Machine learning assisted renewable energy penetration and component size three-level optimization system and method

Abstract

The application belongs to the technical field of renewable energy fixed capacity system planning and optimization, and discloses a machine learning assisted renewable energy penetration and component size three-level optimization system and method. The system comprises an input data module, a machine learning prediction layer, a penetration optimization layer, a component size optimization layer and a coordination control module. By inputting load curves, renewable resource data, cost parameters and reliability indicators, a machine learning model trained by historical simulation and optimization data is used to predict the feasible penetration interval and component capacity interval, and the optimization search space is intelligently reduced; the optimal renewable energy penetration rate is determined within the predicted interval, and the photovoltaic, wind power, battery energy storage and hydrogen energy storage capacity is optimized under the constraint, so as to minimize the horizontal energy cost and meet the unsatisfied load fraction and energy balance constraint conditions. The application reduces the calculation complexity, improves the optimization convergence speed and system planning accuracy, and is suitable for engineering deployment in different geographical regions.

Description

Technical Field

[0001] This application relates to the field of renewable energy fixed capacity system planning and optimization, and in particular to a machine learning-assisted three-level optimization system and method for renewable energy penetration and component size. Background Technology

[0002] The current technological landscape indicates that the global energy system is undergoing a profound transformation. Continued population growth, accelerated industrialization and urbanization, widespread electrification in transportation and buildings, and the rapid development of digital technologies such as data centers, artificial intelligence, and high-performance computing have all significantly driven the sustained rise in electricity demand. While energy demand is increasing, climate change is becoming increasingly severe, with countries successively proposing carbon peaking and carbon neutrality targets, and international climate agreements imposing stricter constraints on greenhouse gas emissions. Against this backdrop, renewable energy sources, represented by solar and wind power, are gradually becoming the core pathway for energy structure adjustment and the low-carbon transformation of the power system.

[0003] The large-scale integration of distributed energy resources (DERs) has become an important trend in power system development. Among them, solar photovoltaic (SPV) systems and wind power systems have been widely adopted globally due to their advantages such as modular design, flexible installation, short construction period, and rapidly decreasing costs. However, compared with traditional dispatchable power sources such as thermal power and hydropower, renewable energy power generation has significant intermittency, randomness, and weather dependence. Solar irradiance fluctuates significantly with diurnal and weather changes, and wind speed has high uncertainty, resulting in non-stationary and nonlinear characteristics in output power.

[0004] When a large amount of renewable energy is directly connected to the distribution network, it may trigger a series of technical problems, such as voltage fluctuations, frequency deviations, power quality degradation, reverse power flow, relay protection malfunctions, decreased system inertia, and insufficient short-circuit capacity. These problems are more pronounced in weak grids, rural grids, or isolated microgrids. As the penetration rate of renewable energy increases, the safety margin of grid operation gradually decreases, posing a severe challenge to system stability and reliability.

[0005] To address the instability issues arising from the large-scale integration of renewable energy, Renewable Firm Capacity (RFC) systems have gradually become an important direction for research and engineering applications. RFC systems combine renewable energy generation units with energy storage systems or auxiliary peak-shaving power sources to achieve smooth power output and dispatchable operation amidst fluctuations. By rationally configuring energy storage capacity and control strategies, the inherent volatility of renewable energy can be converted into stable, continuous, and controllable equivalent power generation capacity, thereby improving grid resilience and enhancing the system's resistance to disturbances.

[0006] In the RFC system, energy storage technology plays a crucial supporting role. Battery storage systems (BSS), especially lithium-ion batteries, offer advantages such as fast response, high control precision, and high power density, and are widely used in peak shaving, frequency regulation, and short-term balance control. On the other hand, hydrogen storage systems (HSS) convert excess electrical energy into hydrogen through water electrolysis for storage. During peak load periods or energy shortages, they generate electricity through fuel cells, making them suitable for medium- to long-term or seasonal energy storage. The complementary nature of BSS and HSS gives them a significant synergistic advantage in energy balance across multiple time scales.

[0007] Therefore, in the RFC system composed of solar energy, wind energy and multiple types of energy storage technologies, how to achieve a reasonable renewable energy penetration design and component capacity configuration has become a key technical issue in achieving a balance between system economy and reliability.

[0008] Existing technologies face the following technical problems and limitations: Despite continuous advancements in renewable energy and energy storage technologies, the planning and optimization of RFC systems still present complex challenges. The core issue in system planning lies in determining the optimal renewable energy penetration rate and the capacity scale of various power generation and energy storage components, given load demand, resource conditions, and economic parameter constraints, so that the system can operate at the lowest cost while meeting reliability constraints.

[0009] First, this problem is highly nonlinear, involves multivariate coupling, and is subject to multiple constraints. Increased renewable energy penetration typically reduces fuel costs but increases energy storage demand and system scheduling complexity; increased module capacity may improve reliability but also significantly increases initial investment and maintenance costs. Therefore, there is a strong coupling relationship between penetration rate and module size, necessitating comprehensive system-level optimization.

