An optimal design method for a source-grid-load-storage system

Abstract

This invention provides an optimal design method for a power generation, grid, load, and storage system, belonging to the field of new energy power generation and energy storage technology. This invention introduces the NSGA-II algorithm for multi-objective iterative optimization, aiming to minimize payback time and maximize annual revenue. The results are adjusted using manual calculation methods, outputting an optimal configuration scheme that includes the optimal combination of photovoltaic and wind power capacity, energy storage system capacity, and flexible load dispatch ratio. This invention improves the NSGA-II algorithm in areas such as initial population, non-dominated sorting, selection of superior parent individuals, and crossover and mutation operations. The optimized algorithm significantly improves global search efficiency and the quality of offspring solutions. This invention achieves global optimization of wind and solar power generation, energy storage systems, grid dispatch, and load demand, ensuring a supply-demand balance between power generation and load demand, and significantly improving the system's economic efficiency and resource utilization.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation and energy storage technology, specifically to an optimal design method for a source-grid-load-storage system. Background Technology

[0002] New energy power generation has attracted widespread attention due to its green, environmentally friendly, low-carbon, and pollution-free advantages. However, its power generation capacity is highly volatile and uncertain, significantly affected by time, climate, and regional conditions. As the core hub for energy transmission, the power grid's capacity and regulation capabilities are limited by technological and economic constraints, making it difficult to fully adapt to the highly volatile and random characteristics of new energy power generation. Furthermore, load-side electricity demand exhibits distinct temporal and diversified characteristics, failing to effectively match the power generation capacity, grid regulation capabilities, and energy storage system capacity. While energy storage systems can mitigate fluctuations and improve power system stability to some extent, their high investment costs and limited dispatch efficiency make the efficient allocation of energy storage resources a crucial challenge for optimizing system operation.

[0003] Ultimately, due to the lack of effective coordinated design among the generation (source), transmission and distribution (grid), energy storage (storage), and load (load) sides, the energy system faces the following problems: 1. Supply and demand imbalance: Large fluctuations in generation, difficulty in matching load demand, and insufficient regulation capacity of the grid and energy storage lead to energy waste or power shortages. 2. Low resource utilization: During off-peak periods, wind and solar power curtailment is prominent, while during peak periods, power supply pressure may arise due to insufficient regulation by energy storage and the grid. 3. Poor economic efficiency: Optimization of a single link fails to reduce the overall operating cost of the system, resulting in a long payback period and difficulty in improving economic efficiency. 4. Lack of global optimization: Traditional design methods only optimize a single link, such as generation capacity design or load-side demand response, failing to fully coordinate the overall configuration relationship of source, grid, storage, and load, affecting system operating efficiency and economy. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides an optimal design method for a power-grid-load-storage system. This invention combines deep learning prediction with manual verification strategies, introducing an improved NSGA-II non-dominated sorting multi-objective optimization genetic algorithm to achieve rational allocation of capacity for wind and solar power generation systems and energy storage systems. It comprehensively considers multiple objectives such as the volatility of wind and solar power generation, energy storage charging and discharging efficiency, load forecasting results, system construction costs, and operating revenue. It globally optimizes wind and solar power generation, energy storage systems, grid dispatching, and load demand to maximize annual system revenue while minimizing payback period, shortening the investment payback period, and increasing annual returns. This invention can significantly improve the system's economic efficiency, enhance resource utilization, and ensure supply and demand balance between power generation and load demand.

[0005] To achieve the above technical objectives, this invention proposes an optimal design method for a source-grid-load-storage system, comprising the following steps:

[0006] Step S1: Prepare relevant data on photovoltaic and wind power generation in a certain region, and perform collaborative modeling based on the relevant data on photovoltaic and wind power generation to obtain a dynamic joint output model of the two power sources;

[0007] Step S2: Using payback period and annual return as objective functions, the NSGA-II algorithm is used to perform multi-objective iterative optimization on the dynamic combined output model, and the output includes the optimal configuration scheme of photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load call ratio.

[0008] Step S3: Based on the relevant data and dynamic combined output model of photovoltaic and wind power generation in a certain region, manually calculate the payback time and annual income under different system scales, obtain multiple settings combinations of photovoltaic capacity, wind power capacity, and energy storage system capacity, and draw curves of annual income and payback time for different energy storage construction scales based on the manual calculation results.

[0009] Step S4: Compare the results of the manual calculation in Step S3 with the output results of the NSGA-II algorithm in Step S2, adjust the output results of the NSGA-II algorithm, and output the optimal configuration scheme of the source-grid-load-storage system, including the optimal combination of photovoltaic and wind power capacity, energy storage system capacity, and flexible load dispatch ratio.

[0010] Preferably, in step S1, a dynamic joint output model of photovoltaic and wind power is obtained by introducing an LSTM model for collaborative modeling, including the following steps:

[0011] Step S11: Preprocess the collected historical power generation data of photovoltaic and wind power and related meteorological data;

[0012] Step S12: Train the LSTM model using preprocessed historical power generation data and relevant meteorological data, optimize the LSTM model parameters, and output the trained LSTM model. The LSTM model includes a sequentially connected input layer, hidden layer and output layer. The input layer is time series features, including historical power generation, sunshine and wind speed. The hidden layer includes multiple LSTM units. The output layer is the predicted value of future power generation.

[0013] Step S13: Use the trained LSTM model to predict the output of photovoltaic and wind power in future periods, and combine the power output prediction results of the time series to construct a dynamic joint output model of photovoltaic and wind power sources, specifically:

[0014] P total (t)=P PV (t)+P Wind(t)

[0015] Among them, P PV (t): The photovoltaic power output predicted by the LSTM model at time t; P Wind (t): Wind power generation output predicted by the LSTM model at time t; P total (t): The combined output of the two power sources at time t.

[0016] Preferably, for the obtained dynamic joint output model of photovoltaic and wind power generation, a dynamic complementarity index R is introduced. complement To quantify the coordination between photovoltaic and wind power output, a load response rate index R is introduced. load This reflects the dynamic matching degree between photovoltaic (PV) and wind power generation output and load demand. The scheduling optimization of PV and wind power generation is based on these two indicators; the dynamic complementarity index R... complement The expression is:

[0017]

[0018] Among them, P PV (t): The photovoltaic power output predicted by the LSTM model at time t; P Wind (t): Wind power generation output predicted by the LSTM model at time t;

[0019] Load response rate index R load The expression is:

[0020]

[0021] Where: P demand (t): Load demand at time t; T is the total statistical duration.

[0022] Preferably, step S2 involves using the NSGA-II algorithm to perform multi-objective iterative optimization on the dynamic combined output model, outputting an optimal configuration scheme that includes the combination of photovoltaic and wind power capacity, energy storage system capacity, and flexible load dispatch ratio. Specifically, this includes the following steps:

[0023] Step S21: Initialize the population, set grid parameter constraints, generate an initial population of individuals with photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load dispatch ratio, and use the initial population as the parent population;

[0024] Step S22: Construct a fitness function based on the set objective function, calculate the fitness value of each individual in the parent population, and perform non-dominated sorting of individuals in the parent population based on the fitness values, dividing the parent population into multiple non-dominated levels.

[0025] Step S23: Calculate the crowding distance for each individual in the non-dominant tier and sort them;

[0026] Step S24: Based on the non-dominance level and crowding distance results, select parent individuals from the parent population for crossover and mutation operations to generate new individuals to form the offspring population.

[0027] Step S25: Merge the parent population and the offspring population, recalculate the non-dominated sorting and crowding distance, and select individuals from the merged population to form a new parent population.

[0028] Step S26: Repeat steps S22-S25 until the preset number of iterations is reached or the termination condition is met, and output the optimal solution, which includes the optimal configuration scheme of photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load dispatch ratio.

[0029] Preferably, in step S21, an adaptive sampling initialization strategy is used to initialize the population. Specifically, this initialization strategy uses photovoltaic and wind power capacity, grid parameter constraints, energy storage system capacity, and flexible load ratio as multidimensional variables. First, an initial population is randomly generated, and then dynamically adjusted using a perturbation factor, so that the initialization of the population is concentrated in the potential optimal solution region. The initialization calculation formula for the population is as follows:

[0030]

[0031] in, The initial solution vector for the i-th individual; X rand : Randomly generated population solution vector; E init : A disturbance factor dynamically adjusted based on historical data or forecast results; P: A forecast distribution matrix related to the physical constraints of the energy storage system;

[0032] Disturbance factor E init The calculation formula is:

[0033] E init =k·σ data +μ data

[0034] Where, σ data The standard deviation of historical data reflects the volatility of photovoltaic and wind power generation or load demand; μ data : The mean of historical data; k: Adjustment coefficient, usually ranging from 0.1 to 0.2, dynamically adjusted according to the needs of the scenario.

[0035] Preferably, for the non-dominated sorting in step S22, during the population evolution process, the weights of each objective function are dynamically adjusted according to the differences in the importance of different optimization objectives at each stage of population evolution. The objective function mapping formula in the non-dominated sorting is:

[0036] fmapped =w1(m)·f1+w2(m)·f2

[0037] Where: w1(m): dynamic weight of payback period; w2(m): dynamic weight of annual return; w1(m) + w2(m) = 1; f1: objective function to minimize payback period; f2: objective function to maximize annual return; f mapped : Objective function mapping;

[0038] The formulas for calculating the dynamic weights w1(m) of payback period and w2(m) of annual return are as follows:

[0039]

[0040] Where: M is the maximum iteration number; m is the current iteration number; ω min : Initial weight of payback period; ω max : The maximum weight of payback period.

[0041] Preferably, for step S24, selecting parent individuals from the parent population through a multi-layered selection mechanism based on elite protection specifically includes the following steps:

[0042] Step S241: Based on fitness values, divide the parent population into two parts: the elite population, which consists of the top N individuals by fitness value. e Individuals; non-elite population, representing the remaining NN e The individuals, where N is the total number of individuals in the parent population;

[0043] Step S242: For the elite population, directly preserve all individuals to the next generation;

[0044] Step S243: For the non-elite population, select individuals to be retained in the next generation population using roulette wheel or random selection methods; the probability formula for selecting an individual from the non-elite population is as follows:

[0045]

[0046] Among them, P roulette f: The probability of an individual in a non-elite population being selected; i : Fitness value of the i-th non-elite individual; non-elite: Set of non-elite individuals;

[0047] Step S244: Merge the retained elite and non-elite populations proportionally to obtain the selected parent individuals, where the proportion of the elite population is R. e The formula is dynamically adjusted according to the algebra, and the specific calculation formula is as follows:

[0048]

[0049] Where: R min R max These are the minimum and maximum values ​​of the elite population proportion, respectively; M is the maximum number of iterations; and m is the current number of iterations.

