuting habits, annual mileage, electricity price sensitivity, etc.) and decision-making logic. Simultaneously, a virtual environment is constructed, including the park's geographical layout, the distribution and number of charging piles, real-time electricity price signals, and V2G scheduling instructions. Before optimization begins, a large-scale (e.g., simulating for several weeks or months) ABM simulation is run. Each EV agent moves within the virtual environment according to a preset travel chain. When the vehicle stops at a parking space equipped with V2G, a decision is made based on a utility function that comprehensively considers the current SOC, expected dwell time, V2G charging and discharging benefits, battery wear costs, and range anxiety. By statistically analyzing the individual decision results of all EVs in the simulation, the results at each time point can be obtained. t Maximum charging and discharging power boundaries of the entire park's V2G cluster Dispatchable energy capacity boundary The corresponding marginal cost curves, and these dynamic boundaries and cost parameters will be used as constraints and cost terms input into the master optimization model.

[0048] Furthermore, the heat pump COP The model and methodology include: using a multivariate nonlinear regression method to fit heat pump data based on equipment manufacturer data or TRNSYS simulation data. COP The performance surface.

[0049] Specifically, the aforementioned COP It can be represented as:

[0050] In the formula, The energy efficiency ratio of a heat pump. For COP performance functions, The temperature on the heat pump source side. The outlet water temperature on the load side of the heat pump. This refers to the heat pump partial load rate.

[0051] Furthermore, the electro-hydrogen-electric chain model and method include: the PEM electrolyzer uses the Butler-Volmer equation to describe the nonlinear relationship between overpotential and current density to obtain an accurate hydrogen production efficiency curve; the hydrogen storage tank uses the Peng-Robinson or Redlich-Kwong equation of state under high pressure to accurately calculate the hydrogen storage capacity; and the PEM fuel cell uses a voltage model that includes activation, ohmic and concentration polarization to describe the power generation efficiency under different loads.

[0052] Furthermore, the construction of the multi-objective two-stage stochastic programming optimization model specifically includes: Define decision variables and divide them into first-stage variables and second-stage variables; An objective function is constructed with "economic efficiency, environmental friendliness, and resilience" as the three-dimensional optimization objectives; Set the main constraints.

[0053] Furthermore, the variables in the first stage include: device capacity configuration vector. X Including photovoltaics Energy storage capacity Energy storage heat pump Continuous variables and discrete variables such as equipment location and model.

[0054] Furthermore, the variables in the second stage include: various scenarios s At each moment t The running strategy vector Power purchased Energy storage charging power Boiler start-up and shutdown Continuous variables such as the output of each device, the charging and discharging power of energy storage, and the power interacting with the grid, as well as 0-1 integer variables representing the start-up and shutdown status of the devices.

[0055] Furthermore, the objective function constructed using "economic efficiency, environmental friendliness, and resilience" as the three-dimensional optimization objectives includes: Minimize the expected value of total life cycle cost (LCC) :

[0056] In the formula, Expected total lifecycle cost For total lifecycle cost, X The equipment capacity configuration vector is used as the first-stage decision variable. The initial investment cost of the equipment. For the planning period, The operating strategy is the decision variable for the second stage. For the first y Expected annual operating costs The equipment replacement cost for year y. is the discount rate.

[0057] Minimize the expected annual carbon emissions :

[0058] In the formula, This represents the expected annual carbon emissions. This refers to the total annual carbon emissions. For the scene s Next t The time-tracking system purchases power from the municipal power grid. For the first t Marginal carbon emission factor of municipal power grid at all times For the scene s Next t The natural gas consumption of the gas-fired boiler at all times. It is a carbon emission factor for natural gas.

[0059] Maximize the overall system resilience index :

[0060] In the formula, As a comprehensive system resilience index, For the probability-weighted expectation operator of extreme events, For extreme events e The system resilience performance value under the following conditions Index for extreme events, Let e ​​be the probability of extreme event e occurring. For the critical load set, For critical load indexing, For the first Weighting coefficients for critical loads, For extreme events Next Actual supply of critical loads For extreme events Next Total demand for critical loads.