[0010] In existing technologies, RFC system planning typically employs heuristic algorithms, metaheuristic algorithms (such as genetic algorithms, particle swarm optimization, differential evolution, etc.), or traditional optimization methods based on iterative search. These methods rely on extensive time-series simulation calculations, repeatedly searching and evaluating within a high-dimensional solution space. Due to the vast solution space containing numerous infeasible regions, the algorithms often require a large number of iterations to converge, resulting in heavy computational burden, long optimization time, and low efficiency in engineering applications.

[0011] Secondly, existing methods typically treat "determining the renewable energy penetration rate" and "optimizing module capacity" as independent steps. That is, a certain penetration rate is assumed first, and then the module capacity is optimized under that condition, or vice versa. This weakly coupled or separate optimization approach easily leads to local optima and cannot guarantee global synergistic optimization. If the penetration rate is set too high, it may result in excessive investment in energy storage and increased system costs; if the penetration rate is set too low, it may fail to fully utilize local renewable resources, reducing the system's economic efficiency.

[0012] Furthermore, traditional optimization methods lack effective utilization of historical planning data. Each planning iteration typically begins with a fixed or empirically defined search interval, failing to leverage existing optimization results to guide future planning. This "memoryless" optimization approach ignores the value of data, preventing the system from achieving continuous learning and adaptive improvement.

[0013] Furthermore, in terms of search space initialization, existing methods often use a wide or empirical range as the search boundary, resulting in a significant waste of computational resources on obviously infeasible or suboptimal regions. This not only increases computation time but may also affect the stability and robustness of the algorithm. Traditional methods exhibit poor adaptability when facing different geographical regions, different load characteristics, and different cost structures, requiring the resetting of parameters and search range, thus lacking practical engineering applicability.

[0014] In practical applications, as power grids expand and system complexity increases, traditional single-layer optimization methods struggle to meet the demands for rapid planning and real-time decision-making. This is particularly true in regional energy planning or distributed energy cluster configurations, where computational efficiency and planning accuracy become critical factors hindering technology adoption.

[0015] In summary, existing RFC system planning technologies have the following limitations: lack of system-level collaborative mechanisms for penetration rate and component capacity optimization; large search space, high computational complexity, and low convergence efficiency; inability to utilize historical data for intelligent prediction and search space reduction; lack of adaptive learning capabilities in the planning process; and insufficient scalability and adaptability under different regional conditions. Summary of the Invention

[0016] To address the aforementioned shortcomings in existing technologies, this application provides a machine learning-assisted three-level optimization system and method for renewable energy penetration and component size.

[0017] The machine learning-assisted three-level optimization system for renewable energy penetration and component size provided in this application adopts the following technical solution: A machine learning-assisted three-level optimization system for renewable energy penetration and component size. include: The input data module is used to collect or import engineering parameter data of renewable energy fixed capacity systems. The engineering parameter data includes at least load demand curves, renewable resource availability data, component cost parameters, and reliability constraint parameters. The machine learning prediction layer is connected to the input data module and is used to perform predictive analysis on the engineering parameter data based on the machine learning model trained on historical simulation or optimization data, and output the feasible range of renewable energy penetration rate and the predicted range of system component capacity. A penetration optimization layer, which is connected to the machine learning prediction layer, is used to perform optimization calculations within the feasible range of the renewable energy penetration rate to determine the optimal renewable energy penetration rate that meets the techno-economic constraints. A component size optimization layer, which is connected to the penetration optimization layer, is used to perform capacity optimization calculations within the system component capacity prediction range under the constraint of the optimal renewable energy penetration rate, and to determine the optimal installed capacity of renewable energy power generation components and energy storage components. The coordination and control module is used to establish a data linkage and constraint transmission mechanism among the machine learning prediction layer, the penetration optimization layer and the component size optimization layer, so as to realize the hierarchical optimization and collaborative operation. The optimization objective of the component size optimization layer is to minimize the system levelized cost of energy (LCOE) while satisfying reliability constraints and energy balance constraints, and to output system configuration parameters including the proportion of renewable energy, the installed capacity of each component, dispatchable power generation capacity, reliability indicators, and economic indicators.

[0018] By adopting the above technical solutions, an intelligent system-level planning framework with data-driven approach and hierarchical optimization as its path is constructed. This framework achieves coordinated optimization of renewable energy penetration rate and component capacity configuration. The machine learning prediction layer uses artificial neural networks to train on historical simulation and optimization data, extracting nonlinear mapping relationships from multi-dimensional engineering parameters such as load curve characteristics, solar irradiance, wind speed, component cost, and reliability thresholds. This allows for rapid prediction of reasonable penetration rate ranges and capacity ranges, effectively narrowing the solution space before optimization. Secondly, the penetration optimization layer performs constrained searches within the prediction interval, avoiding invalid traversals of obviously infeasible or inefficient areas. This improves convergence speed and stability; the component size optimization layer performs refined optimization of photovoltaic, wind power, and energy storage capacity under optimal penetration constraints, enabling the system to minimize horizontal energy costs while meeting constraints such as unmet load fraction, energy balance, and energy storage operation boundaries; the coordination control module enables data linkage and dynamic constraint transmission between layers, enhancing the overall consistency and robustness of the system; it effectively reduces computational complexity, improves optimization efficiency and planning accuracy, and enhances the system's adaptability to different regional resource conditions and load structures, providing an efficient, reliable, and scalable intelligent solution for the engineering design of renewable energy fixed capacity systems.

