Island microgrid multi-time scale optimization scheduling method and device

Abstract

The invention provides an island microgrid multi-time scale optimization scheduling method and device. The method belongs to the technical field of microgrid scheduling. The method comprises: according to a season to which a target scheduling day belongs, selecting a scene subset of a corresponding season from a pre-constructed seasonal wind and light scene set, the scene comprising a wind speed sequence and an irradiance sequence; mapping the wind speed sequence and the irradiance sequence into a wind power available output upper limit and a photovoltaic available output upper limit, inputting the wind power available output upper limit and the photovoltaic available output upper limit into a day-ahead optimization scheduling model, and solving to obtain a day-ahead plan; and repeatedly executing the steps in a rolling mode: solving the intra-day rolling optimization model by taking the minimum output deviation and the minimum operation cost of the current time period relative to the time period corresponding to the day-ahead plan as targets, and obtaining the output plan of the current time period. According to the method, a modeling system fusing seasonal features and extreme scenes is constructed, a day-ahead and intra-day rolling optimization architecture is established, and the problems of large scheduling deviation and response lag are solved while the power supply reliability and the electric energy quality are ensured.

Description

Technical Field

[0001] This invention belongs to the field of microgrid dispatching technology, and in particular relates to a multi-time-scale optimized dispatching method and device for islanded microgrids. Background Technology

[0002] An isolated microgrid is a localized power system capable of operating independently of the main power grid. It typically consists of distributed power sources, energy storage devices, diesel generators, and loads. Remote areas, offshore platforms, and military bases face difficulties connecting to the main grid or extremely high connection costs, necessitating self-sufficiency through local energy resources. In the event of main grid failures or extreme weather events, isolated microgrids can continue to supply power to critical loads, improving power reliability. Simultaneously, regions rich in wind and solar resources can utilize microgrids for local consumption, reducing dependence on fossil fuels and promoting low-carbon development. With the increasing penetration rate of renewable energy, isolated microgrids are widely used as independent power supply systems in remote areas, islands, and offshore platforms.

[0003] However, due to the unique operating environment and complex energy composition, isolated microgrids face severe challenges from multiple uncertainties. On the one hand, wind and solar resources are significantly intermittent and volatile, greatly affected by weather changes and geographical conditions, resulting in large deviations in power generation forecasts. On the other hand, load demand, especially residential and industrial electricity consumption, is random and time-varying, further increasing the difficulty of power balancing. These factors work together to create a dynamic optimization problem with high dimensions and multiple time scales, which traditional centralized dispatching methods based on deterministic models cannot effectively address. This often leads to problems such as frequency deviation, voltage exceeding limits, and excessive energy storage losses, affecting not only power supply reliability and power quality but also increasing operating costs and wasting renewable energy.

[0004] These factors combined create a challenging dynamic optimization problem involving high dimensions and multiple time scales, which traditional centralized scheduling methods based on deterministic models struggle to address effectively. Existing technologies have significant limitations in handling extreme scenario modeling and multi-time scale coordination, resulting in high system operating costs and low renewable energy integration rates. Therefore, it is necessary to provide a scheduling method for isolated microgrids that can balance typical climate characteristics with extreme operating condition boundaries and achieve decoupled optimization across multiple time scales. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-timescale optimization scheduling method and device for isolated microgrids. By constructing an uncertainty modeling system that integrates seasonal characteristics and extreme scenario screening, a day-ahead-intraday rolling coordination optimization architecture is established. While ensuring power supply reliability and power quality under extreme operating conditions, this invention solves the problems of large scheduling deviation and response lag caused by multi-timescale uncertainty coupling, thereby improving the system's operating economy and renewable energy absorption capacity.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a multi-time-scale optimized scheduling method for isolated microgrids, comprising: Based on the season of the target scheduling day, a subset of scenes corresponding to the season is selected from a pre-constructed set of seasonal landscape scenes. The subset of scenes includes: typical scenes and extreme scenes representing the changing characteristics of seasonal landscape resources. Each scene includes: a wind speed sequence that varies with time and an irradiance sequence that varies with time. Based on the output characteristics of wind and solar power units within the isolated microgrid, the wind speed sequence and irradiance sequence are mapped to obtain the time-varying sequences of the upper limits of available wind and solar power output on the target scheduling day. The time-varying sequences of the upper limits of available wind and solar power output, along with load forecasts, are input into the day-ahead optimization scheduling model. The day-ahead optimization scheduling model is solved with the goal of minimizing operating costs to obtain the day-ahead output plans for diesel generators, wind power, solar power, and shared energy storage on the target scheduling day. Within the target scheduling day, the following steps are repeated in a rolling manner according to time sequence: the wind power available output forecast, photovoltaic available output forecast and load forecast for the current time period are input into the intraday rolling optimization model. With the goal of minimizing the output deviation and operating cost relative to the previous day's output plan, the intraday rolling optimization model is solved to obtain the intraday output plan for diesel engines, wind power, photovoltaics and shared energy storage for the current time period.