[0061] Furthermore, the setting of the main constraints includes: At any given time, various energy networks (electricity, heat, cooling, hydrogen) t Various scenarios s Under the power / energy balance constraint; Photovoltaic (PV), energy storage (BESS), heat pump (HP), boiler, V2G cluster, hydrogen system, and other equipment and their capacity X Output upper and lower limit constraints that are dynamically linked to dynamic efficiency and SOH state; State transition (SOC) equations and safe operating boundary constraints for energy storage systems (BESS, thermal / cold storage tanks, hydrogen storage tanks); The dynamic power boundary provided by the V2G aggregation model With the energy boundary constraint; Contractual capacity constraints on the power interaction with the municipal power grid; Dynamic constraints on equipment start / stop logic, minimum running / downtime, ramp rate, etc. Minimum requirements for performance indicators such as renewable energy integration rate and system self-sufficiency rate.

[0062] Furthermore, the adaptive multi-objective evolution algorithm assisted by the surrogate model specifically includes: The core of the efficient solution algorithm framework is designed with "inexpensive approximate evaluation replacing expensive precise evaluation". The upper layer adopts the improved NSGA-III with reference point mechanism as the master optimizer to establish global search capability, so as to efficiently handle high-dimensional multi-objective optimization problems and guide the population to converge uniformly to the Pareto front. Based on the upper-layer optimized architecture, initial sample points are generated in the capacity variable space X through LHS and high-fidelity data is obtained through full-scene MIP real solution. Then, three independent GPR proxy models that can output the objective function prediction value and uncertainty variance are trained to complete the initialization preparation of the lower-layer adaptive proxy model evaluation mechanism. Based on the initialization results, a lower-level dynamic iterative adaptive strategy is constructed in the NSGA-III iteration. The lower-level adaptive evaluation and model optimization are achieved by quickly predicting new candidate solutions through the GPR surrogate model, screening key candidate solutions by EHVI value, performing MIP true solution on a small number of key candidate solutions and updating the GPR model. By alternating between upper-level NSGA-III global search and lower-level EHVI adaptive evaluation, the majority of fitness evaluations are completed with a low-cost surrogate model, and the real solver is called only at key nodes, which greatly shortens the computation time. When the convergence condition is met, the final Pareto optimal frontier is output.

[0063] Furthermore, the hybrid multi-attribute decision-making based on fuzzy hierarchical analysis and VIKOR specifically includes: Construct a three-level decision-making model; A decision-making group consisting of school administrators, investors, and other stakeholders was formed. The triangular fuzzy pair criterion was used to compare the importance of each pair of factors, and a fuzzy judgment matrix was constructed to capture the fuzziness and uncertainty of human judgment. The Chang extended analysis method is used to integrate the expert judgment matrix, calculate the fuzzy weights of each criterion, and after consistency verification, defuzzification is performed using the centroid method to obtain a clear and objective group weight vector. w ; extract k The three target performance values ​​of the alternative schemes were normalized to eliminate dimensional differences and construct... k A standardized decision matrix of ×3 provides data support for indicator calculation; Calculate the VIKOR index for each scheme; The solutions are sorted in ascending order of Q value. The one ranked first is the preliminary optimal compromise solution. We will verify whether it meets the requirements of "acceptable advantage" and "acceptable stability". If it meets the requirements, we will provide alternative solutions and performance interpretation reports for decision-makers to make the final decision.

[0064] Furthermore, the three-layer decision-making model is constructed, including the objective layer, the criterion layer, and the alternative layer.

[0065] Specifically, the objective layer is to select the optimal IES solution, and the criterion layer includes economic considerations. f 1) Environmental friendliness ( f 2) Toughness ( f 3) The solution layer is the Pareto frontier. k A representative non-inferior solution.

