A Multi-Objective Scheduling Method and System for All Renewable Energy Systems Based on PI-GAN
By constructing photovoltaic power output scenarios using a PI-GAN generator and discriminator, and combining scenario reduction and multi-objective scheduling models, the problems of photovoltaic power output uncertainty and the singleness of system optimization strategies are solved. This achieves balanced scheduling of economic and environmental benefits for the biomass-photovoltaic-hydrogen all-renewable energy system, and improves the stability and reliability of the system.
Patent Information
- Application Number
- CN202611427700.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-15
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies are unable to effectively handle the uncertainty of photovoltaic output, and traditional single optimization strategies cannot take into account the economic and environmental benefits of a fully renewable energy system of biomass-photovoltaic-hydrogen, leading to system instability and failure of optimization decisions.
A multi-objective day-ahead scheduling optimization method based on physical information generative adversarial networks (PI-GAN) is adopted. A photovoltaic power output scenario is constructed through a generator and a discriminator. Combined with a scenario reduction method, a multi-objective scheduling model is constructed. An improved non-dominated sorting genetic algorithm is used to obtain the Pareto optimal frontier, so as to achieve a balanced scheduling of economic and environmental benefits.
It significantly improves the physical reliability of photovoltaic power generation scenarios and the reliability of scheduling optimization, reduces system operating costs, reduces carbon emissions, and enhances system safety, stability, and the comprehensive utilization rate of renewable energy.
Smart Images

Figure CN123117698A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of integrated energy system optimization and scheduling technology, and particularly relates to a multi-objective day-ahead scheduling optimization method and system for a biomass-photovoltaic-hydrogen all-renewable energy system based on a Physics-informed Generative Adversarial Network (PI-GAN). Background Technology
[0002] Driven by the "dual carbon" goals and energy transition strategy, building a clean and low-carbon energy system is crucial. Biomass energy, with its dispatchable output and closed-loop carbon cycle, can serve as a stable baseload in multi-energy complementary systems; hydrogen energy, with its high energy density, is an effective carrier for large-scale renewable energy storage. Therefore, constructing a Biomass-Photovoltaic-Hydrogen Fully Renewable Energy System (BPH-FRES) is of great significance for solving the challenges of energy transition.
[0003] However, optimizing the operation of BPH-FRES still faces two key challenges:
[0004] First, the output of renewable energy sources such as photovoltaics is subject to significant uncertainty, becoming a major bottleneck restricting the safe, stable, and efficient operation of the system. Traditional statistical models struggle to accurately describe their high-dimensional, non-Gaussian, and nonlinear output characteristics. Although deep learning methods such as Generative Adversarial Networks (GANs) have become a research hotspot for uncertainty modeling, their purely data-driven modeling approach lacks physical constraints, easily leading to physical inconsistencies such as output exceeding limits and violations of the law of conservation of energy in generated scenarios, thus affecting the effectiveness of subsequent optimization decisions.
[0005] Second, the system involves highly coupled energy flows such as electricity, heat, and hydrogen, and complex dynamic interactions between devices. A single optimization strategy aimed solely at minimizing operating costs or carbon emissions cannot fully quantify its synergistic effects and potential conflicts. Existing research largely focuses on general-purpose integrated energy systems or specific device combinations; further research is needed on the multi-objective trade-offs and dynamic synergistic mechanisms unique to BPH-FRES.
[0006] Therefore, there is an urgent need for a method for scheduling and optimizing a fully renewable energy system that can effectively handle the uncertainty of photovoltaic output while taking into account multiple economic and environmental objectives. Summary of the Invention
[0007] The purpose of this invention is to provide a multi-objective day-ahead scheduling optimization method and system for all renewable energy systems based on physical information generative adversarial networks (PI-GAN) to solve the physical mismatch problem in the generation of uncertain photovoltaic power output scenarios, and to achieve day-ahead scheduling of biomass-photovoltaic-hydrogen all renewable energy systems that takes into account both economic and environmental benefits.