[0019] Optionally, the machine learning prediction layer employs an artificial neural network model, which takes load curve features, solar irradiance data, wind speed data, component unit cost, and reliability threshold as input features.

[0020] By adopting the above technical solutions and training and mapping multidimensional engineering features using artificial neural network models, the nonlinear relationship between load characteristics, resource fluctuations and economic parameters can be automatically mined, the feasible range of renewable energy penetration rate and component capacity can be accurately predicted, the optimization search space can be intelligently reduced, and the planning accuracy and calculation efficiency can be improved.

[0021] Optionally, the prediction results output by the machine learning prediction layer include upper and lower limits of renewable energy penetration, photovoltaic capacity range, battery energy storage capacity range, and hydrogen energy storage capacity range.

[0022] By adopting the above technical solutions, clear boundaries and initialization basis can be provided for subsequent optimization, effectively limiting the search range, reducing invalid calculations, improving optimization convergence speed and result stability, and enhancing the pertinence and engineering feasibility of system planning.

[0023] Optionally, the reliability constraint is limited by the unmet load fraction (UF), satisfying the following relationship: UF ≤ UFmax, Where UF is the ratio of unmet load energy to total load demand, and UFmax is the preset maximum allowable threshold.

[0024] By adopting the above technical solution, using the unmet load fraction UF as a reliability constraint index and limiting UF≤UFmax, the system power supply reliability level can be quantified, ensuring that the power generation and energy storage configuration meets the load demand during the optimization process, preventing power supply shortage due to excessive capacity compression, and improving system operation safety and engineering feasibility.

[0025] Optionally, the energy balance constraint satisfies the following condition: within any time step, the sum of renewable energy power generation, energy storage discharge, and external grid support power is greater than or equal to the load demand; The energy storage system includes a battery energy storage system and a hydrogen energy storage system, wherein: Battery energy storage systems satisfy the state of charge constraint: SOCmin ≤ SOC(t) ≤ SOCmax; The hydrogen energy storage system satisfies the hydrogen storage state constraint: SOHmin ≤ SOH(t) ≤ SOHmax; It also includes a feedback learning module, which is used to retrain or update the parameters of the machine learning model based on the optimization results of the penetration optimization layer and the component size optimization layer, so as to achieve adaptive learning; The optimization process is performed within the constrained search space generated by predictive decision support information to avoid exhaustive searches of infeasible or suboptimal regions. The machine learning model is any one or a combination of support vector machine, decision tree, random forest, deep learning model or hybrid statistical learning model; The system is implemented with a unified software architecture and deployed on a local computing platform, cloud computing platform, or real-time energy management system.

[0026] By adopting the above technical solutions and setting energy balance constraints, the sum of power generation and energy storage output is ensured to meet load demand at any given time step, guaranteeing the continuity and stability of system power supply from the operational mechanism level. Simultaneously, upper and lower limits are imposed on the State of Charge (SOC) of the battery energy storage system and the State of Health (SOH) of the hydrogen energy storage system, preventing overcharging, over-discharging, or excessive consumption of energy storage resources, thus improving equipment safety and lifespan reliability. Based on this, a feedback learning module is introduced to retrain or update the machine learning model based on the results of penetration optimization and capacity optimization, enabling the system to continuously learn and dynamically adapt, and continuously improve prediction accuracy with data accumulation. By performing the optimization process within the constrained search space generated by predictive decision support information, invalid traversal of infeasible or inefficient regions is effectively avoided, significantly reducing computational complexity and improving convergence efficiency. The system supports the substitution and combination of various machine learning model structures, improving algorithm flexibility and adaptability. Deploying the system in a unified software architecture on a local or cloud platform enhances the scalability of engineering implementation and its practical application value.

[0027] A machine learning-assisted three-level optimization method for renewable energy penetration and component size. Includes the following steps: Step 1: Collect or import load demand data, renewable resource data, component cost parameters, and reliability requirements of renewable energy fixed capacity systems; Step 2: Based on historical simulation and optimization results, construct and train a machine learning model, using the engineering parameters from Step 1 as input, to predict the feasible range of renewable energy penetration rate and the predicted range of system component capacity. Step 3: Perform penetration optimization calculations within the feasible penetration range to determine the optimal renewable energy penetration rate by satisfying the technical and economic constraints; Step 4: Under the constraint of the optimal renewable energy penetration rate, perform component capacity optimization calculation within the component capacity prediction range to minimize the system levelized cost of energy (LCOE). Step 5: Output the optimal renewable energy ratio, photovoltaic system capacity, wind power generation capacity, battery energy storage capacity, hydrogen energy storage capacity, system reliability indicators, and economic indicators; In step two, the search space is reduced by predicting constraints, which reduces the computational complexity of steps three and four and improves the optimization convergence efficiency.