[0066] Furthermore, the calculation of the VIKOR index for each scheme includes calculating the group benefit ( S ), individual regrets R ) and compromise quantity ( Q ).

[0067] Specifically, the calculation of group benefits ( S ), individual regrets R ) and compromise quantity ( Q This can be represented as:

[0068]

[0069]

[0070] In the formula, For the first i The group benefit value of each alternative plan The index takes values ​​of 1, 2, and 3 as criteria. For the first The optimal target value of each criterion. No. i The first scheme is in the Normalized performance values ​​based on a criterion, No. j The worst-case objective value of each criterion For the first i Individual regret value of each alternative option For the first i The compromise value of each alternative solution. Decision preference coefficient Group benefits among all alternative options The optimal value, Group benefits among all alternative options The worst value, Individual regrets among all alternative options The optimal value, Individual regrets among all alternative options The worst value.

[0071] The present invention will be further illustrated below with a specific example.

[0072] like Figure 1 As shown, this embodiment illustrates the comprehensive energy system planning for a newly built campus of a research university in eastern China. Located in the subtropical monsoon climate zone of eastern China, the campus experiences four distinct seasons: hot and humid summers, mild and dry winters, and frequent extreme weather events (such as continuous rain during the plum rain season and extreme summer temperatures), placing high demands on the robustness and adaptability of the energy system. The campus covers a total area of ​​1000 mu (approximately 67 hectares), encompassing 15 teaching and research buildings, 20 student dormitories, a university-level data center, a gymnasium, and supporting commercial facilities, with a total building area of ​​approximately 500,000 square meters. The expected full-time student and faculty population is 25,000. From the perspective of energy demand characteristics, the campus exhibits "diversified, highly volatile, and rigid" features. The teaching and research buildings and student dormitories experience significant diurnal and semester-specific load fluctuations (load differences between weekdays and holidays reaching 40%). The university-level data center operates continuously 24 / 7, requiring extremely high power supply reliability (allowable interruption time ≤ 5 minutes). The gymnasium and commercial facilities experience intermittent high-load demands (such as during sporting events and peak holiday consumption periods). Based on the above requirements, the power and cooling loads of the data center and core laboratory are defined as "critical loads," which must be prioritized for protection under extreme events; other loads (such as general lighting and non-core air conditioning) are defined as "adjustable loads," which can participate in system optimization scheduling. In this embodiment, the system planning includes the following steps: Integrated energy system of the park (100); The energy input interface has a 50MW power supply contract with the municipal power grid (101), implementing time-of-use pricing and demand-based pricing policies. It is connected to the municipal natural gas pipeline network (102), with a fixed gas price. The renewable energy unit plans to deploy photovoltaic (PV) systems on a large scale on building rooftops and parking lots (103), the capacity of which is a decision variable to be optimized; The energy storage unit is planned to be a centralized electrochemical energy storage station (BESS) (104), with lithium iron phosphate batteries as the selected technology. The rated power and capacity are the decision variables to be optimized. Meanwhile, considering the long-term backup power requirements of critical loads such as data centers and seasonal energy storage, a hydrogen energy system (105) is considered as an alternative technology. The hydrogen energy system includes a PEM electrolyzer (105a), a 35MPa high-pressure gaseous hydrogen storage tank (105b), and a PEM fuel cell (105c). The capacity of all three is the decision variable to be optimized. The energy conversion unit uses a ground source heat pump (HP) (106) as the basic heat source and utilizes shallow geothermal energy. A gas-fired boiler (107) is also configured as a peak-shaving and backup heat source, and its capacity is a decision variable to be optimized. Flexible resources: the campus is planned to have more than 2,000 parking spaces, of which 500 are planned to be equipped with V2G bidirectional charging piles to form a V2G vehicle network (108). Hot / cold storage and network, constructing large-scale thermal / cold water storage tanks (TS) (109), the capacity of which is a decision variable to be optimized, and supplying energy to each building through hot and cold water pipe networks; The load side includes the school's electrical load (110) and cooling / heating load (111).