[0008] To achieve the above objectives, the present invention adopts the following technical solution: A multi-objective day-ahead scheduling optimization method for all renewable energy systems based on PI-GAN includes: S1. Construct a multi-objective day-ahead scheduling optimization model for a biomass-photovoltaic-hydrogen all-renewable energy system (BPH-FRES); S2. Generate photovoltaic power output scenarios using Physical Information Generative Adversarial Network (PI-GAN) and extract typical scenarios using scenario reduction methods; S3. Based on the day-ahead forecast data and the generated typical photovoltaic power output scenarios, the day-ahead scheduling of the biomass-photovoltaic-hydrogen all-renewable energy system is achieved by using the multi-objective day-ahead scheduling optimization model, taking into account both economic and environmental benefits.
[0009] Preferably, the Physical Information Generative Adversarial Network (PI-GAN) described in step S2 includes a generator and a discriminator; The generator takes a random noise vector and meteorological time series data as input and outputs a photovoltaic power output sequence. A differentiable physical constraint layer is introduced at the end of the generator. The photovoltaic power output sequence generated by the differentiable physical constraint layer based on the photovoltaic single diode physical model satisfies the equipment power output boundary. The discriminator adopts a one-dimensional convolutional neural network structure. The input is the generated photovoltaic power output sequence, and the output is a probability estimate of authenticity. The discriminator adds a physical consistency branch and calculates the residual between the generated sequence and the physical model as a physical regularization term.
[0010] Preferably, the mathematical expression of the photovoltaic single diode physical model is:
[0011] in, for Solar power output during specific time periods For photovoltaic conversion efficiency, For the area of the photovoltaic panel, for Time-of-use irradiance, For temperature coefficient, for Cell temperature during the period Standard test temperature; Preferably, the constraint expression of the differentiable physical constraint layer is:
[0012] This operation forces the generation of a photovoltaic power output sequence to strictly meet the equipment output boundary, while using gradient stopping technology to ensure the effectiveness of backpropagation.
[0013] Preferably, the total loss function of the Physical Information Generative Adversarial Network (PI-GAN) consists of an adversarial loss and a physical regularization term:
[0014] in, For the total loss function, The discriminator's expected output of the real data. Let $\mathbf{ ... For the generator based on noise The generated samples, The output probability of the discriminator. For physical constraint weights, The physical consistency residual is expressed as follows: .
[0015] Preferably, the scene reduction method in step S2 employs the K-means++ clustering algorithm, specifically including: First, the influence of dimensions is eliminated through z-score standardization:
[0016] in, For standardized photovoltaic power output scenarios, For original photovoltaic power generation scenarios, , Historical data in Mean and standard deviation for the time period; Secondly, the optimal number of clusters is determined based on a combination of the elbow rule and the profile coefficient. The K-means++ algorithm is used to initialize the cluster centers, the Euclidean distance is used to minimize the allocation scenario, and the cluster centers are iteratively updated until convergence. Finally, extract typical scene sets. and their weights :
[0017] in, For the first A typical scenario For the first Weights for a typical scenario For the first The center vector corresponding to a typical scenario For clusters Internal sample size This represents the total number of scenes generated.
[0018] Preferably, the multi-objective function of the multi-objective day-ahead scheduling optimization model for the biomass-photovoltaic-hydrogen all-renewable energy system is: Economic objective – Minimize the expected total operating cost of the system:
[0019] in, Total operating cost, for In the scene The cost of purchasing electricity from the power grid during a given time period. for In the scene Periodic fuel consumption costs for In the scene Equipment maintenance costs during specific time periods for In the scene Hydrogen sales revenue during the period This represents the total number of typical scenarios. This represents the total number of scheduling periods; Environmental objective – minimizing the expected total carbon emissions of the system:
[0020] in, For total carbon emissions, for In the scene Carbon emissions from electricity consumption during certain periods for In the scene Carbon emissions from biogas combustion during certain periods.