[0028] By adopting the above technical solution, a machine learning-assisted three-level optimization method is constructed. First, the model is trained using historical simulation and optimization data to predict the renewable energy penetration rate and component capacity range, achieving pre-screening and intelligent reduction of the traditional large-scale search space, reducing ineffective calculation areas from the source. Based on this, penetration optimization calculations are performed within the predicted feasible penetration range to achieve optimal matching of the renewable energy share while meeting technical constraints and economic requirements, avoiding excessive penetration leading to a surge in energy storage and reserve capacity, or excessively low penetration resulting in insufficient utilization of clean energy. Subsequently, component capacity optimization is performed under the constraint of optimal penetration rate, aiming to minimize the horizontal energy cost of the system while considering reliability indicators and operational constraints, achieving coordinated configuration of generation and storage capacity. This hierarchical and progressive optimization process significantly reduces computational complexity, improves algorithm convergence speed and stability, and enhances the engineering feasibility and economic rationality of the optimization results. The final output system configuration parameters cover energy share, installed capacity, reliability, and economic indicators, providing systematic, quantifiable, and high-precision decision support for planning, design, and investment decisions, enabling efficient planning and intelligent design of renewable energy fixed capacity systems.

[0029] Optionally, the penetration optimization layer employs an improved binary search algorithm to optimize the penetration rate of renewable energy.

[0030] By adopting the above technical solution and using an improved binary search algorithm to perform interval iterative optimization of renewable energy penetration rate, the optimal solution can be quickly approximated within the defined feasible range, reducing the number of invalid searches, improving convergence speed and computational efficiency, and enhancing the stability and accuracy of the optimization process, making it suitable for large-scale engineering applications.

[0031] Optionally, the component size optimization layer uses an enhanced particle swarm optimization algorithm or other metaheuristic algorithms to perform capacity optimization calculations.

[0032] By adopting the above technical solutions and using enhanced particle swarm optimization algorithms or other metaheuristic algorithms for capacity optimization, the global search capability and the ability to escape local optima can be improved, enabling efficient solutions to multivariable and multi-constraint nonlinear problems, and enhancing the accuracy, stability, and economic optimality of capacity configuration results.

[0033] Optionally, the optimal renewable energy penetration rate is a region-specific optimal value determined independently for a specific geographical area, used to avoid excessive or insufficient penetration of renewable energy resources.

[0034] By adopting the above technical solutions, the optimal renewable energy penetration rate is set as a regionally specific optimal value determined independently for a specific geographical area. Differentiated configurations can be made based on local resource endowment, load characteristics, grid structure, and absorption capacity, avoiding resource waste or insufficient installed capacity caused by uniform standards. This improves the economic efficiency and stability of system operation, enhances the utilization efficiency of renewable energy, and ensures the optimization of regional energy structure and the safe and reliable operation of the power system.

[0035] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method as described in claim 6.

[0036] By adopting the above technical solution, the three-level optimization method is stored in a computer-readable storage medium in the form of a computer program. When the processor executes the program, it can automatically complete the entire process of calculation, including data acquisition, machine learning prediction, penetration rate optimization, and component capacity optimization. This achieves software-based and automated operation of the method steps, encapsulates complex system planning algorithms into standardized program modules, improves the flexibility and portability of system deployment, and facilitates rapid integration and invocation in different hardware platforms, energy management systems, or cloud environments. At the same time, it helps reduce human intervention and human error, improves the consistency and repeatability of calculation results, and through programmatic implementation, it can also support subsequent functional expansion and model upgrades, enhance the system's continuous optimization and iteration capabilities, and provide stable, efficient, and scalable technical support for the engineering design and operation decision-making of renewable energy fixed capacity systems.

[0037] In summary, this application includes at least one of the following beneficial technical effects: By introducing a machine learning prediction mechanism, the penetration rate of renewable energy and the capacity range of components are predicted in advance and the constraints are reduced, which effectively compresses the search space of traditional optimization algorithms, significantly reduces computational complexity, and improves optimization convergence speed and solution efficiency.

[0038] A three-tiered, hierarchical collaborative architecture of "prediction-penetration optimization-capacity optimization" is constructed to achieve coordinated optimization between the proportion of renewable energy and the capacity of power generation and energy storage. This avoids the problems of over-penetration or under-configuration caused by parameter separation design in traditional methods, and improves the overall coordination and planning accuracy of the system. By setting unmet load fraction constraints, energy balance constraints, and energy storage operation boundary constraints, the system can ensure stable and reliable power supply capabilities while meeting economic objectives, thereby improving the safety and feasibility of project implementation.