[0073] The following is a detailed description of this embodiment, such as Figure 2 As shown: Step S1: Generate multidimensional uncertain temporal scenarios based on conditional generative adversarial networks (CGAN).

[0074] Step S11: Collect hourly meteorological data of the campus location over the past 10 years and historical load data of similar functional areas in the surrounding area. Through questionnaires for teachers and students and traffic flow simulation, establish a probability distribution model of EV users' commuting patterns, average daily driving mileage, and dwell time. The hourly meteorological data of the campus location over the past 10 years includes solar irradiance, ambient temperature, and wind speed. The historical load data of similar functional areas in the surrounding area includes electricity, cooling, and heating.

[0075] Step S12: Build an LSTM-based CGAN model. The generator G is a 3-layer LSTM network, using the past 24 hours of weather, load data, and time encoding as conditions. c The algorithm generates multidimensional uncertainty variables for the next hour. The discriminator D is a 1D-CNN network. It uses the WGAN-GP loss function and is trained on an NVIDIA A100 GPU server, taking approximately 48 hours.

[0076] Step S13: After training, using historical data as initial conditions, 1000 time-series scenarios covering the next year (8760 hours) are generated through Monte Carlo sampling, incorporating multi-dimensional variables such as photovoltaic output, electricity / cooling / heating load, and the number of available V2G vehicles. Subsequently, a K-medoids clustering algorithm based on DTW distance is used to reduce the 1000 scenarios to 30 typical scenarios. s And calculate the probability of occurrence for each scenario. .

[0077] Step S2: Establish a high-fidelity dynamic behavior model library.

[0078] Step S3: Construct a multi-objective two-stage stochastic programming optimization model.

[0079] Step S31: Build the model in the Python environment using the Pyomo optimization modeling framework. The decision variables are divided according to a two-stage stochastic programming logic. The first stage is a unified device capacity configuration vector across the entire scenario. X The first stage includes 10 non-negative continuous variables, such as photovoltaics, battery energy storage, ground source heat pumps, gas boilers, electrolyzers, hydrogen storage tanks, fuel cells, and thermal / cold water storage tanks. The value range is determined by combining campus requirements and equipment parameters, and the decision results remain consistent across all scenarios. The second stage is a scenario-dependent operation strategy vector. The system is defined for 30 typical scenarios and 8760 hours per year, including continuous operation variables such as power grid purchase and equipment output, as well as 0-1 integer variables for equipment start and stop. Due to the large number of scenarios and time periods, the total number of variables reaches millions, which correspond to the decision-making needs of the planning and scheduling layers.

[0080] Step S32: Using "economic efficiency, environmental friendliness, and resilience" as the three-dimensional optimization objectives, the expected values ​​are calculated based on scenario probability weighting. Objective 1 is to minimize the expected value of the total life cycle cost (LCC) over a 20-year planning period with a discount rate of 8%, covering initial investment, annual operation, and equipment replacement costs. Objective 2 is to minimize the expected value of the average annual total carbon emissions, taking into account the carbon emissions implied by grid purchases and natural gas combustion. Objective 3 is to maximize the comprehensive resilience index of the system. Through nested Monte Carlo simulation evaluation, three extreme events are preset: 72 hours of grid outage, 30% reduction in heat pump efficiency and 50% increase in cooling load due to extreme high temperature, and 48 hours of natural gas outage, to quantify the system's ability to ensure critical loads.

[0081] Step S33: Preprocessing, the nonlinear terms such as heat pump COP and electrolytic cell polarization curve in the lower-level operation problem are piecewise linearized and transformed into mixed integer linear programming (MIP). The 8760-hour time series is clustered to reduce the amount of computation. However, even so, the complete evaluation of a single capacity scheme is still time-consuming, highlighting the necessity of using a surrogate model to assist the evolution algorithm to improve the solution efficiency.