[0021] Preferably, the constraints of the multi-objective day-ahead scheduling optimization model for the biomass-photovoltaic-hydrogen all-renewable energy system include: power balance constraints, thermal balance constraints, hydrogen balance constraints, biogas balance constraints, and operating constraints of each device.
[0022] Preferably, in step S3, the improved non-dominated sorting genetic algorithm (NSGA-II) is used to solve the multi-objective day-ahead scheduling optimization model to obtain the Pareto optimal front, and a balanced scheduling strategy is selected based on the Pareto optimal front.
[0023] This invention further protects a multi-objective day-ahead scheduling optimization system for all renewable energy systems based on PI-GAN, including: The model building module is used to construct a multi-objective day-ahead scheduling optimization model for a biomass-photovoltaic-hydrogen all-renewable energy system. The scene generation module is used to generate photovoltaic power output scenes using physical information generative adversarial networks and to extract typical scenes using scene reduction methods. The scheduling optimization module is used to achieve day-ahead scheduling of the biomass-photovoltaic-hydrogen all-renewable energy system, taking into account both economic and environmental benefits, based on day-ahead forecast data and generated typical photovoltaic power output scenarios, through the multi-objective day-ahead scheduling optimization model.
[0024] The present invention further protects an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the multi-objective day-ahead scheduling optimization method as described in any one of claims 1 to 8.
[0025] The present invention further protects a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-objective day-ahead scheduling optimization method as described in any one of claims 1 to 8.
[0026] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention embeds the photovoltaic single diode physical model into the generator end to form a differentiable physical constraint layer, and adopts gradient stopping technology to force the generated photovoltaic power output sequence to meet the equipment power output boundary without affecting the backpropagation training. At the same time, a physical consistency branch is added to the discriminator, and the residual between the generated sequence and the physical model is used as the physical regularization term, thereby significantly suppressing the physical inconsistency problems such as power output over-limit and violation of energy conservation that are prone to occur in pure data-driven GANs, and improving the physical credibility and usability of the generated scene.
[0027] (2) This invention takes random noise vector and meteorological time series data as input, outputs photovoltaic power output sequence through generator, and uses a one-dimensional convolutional neural network to construct discriminator, which can effectively capture the high-dimensional, non-Gaussian and nonlinear time series characteristics of photovoltaic power output, generate a day-ahead photovoltaic power output scenario that is closer to the actual operating characteristics, and provide more reliable uncertainty input for subsequent scheduling optimization.
[0028] (3) This invention uses z-score standardization to eliminate the influence of dimensions, determines the optimal number of clusters based on the elbow rule and the contour coefficient, and uses the K-means++ algorithm to cluster and reduce the scene, extract the typical scene set and its weight, effectively compress the number of scenes while retaining the statistical characteristics of photovoltaic uncertainty, reduce the solution scale of the multi-objective day-ahead scheduling model, and improve the computational efficiency.
[0029] (4) This invention constructs a multi-objective day-ahead scheduling optimization model for a biomass-photovoltaic-hydrogen all-renewable energy system. The model aims to minimize the expected value of the total operating cost and the expected value of the total carbon emissions. It comprehensively considers factors such as grid electricity purchase cost, fuel consumption cost, equipment operation and maintenance cost, hydrogen sales revenue, grid electricity carbon emissions and biogas combustion carbon emissions. It can quantify the synergistic effect and objective conflict under the coupling of multiple energy flows such as electricity, heat, hydrogen and biogas, and avoid the one-sidedness brought about by single objective optimization.
[0030] (5) The present invention uses an improved non-dominated sorting genetic algorithm to solve the multi-objective day-ahead scheduling optimization model and obtain the Pareto optimal frontier, so that decision-makers can select a balanced scheduling strategy that takes into account both economic and environmental benefits from the Pareto frontier according to actual operating preferences, thereby improving the flexibility and engineering practicality of scheduling decisions.
[0031] (6) This invention optimizes scheduling based on day-ahead forecast data and generated typical photovoltaic power output scenarios, and introduces power balance constraints, thermal balance constraints, hydrogen balance constraints, biogas balance constraints and equipment operation constraints into the model. It can effectively cope with the uncertainty of photovoltaic power output, improve the safe and stable operation level of the system, and promote photovoltaic consumption and hydrogen energy storage and utilization.