[0007] Furthermore, the steps for constructing the seasonal landscape scene set include: Based on historical wind speed and irradiance data, multiple historical scene sequences were obtained by dividing the data into preset time windows, forming a set of historical scene sequences. ,in Indicates the first A sequence of historical scenes; Historical scene sequence The wind speed and irradiance at each time point were normalized to obtain the corresponding normalized feature sequences. ; Based on each historical scene sequence Constructing seasonal tag vectors from date information ,in, The value is an integer code representing summer, transitional season, or winter; Normalized feature sequences and their corresponding seasonal tags As input to the conditional generative adversarial network, the adversarial objective function is:

[0008] in, For generator, Let E be the discriminator and E be the expectation. As a joint sample of real landscapes, For the true sample distribution, For random noise samples, This is a vector of seasonal labels; By alternating training of the generator With discriminator Until the given seasonal label vector Under these conditions, the distribution of fake samples generated by the generator approximates the real wind-sun joint output distribution, resulting in a large-scale wind-sun joint candidate scene set generated under the drive of random noise and seasonal labels. The candidate scene set is divided into three seasonal subsets: summer, transitional season, and winter, according to seasonal labels. For each seasonal subset, several representative scene curves are automatically selected from the candidate scenes of that season based on a preset clustering algorithm, and the representative scenes are used as typical scenes of that season. Within each seasonal subset, the maximum and minimum values ​​of wind speed and irradiance for all candidate scenes in that season are taken at each moment to obtain the upper and lower envelope curves of wind speed and irradiance. Based on the upper and lower envelope curves, extreme scenes reflecting maximum and minimum power output are constructed. By combining all typical scenes obtained in the same season with the extreme scenes, a subset of seasonal landscape scenes for that season is obtained.

[0009] Furthermore, the steps for solving the day-ahead optimization scheduling model include: Using the time-varying sequences of wind power available output upper limits and photovoltaic available output upper limits corresponding to each scenario in the seasonal wind and solar scene subset, along with the load forecast for the corresponding time period, as input, the objective function of the day-ahead optimization scheduling model is solved to obtain the day-ahead output plan. The objective function of the day-ahead optimization scheduling model is:

[0010] in, For scene indexing, A collection of seasonal landscape scenes. For the scene The probability of occurrence, The total number of time periods for the target scheduling day. For the time period index within the target scheduling day, , , , Scenes Time period The actual output of diesel engines, wind power, photovoltaics, and shared energy storage; For the scene Time period The power of wind and solar power curtailment For the scene Time period The load reduction amount, For diesel engine fuel cost function, The operation and maintenance cost function for wind power, solar power, and shared energy storage. The cost function for the penalty of curtailing wind and solar power. The load reduction penalty cost function; The constraints of the day-ahead optimization scheduling model are:

[0011]

[0012]

[0013]

[0014] in, For device indexing, , For the scene Time period Lower device of efforts, and respectively equipment The minimum and maximum allowable output, and Scenes Time period The charging and discharging power of shared energy storage Time period The system load demand, For the scene Time period State of charge of shared energy storage and These represent the minimum and maximum states of charge allowed for shared energy storage, respectively. and These represent the charging efficiency and discharging efficiency of shared energy storage, respectively.