[0082] like Figure 3 As shown, step S4: Solve the problem using an adaptive multi-objective evolution algorithm assisted by a surrogate model.

[0083] Step S41: Set the algorithm, with the NSGA-III population size set to 100, the maximum number of iterations to 200 generations, and Gaussian process regression (GPR) chosen as the surrogate model because it can naturally provide prediction uncertainty. The lower-level optimization problem (after linearization) is solved using the Gurobi solver.

[0084] Step S42: Initialization is set up. Latin hypercube sampling generates 30 initial capacity configuration schemes. For each scheme, Gurobi is called to solve for runtime optimization under 30 scenarios, and the actual objective function value is calculated. This step is computed in parallel and takes approximately 5 hours.

[0085] Step S43: Use these 30 high-fidelity samples to train the GPR proxy model.

[0086] Step S44: NSGA-III performs genetic operations (selection, crossover, mutation) to generate 100 new offspring individuals. For these 100 new individuals, the GPR surrogate model is used to make a rapid prediction and calculate the expected hypervolume improvement (EHVI) of each new individual to the current Pareto frontier.

[0087] Step S45: Select the 5 individuals (decision points) with the highest EHVI values ​​and determine that they are "worth" being evaluated.

[0088] Step S46: Feed these 5 individuals into the lower-level Gurobi solver for accurate evaluation. After obtaining the true values, add these 5 new high-fidelity sample points to the training set and update the GPR surrogate model.

[0089] Step S47: Use the fitness of 100 offspring individuals together with the parents to perform non-dominated ranking and elite retention to form a new generation population.

[0090] Step S48: Repeat the main loop 200 times.

[0091] Step S49: Obtain a set containing 100 Pareto optimal solutions.

[0092] like Figure 4 As shown, step S5: Hybrid multi-attribute decision based on fuzzy hierarchical analysis (Fuzzy AHP) and VIKOR.

[0093] Step S51: Invite school leaders, the finance department, the logistics group, and professors from the School of Energy to form a decision-making expert group.

[0094] Step S52: The expert panel used the triangular fuzzy number form of the "(1-9) scaling method" to perform pairwise comparisons of "economic efficiency (LCC)," "environmental friendliness (CO2)," and "resilience." The comprehensive weight vector was obtained through FuzzyAHP calculation. w .

[0095] Step S53: Select 20 representative solutions from the 100 Pareto solutions. Construct the decision matrix and calculate the decision value of each solution. S , R , Q value.

[0096] Step S54: Sort the options according to their Q values, and at the same time, select the second and third ranked options as alternatives, and provide a detailed equipment configuration list, expected annual cash flow, carbon footprint analysis and resilience assessment report for the decision-making level to make the final decision.

[0097] Through the complete implementation process described above, this invention provides the university's new campus with a scientific, efficient, and robust integrated energy system planning scheme, fully demonstrating the advanced nature and practical value of this invention.

[0098] This invention proposes a multi-objective collaborative optimization method for integrated energy systems based on conditional generative adversarial networks (CGANs) and surrogate model-assisted evolutionary algorithms. Addressing issues such as distorted uncertainties, simplified equipment models, separation of planning and operation, low solution efficiency, and strong decision-making subjectivity in existing integrated energy system planning, this method constructs a complete technology chain: "data-driven scenario generation—mechanism and data-integrated dynamic modeling—multi-dimensional objective integrated optimization—algorithm-level computational acceleration—scientific multi-attribute decision-making." Specifically, it generates long-term uncertainties with high-fidelity spatiotemporal correlations using CGANs and reduces them through clustering; establishes a high-fidelity model library covering equipment dynamic characteristics; constructs a two-stage stochastic programming model with equipment capacity configuration as the first stage and scenario operation scheduling as the second stage, considering economy, environmental protection, and resilience; employs an adaptive multi-objective evolutionary algorithm assisted by a surrogate model to achieve efficient solution and output the Pareto optimal front; and combines fuzzy hierarchical analysis (Fuzzy AHP) and VIKOR to complete scientific decision-making integrating multiple preferences. This method significantly improves the robustness, life-cycle optimization, and decision-making transparency of planning schemes, and significantly increases solution efficiency. It is applicable to the planning and design of various park-level integrated energy systems and has strong engineering application value and promotion potential.