[0032] (7) This invention utilizes the dispatchability of biomass energy output and the high energy density storage characteristics of hydrogen energy, combined with photovoltaic power output scenario generation and multi-objective scheduling optimization, which is conducive to reducing system operating costs, reducing carbon emissions, and improving the comprehensive utilization rate of renewable energy, providing a feasible and scalable technical solution for day-ahead scheduling of all renewable energy systems.
[0033] In summary, this invention has significant technical advantages in physical consistency scenario generation, uncertainty modeling, scenario reduction, multi-objective collaborative scheduling, and Pareto optimal decision-making. It can achieve day-ahead scheduling optimization of biomass-photovoltaic-hydrogen all-renewable energy systems that takes into account economic, environmental, and reliable operation. Attached Figure Description
[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0035] Figure 1 This is a schematic diagram of the structure of the biomass-photovoltaic-hydrogen all-renewable energy system (BPH-FRES) proposed in Embodiment 1 of the present invention; Figure 2This is a schematic diagram illustrating the entire process of uncertainty characterization in the Physical Information Generative Adversarial Network (PI-GAN) proposed in Embodiment 1 of the present invention. Detailed Implementation
[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0037] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0038] Example 1: This embodiment proposes a multi-objective scheduling method and system for a fully renewable energy system based on PI-GAN, specifically including the following contents.
[0039] 1. System Structure and Parameter Settings This embodiment selects a 5×10 area in northern China. 4 m 2 The typical biomass-photovoltaic-hydrogen all-renewable energy system (BPH-FRES) is used as the research object. The system structure is as follows: Figure 1 As shown, the system is powered by a combination of photovoltaic, biomass energy, and the upstream power grid. Key equipment includes electrolyzers, biogas digesters, biogas boilers, biogas generator sets, hydrogen fuel cells, and ground source heat pumps.
[0040] The core equipment capacity configuration includes: 800 kW photovoltaic power, a biogas digester with a daily processing capacity of 5 t straw, and a 500 kg hydrogen storage tank. The study uses a 24-hour scheduling cycle and optimizes the calculations with a 1-hour simulation step.
[0041] The relevant equipment parameters of the system are shown in Table 1.
[0042] Table 1 Key Equipment Parameters of the System
[0043] Electricity is priced on a time-of-use basis: 0.85 yuan / kWh during peak hours (8:00-11:00, 17:00-21:00); 0.52 yuan / kWh during normal hours (6:00-8:00, 11:00-17:00, 21:00-22:00); and 0.35 yuan / kWh during off-peak hours (22:00-6:00 the next day). Hydrogen is priced at 5.2 yuan / kg, and straw is priced at 350 yuan / t.
[0044] 2. Generation of photovoltaic power output scenarios Based on the constructed PI-GAN framework, 1000 initial photovoltaic power output scenarios were first generated. The generator consisted of 8 convolutional layers with a hidden layer dimension of 128 and a kernel size of 5; the discriminator comprised 3 convolutional blocks with kernel sizes of [8, 5, 3]. The training process involved 10,000 iterations, with a batch size of 64 and a learning rate of 2×10⁻⁶. -4 The physical regularization weight λ = 0.5.
[0045] Further, the improved K-means++ algorithm was used to perform cluster analysis on the 1000 groups of photovoltaic power output scenarios generated, and finally five typical scenarios with significant differences were extracted. Their distribution characteristics are shown in Table 2.
[0046] Table 2. Distribution characteristics of typical scenarios
[0047] Five typical scenarios comprehensively cover the key characteristics of photovoltaic power output under different weather conditions. Their probability distribution reflects the statistical regularity of historical data and will serve as the basic input for multi-scenario stochastic optimization.
[0048] To verify the advantages of the PI-GAN method, a comparative analysis was conducted from three dimensions: physical consistency, statistical distribution fidelity, and ability to characterize extreme scenes. (1) Physical constraint verification. The comparison of the three methods on key physical indicators is shown in Table 3.