[0039] By introducing a feedback learning mechanism, the model can be continuously updated and adaptively adjusted based on the optimization results, thereby enhancing the system's adaptability and robustness under different regional resource conditions and load structures.

[0040] By implementing a unified software architecture and readable storage media for programmatic deployment, the system's scalability, portability, and standardization are enhanced, providing an efficient, intelligent, and scalable technical solution for the planning, design, and decision support of renewable energy fixed-capacity systems. Attached Figure Description

[0041] Figure 1 This is a system architecture diagram of an embodiment of this application.

[0042] Figure 2 This is an operation flowchart of an embodiment of this application. Detailed Implementation

[0043] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0044] This application presents a three-layer intelligent optimization architecture for RFC systems that combines machine learning and hierarchical optimization. It utilizes an artificial neural network (ANN) model to intelligently predict feasible RFC penetration levels, minimizes computational burden by reducing the search space of the optimization algorithm, achieves data-driven initialization and adaptive search space reduction, improves the economy and reliability of the RFC system, and provides a MATLAB-level software architecture suitable for practical deployment. The system's output data includes: the percentage of renewable energy (%), the installed capacity (kW / kWh) of photovoltaic, wind, battery storage, and hydrogen storage components, dispatchable power generation capacity, reliability indicators, and economic benefit indicators.

[0045] A machine learning-assisted three-level optimization system for renewable energy penetration and component size. Includes: an input data module for collecting or importing engineering parameter data of renewable energy fixed capacity systems, wherein the engineering parameter data includes at least load demand curves, renewable resource availability data, component cost parameters, and reliability constraint parameters; The machine learning prediction layer is connected to the input data module and is used to perform predictive analysis on the engineering parameter data based on the machine learning model trained on historical simulation or optimization data, and output the feasible range of renewable energy penetration rate and the predicted range of system component capacity. A penetration optimization layer, which is connected to the machine learning prediction layer, is used to perform optimization calculations within the feasible range of the renewable energy penetration rate to determine the optimal renewable energy penetration rate that meets the techno-economic constraints. A component size optimization layer, which is connected to the penetration optimization layer, is used to perform capacity optimization calculations within the system component capacity prediction range under the constraint of the optimal renewable energy penetration rate, and to determine the optimal installed capacity of renewable energy power generation components and energy storage components. The coordination and control module is used to establish a data linkage and constraint transmission mechanism among the machine learning prediction layer, the penetration optimization layer and the component size optimization layer, so as to realize the hierarchical optimization and collaborative operation. The optimization objective of the component size optimization layer is to minimize the system levelized cost of energy (LCOE) while satisfying reliability constraints and energy balance constraints, and to output system configuration parameters including the proportion of renewable energy, the installed capacity of each component, dispatchable power generation capacity, reliability indicators, and economic indicators.

[0046] Reference Figure 1 , Figure 2 The system is implemented using the MATLAB unified planning and computing platform and adopts a three-layer intelligent optimization architecture, which includes the following functional modules: input data module, machine learning prediction layer, penetration optimization layer, and component size optimization layer. Each module works together through functional interconnection to achieve intelligent optimization. The installation relationship of each module is as follows: the machine learning prediction layer is located upstream of the optimization layer and is connected to the input data module. It is used to receive historical simulation data, system configuration parameters and previous optimization results; the penetration depth optimization layer receives the predicted penetration depth range from the machine learning prediction layer; the component size optimization layer is located downstream of the penetration optimization layer, receives the component size prediction range output by the machine learning prediction layer, and runs under the penetration constraints set by the penetration optimization layer.

[0047] The working principle of the technical solution in this application is as follows: the framework consists of three interrelated layers. The first layer is the machine learning prediction layer, which serves as a decision support system. It uses an artificial neural network (ANN) to train the historical data of the RFC system simulation and optimization. ANN uses engineering characteristics as input parameters, specifically including: load demand curve, availability of renewable resources (such as solar irradiance and wind speed), component cost parameters, and reliability requirements. Artificial neural networks (ANNs) output engineering planning parameters, including: renewable energy penetration range, The approximate optimal capacity range and expected level of energy cost (LCOE) range for system components (such as solar photovoltaic (SPV), battery storage system (BSS), and hydrogen storage system (HSS)). Expected values ​​for system reliability indicators; The output is not the final optimized solution, but rather provides guiding constraints and initial reference values. By intelligently narrowing down the feasible penetration depth and size range, this layer can eliminate infeasible regions, reduce the search space of subsequent layers, and thus improve overall computational efficiency and planning accuracy. The predicted parameters will directly affect the engineering design and physical configuration of the RFC system.

[0048] The second layer is the penetration optimization layer, which determines the optimal penetration level of renewable energy through optimization algorithms. In the preferred implementation, an improved binary search algorithm (BSA) is used. Unlike traditional penetration optimization methods with a fixed search space, the search range of this layer is dynamically limited by the feasible penetration interval predicted by the decision support layer. During the optimization process, the system applies techno-economic constraints, including an economic constraint ensuring that the LCOE of the renewable energy distribution network (RFC) system does not exceed the grid electricity price. This constraint layer effectively avoids infeasible, over-penetrated, or under-penetrated solutions, while rapidly converging to the optimal penetration level.