[0050] Preferably, in step S24, when performing crossover on the selected parent individuals, the directional weights of the parent individuals are dynamically adjusted based on the relative differences in the objective function values ​​of the parent individuals, so that the offspring solution is closer to the parent solution with the better objective function value; the formula for generating the next generation after adjusting the parent direction is:

[0051] P new =α·P parent1 +(1-α)·P parent2

[0052] Where: P new : The generated offspring individuals; P parent1 P parent2 : Two parent individuals; α: Crossover coefficient, dynamically adjusted according to the changing trend of the objective function, calculated using the following formula:

[0053]

[0054] Among them, f parent1 f parent2 : The objective function value of the parent individual; ∈: A small constant to prevent the denominator from being zero, usually taken as 10. -6 ;

[0055] When performing mutation operations on selected parent individuals, an adaptive mutation operator based on local dissimilarity is introduced, allowing the mutation probability to be dynamically adjusted according to the local dissimilarity of the population. The mutation probability P mut The adjustment formula is:

[0056] P mut =P min_mut +λ conv ·Δf local

[0057] Where: P mut : Current mutation probability; P min_mut λ: Lower bound of the mutation probability; conv : Convergence rate parameter, used to control the adjustment range of mutation probability; Δf lobal Local dissimilarity measure, reflecting the local distribution of the current solution in the population, is calculated using the following formula:

[0058]

[0059] Among them, f neighbor f is the average fitness value of the solutions in the neighborhood of the current solution. currentIt is the fitness value of the current solution.

[0060] Preferably, step S3 involves manually calculating the payback period and annual return for different system scales to obtain multiple combinations of photovoltaic capacity, wind power capacity, and energy storage system capacity. Based on the manual calculation results, curves showing the annual return and payback period for different energy storage construction scales are plotted, including the following steps:

[0061] Step S31: Based on the dynamic combined output model, construct the combined output curve of the power generation side during future operating periods;

[0062] Step S32: Compare the combined output curve data with the load demand during the electricity consumption period in real time to obtain the dynamic matching relationship between power generation and load demand, and dynamically schedule the load demand and energy storage system according to the dynamic matching relationship.

[0063] Step S33: Based on the scale range of photovoltaic and wind power generation and energy storage systems, calculate the payback time and annual income under different scale ranges to obtain multiple settings combinations of photovoltaic capacity, wind power capacity and energy storage system capacity;

[0064] Step S34: Based on the calculation results of payback time and annual income corresponding to different configuration combinations, and combined with the dynamic scheduling relationship between load demand and energy storage system, plot the curves of annual income and payback time for different energy storage construction scales.

[0065] Preferably, in step S32, the specific strategy for dynamic scheduling of load demand and energy storage system is as follows:

[0066] Step S321: Compare the combined output curve data with the load demand in real time. When the combined output curve data is lower than the load demand, prioritize the adjustment of flexible loads and reduce power demand by removing or postponing some adjustable loads.

[0067] Step S322: If the adjusted load demand still exceeds the combined output curve data, then activate the energy storage system to supplement power;

[0068] Step S323: When the combined output curve data is higher than the load demand, the system prioritizes restoring the operation of flexible loads to improve power efficiency;

[0069] Step S324: If there is still residual power after the flexible load is restored, the residual power will be used to charge the energy storage device to maximize power utilization.

[0070] Compared with the prior art, the beneficial effects of the present invention are:

[0071] 1. This invention combines deep learning prediction with manual verification strategies. By introducing an improved NSGA-II non-dominated sorting multi-objective optimization genetic algorithm, it achieves rational allocation of capacity for wind and solar power generation systems and energy storage systems. It comprehensively considers multiple objectives such as the volatility of wind and solar power generation, energy storage charging and discharging efficiency, load forecasting results, system construction costs, and operating revenue. It globally optimizes wind and solar power generation, energy storage systems, grid dispatch, and load demand to maximize system annual revenue while minimizing payback time, shortening the investment payback period, and increasing annual returns. This invention can significantly improve the economic efficiency of the system, enhance resource utilization, and ensure supply and demand balance between power generation and load demand.

[0072] 2. This invention integrates resources from the generation side, grid side, energy storage side, and load side, comprehensively considering multiple objectives such as the volatility of wind and solar power generation, energy storage charging and discharging efficiency, load forecasting results, system construction costs, and operating revenue. It achieves optimized configuration of wind and solar power generation, energy storage systems, grid dispatching, and load demand across the entire system. By introducing an improved NSGA-II algorithm for multi-objective iterative optimization, with the optimization objective of minimizing payback time and maximizing annual revenue, and comprehensively considering system operating costs, payback time, and revenue performance, it achieves a reasonable configuration of wind and solar power generation and energy storage system capacity. This invention has the following advantages: 1) Matching generation volatility with load demand: By coordinating the generation volatility and load demand curves through the source-grid-load-storage configuration ratio, it avoids energy curtailment and improves resource utilization efficiency. 2) Enhancing the flexibility of energy storage and the grid: Optimizing energy storage charging and discharging strategies and grid transmission and distribution capabilities ensures supply and demand balance between the generation side and load demand, reducing dispatching pressure. 3) Optimizing return on investment: By comprehensively designing the configuration of generation, energy storage, and transmission and distribution capacity, it significantly shortens the payback period and improves overall returns. 4) Promote the green transformation of the energy structure: reduce dependence on traditional fossil fuels, increase the proportion of clean energy in the power system, and help achieve the goals of low-carbon economy and sustainable development.

[0073] 3. This invention improves various aspects of the NSGA-II algorithm, mainly including: 1. Adaptive sampling initialization strategy is adopted for initializing the population, dynamically adjusted with the help of perturbation factors, so that the population initialization is concentrated in the potential optimal solution region; 2. A dynamic adjustment method based on the importance of objectives is introduced for non-dominated sorting, and the weights of each objective function are dynamically adjusted according to the differences in the importance of different optimization objectives at each stage of population evolution; 3. An environmental constraint factor is introduced into the crowding distance calculation, incorporating environmental constraints into the crowding calculation; 4. For selecting superior parent individuals, a multi-layer selection mechanism based on elite protection is proposed, dividing the population into elite and non-elite populations, and dynamically adjusting the proportion of elite populations and the selection probability of non-elite populations; 5. Directional crossover operation is performed on parent individuals, and the directional weights of the parents are dynamically adjusted according to the relative differences in the objective function values ​​of the parents; 6. When performing mutation operation on selected parent individuals, an adaptive mutation operator based on local dissimilarity is introduced, so that the mutation probability can be dynamically adjusted according to the local dissimilarity of the population. The optimized algorithm achieves significant improvements in global search efficiency, sub-solution quality, and optimization efficiency.

[0074] 4. In the power generation section, this invention introduces an LSTM model for collaborative modeling to obtain a dynamic joint output model for photovoltaic and wind power, improving the accuracy of photovoltaic and wind power output prediction and thus achieving dynamic collaborative optimization scheduling of the two power sources. In the power generation section, power conservation, voltage conservation, and frequency response models are constructed to optimize the control of the power grid, ensuring power balance, voltage stability, and frequency regulation. In the energy storage system section, efficient management of the energy storage system is achieved. In the load section, a priority-based load scheduling strategy is proposed, which improves the system's economic benefits and energy utilization efficiency through flexible load scheduling. Attached Figure Description

[0075] Figure 1 This is a basic flowchart of an optimal design method for a source-grid-load-storage system according to an embodiment of the present invention;

[0076] Figure 2 This is a detailed flowchart of an optimal design method for a source-grid-load-storage system according to an embodiment of the present invention;

[0077] Figure 3 The source-grid-load-storage curves of the original scheme without using the improved NSGA-II optimization algorithm in this embodiment of the invention;

[0078] Figure 4 The source-grid-load-storage curve of the scheme with the shortest payback time and the highest rate of return using the improved NSGA-II optimization algorithm in this embodiment of the invention;

[0079] Figure 5The source-grid-load-storage curve of the scheme with the maximum total revenue using the improved NSGA-II optimization algorithm in this embodiment of the invention;

[0080] Figure 6 The source-grid-load-storage curve of the minimum investment scheme using the improved NSGA-II optimization algorithm in this embodiment of the invention;

[0081] Figure 7 This is a schematic diagram of the source-grid-load-storage park system according to an embodiment of the present invention. Detailed Implementation

[0082] 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.

[0083] Example 1

[0084] Combination Figure 1 As shown in the figure, this embodiment provides an optimal design method for a source-grid-load-storage system, including the following steps:

[0085] Step S1: Prepare relevant data on photovoltaic and wind power generation in a certain region, and perform collaborative modeling based on the relevant data on photovoltaic and wind power generation to obtain a dynamic joint output model of the two power sources;

[0086] Step S2: Using payback period and annual return as objective functions, the NSGA-II algorithm is used to perform multi-objective iterative optimization on the dynamic combined output model, and the output includes the optimal configuration scheme of photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load call ratio.

[0087] Step S3: Based on the relevant data and dynamic combined output model of photovoltaic and wind power generation in a certain region, manually calculate the payback time and annual income under different system scales, obtain multiple settings combinations of photovoltaic capacity, wind power capacity, and energy storage system capacity, and draw curves of annual income and payback time for different energy storage construction scales based on the manual calculation results.

[0088] Step S4: Compare the results of manual calculation in Step S3 with the results of automatic output by the NSGA-II algorithm in Step S2, adjust the output results of the NSGA-II algorithm, and output the optimal configuration scheme of the source-grid-load-storage system, including the optimal combination of photovoltaic and wind power capacity, energy storage system capacity, and flexible load dispatch ratio.

[0089] In the source-grid-load-storage system, source refers to power generation, specifically photovoltaic and wind power; grid refers to the power transmission and distribution network; load refers to load demand, such as charging loads like electric vehicle charging piles; and storage refers to the energy storage system.