[0049] Table 3 Comparison of key physical indicators of the three methods
[0050] By introducing a differentiable physical constraint layer, PI-GAN achieved a zero output over-limit rate. The mean absolute residual was reduced by 91.8% and 86.9% compared to traditional GAN and Monte Carlo simulation, respectively. The temperature-output correlation coefficient reached 0.89, which is significantly better than the comparison methods.
[0051] (2) Statistical property fidelity assessment. The assessment results are shown in Table 4.
[0052] Table 4 Statistical property fidelity evaluation indicators
[0053] The KL divergence of PI-GAN is only 0.15, and the F norm error of the autocorrelation matrix is reduced by 64.2% and 42.8% compared with traditional GAN and MCS, respectively, demonstrating the best distribution consistency and temporal structure preservation capabilities.
[0054] (3) Extreme scene characterization capability. In the extreme scenario where the irradiance drops sharply from 98.3 kW to 0 kW, PI-GAN can accurately generate a physically believable scene where the output force instantly returns to zero, while traditional GAN and MCS produce distortion phenomena such as non-physical oscillation and linear decay, respectively.
[0055] 3. Multi-objective optimization scheduling results To address the high-dimensional and nonlinear characteristics of the model, an improved NSGA-II algorithm is used for solving the problem. The population size is 200, the maximum number of iterations is 300, the crossover probability is 0.9, and the mutation probability is 0.1.
[0056] (1) Pareto front analysis.
[0057] The optimized non-dominated solution set constitutes a two-dimensional Pareto front. The results show that system operating cost and carbon emissions constitute a set of conflicting bi-objectives. As carbon emissions increase, system operating cost gradually decreases. Based on this front characteristic, scheduling strategies can be categorized into three types: economy-first strategy, low-carbon-first strategy, and balanced scheduling strategy. This embodiment selects the balanced scheduling strategy for result analysis.
[0058] (2) Analysis of the results of the balanced scheduling strategy.
[0059] Based on the Pareto front selection equilibrium scheduling strategy, the system's coordinated scheduling results for multiple energy flows (electricity, heat, and hydrogen) during a typical day are as follows: Regarding load balancing: During peak photovoltaic output periods, the electrolyzer's power consumption significantly increases to 161.3 kW, converting surplus electricity into hydrogen energy; the biogas generator's output decreases accordingly, and grid power purchases are kept to a minimum. At night, electricity demand is mainly met by relying on the biogas generator and grid power purchases.
[0060] Regarding heat load balance: During the day, the system makes full use of waste heat recovery, accounting for 42%, and the heat storage tank releases heat actively, reducing the heating demand of the biogas boiler by about 31%; at night, the biogas boiler undertakes the main heating task, ensuring heating stability.
[0061] Regarding hydrogen energy balance: the peak hydrogen production rate of the daytime electrolyzer reaches 118.9 kg / h, the peak utilization rate of the hydrogen storage tank is about 72%, and the generated hydrogen energy is mostly sold during peak electricity price periods to realize arbitrage profits.
[0062] Regarding biogas balance: the biogas consumption of the biogas generator set changes in tandem with the electricity price during different time periods, the biogas boiler's gas consumption responds dynamically to the heat load demand, and the overall biomass energy utilization rate of the system is optimized to 45.3%.
[0063] (3) Comparative analysis of multi-objective strategies.
[0064] To verify the necessity of multi-objective collaborative optimization, three comparative scenarios were set up: Case 1 considering only the economic objective, Case 2 considering only the low-carbon objective, and Case 3 considering both economic and low-carbon objectives (equilibrium strategy). The results are shown in Table 5.