[0049] The third layer is the component capacity planning layer. After determining the optimal penetration level, the system executes the component capacity optimization layer to determine the optimal capacity of renewable energy generation and storage components. In one embodiment, an enhanced butterfly particle swarm optimization (BF-PSO) algorithm is employed. The predicted component capacity range generated by the decision support layer is used as the initialization parameter for particle positions and defines the upper and lower boundary conditions. This guided initialization accelerates the convergence process, improves solution quality, and avoids unnecessary exploration of non-optimal regions. The optimization objective is to minimize the levelized cost of energy (LCOE) of the system while satisfying key engineering constraints, including: Reliability constraints – System reliability is enforced through unmet load fractions (UF), defined as the ratio of unmet load energy to total load demand. This ratio must satisfy: UF ≤ UFmax， Where UFmax is the maximum permissible reliability threshold; Capacity limitations - Module capacity is subject to preset physical and economic constraints (such as maximum SPV area, upper limit of battery / high-density supercapacitor energy storage capacity). Energy balance constraint - At each time step, the total available energy must meet or exceed the load demand: Renewable energy generation + energy storage discharge + grid support ≥ load demand; Storage operation constraints - Storage systems must adhere to charge and discharge limits to ensure safe and efficient operation: For BSS: SOCmin ≤ SOC(t) ≤ SOCmax, For BSS: SOCmin ≤ SOC(t) ≤ SOCmax, The state of charge (SOC) is dynamically updated based on charging / discharging power, system efficiency, and storage capacity, and a deep discharge limit is imposed.

[0050] For HSS: SOHmin ≤ SOH(t) ≤ SOHmax, For HSS: SOHmin ≤ SOH(t) ≤ SOHmax, When there is excess power generation, hydrogen is produced through an electrolyzer; when there is a shortage of energy, it is consumed through a fuel cell. This system achieves rapid, accurate, and adaptive determination of renewable energy penetration rate and component size through the coordinated operation of the predictive decision support layer and the hierarchical optimization layer, while significantly reducing the computational burden.

[0051] The following two specific embodiments, in conjunction with the technical solution of this application, are provided to further illustrate the system and method of this application, but do not constitute a limitation on the scope of protection of this application.

[0052] Example 1 is an optimization of a fixed-capacity renewable energy system for isolated microgrids; This embodiment takes a coastal island microgrid as an example. The area has no main grid support, and the power system consists of photovoltaic power generation, wind power generation, battery energy storage system and hydrogen energy storage system. The annual peak load is 850kW, the annual average load is 520kW, and the system needs to meet the requirement of uninterrupted power supply throughout the year. The reliability threshold is set at UFmax=1%. Step 1: Data Collection Import the load curve data (hourly level), solar irradiance data, wind speed data and equipment investment and operation and maintenance cost parameters of the area for the past three years through the input data module, and set reliability and energy balance constraints. Step 2: Machine Learning Prediction The machine learning prediction layer is trained using an artificial neural network model. Training samples are derived from historical simulation data and optimization results under different permeability and capacity configurations. Model inputs include load characteristics, peak solar irradiance, average wind speed, unit capacity cost, and UF threshold. Outputs include: The feasible range for renewable energy penetration is [62%, 78%]; The predicted photovoltaic capacity range is [900kW, 1200kW]; The battery energy storage capacity ranges from [1800kWh to 2600kWh]; The hydrogen energy storage capacity range is [3500kWh, 5000kWh]; Through this prediction, the originally set 0-100% full range search was effectively compressed into a locally reasonable range; Step 3: Penetration Optimization The penetration optimization layer employs an improved binary search algorithm, performing a constrained search within the 62%-78% range. Through successive approximation calculations, the optimal renewable energy penetration rate is determined to be 72.4%, provided that the LCOE is no higher than the local diesel power generation cost and UF ≤ 1%. Step 4: Component Capacity Optimization The component size optimization layer employs an enhanced particle swarm optimization algorithm to search for capacity combinations within the predicted capacity range. The algorithm aims to minimize the level of efficiency (LCOE) and converges to the optimal configuration while satisfying energy balance, state of charge (SOC), and state of harmonic equilibrium (SOH) boundary constraints. Photovoltaic capacity 1080kW; Wind power capacity 520kW; Battery energy storage capacity: 2300 kWh; Hydrogen energy storage capacity is 4200kWh.

[0053] The system's average annual LCOE is reduced by approximately 11.3% compared to traditional experience-based configurations, and computation time is shortened by approximately 38%.

[0054] This embodiment demonstrates that the system of this application can achieve stable operation of a high proportion of renewable energy in islanded microgrids, and reduce system costs while ensuring reliability.