[0090] Example 2

[0091] Combination Figure 1 As shown, this embodiment further optimizes and explains step S1 based on embodiment 1.

[0092] The specific steps for preparing relevant data on photovoltaic and wind power generation in a certain region, as described in step S1, are as follows:

[0093] Data preparation for photovoltaic (PV) power generation includes: 1. Collecting historical operating data of the park's PV system: including hourly power generation, weather conditions (such as solar intensity, temperature, humidity, etc.), and parameters such as PV panel tilt angle and azimuth angle. 2. Setting design parameters for the PV system: Selecting the technical parameters of a single PV panel, with dimensions of 1134mm × 2382mm and a power of 550Wp. Based on the actual conditions of the park, the design capacity range for the PV system is set to 3-5MW. Assuming the total number of panels is calculated as design capacity / single panel power / single panel power, the scale of the PV system can be determined. The construction cost of the PV system is extracted: including PV panel procurement, installation costs, and operation and maintenance costs. The unit power cost is set at 2 yuan / Wp. 3. Constructing a PV power generation model: Based on historical data of the PV power generation system, combined with satellite data or historical meteorological data provided by the meteorological station, a model is constructed to analyze the annual average solar irradiance and distribution characteristics of the study area, in order to build a long-term prediction model for PV power generation. Based on the PV power generation prediction model, the hourly power generation under different PV capacities is calculated. The specific calculation formula is as follows:

[0094] P PV (t)=η PV ·G(t)·A

[0095] Where: P PV (t): Power generation of the photovoltaic system during time period t; η PV : Photovoltaic module efficiency; G(t): Solar irradiance at time t; A: Total area of ​​photovoltaic panels. The model simulates photovoltaic power generation under various meteorological conditions, such as sunny and cloudy days, to ensure the reliability and diversity of the predictions.

[0096] Data preparation for wind power generation includes: 1. Collecting historical operating data of the wind power system in the park: This mainly includes collecting wind speed data of the wind power system in the park, specifically using wind speed time series with a 10-minute sampling interval; collecting hourly wind speed and direction data provided by the meteorological station, and analyzing the seasonality, daily periodic fluctuations, and abnormal conditions of wind speed. 2. Setting wind power system design parameters: Determining the design capacity range of the wind power system to be 0.5MW, selecting a wind turbine with a rated power of 500kW based on the rated power curve of the wind turbine generator set; determining the technical parameters of the wind turbine, including the starting wind speed, rated wind speed, and cut-out wind speed of 3m / s, 12m / s, and 25m / s respectively, and a rotor diameter of 80m. 3. Constructing a wind power generation model: Combining the wind turbine power curve and hourly wind speed data, calculating the output power of the wind power system at different wind speeds, using the following formula:

[0097]

[0098] Where: v: wind speed; v cut-in v rated v cut-out : Fan start-up, rated, and cut-off air speeds; P rated Rated power of the wind turbine. A Weber distribution probability model of wind speed distribution is constructed for long-term wind power output assessment and optimization calculation.

[0099] Step S1 describes collaborative modeling based on relevant data from photovoltaic (PV) and wind power generation to obtain a dynamic joint output model for the two power sources. Considering the strong temporal correlation between PV and wind power output, this invention introduces a Long Short-Term Memory (LSTM) model for collaborative modeling to obtain a dynamic joint output model for the two power sources. This enables dynamic prediction of PV and wind power generation, improves the accuracy of PV and wind power output prediction, and thus achieves dynamic collaborative optimization scheduling of the two power sources, maximizing the utilization efficiency of renewable energy. Specifically, the steps include:

[0100] Step S11: Preprocess the collected historical power generation data of photovoltaic and wind power, as well as related meteorological data. Specifically, photovoltaic data includes: power generation, solar irradiance, and ambient temperature; wind power data includes: power generation, wind speed, and wind direction; the time granularity is a 10-minute time series. Noise removal and missing value imputation are performed on these data to ensure data quality.

[0101] Step S12: Train the LSTM model using preprocessed historical power generation data and relevant meteorological data, optimize the LSTM model parameters, and output the trained LSTM model to improve prediction accuracy. The LSTM model consists of sequentially connected input, hidden, and output layers. The input layer contains time-series features (including historical power generation, solar irradiance, and wind speed), the hidden layer contains multiple LSTM units, and the output layer is the predicted future power generation. The LSTM model, through its built-in memory units and forgetting mechanism, can effectively capture the long-term dependencies and short-term fluctuations in power generation data, achieving high-precision predictions of short-term photovoltaic and wind power output. The specific formula for LSTM power generation prediction is as follows:

[0102] h t =σ(W h ·[h t-1 x t ]+b h )

[0103]

[0104] P forecast (t)=W o ·h t +b o

[0105] Where: x t Input data for time t (including historical power generation, weather conditions, load, etc.); h t : The hidden state at time t; c t : Memory unit of time t; P forecast (t): Predicted power generation at time t; W h W o b h b o Model parameters are obtained through optimization during the training process.

[0106] Step S13: Use the trained LSTM model to predict the output of photovoltaic and wind power in future periods, and combine the power output prediction results of the time series to construct a dynamic joint output model for the two power sources, namely photovoltaic and wind power. Specifically:

[0107] P total (t)=P PV (t)+P Wind (t)

[0108] Among them, P PV (t): The photovoltaic power output predicted by the LSTM model at time t; P Wind (t): Wind power generation output predicted by the LSTM model at time t; P total(t): The combined output of the two power sources at time t.

[0109] In addition, this invention introduces a dynamic complementarity index R. complement This quantitative indicator reflects the degree of coordination between photovoltaic and wind power output, guiding the optimization of power generation scheduling. Dynamic complementarity index R complement The expression is:

[0110]

[0111] By introducing the load response rate index R load Load matching assessment is conducted to reflect the dynamic matching degree between photovoltaic and wind power generation output and load demand, with the load response rate index R... load The expression is:

[0112]

[0113] Where: P demand (t): Load demand at time t; T is the total statistical duration.

[0114] The two indicators are the predicted values ​​of photovoltaic power generation and wind power generation. The numerator of the formula is the minimum value of photovoltaic or wind power output, and the denominator is the maximum value of photovoltaic or wind power output. Based on this equation, the dynamic complementarity index is obtained and put into the scheduling program. The program will use wind power and photovoltaic power according to the coefficient ratio. The maximum and minimum values ​​of photovoltaic and wind power will affect the index, thus avoiding the overuse of photovoltaic or wind power.

[0115] When conducting collaborative modeling of photovoltaic and wind power generation, the following points should be noted: 1. Data acquisition must ensure temporal consistency to avoid modeling errors caused by time shifts between data from different sources. 2. Meteorological data should cover multiple years to avoid interference from abnormal weather conditions. 3. The prediction models for photovoltaic and wind power must match the actual equipment parameters in the industrial park to improve the reliability of the prediction results.

[0116] This invention introduces an LSTM model for predicting photovoltaic (PV) and wind power generation. The LSTM model can efficiently capture the temporal characteristics of PV and wind power generation data, improving the accuracy of power output prediction and providing accurate data support for system optimization. This is achieved through the dynamic complementarity index R... complement and load response rate index R load Adjusting the dispatch strategy for photovoltaic and wind power will improve system operating efficiency. Among these adjustments is the dynamic complementarity index R. complement It can quantify the coordination between photovoltaic and wind power, guide the optimization of power generation scheduling, and improve the utilization rate of renewable energy. Load response rate R loadIt can achieve dynamic matching between power generation output and load demand, and through dynamic optimization, it can significantly improve the system's response to load and overall operating efficiency.

[0117] Furthermore, this invention optimizes grid control, energy storage system management, and load demand scheduling in the collaborative optimization design of the power generation, grid, energy storage, and load systems. Specifically:

[0118] In the collaborative optimization design of the power generation, grid, storage, and load system, the power grid is the core component. In some optimization embodiments, this invention also optimizes the control of the power grid by constructing power conservation, voltage conservation, and frequency response models to ensure power balance, voltage stability, and frequency regulation.

[0119] The power conservation model of the power grid ensures the real-time balance of the system by dynamically calculating the relationship between power generation, load demand, energy storage charging and discharging power, and power exchange between the power grid and the external environment. The power balance of the power grid is expressed by the following formula:

[0120] Pgen+Pstorage-Pload-Pgrid=0

[0121] Where: Pgen: Total power output of the power generation source (including renewable energy sources such as wind power and photovoltaics); Pstorage: Charging and discharging power of the energy storage system (positive value indicates discharging, negative value indicates charging); Pload: Total power demand of the grid load; Pgrid: Power exchange between the grid and the external environment (positive value indicates input from the grid, negative value indicates output to the grid). This formula ensures that at any given time, the power interaction between power generation, energy storage, load, and the grid is always in a balanced state, avoiding grid fluctuations caused by supply and demand imbalances.

[0122] The voltage conservation model dynamically adjusts reactive power to control node voltages within a safe range, typically [0.95Vref, 1.05Vref]. Stable node voltages ensure the safe operation of the power grid. For each node i, the voltage-power relationship can be expressed by the following formula:

[0123]

[0124] Where: Vi: voltage value at node i; Vref: reference voltage of the node; ΔVi: voltage offset of the node, determined by reactive power Qi; Ci: capacitance coefficient of the node.

[0125] The frequency regulation model adjusts the frequency offset through the rapid charging and discharging response of the energy storage system, ensuring the frequency remains stable within the standard range, typically with a maximum allowable range of 49.8–50.2 Hz, thereby ensuring stable grid operation. Its formula is:

[0126]

[0127] Where: Δf: frequency offset; ΔP: net change in grid power; M: system inertia constant.

[0128] In the collaborative optimization design of power generation, grid, storage, and load systems, the energy storage system plays a crucial role in balancing power supply and demand and regulating grid stability. In some optimized embodiments, this invention also achieves efficient management of the energy storage system through the following steps:

[0129] Step 1: Capacity allocation of the energy storage system

[0130] Based on historical load data and power generation fluctuations, and considering the design range of the energy storage system (e.g., 1000–8000 kWh), the capacity of the energy storage system is allocated. Capacity allocation must comprehensively consider the system's maximum charging power, maximum discharging power, and charging / discharging efficiency.