[0015] Furthermore, the solution steps for the intraday rolling optimization model include: Using the day-ahead plan, wind power available output forecast, photovoltaic available output forecast, and load forecast for the current time period as inputs, the objective function of the intraday rolling optimization model is solved to obtain the output plan for the current time period. The constraints of the intraday rolling optimization model are the same as those of the day-ahead optimization scheduling model. The objective function of the intraday rolling optimization model is:

[0016] in, The set of time periods contained within the current scrolling window. For the time period index within the scrolling window, Time period Operating costs As a carbon emission cost weight, Time period The output of the diesel engine; Weighting for plan deviation penalties The time period obtained by the day-ahead optimized scheduling model The recent work plan, To optimize the time period within the current scrolling window The adjusted output vector.

[0017] Furthermore, multi-time-scale optimization scheduling methods for isolated microgrids also include: Using the rated power capacity and rated energy capacity of shared energy storage as capacity decision variables, and the daily output plan for each time period within a given cycle as input, with the goal of minimizing the overall life-cycle cost while meeting the upper limit constraints of capacity engineering, the improved gray wolf optimization algorithm is used to solve the capacity collaborative optimization model to obtain the optimal combination of shared energy storage power capacity and energy capacity. The objective function of the capacity collaborative optimization model is:

[0018] in, To share the power capacity of energy storage, To share the energy storage capacity, , and These are the cost coefficients per unit power and per unit capacity, respectively. The upper limit constraint for capacity engineering under this objective function is:

[0019]

[0020] in, and These refer to the maximum power capacity and maximum energy capacity that can be constructed for shared energy storage, respectively. The power capacity and energy capacity of shared energy storage constitute a capacity vector: ,by The fitness function is minimized using an improved gray wolf optimization algorithm to solve the objective function of the capacity collaborative optimization model. The optimal capacity vector is obtained through iterative updates of the positions of the alpha wolf, sub-optimal individuals, and second-best individuals. ,in, and These represent the optimal rated power capacity and optimal rated energy capacity of the shared energy storage, respectively.

[0021] Furthermore, multi-time-scale optimization scheduling methods for isolated microgrids also include: Using the optimal capacity parameters of shared energy storage and Using the intraday power output plan as input, a multi-entity energy trading model is constructed between the shared energy storage operator and multiple microgrid users. With the joint objective of minimizing the total operating cost of users and maximizing the revenue of the shared energy storage operator, an augmented Lagrangian function is constructed by introducing consistency constraints, and the Nash equilibrium solution between microgrid users and shared energy storage is obtained by using the alternating direction multiplier method. In the game step described above, let the user... In time period The charging power and discharging power are respectively ; The global charging power and discharging power of shared energy storage are as follows:

[0022] Define the consistency error as:

[0023] Shared energy storage operators during the time period The payoff function is:

[0024] in Time period set for operators The price of charging and discharging services; user The operating cost function is: .

[0025] Furthermore, the multi-time-scale optimization scheduling method for isolated microgrids also includes: Q-learning-based real-time control steps: State vectors are constructed using real-time measurement information from wind power and photovoltaic power generation units as well as critical loads. The state vector includes at least one or more of the following: current system frequency, wind power output, photovoltaic power output, and shared energy storage charge state. The blade pitch angle, electromagnetic torque, shared energy storage charging and discharging power regulation, and interruptible load connection status are taken as the action. An instant reward is constructed based on the increase in power generation, the reduction in power point tracking error, and the improvement in frequency deviation. The Q-learning method is used to establish the state-action value function. And iteratively update according to the following formula:

[0026] Within each real-time control cycle, based on the updated Q-value function in the current state... Select below The largest action is taken as the optimal control action, and the operating points of the wind power and photovoltaic power generation units are adjusted so that they achieve maximum power point tracking near the planned output for the current time period. When a system frequency deviation is detected Exceeding the preset threshold At that time, the adjustable power of the interruptible load is calculated according to the following formula: And within the set range, the corresponding interruptible load can be cut off or restored; in, For learning rate, As a discount factor, and The first Second and third The Q-value function at the next iteration In the state Next action The new state that is obtained later For candidate actions in the new state, For the first System frequency deviation per control cycle For frequency deviation threshold, This is the frequency-load regulation coefficient. This refers to the interruptible load power that needs to be adjusted within the current control cycle.

[0027] In a second aspect, the present invention provides a multi-time-scale optimized scheduling device for isolated microgrids, comprising: The scene determination module is used to select a subset of seasonal scenery scenes corresponding to the season from a pre-built set of seasonal scenery scenes based on the season to which the target scheduling day belongs. The subset of seasonal scenery scenes includes typical scenes and extreme scenes that represent the changing characteristics of scenery resources in that season, and each scene contains a wind speed sequence that varies with time and an irradiance sequence that varies with time. The day-ahead scheduling module maps wind speed and irradiance sequences to time-varying sequences of wind and solar power units within the isolated microgrid, based on their output characteristics. It then inputs these time-varying sequences, along with load forecasts, into the day-ahead optimization scheduling model. The model is then solved to minimize operating costs, yielding the day-ahead output plans for diesel generators, wind power, solar power, and shared energy storage on the target scheduling day. The intraday scheduling module is used to repeatedly execute the following steps in a rolling manner according to time sequence within the target scheduling day: input the wind power available output forecast, photovoltaic available output forecast and load forecast for the current time period into the intraday rolling optimization model, and solve the intraday rolling optimization model with the goal of minimizing the output deviation and operating cost of the current time period relative to the corresponding time period of the previous day's output plan, to obtain the intraday output plan of diesel engine, wind power, photovoltaic and shared energy storage for the current time period.

[0028] In a third aspect, the present invention provides an electronic device including a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement a multi-time-scale optimized scheduling method for isolated microgrids.

[0029] In a fourth aspect, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements a multi-time-scale optimized scheduling method for isolated microgrids.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: The multi-time-scale optimized scheduling device, electronic device, and computer-readable storage medium for isolated microgrids provided by this invention also solve the problems raised in the background section.

[0031] 1. This invention introduces seasonal wind and solar resource characteristics. Before scheduling begins, it selects typical and extreme scenarios highly consistent with the actual meteorological fluctuations of the target scheduling day based on the season. This ensures that the dynamic changes in wind speed and irradiance are accurately reflected in the scheduling model, fundamentally solving the problem of large scheduling deviations and insufficient robustness caused by "using general scenarios to replace seasonal differences" in traditional methods. During the day-ahead phase, a multi-scenario optimization model is constructed using seasonal wind and solar scenarios on a subset. This model can simultaneously generate coordinated output decisions for diesel engines, wind power, photovoltaics, and shared energy storage under various potential meteorological conditions, giving the scheduling scheme forward-looking and optimal performance in a predictive sense. For unavoidable prediction errors and disturbances during operation, this invention uses real-time wind power forecasts, photovoltaic forecasts, and load forecasts as inputs during the intraday phase. It continuously corrects the aforementioned day-ahead plan through a rolling window, ensuring that intraday output can respond quickly to real-time changes while maintaining the overall trend of following the day-ahead plan. This achieves a consistent, dynamically executable scheduling process across multiple time scales. This invention has good adaptability and feasibility in different seasons and weather conditions, enabling isolated microgrids to achieve safe, economical and stable operation under conditions of high proportion of renewable energy.