[0065] Table 5 Comparison of results for different optimization strategies
[0066] The economic priority strategy has the lowest daily operating cost (8322.3 yuan), but its carbon emissions are as high as 1243.6 kg CO2; the low-carbon priority strategy has the lowest carbon emissions (822.3 kg CO2), but its operating cost rises to 9561.8 yuan, and its load failure rate reaches 6.8%. The equilibrium strategy reduces the operating cost by 6.3% compared to the low-carbon strategy, reduces carbon emissions by 20.8% compared to the economic strategy, and significantly reduces the load failure rate to 0.5%, verifying the necessity of multi-objective synergistic optimization.
[0067] (4) Comparison of scheduling performance of different scene generation methods.
[0068] Table 6 shows the scheduling optimization results based on different scene generation methods when using the same improved NSGA-II algorithm.
[0069] Table 6. Scheduling optimization results for different scene generation methods
[0070] The daily operating cost of the PI-GAN-based system is reduced by RMB 366.4 and RMB 217.3 compared to traditional GANs and MCS, respectively. Carbon emissions are reduced by 16.6% and 6.5%, respectively. The load failure rate is reduced by 92.6% and 88.9%, respectively, and the energy storage utilization rate is increased by 18% and 5.8%, respectively. At the same time, the improved NSGA-II algorithm has a faster convergence speed, and the solution time is shortened by 36.3% and 21.9%, respectively.
[0071] 4. Sensitivity analysis of exogenous factors To investigate the impact of key exogenous parameters on the system's multi-objective scheduling strategy, tests were conducted on electricity prices, hydrogen prices, and biogas costs with fluctuations of ±20% based on a baseline scenario. Regarding electricity prices: A 20% increase in electricity prices led to a 15.3% increase in operating costs, but biomass energy played a stronger substitution effect, resulting in a 9.6% decrease in carbon emissions; a 20% decrease in electricity prices stimulated a 47.5% increase in grid-purchased electricity demand, resulting in a 13.4% increase in carbon emissions.
[0072] Regarding hydrogen prices: A 20% increase in hydrogen prices boosted the utilization rate of electrolyzers, reduced total costs by 8.9%, and decreased carbon emissions by 4.1%; when hydrogen prices fell to 4.16 yuan / kg, the utilization rate of electrolyzers dropped sharply by 48%, and operating costs increased by 14.2%.
[0073] Regarding biogas costs: a 20% increase in biogas costs pushes up operating costs by 18.4%, but the suppression of high-carbon energy use results in an 11.6% carbon emission reduction effect; when costs decrease by 20%, operating costs decrease by 15.1%, but carbon emissions increase by 9.4%.
[0074] The results reveal the nonlinear characteristics of parameter sensitivity in multi-energy flow coupled systems, providing an important decision-making basis for the dynamic optimization of high-proportion renewable energy systems.
[0075] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims
1. A multi-objective day-ahead scheduling optimization method for all renewable energy systems based on PI-GAN, characterized in that, include: S1. Construct a multi-objective day-ahead scheduling optimization model for a biomass-photovoltaic-hydrogen all-renewable energy system; S2. Generate photovoltaic power output scenarios using physical information generative adversarial networks, and extract typical scenarios using scenario reduction methods; S3. Based on the day-ahead forecast data and the generated typical photovoltaic power output scenarios, the day-ahead scheduling of the biomass-photovoltaic-hydrogen all-renewable energy system is achieved by using the multi-objective day-ahead scheduling optimization model, taking into account both economic and environmental benefits.
2. The multi-objective day-ahead scheduling optimization method as described in claim 1, characterized in that, The physical information generative adversarial network described in step S2 includes a generator and a discriminator; The generator takes a random noise vector and meteorological time series data as input and outputs a photovoltaic power output sequence. A differentiable physical constraint layer is introduced at the end of the generator. The photovoltaic power output sequence generated by the differentiable physical constraint layer based on the photovoltaic single diode physical model satisfies the equipment power output boundary. The discriminator adopts a one-dimensional convolutional neural network structure. The input is the generated photovoltaic power output sequence, and the output is a probability estimate of authenticity. The discriminator adds a physical consistency branch and calculates the residual between the generated sequence and the physical model as a physical regularization term.