[0131] Step 2: Dynamic optimization of energy storage charging and discharging

[0132] An energy storage charging and discharging scheduling model is constructed with the goal of minimizing energy storage losses and maximizing system economy. Real-time response to load fluctuations is achieved through the dynamic charging and discharging behavior of the batteries, and the charging and discharging power is dynamically adjusted according to power balance requirements.

[0133] Step 3: Optimization of Energy Storage Lifespan and Cost

[0134] A battery life degradation model is introduced to calculate the number of cycles and capacity loss of the energy storage system under different operating conditions, balancing economic efficiency and energy storage life, and ensuring the long-term stable operation of the overall system.

[0135] The energy storage parameters designed for this system include a capacity range of 1000–8000 kWh, a maximum charge / discharge power designed to complete charging / discharging in 2 hours, and a charge / discharge efficiency of 90%. During dynamic charging / discharging, the target power is adjusted in real time according to the power balance of the power grid. The frequency compensation response time of the energy storage system is less than 1 second, and battery life is extended by maintaining the state of charge (SOC) between 20% and 80%. For lifespan optimization, a battery degradation model with operating temperature, depth of discharge (DoD), and cycle count as the main variables is adopted, along with a comprehensive cost model that combines battery replacement costs and system operating costs.

[0136] In terms of key steps, the energy storage system uses optimization algorithms to schedule charging and discharging behavior in real time to avoid the impact of overcharging and discharging on battery life, while ensuring power balance and frequency stability of the power grid; it dynamically monitors the state of charge (SOC) to ensure that the battery operates within the optimal charging range (20% to 80%) to extend its service life; and in the process of optimizing charging and discharging power, it takes battery life and economic benefits as joint optimization objectives, dynamically adjusts the depth of charging and discharging, and achieves a trade-off between system operating costs and battery replacement costs.

[0137] In the collaborative optimization design of source-grid-storage-load systems, flexible load scheduling is a crucial aspect of system optimization, such as... Figure 7 The diagram shows a schematic of a power-grid-load-storage (PGS) industrial park system. This invention proposes a priority-based load dispatching strategy, dividing the total load into rigid loads and flexible loads. Flexible loads (such as charging piles for new energy vehicles) account for 20% of the total load. Dynamic dispatching optimizes their operating time and power allocation to improve system economy and stability. During dispatching, rigid loads are prioritized, while flexible loads are adjusted based on electricity price signals and generation curves, prioritizing operation during peak clean energy generation periods or low-price periods. This balances system generation and load demand and significantly improves the absorption rate of photovoltaic and wind power. The dispatching objective for flexible loads is to minimize operating costs and dispatching deviations while meeting power upper and lower limits and time window constraints, achieved through the following formula:

[0138] P load =P rigid +P flex

[0139] Among them, P load The total load power of the system, measured in kW, includes both rigid and flexible loads. The total load must ensure a power supply-demand balance within the system. rigid Rigid load power, measured in kW, refers to the fixed load portion that must be met immediately, such as the power consumption demand of industrial production equipment; this portion of the load cannot be flexibly adjusted. flex Flexible load power, measured in kW, refers to the portion of the load that can be flexibly adjusted according to scheduling optimization strategies, such as charging piles for new energy vehicles and air conditioners. Flexible load accounts for 20% of the total load, and its scheduling can be used to balance the fluctuations in system power generation.

[0140] The total load satisfies the power balance constraint, and the optimization objective function for the flexible load is:

[0141]

[0142] In the objective function, C flex The operating cost of flexible loads, expressed in yuan, is mainly composed of the product of electricity price and load power. The calculation formula is as follows:

[0143]

[0144] Among them, C elec (t) represents the electricity price for each time period t (unit: yuan / kWh).

[0145] In the objective function, λ represents the penalty coefficient for scheduling deviation, a non-negative weighting factor used to balance the trade-off between the operating cost of flexible loads and scheduling deviation. A larger value indicates a higher priority for optimizing scheduling deviation. P target (t) represents the target load power (unit: kW), which is the target operating curve of flexible load, usually determined by the power generation forecast curve or system planning.

[0146] |P flex (t)-P target (t)∣: Represents the deviation between the actual power and the target power of the flexible load, which increases the operating cost of scheduling. t represents the time step, a discrete point in time within a scheduling cycle, in hours or minutes. T represents the total length of the scheduling cycle, in hours or minutes.

[0147] The load management of this invention has the following advantages: 1) It introduces flexible loads (such as electric vehicle charging stations) into the dynamic scheduling model, and by adjusting the load operating time and power allocation in real time, combined with dynamic electricity price signal optimization, it maximizes the use of off-peak electricity prices and peak renewable energy generation, significantly reducing system operating costs. 2) It introduces a target load curve P. target (t) and deviation penalty coefficient λ, by dynamically adjusting the weight of λ, an adaptive trade-off is achieved between economy and load balance, improving the accuracy of load dispatching and the stability of system operation. 3) By dividing the load into rigid loads and flexible loads, rigid loads ensure the basic operating requirements of the system, while flexible loads provide adjustment space through optimized dispatching, providing greater flexibility for the grid to cope with fluctuating generation and dynamic demand. 4) Using a comprehensive optimization objective function, while considering the operating cost and dispatch deviation of flexible loads, a smart load management method that balances economy and accuracy is proposed. Through priority dispatching of flexible loads, optimization driven by dynamic electricity prices, and dynamic matching of load and power source, the economic benefits and energy utilization efficiency of the system are significantly improved, providing a new approach for the intelligent operation of the power generation, grid, storage, and load system.

[0148] Example 3

[0149] Combination Figure 1 As shown, this embodiment further optimizes and explains step S2 based on embodiment 1.

[0150] In this embodiment, for step S2, the NSGA-II algorithm is used to perform multi-objective iterative optimization on the dynamic combined output model, and the output is an optimal configuration scheme including the combination of photovoltaic and wind power capacity, energy storage system capacity, and flexible load dispatch ratio. Specifically, it includes the following steps:

[0151] Step S21: Initialize the population, set grid parameter constraints, generate an initial population of individuals with photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load dispatch ratio, and use the initial population as the parent population;

[0152] Step S22: Construct a fitness function based on the set objective function, calculate the fitness value of each individual in the parent population, and perform non-dominated sorting of individuals in the parent population based on the fitness values, dividing the parent population into multiple non-dominated levels.

[0153] Step S23: Calculate the crowding distance for each individual in the non-dominant tier and sort them;

[0154] Step S24: Based on the non-dominance level and crowding distance results, select superior parent individuals from the parent population for crossover and mutation operations to generate new individuals to form the offspring population.

[0155] Step S25: Merge the parent population and the offspring population, recalculate the non-dominated sorting and crowding distance, and select individuals from the merged population to form a new parent population.

[0156] Step S26: Repeat steps S22-S25 until the preset number of iterations is reached or the termination condition is met, and output the optimal solution, which includes the optimal configuration scheme of photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load dispatch ratio.

[0157] The grid constraint parameters are that the voltage, current and frequency should not exceed the limits. When initializing the population, setting grid parameter constraints (such as current and voltage limits) can exclude such populations.

[0158] The two main optimization objectives of constructing the fitness function are: minimizing the payback period and maximizing the annual return. By constructing the fitness function, the payback period and annual return of the system are comprehensively evaluated, thus solving the multi-objective optimization problem. The payback period of the system is mainly calculated based on the construction costs of the power generation sources (photovoltaic and wind power), grid connection costs, the construction costs of the energy storage system, and load management revenue. The payback period of the system is the ratio of the total investment cost to the annual return of the system, and the specific calculation formula is as follows:

[0159]

[0160] Among them, T payback Payback period (in years); C PVConstruction cost of photovoltaic systems; C wind Construction cost of wind power systems; C storage Construction cost of energy storage systems; C grid : Grid connection cost; R annual The system's annual revenue. All engineering costs are already included in the construction costs.

[0161] The system's annual revenue is evaluated through the coordinated operation of its power generation, grid, storage, and load components. This evaluation considers the electricity price revenue from photovoltaic and wind power generation, the peak-valley price difference revenue from energy storage dispatch, and the economic benefits of flexible load dispatch. The specific calculation formula is as follows:

[0162] R annual =R PV +R wind +R storage +R load

[0163] Where: R annual : The system's annual revenue; R PV Annual revenue of photovoltaic systems; R wind R represents the annual revenue of the wind power system. storage Annual revenue of energy storage systems; R load For flexible load benefits.

[0164] To assess long-term returns, the total return over a 20-year period can be calculated and incorporated into the optimization objective:

[0165] R total，20 =20·(R) PV +R wind +R storage +R load )-(C PV +C wind +C storage +C grid )

[0166] To improve the performance of the source-grid-load-storage multi-objective optimization algorithm, this invention proposes a dynamic optimization strategy based on the fitness function in step 22, specifically in the calculation of the fitness value. By introducing evaluation metrics for population convergence and diversity, and dynamically adjusting weights during algorithm iterations, the strategy ensures that convergence or diversity is prioritized at different evolutionary stages, thereby improving population optimization efficiency and solution quality. In this invention, the next-generation population S is based on a fitness function-corrected algorithm. next The selection formula is:

[0167] S next =argmax(w c (t)·C convergence +w d(t)·f diversity )

[0168] The fitness score for dynamic optimization is calculated using the following formula:

[0169] fitness_score=ω c (t)·C convergence +ω d (t)·f diversity

[0170] This formula combines performance evaluations of convergence and diversity, dynamically switching the optimization objective through dynamic weight balancing. In the early stages of the algorithm, the focus is on diversity expansion; in the middle stages, the balance between the two is achieved; and in the later stages, the focus shifts to accelerating convergence optimization.

[0171] C onvergence The convergence index of the population is used to quantify how close the population as a whole is to the current optimal solution, indicating the approximation ability of the algorithm. The convergence C of the population... onvergence The calculation formula is:

[0172]

[0173] Where: f i f represents the objective function value of the i-th solution; best This represents the optimal solution in the current generation population; N is the total number of individuals in the parent population.

[0174] f diversity This is a population diversity index used to measure the evenness of population distribution in the target space. A higher value indicates a more dispersed population distribution, which helps prevent the population from prematurely converging to a local optimum. Population diversity f diversity The calculation formula is:

[0175]

[0176] Where: f i+1 and f i represents the objective function values ​​of adjacent individuals after sorting; N-1 represents the total number of sortable adjacent solutions in the population.