[0099] Example 2 This embodiment discloses a multi-objective collaborative optimization system for integrated energy systems.

[0100] A multi-objective collaborative optimization system for integrated energy systems, comprising: The scenario generation module is configured to generate multi-dimensional uncertain time-series scenarios of a comprehensive energy system based on a conditional generative adversarial network. The model building module is configured to: build a high-fidelity dynamic behavior model library including a BESS health state dual-scale decay model; The optimization model building module is configured to: define decision variables, take the economic efficiency, environmental protection and resilience of the integrated energy system as three-dimensional optimization objectives, and build a multi-objective two-stage stochastic programming optimization model. The solution module is configured to use an adaptive multi-objective evolution algorithm assisted by a surrogate model to solve a multi-objective two-stage stochastic programming optimization model. The decision-making module is configured to: based on the solution results, implement hybrid multi-attribute decision-making using fuzzy hierarchical analysis and multi-criteria compromise solution ranking method, and provide alternative schemes and performance interpretation reports for multi-objective collaborative optimization of the integrated energy system for decision-makers to make final decisions. Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0101] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the multi-objective collaborative optimization method for integrated energy systems as described in Embodiment 1 of this disclosure.

[0102] Example 4 The purpose of this embodiment is to provide an electronic device.

[0103] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the multi-objective collaborative optimization method for integrated energy systems as described in Embodiment 1 of this disclosure.

[0104] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0105] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0106] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for multi-objective collaborative optimization of an integrated energy system, characterized in that, The method comprises the following steps: generating a multi-dimensional uncertainty time sequence of the comprehensive energy system based on a conditional generative adversarial network; establishing a high-fidelity dynamic behavior model library comprising a double-scale attenuation model of a BESS health state; defining decision variables, taking economy, environmental protection and resilience of the comprehensive energy system as three-dimensional optimization objectives, and constructing a multi-objective two-stage stochastic programming optimization model; solving the multi-objective two-stage stochastic programming optimization model by using an agent model assisted adaptive multi-objective evolutionary algorithm; based on the solving result, realizing hybrid multi-attribute decision making based on a fuzzy analytic hierarchy process and a multi-criteria compromise solution sorting method, providing an alternative scheme and performance interpretation report of the multi-objective collaborative optimization of the comprehensive energy system for a decision maker to make a final decision.

2. The integrated energy system multi-objective collaborative optimization method of claim 1, wherein, The method for generating a multi-dimensional uncertainty time sequence of the comprehensive energy system based on a conditional generative adversarial network comprises the following steps: the conditional generative adversarial network CGAN model comprises a generator and a discriminator; the generator adopts a long short-term memory network LSTM or a Transformer architecture, and generates output and load data at a future time based on historical meteorological and load data sequences at a time; the discriminator is used to distinguish between generated data-condition pairs and real historical data-condition pairs; through an adversarial training method of a Wasserstein distance plus gradient penalty, the generator can stably learn and reproduce nonlinear time sequence correlations and statistical characteristics of extreme events between uncertainty variables; alternatively, the method further comprises scene generation and reduction: After the conditional generative adversarial network CGAN is trained, different random noises are input to the generator z and historical conditions c , and annual operation scenarios with statistical characteristics highly consistent with real data, time correlation and variable coupling are generated; Subsequently, a scene reduction technique based on dynamic time warping distance and K-medoids clustering is adopted to cluster the massive scenes into representative typical scenes, and the occurrence probability of each scene s is calculated to balance the uncertainty information retention and the computational complexity control requirements.