[0055] The beneficial effects of the technical solution in this application are as follows: This application pertains to RFC system planning and design technology solutions, providing a technical solution for determining the physical system configuration parameters of an RFC system, offering an optimized techno-economic design scheme for the RFC system, ensuring the best balance between cost, performance, and reliability, achieving high-precision optimization of renewable energy penetration and component selection through coordinated and consistent data-driven decision-making, reducing energy leveling costs and improving economic efficiency, and achieving adaptive learning capabilities through continuous artificial neural network (ANN) retraining. To improve system reliability and scalability and achieve efficient integration of renewable energy, the system can intelligently predict the feasible penetration range of renewable energy resources in specific geographical areas to prevent over-penetration or under-penetration, and significantly reduce computation time by using machine learning to predict and optimize the search space.

[0056] The key technical points of this application's technical solution are: by integrating machine learning prediction and hierarchical optimization systems, a solution is developed that can intelligently reduce the search space, thereby quickly and accurately determining the optimal penetration range and size of the RFC system and achieving the optimal configuration of the RFC system.

[0057] The technical protection points of the technical solution in this application are: Protection Point 1: System-level protection (core protection) A system for optimizing renewable energy penetration and RFC system component selection includes a penetration rate determination module, a component selection module, and a coordination mechanism. The coordination mechanism enables the linkage between penetration rate determination and component selection within a unified system framework.

[0058] Protection Point 2: Intelligent Search Space Reduction The system's penetration determination and component size determination process is performed within an intelligent constraint search space generated based on predictive decision support information.

[0059] Protection Point 3: Machine Learning as a Supporting Module The predictive decision support information of the system is generated by machine learning models trained on historical system simulation or optimization data.

[0060] Protection Point 4: Hierarchical Optimization Architecture The system's penetration optimization and component size optimization are performed in a layered manner, so that penetration optimization is performed before component size optimization.

[0061] Protection Point 5: Independence of Optimization Algorithms The hierarchical optimization layer of the system employs one or more optimization algorithms, which are selected from search algorithms, heuristic algorithms, or metaheuristic algorithms.

[0062] Protection Point 6: Economic and Reliability Constraints The optimization process of the system must meet the technical, economic and reliability constraints, including minimizing the levelization cost (LCOE).

[0063] Protection Point 7: Adaptive Learning Capability The system includes a feedback learning module, which is configured to update the predictive decision support model based on the optimization results.

[0064] Protection point 8: Location-specific protection The optimal penetration range of system d is specifically determined to prevent over- or under-penetration of renewable energy resources in a given geographic location.

[0065] Protection Point 9: Computer-Implemented Protection A computer-implemented method or software system is used to implement any of the above system claims.

[0066] Alternative solutions to the technical solution of this application: 1. Alternative Predictive Decision Support Models: This system can use other data-driven or rule-based predictive models to replace artificial neural networks, including but not limited to: Support Vector Machine (SVM) Decision Tree or Random Forest Deep learning models Regression models or mixed statistical-learning models generate feasible penetration ranges and initial component size guidelines.

[0067] 2. Alternative Optimization Algorithms -- Hierarchical optimization systems can employ various alternative optimization techniques, such as genetic algorithms, differential evolution, and ant colony optimization algorithms, provided that penetration optimization and component size optimization are carried out in a coordinated or hierarchical manner.

[0068] 3. Optional hierarchical architecture -- Without deviating from the core system-level optimization goals, the penetration testing and component size determination process can adopt alternative hierarchical or sequential execution strategies such as multi-stage, two-level, or iterative coordination frameworks.

[0069] 4. Alternative search space constraint mechanisms: Compared to machine learning-based prediction methods, the search space can be constrained by expert rules, adaptive thresholds, historical performance boundaries, sensitivity analysis, or statistical confidence intervals.

[0070] 5. Alternative Implementation Scheme - The system can be deployed using different software or hardware platforms, programming languages, or embedded systems (such as cloud-based computing platforms or real-time energy management systems) without affecting the technical effectiveness of determining the optimal penetration rate and scale.

[0071] The implementation principle of a machine learning-assisted three-level optimization system and method for renewable energy penetration and component size in this application embodiment is as follows: By constructing a collaborative architecture of "data-driven prediction - hierarchical constraint optimization - feedback adaptive update", the machine learning model and the phased optimization algorithm are deeply integrated to achieve efficient collaborative solution of renewable energy penetration ratio and key component capacity. First, the machine learning model is trained based on historical simulation data and multi-scenario optimization samples, enabling it to learn nonlinear mapping relationships from load characteristics, resource conditions, cost parameters and reliability indicators, and predict the feasible range of renewable energy penetration rate and component capacity, thereby compressing the original large-scale search space. Second, within the feasible range after the prediction constraints are reduced, penetration rate optimization and capacity optimization calculations are performed respectively. By satisfying energy balance, reliability constraints and energy storage operation boundary constraints, the comprehensive optimization of technical and economic goals is achieved. Finally, through the feedback learning mechanism, the optimization results are used back to update the model parameters, enabling the system to have continuous learning and adaptive capabilities in different regions and operating scenarios. This realizes a closed-loop operation mechanism of prediction-guided optimization and optimization feeding back to prediction, improving computational efficiency and planning accuracy while ensuring power supply reliability.