[0177] wc is the convergence weight, and wd is the diversity weight. The optimization objective is balanced in stages by adjusting the weights wc and wd. The calculation formula is as follows:

[0178]

[0179] Where: ω c (t) represents the convergence weight of the t-th generation; ω d(t) represents the diversity weight in generation t; T is the maximum number of iterations; t is the current number of iterations. In the early stages of the algorithm (when the initial number of iterations is small), the diversity weight ω d The initial weights are relatively high, aiming to expand the search range and increase the diversity of solutions; in the middle stage of the algorithm, the weights gradually transition to convergence ω. c and diversity ω d The balance; in the later stages of the algorithm (when the number of iterations approaches the maximum value), the convergence weight ω c Significantly improved, guiding the population to converge rapidly toward the optimal solution region.

[0180] This invention incorporates convergence and diversity—two core elements—into the evaluation mechanism through dynamic optimization of the fitness function. By dynamically adjusting weights according to different stages, the algorithm effectively prevents itself from getting trapped in local optima in the early stages through a diversity maintenance mechanism. In the later stages, it rapidly approaches the optimal solution region, reducing computation time and ensuring a balance between convergence and diversity at different stages. This enhances the algorithm's adaptability to complex multi-objective problems. This method accelerates convergence to the optimal solution while maintaining population diversity, improving the quality of the solution and the convergence speed of the algorithm. It achieves precise control over convergence and diversity in the optimization problem of source-grid-storage-load systems, providing theoretical support and practical solutions for the efficient optimization of complex energy systems.

[0181] To address the complex multi-objective optimization requirements of source-grid-load-storage systems, this invention proposes an adaptive sampling initialization strategy for the initial population in step S21 of the NSGA-II algorithm. Specifically, this strategy generates an initial solution vector covering the variables of the source-grid-storage-load system through adaptive sampling. The initial population not only randomly generates values ​​for multi-dimensional variables such as source (PV and wind power capacity), grid (grid parameter constraints), storage (energy storage system capacity), and load (flexible load ratio), but also combines historical operating data, load forecasting models, and meteorological data to comprehensively predict PV power generation, wind power output, and load demand, thus pre-estimating potential high-efficiency areas. This method dynamically adjusts the population distribution using a perturbation factor, making the initial population distribution more concentrated in potential optimal solution regions, thereby accelerating the optimization process. The initialization calculation formula for the population is as follows:

[0182]

[0183] in, The initial solution vector for the i-th individual includes the source (photovoltaic and wind power capacity), grid (grid parameter constraints), storage (energy storage system capacity), and load (flexible load ratio); X rand : Randomly generated population solution vector; E init: A disturbance factor dynamically adjusted based on historical data or forecast results to optimize the diversity of solutions; P: A forecast distribution matrix related to the physical constraints of the energy storage system (e.g., the forecast range of photovoltaic and wind power capacity, load forecast curves, etc.).

[0184] Among them, the disturbance factor E init The calculation formula is:

[0185] E init =k·σ data +μ data

[0186] Where, σ data The standard deviation of historical data reflects the volatility of photovoltaic and wind power generation or load demand; μ data : The mean of historical data; k: Adjustment coefficient, usually ranging from 0.1 to 0.2, dynamically adjusted according to the needs of the scenario.

[0187] The adjustment of the perturbation factor is based on three considerations: First, based on historical data volatility: the perturbation range is estimated using the mean and standard deviation to ensure the coverage of the initial population. For example, for scenarios with large fluctuations in historical photovoltaic power generation, k is appropriately increased to expand the search space; while for scenarios with smaller fluctuations, k is decreased to concentrate the population in the high-efficiency region. Second, based on the sensitivity of the objective function: when the objective function is sensitive to changes in the search space, k is decreased. init The value of ; while when the objective function requires a broad search, increasing ∈ init Based on physical constraints: The perturbation amplitude is dynamically adjusted according to the upper and lower limits of the capacity and load demand of photovoltaic, wind power, and energy storage systems to ensure the effectiveness and feasibility of the search space.

[0188] For example, in the specific implementation of the algorithm's initial population, the ranges for photovoltaic and wind power capacity are set to 3-5MW and 1-2MW respectively, the energy storage system capacity is set to 1000-8000kWh, and the flexible load ratio range is set to 10%-30%. Historical photovoltaic power generation curves, wind power forecast data, and load curves are used to dynamically adjust the disturbance factors, concentrating the population in the possible optimal solution region. For instance, when historical data indicates a large peak load demand, the initial range of the energy storage system capacity can be appropriately increased, while simultaneously balancing the disturbances in photovoltaic and wind power capacity to ensure system supply and demand balance.

[0189] This invention introduces an adaptive sampling initialization strategy, combining historical data, prediction models, and physical constraints to improve the quality and distribution range of the initial population. This avoids complete reliance on randomly generated populations, allowing the algorithm to directly initiate the search within potential optimal solution regions, thereby accelerating the optimization process and making population initialization more targeted and efficient. Specifically, this is reflected in the following three aspects: 1) Multi-dimensional variable fusion: Variables such as photovoltaic capacity, wind power capacity, energy storage system capacity, and load flexibility ratio are simultaneously incorporated into the population initialization, fully considering the multi-dimensional coupling characteristics of the source-grid-load-storage system. 2) Data-driven initialization: The population distribution is dynamically adjusted using historical operating data, prediction models, and physical constraints, reducing the impact of randomness on algorithm efficiency. 3) Concentration of optimal solution regions: Through dynamic adjustment of perturbation factors, population initialization is concentrated in potential optimal solution regions, avoiding redundant calculations and accelerating convergence.

[0190] This invention addresses the complexity of optimizing source-grid-storage-load systems by introducing a non-dominated sorting method based on dynamic adjustment of objective importance for the non-dominated sorting in step S22 of the NSGA-II algorithm. During the population evolution process, the weights of each objective function are dynamically adjusted according to the differences in the importance of different optimization objectives at each stage of population evolution, achieving adaptive balance among different optimization objectives to improve the accuracy and efficiency of multi-objective optimization. The two main optimization objectives of the source-grid-storage-load system of this invention are: Objective 1: Minimize the payback period f1; Objective 2: Maximize the annual return f2. The designed objective function mapping formula is:

[0191] f mapped =w1(m)·f1+w2(m)·f2

[0192] Where: w1(m): dynamic weight of payback period; w2(m): dynamic weight of annual return; w1(m) + w2(m) = 1; f1: objective function to minimize payback period; f2: objective function to maximize annual return; f mapped : Objective function mapping;

[0193] The formulas for calculating the dynamic weights w1(m) of payback period and w2(m) of annual return are as follows:

[0194]

[0195] Where: M is the maximum iteration number; m is the current iteration number; ω min The initial weight for payback period is typically set at 0.3-0.4 to ensure a high focus on the annual return target during the initial optimization phase; ω max : The maximum weight of payback time, usually taken as 0.7-0.8, to ensure that the focus on payback time is enhanced in the later stages of optimization; ω2(m): dynamically calculated based on w1(m), gradually decreasing as m increases.

[0196] In the early stages of algorithm optimization (global search stage), the weight ω2(m) is relatively large, and the algorithm focuses more on maximizing returns, exploring potential high-return regions. At this time, the contribution of the annual return objective f2 is dominant through the objective mapping formula. In the later stages of algorithm optimization (local convergence stage), as the algebra m increases, the weight ω1(m) gradually increases, and the objective of minimizing the payback period f1 is given priority. The algorithm focuses on finding an equilibrium solution to achieve final convergence.

[0197] The non-dominated ranking method based on dynamic adjustment of target importance in this invention can dynamically switch the focus of optimization objectives according to changes in the population evolution stage, avoiding the limitations of traditional fixed-weight methods. In multi-objective optimization of source-grid-storage-load systems, by dynamically adjusting weights, it effectively solves the problem of the importance of different objective functions changing with the stage, enabling the algorithm to exhibit good adaptability in both global search and local convergence stages, improving the exploration capability of the solution space and the accuracy of local search. Furthermore, by combining the objective function mapping with the physical constraints of the source-grid-storage-load system, the practical feasibility of the optimization solution is improved.

[0198] This invention proposes a congestion distance calculation method based on environmental constraints for step S23 of the NSGA-II algorithm. Based on the traditional congestion calculation formula, the congestion degree is modified by combining the degree to which an individual meets environmental constraints (such as power grid limits, equipment capacity constraints, etc.), which significantly improves the practical feasibility and optimization efficiency of the solution.

[0199] Improved congestion distance d i The calculation formula is:

[0200]

[0201] Where, d i The traditional formula for calculating the crowding distance of the i-th individual is: M: The number of objective functions; f k,i+1 f k,i-1 : Under the k-th objective, the objective function values ​​of the current individual's immediate and next neighbors; f k,max f k,min : The maximum and minimum values ​​of the k-th target; C env,i : The environmental constraints satisfied by the i-th individual, such as grid power limits or equipment capacity; C env,max : The maximum allowable value of environmental constraints; α: Environmental constraint weighting coefficient, used to balance the influence of the objective function distribution and environmental constraints. When the individual's environmental constraint satisfaction level is low (i.e., C... env,i Approaching C env,max (time), d iThe crowding distance will decrease significantly, thus lowering the individual's priority; conversely, when an individual's environmental constraints are met to a high degree, their crowding distance will increase, and the individual will obtain a higher priority.

[0202] This invention introduces an environmental constraint factor C. env,i By incorporating environmental constraints into the congestion calculation, individuals that meet these constraints receive higher priority. Furthermore, to balance the relationship between the objective function optimization and environmental constraints, and to avoid excessive bias towards one objective, an environmental constraint weighting coefficient α is used to further adjust the congestion calculation. In the optimal configuration of a power generation, grid, and energy storage system, excessively large photovoltaic and energy storage system capacities may lead to grid overload and failure to meet environmental constraints; conversely, insufficient capacity may result in inadequate returns. By introducing an improved congestion distance calculation method, when a photovoltaic and energy storage system with excessively large capacities is configured, even if its returns are high, its congestion will decrease due to violating environmental constraints, thus eliminating it from the algorithm. Conversely, solutions that meet environmental constraints will receive higher priority during the selection process and are more likely to be retained. Incorporating environmental constraints into the multi-objective optimization evaluation system makes the optimization results more aligned with actual needs, significantly improving the practical feasibility and reliability of the optimization solutions; enhancing the algorithm's robustness, enabling it to effectively find the optimal solution that meets actual constraints even in complex power generation, grid, and energy storage scenarios; and assigning low priority to solutions that do not meet actual environmental constraints, avoiding computational waste of invalid solutions and improving the algorithm's optimization efficiency.