[0032] 2. This invention constructs a set of typical and extreme seasonal wind and solar scenarios, enabling the scheduling model to simultaneously consider common weather patterns and extreme meteorological conditions. It comprehensively characterizes the random fluctuations, rapid changes, and extreme boundaries of wind and solar resources, significantly improving the robustness of scheduling to uncertainties. During the day-ahead scheduling phase, by solving for the probability-weighted solutions of seasonal scenarios, the planned output of diesel engines, wind power, photovoltaics, and shared energy storage minimizes costs under various possible wind and solar conditions. This improves upon the problems of insufficient safety margins, frequent wind and solar curtailment, and high costs caused by traditional single-scenario or deterministic forecasting scheduling. During the intraday phase, this invention employs a rolling optimization approach. Whenever new forecast information arrives, the optimal output plan within the current window is re-solved, and the day-ahead plan is added as a deviation reference term to the objective function. This ensures that the operating plan no longer relies on static forecasts but dynamically aligns with real-time measurements and the latest forecasts, guaranteeing both the global consistency of day-ahead scheduling and the response speed of intraday scheduling. Through this multi-timescale, seasonally sensitive joint optimization mechanism, this invention can maintain low operating costs and stable renewable energy utilization even under conditions of severe wind and solar fluctuations and large forecast deviations.

[0033] 3. The shared energy storage capacity collaborative optimization and trading game mechanism proposed in this invention balances long-term investment planning with short-term operational economics in isolated microgrid scenarios, effectively improving the rationality and utilization efficiency of energy storage configuration. By using the sum of capacity investment cost and the total daily operating cost obtained based on day-ahead and intraday scheduling results as the optimization objective, and the rated power capacity and energy capacity of shared energy storage as decision variables, this invention can automatically search for the optimal energy storage scale that meets engineering constraints, avoiding over-investment or insufficient capacity problems caused by traditional manual experience-based configuration. After capacity configuration is completed, this invention further constructs an energy trading model between the operator and multiple microgrid users. Through the alternating direction multiplier method, it achieves a balance between local optima and global consistency in the charging and discharging decisions between each user and shared energy storage, thereby forming the optimal trading strategy in the sense of Nash equilibrium. This mechanism ensures that shared energy storage obtains stable returns while ensuring that different users participate in energy storage scheduling reasonably according to marginal benefits, further reducing the overall operating cost. Through the integrated design of capacity planning and transaction scheduling, this invention enables isolated microgrids to be optimized collaboratively at the capacity and operation levels, significantly improving system economy, shared energy storage utilization, and fairness among multiple stakeholders. Attached Figure Description

[0034] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of the multi-time-scale optimization scheduling method for isolated microgrids according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a multi-time-scale optimized scheduling device for isolated microgrids according to an embodiment of the present invention; Figure 3 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0035] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0036] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0037] Example 1 The following examples are based on Figure 1The method structure shown, based on the operational background of isolated microgrids, elaborates on the entire process of constructing wind-solar combined power output scenarios, extracting typical and extreme scenarios, multi-timescale scheduling, energy storage capacity optimization, and energy storage trading coordination. The entire method presents a complete chain from data preprocessing, model training, typical scenario generation, scheduling optimization to real-time control, capacity coordination, and economic trading.

[0038] The multi-time-scale optimization scheduling method for isolated microgrids provided by this invention includes the following steps S1 to S3.

[0039] In step S1, based on the season to which the target scheduling day belongs, a subset of seasonal landscape scenes corresponding to the season is selected from the pre-constructed set of seasonal landscape scenes. The subset of seasonal landscape scenes includes typical scenes and extreme scenes that represent the changing characteristics of landscape resources in that season, and each scene contains a wind speed sequence that varies with time and an irradiance sequence that varies with time.

[0040] The implementation process begins by constructing a set of historical scene sequences from historical meteorological data such as wind speed and irradiance. Based on historical wind speed and irradiance data, multiple historical scene sequences are obtained by dividing the data into preset time windows, forming a set of historical scene sequences. The historical data covers multiple seasons, and each data point consists of wind speed and irradiance values ​​within a continuous time window. This dataset R is represented as:

[0041] in Indicates the first A sequence of historical scenes, Represented as:

[0042] Data for each historical scene sequence Include There are 1 time points, and the data at each time point is a two-dimensional vector. :

[0043] in This represents the wind speed at time i in the historical scene sequence n. This indicates the light intensity at the same moment.