[0072] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A machine learning-assisted three-level optimization system for renewable energy penetration and component size. characterized in that include: The input data module is used to collect or import engineering parameter data of renewable energy fixed capacity systems. The engineering parameter data includes at least load demand curves, renewable resource availability data, component cost parameters, and reliability constraint parameters. The machine learning prediction layer is connected to the input data module and is used to perform predictive analysis on the engineering parameter data based on the machine learning model trained on historical simulation or optimization data, and output the feasible range of renewable energy penetration rate and the predicted range of system component capacity. A penetration optimization layer, which is connected to the machine learning prediction layer, is used to perform optimization calculations within the feasible range of the renewable energy penetration rate to determine the optimal renewable energy penetration rate that meets the techno-economic constraints. A component size optimization layer, which is connected to the penetration optimization layer, is used to perform capacity optimization calculations within the system component capacity prediction range under the constraint of the optimal renewable energy penetration rate, and to determine the optimal installed capacity of renewable energy power generation components and energy storage components. The coordination and control module is used to establish a data linkage and constraint transmission mechanism among the machine learning prediction layer, the penetration optimization layer and the component size optimization layer, so as to realize the hierarchical optimization and collaborative operation. The optimization objective of the component size optimization layer is to minimize the system levelized cost of energy (LCOE) while satisfying reliability constraints and energy balance constraints, and to output system configuration parameters including the proportion of renewable energy, the installed capacity of each component, dispatchable power generation capacity, reliability indicators, and economic indicators.

2. The system of claim 1, wherein, The machine learning prediction layer employs an artificial neural network model, which takes load curve features, solar irradiance data, wind speed data, component unit cost, and reliability threshold as input features.

3. The system of claim 1, wherein, The prediction results output by the machine learning prediction layer include upper and lower limits of renewable energy penetration, photovoltaic capacity range, battery energy storage capacity range, and hydrogen energy storage capacity range.

4. The system of claim 1, wherein, The reliability constraints are limited by unmet load fractions (UF) and satisfy the following relationship: UF ≤ UFmax, Where UF is the ratio of unmet load energy to total load demand, and UFmax is the preset maximum allowable threshold.

5. The system of claim 1, wherein, The energy balance constraint satisfies the following condition: within any time step, the sum of renewable energy power generation, energy storage discharge, and external grid support power is greater than or equal to the load demand. The energy storage system includes a battery energy storage system and a hydrogen energy storage system, wherein: Battery energy storage systems satisfy the state of charge constraint: SOCmin ≤ SOC(t) ≤ SOCmax; The hydrogen energy storage system satisfies the hydrogen storage state constraint: SOHmin ≤ SOH(t) ≤ SOHmax; It also includes a feedback learning module, which is used to retrain or update the parameters of the machine learning model based on the optimization results of the penetration optimization layer and the component size optimization layer, so as to achieve adaptive learning; The optimization process is performed within the constrained search space generated by predictive decision support information to avoid exhaustive searches of infeasible or suboptimal regions. The machine learning model is any one or a combination of support vector machine, decision tree, random forest, deep learning model or hybrid statistical learning model; The system is implemented with a unified software architecture and deployed on a local computing platform, cloud computing platform, or real-time energy management system.

6. A machine learning-assisted three-level optimization method for renewable energy penetration and component size. characterized in that Includes the following steps: Step 1: Collect or import load demand data, renewable resource data, component cost parameters, and reliability requirements of renewable energy fixed capacity systems; Step 2: Based on historical simulation and optimization results, construct and train a machine learning model, using the engineering parameters from Step 1 as input, to predict the feasible range of renewable energy penetration rate and the predicted range of system component capacity. Step 3: Perform penetration optimization calculations within the feasible penetration range to determine the optimal renewable energy penetration rate by satisfying the technical and economic constraints; Step 4: Under the constraint of the optimal renewable energy penetration rate, perform component capacity optimization calculation within the component capacity prediction range to minimize the system levelized cost of energy (LCOE). Step 5: Output the optimal renewable energy ratio, photovoltaic system capacity, wind power generation capacity, battery energy storage capacity, hydrogen energy storage capacity, system reliability indicators, and economic indicators; In step two, the search space is reduced by predicting constraints, which reduces the computational complexity of steps three and four and improves the optimization convergence efficiency.

7. The system or method of claim 1 or 6, wherein, The penetration optimization layer employs an improved binary search algorithm to optimize the penetration rate of renewable energy.

8. The system or method of claim 1 or 6, wherein, The component size optimization layer uses an enhanced particle swarm optimization algorithm or other metaheuristic algorithms to perform capacity optimization calculations.

9. The system or method of claim 1 or 6, wherein, The optimal renewable energy penetration rate is a region-specific optimal value determined independently for a specific geographical area, used to avoid excessive or insufficient penetration of renewable energy resources.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method as described in claim 6.

Images