[0203] For step S24 of the NSGA-II algorithm, which involves selecting superior parent individuals from the parent population, this invention proposes a multi-level selection mechanism based on elite protection. This multi-level selection mechanism divides the population into an elite population and a non-elite population, prioritizing the protection of individuals with high fitness while selecting individuals from the non-elite population to supplement the next generation. This enhances global search capabilities while accelerating population convergence. Specifically, it includes the following steps:

[0204] Step S241: Based on fitness values, divide the parent population into two parts: the elite population (top N in fitness ranking) e (individuals) and non-elite populations (remaining NN) e (individuals);

[0205] Step S242: For the elite population, all individuals are directly retained to the next generation to ensure the inheritance of the optimal solution;

[0206] Step S243: For the non-elite population, select individuals to be retained in the next generation population using roulette wheel or random selection methods; the probability formula for selecting an individual from the non-elite population is as follows:

[0207]

[0208] Among them, P roulette f: The probability of an individual in a non-elite population being selected; i : Fitness value of the i-th non-elite individual; non-elite: Set of non-elite individuals;

[0209] Step S244: Merge the retained elite and non-elite populations proportionally to obtain superior parent individuals. To balance convergence and diversity, the proportion of the elite population R is set as follows: e The formula is dynamically adjusted according to the algebra, and the specific calculation formula is as follows:

[0210]

[0211] Where: R min R max These are the minimum and maximum values ​​of the elite population proportion, respectively; M is the maximum number of iterations; and m is the current number of iterations.

[0212] Compared to traditional single-selection methods, this invention proposes a two-layer selection mechanism based on elite protection, achieving a dynamic balance between global exploration and local convergence, thus improving the genetic quality of the solution. Through the elite protection mechanism, individuals with high fitness are directly preserved to the next generation, ensuring the inheritance of superior genes and improving the convergence speed of the population. By dynamically adjusting the proportion Re of the elite population and the selection probability of the non-elite population, convergence and diversity are balanced. The proportion Re of the elite population can be dynamically adjusted with each generation. In the early stages of iteration, the proportion of elites is reduced, increasing the exploration space for non-elite individuals and improving population diversity; in the later stages of iteration, the proportion of elites is gradually increased, accelerating convergence towards the optimal solution. For the selection of the non-elite population, a weighted probability selection is performed based on fitness values ​​to avoid the complete elimination of low-fitness individuals, ensuring population diversity, especially in the early stages of optimization, enabling better exploration of the global solution space. This mechanism is designed for multi-objective optimization problems involving power generation, grid, load, and storage, and can better coordinate the complex relationships between power generation, energy storage, the grid, and the load. The elite protection mechanism ensures that the optimal solution can meet the dynamic balance of energy supply and demand; the diversity selection of non-elite populations enhances the adaptability to fluctuating loads and energy storage scheduling problems.

[0213] For the crossover operation in step S24 of the NSGA-II algorithm, this invention proposes a directional crossover operation. When performing the crossover operation on the selected parent individuals, the directional weights of the parent individuals are dynamically adjusted according to the relative differences in the objective function values ​​of the parent individuals. This makes the child solutions closer to the parent solutions with better objective function values, thereby guiding the solutions towards the target region during the search, achieving a more accurate search, and improving the quality and optimization efficiency of the child solutions. The next generation generation formula is:

[0214] P new =α·P parent1+(1-α)·P parent2

[0215] Where: P new : The next generation of individuals generated; P parent1 P parent2 : Two parent individuals; α: Crossover coefficient, dynamically adjusted according to the changing trend of the objective function, calculated using the following formula:

[0216]

[0217] Among them, f parent1 f parent2 : The objective function value of the parent individual; ∈: A small constant to prevent the denominator from being zero, usually taken as 10. -6 .

[0218] The directional crossover operation of this invention dynamically adjusts the position of the crossover point through the gradient information of the objective function, ensuring that the offspring solution moves closer to the optimal solution region and guiding the search towards a direction with a better objective function value. This effectively enhances the optimization capability of the algorithm, making the generation of offspring solutions more directional, the crossover operation more precise, and avoiding a large number of invalid searches. By dynamically calculating the crossover coefficient α, the offspring solution is more inclined to inherit the characteristics of the parent solution with a better objective function value during the offspring generation process, which helps to retain better solution information in the genetic operation, thereby improving the optimization effect. At the same time, it improves the search efficiency of the algorithm and the quality of the offspring solutions. In the optimization design of source-grid-storage-load systems, directional crossover can dynamically adapt to the multi-objective optimization needs of the system. Especially when dealing with capacity configuration and load scheduling problems of photovoltaic, wind power, and energy storage systems, it can quickly guide the solution to converge towards a direction with a better objective function value, significantly improving the optimization efficiency and practical application effect of the algorithm.

[0219] For the mutation operation in step S24 of the NSGA-II algorithm, this invention introduces an adaptive mutation operator based on local dissimilarity, enabling the mutation probability to be dynamically adjusted according to the local dissimilarity of the population, thereby improving global search capability and local search efficiency. Mutation probability P mut The adjustment formula is:

[0220] P mut =P min_mut +λ conv ·△f local

[0221] Where: P mut : Current mutation probability; P min_mut λ: The lower limit of the mutation probability, usually set to a small value (e.g., 0.01); conv : Convergence rate parameter, used to control the adjustment range of mutation probability; Δf localLocal dissimilarity measure, reflecting the local distribution of the current solution in the population, is calculated using the following formula:

[0222]

[0223] Among them, f neighbor f is the average fitness value of the solutions in the neighborhood of the current solution. current This is the fitness value of the current solution. Using the above formula, when the population convergence speed is slow, local differences are large, and the mutation probability P... mut This enhances global search capabilities; as the population gradually converges in a local area, local differences decrease, the probability of mutation decreases, and the efficiency of fine-grained local search is improved.

[0224] During the local convergence phase, to avoid getting trapped in local optima, directed mutation is introduced. This mutation is guided by the difference between the current solution and the neighborhood optimum, making the mutation more targeted.

[0225] x mut =x current +α·(x best -x current )+β·randn()

[0226] Where: x mut : The value of the mutated solution variable; x current : The value of the variable in the current solution; x best : The local optimum in the current neighborhood; α: Directed mutation parameter, controlling the size of the mutation vector; β: Random perturbation factor, used to increase the diversity of solutions; randn(): Standard normal distribution random number. Through the above formula, when an individual is far from the local optimum, directed mutation can guide it closer to the local optimum; at the same time, the random perturbation factor β·randn() can prevent the solution from getting trapped in local optima, enhancing the global search capability.

[0227] This invention enhances the algorithm's exploration capability in the early stages and its convergence capability in the later stages by dynamically adjusting the mutation probability based on local dissimilarity. Simultaneously, by combining the difference between the current solution and the neighborhood optimal solution, directed mutation guides the search process, making the mutation more targeted and improving search efficiency. To maintain solution diversity, appropriate random perturbation is incorporated to prevent the algorithm from getting trapped in local optima. The combination of random perturbation and directed mutation allows the algorithm to both escape local optima and quickly converge to the global optimum, significantly improving both search efficiency and solution quality.

[0228] Example 4

[0229] Combination Figure 1-2 As shown, this embodiment further optimizes and explains steps S3 and S4 based on embodiment 1.

[0230] In this embodiment, step S3 involves manually calculating the payback period and annual return for different system scales based on relevant data and dynamic combined output models of photovoltaic and wind power generation in a certain region, obtaining multiple combinations of photovoltaic capacity, wind power capacity, and energy storage system capacity, and plotting curves of annual return and payback period for different energy storage construction scales based on the manual calculation results. Specifically, this includes the following steps:

[0231] Step S31: Based on the dynamic combined output model, construct the combined output curve of the power generation side during future operation. This curve dynamically reflects the power supply capacity of the power generation side under different meteorological conditions and load demands. The formula for expressing the combined output curve is:

[0232] P gen (t)=P PV (t)+P wind (t)

[0233] Where: Pgen(t): combined power output (kW) in time period t; PPV(t): photovoltaic power generation (kW) in time period t; Pwind(t): wind power generation (kW) in time period t.

[0234] Step S32: Transfer the combined output curve data P gen (t) and load demand P during the electricity consumption period load (t) Real-time comparison is performed to obtain the dynamic matching relationship between power generation and load demand, and dynamic scheduling of load demand and energy storage system is performed based on the dynamic matching relationship. The dynamic scheduling strategy between load demand and energy storage system is as follows:

[0235] Step S321: Compare the combined output curve data with the load demand in real time. When the combined output curve data is lower than the load demand (P... gen (t) <P load When (t) occurs, priority should be given to adjusting flexible loads. This involves reducing power demand by removing or postponing some adjustable loads, such as partially or completely shutting down electric vehicle charging stations, air conditioners, and other adjustable loads, thereby reducing the discharge pressure on the energy storage system. The adjusted load demand is as follows:

[0236] P load,new (t)=P load (t)-P flex (t)

[0237] Among them, P load,new (t): Adjusted load demand (kW) for time period t; P load (t): Load demand (kW) before adjustment at time period t; P flex (t): Flexible load (kW) adjusted for time period t.

[0238] Step S322: If the adjusted load demand still exceeds the combined output curve, the energy storage system is activated to supplement power, ensuring reliable fulfillment of load demand until the energy storage battery discharges to its minimum charge limit (SOCmin). This is suitable for periods with low electricity prices or scenarios with small peak-valley price differences, maximizing the proportion of self-generated electricity and reducing reliance on purchased power. The discharge power of the energy storage system is:

[0239] P storage (t)=min(P load,new (t)-P gen (t),P storage,max )

[0240] Among them, P storage (t): Discharge power (kW) of the energy storage system during time period t; P load,new (t): Adjusted load demand (kW) for time period t; P storage，max (t): Maximum energy storage capacity of the energy storage system (kW).