3. The integrated energy system multi-objective collaborative optimization method of claim 2, wherein, the double-scale attenuation model of the BESS health state comprises the following steps: in the hourly operation scheduling, the equivalent capacity attenuation amount and the internal resistance increase amount of a single cycle are calculated by using an empirical formula or an electrochemical model based on experimental data fitting, according to the real-time charge-discharge rate, the environmental temperature, the state of charge SOC and the depth of discharge DOD of the battery; in the annual scale planning iteration, the cycle attenuation expectation value under all scenes is accumulated, and the cycle attenuation expectation value is added to the calendar attenuation amount calculated based on the annual average temperature and the annual average state of charge, to update the SOH parameters of the battery in the next year, including the capacity and the internal resistance; the updated SOH parameters are fed back to the scheduling model of the next year as the device capacity constraint and the efficiency parameter of the next year, to form a closed-loop coupling of planning and device health evolution.

4. The integrated energy system multi-objective collaborative optimization method of claim 1, wherein, an expected value of a full life cycle cost LCC reflecting economy of the comprehensive energy system is established: the expected value of the full life cycle cost LCC comprises the sum of all costs weighted by probability under all typical scenes and discounted over the entire planning period, including an initial investment cost of equipment, a device replacement cost based on the SOH model, an annual operation and maintenance cost, a natural gas fuel cost, an electricity grid interaction cost considering time-of-use electricity prices and demand electricity charges, a carbon trading or carbon tax cost, and a V2G scheduling compensation cost paid to EV users; alternatively, a comprehensive resilience index reflecting resilience of the comprehensive energy system is established: the comprehensive resilience index of the comprehensive energy system is evaluated by nested Monte Carlo simulation: For each candidate capacity configuration on the Pareto front, a variety of pre-defined disturbance events are simulated, including long-term grid outage, sudden failure of key equipment, and continuous days of renewable energy output far below expectations under extreme weather; For each disturbance event, an emergency operation optimization sub-problem is solved to minimize the key load loss, and the key load satisfaction rate under this disturbance event is calculated; The final comprehensive resilience index is the probability-weighted average of the key load satisfaction rate under all simulated events.

5. The integrated energy system multi-objective collaborative optimization method of claim 1, wherein, A multi-objective two-stage stochastic programming optimization model is constructed by defining decision variables and taking economic efficiency, environmental protection and resilience of the integrated energy system as three-dimensional optimization objectives, specifically including: The decision variables are divided into first-stage variables and second-stage variables: The first stage variables include: device capacity configuration vector X , including photovoltaic , energy storage power , energy storage energy , heat pump Continuous variables and device siting, model; The second stage variables include: each scene s Each time t The operation strategy vector , the power purchase , the energy storage charging power , the boiler start-stop Each device output, energy storage charging and discharging power, and continuous variables such as power interaction with the grid and 0-1 integer variables of device start-stop state. The three-dimensional optimization objectives specifically include: Minimizing life cycle cost LCC expectation : ； In the formula, This represents the expected value of the total lifecycle cost. Cost over the entire lifecycle; X Configure a vector for equipment capacity, which serves as the decision variable for the first stage; The initial investment cost of the equipment. For the planning period, The operating strategy is the decision variable for the second stage. For the first y Expected annual operating costs The equipment replacement cost for year y; The discount rate; Indicates the first y year; This is a probability-weighted expectation operator for all typical scenarios; Minimizing annual average carbon emission expectations : ； wherein is the annual average carbon emission expectation, is the total annual carbon emission, is the scenario s is the t is the power purchased from the municipal grid at time t, is the t is the marginal carbon emission factor of the municipal grid at time t, is the scenario s is the t is the natural gas consumption of the gas boiler at time t, is the carbon emission factor of natural gas; Maximizing system-wide resilience index : ； In the formula, is the system comprehensive resilience index, is the extreme event probability weighted expectation operator, is the system resilience performance value under the extreme event e , is the extreme event index, is the occurrence probability of the extreme event e, is the critical load set, is the critical load index, is the weight coefficient of the first class critical load, is the actual supply amount of the first class critical load under the extreme event , is the total demand amount of the first class critical load under the extreme event .