3. The multi-objective day-ahead scheduling optimization method as described in claim 2, characterized in that, The mathematical expression for the physical model of the photovoltaic single diode is: in, for Solar power output during specific time periods For photovoltaic conversion efficiency, For the area of the photovoltaic panel, for Time-of-use irradiance, For temperature coefficient, for Cell temperature during the period This is the standard test temperature.
4. The multi-objective day-ahead scheduling optimization method as described in claim 2, characterized in that, The constraint expression for the differentiable physical constraint layer is: This operation forces the generation of a photovoltaic power output sequence to strictly meet the equipment output boundary, while using gradient stopping technology to ensure the effectiveness of backpropagation.
5. The multi-objective day-ahead scheduling optimization method as described in claim 2, characterized in that, The total loss function of the physical information generative adversarial network consists of an adversarial loss and a physical regularization term: in, For the total loss function, The discriminator's expected output of the real data. Let $\mathbf{ ... For the generator based on noise The generated samples, The output probability of the discriminator. For physical constraint weights, The physical consistency residual is expressed as follows: 。 6. The multi-objective day-ahead scheduling optimization method as described in claim 1, characterized in that, The scene reduction method described in step S2 employs the K-means++ clustering algorithm, specifically including: First, the influence of dimensions is eliminated through z-score standardization: in, For standardized photovoltaic power output scenarios, For original photovoltaic power generation scenarios, , Historical data in Mean and standard deviation for the time period; Secondly, the optimal number of clusters is determined based on a combination of the elbow rule and the profile coefficient. The K-means++ algorithm is used to initialize the cluster centers, the Euclidean distance is used to minimize the allocation scenario, and the cluster centers are iteratively updated until convergence. Finally, extract typical scene sets. and their weights : in, For the first A typical scenario For the first Weights for a typical scenario For the first The center vector corresponding to a typical scenario For clusters Internal sample size This represents the total number of scenes generated.
7. The multi-objective day-ahead scheduling optimization method as described in claim 1, characterized in that, The multi-objective function of the multi-objective daytime scheduling optimization model for the biomass-photovoltaic-hydrogen all-renewable energy system is: Economic objective – Minimize the expected total operating cost of the system: in, Total operating cost, for In the scene The cost of purchasing electricity from the power grid during a given time period. for In the scene Periodic fuel consumption costs for In the scene Equipment maintenance costs during specific time periods for In the scene Hydrogen sales revenue during the period This represents the total number of typical scenarios. This represents the total number of scheduling periods; Environmental objective – minimizing the expected total carbon emissions of the system: in, For total carbon emissions, for In the scene Carbon emissions from electricity consumption during certain periods for In the scene Carbon emissions from biogas combustion during certain periods.
8. The multi-objective day-ahead scheduling optimization method as described in claim 1, characterized in that, The constraints of the multi-objective day-ahead scheduling optimization model for the biomass-photovoltaic-hydrogen all-renewable energy system include: power balance constraints, thermal balance constraints, hydrogen balance constraints, biogas balance constraints, and operational constraints of each device.
9. The multi-objective day-ahead scheduling optimization method as described in claim 1, characterized in that, In step S3, an improved non-dominated sorting genetic algorithm is used to solve the multi-objective day-ahead scheduling optimization model to obtain the Pareto optimal front, and a balanced scheduling strategy is selected based on the Pareto optimal front.
10. A PI-GAN-based multi-objective day-ahead scheduling optimization system for all renewable energy systems, used to implement the method of any one of claims 1-9, characterized in that, include: The model building module is used to construct a multi-objective day-ahead scheduling optimization model for a biomass-photovoltaic-hydrogen all-renewable energy system; The scene generation module is used to generate photovoltaic power output scenes using physical information generative adversarial networks and to extract typical scenes using scene reduction methods. The scheduling optimization module is used to achieve day-ahead scheduling of the biomass-photovoltaic-hydrogen all-renewable energy system, taking into account both economic and environmental benefits, based on day-ahead forecast data and generated typical photovoltaic power output scenarios, through the multi-objective day-ahead scheduling optimization model.