[0241] Step S323: When the combined output curve data is higher than the load demand (P) gen (t)>P load When (t) occurs, the system prioritizes restoring the operation of flexible loads to improve power efficiency. The adjusted load is:

[0242] P load,new (t)=P load (t)+P flex (t)

[0243] Step S324: If there is still residual power after the flexible load is restored, the residual power will be used to charge the energy storage device to maximize power utilization. The charging power for the energy storage system is:

[0244] P storage,charge (t)=min(P gen (t)-P load,new (t),P storage,charge_max )

[0245] Among them, P storage,charge (t): Charging power of the energy storage system in time period t (kW); Pgen(t): Combined output power in time period t (kW); P storage,charge_max (t): The maximum rechargeable power (kW) of the energy storage system.

[0246] For the dynamic scheduling logic described above, the payback period and annual return are calculated respectively. The specific return formula is as follows:

[0247]

[0248] Among them, Rannual : Annual revenue (RMB); Pgen(t): Combined power output at time t (kW); P storage,charge (t): Charging power (kW) of the energy storage system during time period t; P storage (t): Discharge power (kW) of the energy storage system during time period t; p market (t): Dynamic market electricity price at time t (yuan / kW).

[0249] Step S33: Based on the scale range of photovoltaic and wind power generation and energy storage systems, calculate the payback time and annual return under different scale ranges to obtain multiple combinations of photovoltaic capacity, wind power capacity, and energy storage system capacity settings. Specifically, for the set scale ranges of photovoltaic systems and wind power systems, gradually increase the scale ranges by a fixed step size to obtain multiple photovoltaic capacity settings, wind power capacity settings, and energy storage system capacity settings. Combine these settings to obtain various combination schemes of photovoltaic capacity, wind power capacity, and energy storage system capacity to cover multiple possible scenarios in practical applications for different scale ranges. For each combination scheme of photovoltaic capacity, wind power capacity, and energy storage system capacity, calculate the system's payback time and annual return. Among them, three energy storage consumption methods are set for each energy storage system capacity to examine the performance of returns and payback time under different energy storage configurations. The specific consumption methods include:

[0250] Full absorption: When the output curves of photovoltaic and wind power exceed the electricity demand, the energy is stored. Subsequently, the energy storage discharges and photovoltaic and wind power support the load of the entire park until the electricity price is at its lowest point.

[0251] Peak load absorption: When the output curves of photovoltaic and wind power exceed the electricity demand, the energy is stored. Subsequently, the energy storage discharges and photovoltaic and wind power support the load of the entire park until the electricity price reaches parity.

[0252] Partial consumption: When the output curves of photovoltaic and wind power exceed the electricity demand, the portion of electrical energy is stored. Subsequently, the stored energy is discharged to consume the remaining portion of the electricity.

[0253] Specifically, with photovoltaic capacity gradually increasing from 3.5MW to 5MW and wind power capacity fixed at 0.5MW, the energy storage system capacity is set from 1000kWh to 8000kWh to cover various scenarios in practical applications with different energy storage scales. Meanwhile, considering the fluctuating characteristics of load demand, a 20% flexible load component is taken into account, and the system operation is further optimized by dynamically adjusting the load dispatch strategy.

[0254] Step S34: Based on the calculated payback time and annual revenue for different configuration combinations, and considering the dynamic scheduling relationship between load demand and the energy storage system, curves showing the annual revenue and payback time for different energy storage construction scales are plotted. Specifically, curves showing the relationship between energy storage system capacity and annual revenue, and energy storage system capacity and payback time, are drawn. Furthermore, to more comprehensively assess the impact of flexible loads, curve changes under flexible load activation and deactivation conditions are analyzed separately. These curves reveal the mechanism of energy storage systems in source-grid-storage-load synergy scenarios, as well as the combined impact of photovoltaic, wind power, energy storage scale, and flexible loads on system performance. Through this step, the sensitivity of changes in energy storage system capacity to system revenue and payback time can be intuitively observed. For example, with low energy storage system capacity, the system payback time is longer and the annual revenue is lower. As the energy storage system capacity increases, the revenue gradually increases, but the growth slows down after approaching the energy storage economic saturation point. Furthermore, the participation of flexible loads significantly improves the system's annual revenue and payback period, particularly during peak electricity price periods when it can fully utilize energy storage discharge revenue, while optimizing charging strategies during off-peak electricity price periods to reduce wind and solar curtailment losses. These curves provide important basis for the optimal configuration of the power generation, grid, energy storage, and load system, helping to identify the best combination of different power sources, energy storage scales, and load management strategies, and providing scientific guidance for practical engineering design and investment decisions.

[0255] In this embodiment, for step S4, the results of manual calculation are compared with the results automatically output by the NSGA-II algorithm, the output results of the NSGA-II algorithm are adjusted, and the optimal configuration scheme of the source-grid-load-storage system is output, including the optimal power generation construction scale (combination of photovoltaic and wind power capacity), energy storage system capacity, and flexible load call ratio.

[0256] To verify the effectiveness of the algorithm in the source-grid-load-storage scenario, discrete data points obtained through manual calculation are compared and analyzed with the continuous optimization curves generated by the algorithm. During the comparison, based on the actual scenario of source-grid-load-storage coordinated operation, if the discrete data points such as power generation output, energy storage charging and discharging, and load dispatch obtained through manual calculation match the curves of the algorithm's optimization results, it indicates that the algorithm can effectively solve for the optimal configuration of the source-grid-load-storage system and has high accuracy.

[0257] To ensure the reliability of the algorithm's optimization results, verification is also performed here. For example, under different power generation configurations (combinations of photovoltaic and wind power capacity), energy storage system capacity, and flexible load dispatch modes, the dynamic process of matching power generation with load demand is manually calculated and compared to verify whether the optimization results can meet the requirements of grid power balance, voltage stability, and constraints (such as the energy storage SoC range). Simultaneously, the algorithm's convergence performance on the objective function and its ability to balance multiple objectives are further evaluated at different stages in the calculation of revenue and payback period.

[0258] In the process of adjusting and optimizing the results, we focus on analyzing the changes in payback time and annual income under different configuration schemes, while also considering constraints such as grid power balance, voltage stability and economic benefits.

[0259] Specifically, for the coordinated operation scenario of source, grid, load, and storage, the following content is provided based on the optimization results:

[0260] 1. Optimal power generation configuration: The optimal construction scale to meet load demand under different combinations of photovoltaic and wind power capacity.

[0261] 2. Optimal energy storage configuration: Determine the optimal energy storage scale that satisfies both economic benefits and grid stability within different energy storage system capacity ranges.

[0262] 3. Flexible load dispatch strategy: By combining the load curve and the power generation output curve, the optimal flexible load dispatch ratio is output to improve the system utilization efficiency and economic benefits.

[0263] Furthermore, by comparing the payback period with the annual returns of the optimization results, a scientific basis is provided for decision-makers. For example, under budget constraints, by comparing the shortest payback period with the highest return, the impact of different configuration schemes on the return on investment is assessed, providing feasibility guidance for the actual engineering design of the power generation, grid, load, and storage system. This process ensures that the optimization results not only meet technical requirements but also have high practical application value and economic decision-making significance.

[0264] Example 5

[0265] The following example illustrates the effectiveness of this invention. Specifically, the optimal configuration scheme of the source-grid-load-storage system obtained by the improved NSGA-II algorithm will be compared with the configuration scheme obtained by the traditional method. Before using the improved NSGA-II algorithm, the park had a 3MW photovoltaic power generation system, a 1.5MW wind power generation system, and a 3MW / 1.5MWh energy storage system. The energy storage system was designed to cover peak evening electricity demand. The total investment cost of the entire system was 18,000,000 yuan, the rate of return was 16.07%, the annual income was 2,870,000 yuan, the payback period was 6.22 years, and the profit over 20 years of operation would be 39,600,000 yuan. The source-grid-load-storage curve of the original scheme is shown below. Figure 3 As shown.

[0266] Using the improved NSGA-II algorithm, if the primary focus of the scheme is on payback time: the algorithm calculates that with a 3MW photovoltaic scale, a 1MW wind power system, and a 3.5MW / 1.75MWh energy storage system, the energy storage system should be able to cover peak evening electricity demand and adjust for a 20% flexible load. This scheme can achieve the shortest payback time. The total investment cost of the entire system is 15,750,000 yuan, the rate of return is 19.71%, the annual return is 3,100,000 yuan, the payback period is 5.07 years, and the profit over 20 years is 46,330,000 yuan. The payback period is shortened by 18%, the rate of return is increased by 3.64%, and the total profit is increased by 17%. Furthermore, this scheme also has the highest rate of return. The source-grid-load-storage curve for the scheme with the shortest payback time and highest rate of return is shown below. Figure 4 As shown.

[0267] Using the improved NSGA-II algorithm, if the main focus of the scheme is on total profit: The algorithm calculates that with a 4MW photovoltaic scale, a 1.5MW wind power system, a 4.2MW / 2.1MWh energy storage system, and a 20% flexible load adjustment, the scheme can achieve the maximum total profit. The total investment cost of the entire system is 21,600,000 yuan, the rate of return is 19.18%, the annual return is 4,143,000 yuan, the payback period is 5.21 years, and over 20 years of operation, it can earn 61,300,000 yuan. This represents a 16% reduction in payback period, a 3.11% increase in rate of return, and a 32% increase in total profit. The electricity generated by the photovoltaic and wind power systems can completely cover the period from peak to off-peak demand. This scheme has the highest total profit, but it also has the highest total investment. The source-grid-load-storage curve for the scheme with the highest total profit is shown below. Figure 5 As shown.

[0268] Using the improved NSGA-II algorithm, to achieve a relatively good return with a small total investment: The algorithm calculates that with a 3MW photovoltaic system, a 1MW wind power system, and a 1.6MW / 0.8MWh energy storage system, the energy storage system only needs to discharge all the electricity generated by the photovoltaic and wind power. Adjusting the flexible load by 20% can achieve the goal of a small total investment while maintaining a relatively good return. The total investment cost of this scheme is 14,690,000 yuan, 18% lower than the original scheme, with a return rate of 19.18%, an annual return of 4,143,436 yuan, a payback period of 5.67 years, and a profit of 46,325,544 yuan over 20 years. Specifically, the payback period is shortened by 9%, the return rate is increased by 1.5%, and the total return is reduced by 6%. This scheme represents the lowest total investment while maintaining a good return. The source-grid-load-storage curve for the minimum investment scheme is shown below. Figure 6 As shown.