6. The integrated energy system multi-objective collaborative optimization method of claim 4, wherein, An adaptive multi-objective evolutionary algorithm assisted by a surrogate model is used to solve the multi-objective two-stage stochastic programming optimization model, specifically including: An efficient solution algorithm framework is designed, and an improved NSGA-III with reference point mechanism is used as the main optimizer in the upper layer to guide the population to converge uniformly to the Pareto front; Based on the upper optimization architecture, initial sample points are generated in the capacity variable space X and high-fidelity data is obtained by full-scenario MIP real solution, and then three independent GPR surrogate models that can output target function prediction values and uncertainty variance are trained; In the NSGA-III iteration, a dynamic iterative adaptive strategy is constructed, which includes steps of predicting new candidate solutions by GPR surrogate model, screening key candidate solutions by EHVI value, performing MIP real solution on key candidate solutions and updating GPR model, to realize adaptive evaluation and model optimization in the lower layer; Through the alternative iteration of global search in the upper NSGA-III and adaptive evaluation in the lower EHVI, most of the fitness evaluations are completed by the low-cost surrogate model and the real solver is called only at key nodes, and when the convergence condition is met, the final Pareto optimal front is output; Wherein: The EHVI value is used to screen key candidate solutions, specifically including: For a candidate solution, calculate the expected increment of the hyper-volume formed by the current Pareto front that can be dominated and expanded under the probability distribution of the target function value predicted by the surrogate model; The candidate solution with the largest EHVI value is preferentially selected for accurate lower-layer solution, so as to achieve an efficient balance between exploring unknown high-uncertainty regions and utilizing known promising regions.

7. The integrated energy system multi-objective collaborative optimization method of claim 1, wherein, When the multi-criteria compromise solution ranking method is used to calculate the compromise value Q , a decision mechanism coefficient set by the decision maker is introduced v ∈[0,1], which reflects the decision maker's preference degree between maximizing group benefit and minimizing individual regret. v 0.5 means more inclined to the former, v <0.5 means more inclined to the latter, v =0.5 means neutral. Or, The high-fidelity dynamic behavior model library also includes multi-agent V2G aggregation model and heat pump COP model electric-hydrogen-electric chain model.

8. A multi-objective collaborative optimization system for integrated energy systems, characterized in that, Including: The scene generation module is configured to generate a multi-dimensional uncertainty time sequence scene of the integrated energy system based on a conditional generative adversarial network; The model establishment module is configured to establish a high-fidelity dynamic behavior model library including a BESS health state double-scale decay model; The optimization model construction module is configured to define decision variables, take economic efficiency, environmental protection and resilience of the integrated energy system as three-dimensional optimization objectives, and construct a multi-objective two-stage stochastic programming optimization model; The solving module is configured to solve the multi-objective two-stage stochastic programming optimization model by using an agent model assisted adaptive multi-objective evolutionary algorithm. The decision module is configured to realize hybrid multi-attribute decision-making based on a fuzzy analytic hierarchy process and a multi-criteria compromise solution ordering method according to the solving result, provide an alternative scheme and a performance interpretation report of the multi-objective collaborative optimization of the comprehensive energy system, and provide a final decision for a decision maker.

9. A computer-readable storage medium having stored thereon a program, characterized in that, The program is executed by the processor to implement the steps in the comprehensive energy system multi-objective collaborative optimization method according to any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps in the comprehensive energy system multi-objective collaborative optimization method according to any one of claims 1-7.

Images