[0269] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. An optimal design method for a source-grid-load-storage system, characterized in that, Includes the following steps: Step S1: Prepare relevant data on photovoltaic and wind power generation in a certain region, and perform collaborative modeling based on the relevant data on photovoltaic and wind power generation to obtain a dynamic joint output model of the two power sources; Step S2: Using payback period and annual return as objective functions, the NSGA-II algorithm is used to perform multi-objective iterative optimization of the dynamic joint output model, and the output includes the optimal configuration scheme of photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load call ratio. Step S3: Based on the relevant data and dynamic combined output model of photovoltaic and wind power generation in a certain region, manually calculate the payback time and annual income under different system scales, obtain multiple settings combinations of photovoltaic capacity, wind power capacity, and energy storage system capacity, and draw curves of annual income and payback time for different energy storage construction scales based on the manual calculation results. Step S4: Compare the manual calculation results of Step S3 with the output results of the NSGA-II algorithm in Step S2, adjust the output results of the NSGA-II algorithm, and output the optimal configuration scheme of the source-grid-load-storage system, including the optimal combination of photovoltaic and wind power capacity, energy storage system capacity, and flexible load dispatch ratio. In step S1, a dynamic joint output model of photovoltaic and wind power is obtained by introducing an LSTM model for collaborative modeling. For the obtained dynamic joint output model, a dynamic complementarity index is introduced. R complement To quantify the coordination between photovoltaic and wind power output, a load response rate index is introduced. R load It reflects the dynamic matching degree between the output of photovoltaic and wind power generation and the load demand, and optimizes the scheduling of photovoltaic and wind power generation based on the two indicators; Dynamic complementarity index R complement The expression is: Load response rate index R load The expression is: in: P demand (t): Load demand at time t; T is the total statistical duration; P PV (t): The LSTM model predicts at time... t Photovoltaic power generation output; P Wind (t): Wind power generation output predicted by the LSTM model at time t; P total (t): The time difference between the two power sources t The combined efforts .

2. The optimal design method for a source-grid-load-storage system according to claim 1, characterized in that, In step S1, a dynamic joint output model of photovoltaic and wind power is obtained by introducing an LSTM model for collaborative modeling, including the following steps: Step S11: Preprocess the collected historical power generation data of photovoltaic and wind power and related meteorological data; Step S12: Train the LSTM model using preprocessed historical power generation data and relevant meteorological data, optimize the LSTM model parameters, and output the trained LSTM model. The LSTM model includes a sequentially connected input layer, hidden layer and output layer. The input layer is time series features, including historical power generation, sunshine and wind speed. The hidden layer includes multiple LSTM units. The output layer is the predicted value of future power generation. Step S13: Use the trained LSTM model to predict the output of photovoltaic and wind power in future periods, and combine the power output prediction results of time series to construct a dynamic joint output model of photovoltaic and wind power.

3. The optimal design method for a source-grid-load-storage system according to claim 1, characterized in that, Step S2 involves using the NSGA-II algorithm to perform multi-objective iterative optimization of the dynamic combined power output model, outputting an optimal configuration scheme that includes the combination of photovoltaic and wind power capacity, energy storage system capacity, and flexible load dispatch ratio. This specifically includes the following steps: Step S21: Initialize the population, set grid parameter constraints, generate an initial population of individuals with photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load dispatch ratio, and use the initial population as the parent population; Step S22: Construct a fitness function based on the set objective function, calculate the fitness value of each individual in the parent population, and perform non-dominated sorting of individuals in the parent population based on the fitness values, dividing the parent population into multiple non-dominated levels. Step S23: Calculate the crowding distance for each individual in the non-dominant tier and sort them; Step S24: Based on the non-dominance level and crowding distance results, select parent individuals from the parent population for crossover and mutation operations to generate new individuals to form the offspring population. Step S25: Merge the parent population and the offspring population, recalculate the non-dominated sorting and crowding distance, and select individuals from the merged population to form a new parent population. Step S26: Repeat steps S22-S25 until the preset number of iterations is reached or the termination condition is met, and output the optimal solution, which includes the optimal configuration scheme of photovoltaic and wind power capacity combination, energy storage system capacity, and flexible load dispatch ratio.

4. The optimal design method for a source-grid-load-storage system according to claim 3, characterized in that, In step S21, an adaptive sampling initialization strategy is used to initialize the population. Specifically, the initialization strategy involves using photovoltaic and wind power capacity, grid parameter constraints, energy storage system capacity, and flexible load ratio as multidimensional variables. First, an initial population is randomly generated, and then dynamically adjusted using a perturbation factor. The initialization calculation formula for the population is as follows: in, : No. i The initial solution vector for each individual; : Randomly generated population solution vector; : A disturbance factor dynamically adjusted based on historical data or forecast results; P: A forecast distribution matrix related to the physical constraints of the energy storage system; Disturbance factor The calculation formula is: in, σ data The standard deviation of historical data reflects the volatility of photovoltaic and wind power generation or load demand. μ data The mean of historical data; k : Adjustment coefficient, with a value range of 0.1-0.2, dynamically adjusted according to scenario requirements.

5. The optimal design method for a source-grid-load-storage system according to claim 3, characterized in that, For the non-dominated sorting in step S22, during the population evolution process, the weights of each objective function are dynamically adjusted according to the differences in the importance of different optimization objectives at each stage of population evolution. The objective function mapping formula in the non-dominated sorting is: in: Dynamic weighting of payback period; : Dynamic weighting of annual returns; + =1; The objective function is to minimize the payback period. The objective function that maximizes annual returns; : Objective function mapping; Dynamic weighting of payback period Dynamic weighting of annual returns The calculation formula is: in: M The maximum number of iterations; m Let this be the current iteration algebra; ω min Initial weight of payback period; ω max : The maximum weight of payback period.

6. The optimal design method for a source-grid-load-storage system according to claim 3, characterized in that, For step S24, parent individuals are selected from the parent population through a multi-layered selection mechanism based on elite protection, specifically including the following steps: Step S241: Based on fitness values, divide the parent population into two parts: the elite population, which consists of the top N individuals by fitness value. e Individuals; non-elite population, representing the remaining N N e The individuals, where N is the total number of individuals in the parent population; Step S242: For the elite population, directly preserve all individuals to the next generation; Step S243: For the non-elite population, select individuals to be retained in the next generation population using roulette wheel or random selection methods; the probability formula for selecting an individual from the non-elite population is as follows: ， in, P roulette The probability that an individual in a non-elite population will be selected; f i : No. i Fitness values ​​of non-elite individuals; non-elite A collection of non-elite populations; Step S244: Merge the retained elite and non-elite populations proportionally to obtain the selected parent individuals, wherein the proportion of the elite population is... R e The calculation formula is dynamically adjusted according to the current iteration algebra: in: R min , R max These are the minimum and maximum values ​​of the proportion of the elite population, respectively. M The maximum number of iterations; m Let be the current iteration algebra.

7. The optimal design method for a source-grid-load-storage system according to claim 3, characterized in that, In step S24, when performing the crossover operation on the selected parent individuals, the directional weights of the parent individuals are dynamically adjusted based on the relative differences in their objective function values. After the parent individuals' directions are adjusted, the generation formula for the next generation is: in: P new The next generation of individuals generated; P parent1 , P parent2 Each has two parent individuals; α Cross-correlation coefficient, dynamically adjusted according to the changing trend of the objective function, is calculated using the following formula: in, f parent1 , f parent2 These are the objective function values ​​for the two parent individuals, respectively. A constant to prevent the denominator from being zero; it is usually taken as 10. 6 ; When performing mutation operations on selected parent individuals, an adaptive mutation operator based on local dissimilarity is introduced, allowing the mutation probability to be dynamically adjusted according to the local dissimilarity of the population. P mut The adjustment formula is: in: P mut : The current mutation probability; P min_mut λ: Lower bound of the mutation probability; conv : Convergence rate parameter, used to control the adjustment range of mutation probability; Local dissimilarity measure, reflecting the local distribution of the current solution in the population, is calculated using the following formula: in, f neighbor It is the average fitness of the solutions in the neighborhood of the current solution. f current It is the fitness value of the current solution.

8. The optimal design method for a source-grid-load-storage system according to claim 1, characterized in that, Step S3 involves manually calculating the payback period and annual return for different system scales, resulting in multiple combinations of photovoltaic capacity settings, wind power capacity settings, and energy storage system capacity settings. Based on the manual calculation results, curves showing the annual return and payback period for different energy storage construction scales are plotted, including the following steps: Step S31: Based on the dynamic combined output model, construct the combined output curve of the power generation side during future operating periods; Step S32: Compare the combined output curve data with the load demand during the electricity consumption period in real time to obtain the dynamic matching relationship between power generation and load demand, and dynamically schedule the load demand and energy storage system according to the dynamic matching relationship. Step S33: Based on the scale range of photovoltaic and wind power generation and energy storage systems, calculate the payback time and annual income under different scale ranges to obtain multiple settings combinations of photovoltaic capacity, wind power capacity and energy storage system capacity; Step S34: Based on the calculation results of payback time and annual income corresponding to different configuration combinations, and combined with the dynamic scheduling relationship between load demand and energy storage system, plot the curves of annual income and payback time for different energy storage construction scales.

9. The optimal design method for a source-grid-load-storage system according to claim 8, characterized in that, Step S32, the specific strategy for dynamic scheduling of load demand and energy storage system is as follows: Step S321: Compare the combined output curve data with the load demand in real time. When the combined output curve data is lower than the load demand, prioritize the adjustment of flexible loads by removing or postponing some adjustable loads. Step S322: If the adjusted load demand still exceeds the combined output curve data, then activate the energy storage system to supplement power; Step S323: When the combined output curve data is higher than the load demand, the system prioritizes restoring the operation of flexible loads; Step S324: If there is still residual power after the flexible load is restored, the residual power will be used to charge the energy storage device.

Images