[0044] Historical scene sequence The wind speed and irradiance at each time point were normalized to obtain the corresponding normalized feature sequences. Specifically, to eliminate the differences in dimensions of different physical quantities, wind speed and light intensity are normalized separately. The normalization formula is written as:

[0045] Normalized two-dimensional feature vector for:

[0046] Normalized data from all scenarios constitute a new dataset. This is used for training subsequent models.

[0047] To account for seasonal differences, each historical scene is assigned a seasonal label L based on its time tag, encoded using integers: 0 for summer, 1 for transitional seasons, and 2 for winter. This is represented as a vector:

[0048] This label serves as input to the subsequent conditional generative adversarial network, enabling the generator to learn the conditional distribution of wind turbine and photovoltaic power output variations under different seasons.

[0049] After data normalization and seasonal label construction, a conditional generative adversarial network (CGAN) is introduced to fit the probability distribution of combined wind and solar power output under different seasonal conditions. The CGAN consists of a generator G and a discriminator D, and its adversarial value function is expressed as follows in this embodiment:

[0050] in, This represents the expectation operation. This represents the probability distribution of real landscape scene data, corresponding to the normalized distribution of the joint samples of historical wind speed and illumination. This represents a real-world landscape scene sampled from this distribution; These are conditional variables used to represent seasonal information (such as summer, transitional season, winter, etc.) and to control the seasonal characteristics of the generated data. It is random noise; In seasonal conditions The following are fake samples generated by the generator. The probability distribution for generating samples; For the discriminator under seasonal conditions The discrimination score against the real sample is given below. Let G be the discrimination score of the discriminator on the generated sample. The meaning of the above adversarial objective is: with the generator G fixed, the discriminator D enables itself to distinguish between real and fake scenes by maximizing the score of real samples and minimizing the score of generated samples; with the discriminator D fixed, the generator G minimizes this objective function to make the discrimination score of the generated samples as close as possible to the real samples, so that the generated landscape scene gradually approaches the distribution of the real scene under different seasonal conditions.

[0051] After training CGAN, the trained generator G is used as a scene generator. During the runtime phase, a random noise vector Z and the corresponding season label L are prepared for each scene to be generated. These two are input into the generator G, which will then generate the scene based on... This combination outputs a landscape scene that matches the characteristics of the season. Repeating the above steps allows for the batch generation of scene sets. A large number of samples. The output of generator G is usually normalized data, so this data needs to be denormalized according to the aforementioned rules to restore the wind speed and illumination values ​​in actual units. Generated scene set As a large-scale, seasonally representative collection of combined wind and light landscape scenes, its core content can be understood as a data table: each row represents a combined wind and light landscape output scene curve, and each column represents a specific point in time.

[0052] Through iterative optimization of the aforementioned value function, the generator G learns the distribution characteristics of the real scene under the condition of L. After training, by inputting random noise Z and seasonal label L during the runtime phase, a large-scale set of seasonally representative landscape scenes can be generated. To ensure the physical availability of the generated data, the output sequence is denormalized, thus obtaining large-scale joint wind and light scene samples that meet the characteristics of different seasons.

[0053] After labeling historical landscape data with seasonal tags by date and normalizing it, a conditional generative adversarial network is used to generate corresponding normalized landscape scene sets under summer, transitional season, and winter conditions, denoted as . , and Subsequently, K-means clustering and typical / extreme scene construction were performed on the scene sets for the three seasons respectively to obtain the final seasonal landscape scene sets. This includes scene subsets corresponding to summer, transitional season, and winter. , and .

[0054] Specifically, in the large-scale scene set output by the generator In the generated scene set, there may be high redundancy among the sequences, thus requiring clustering for dimensionality reduction. (include: , and After normalization, the K-means algorithm is used to process the data. , and Clustering is performed separately, using the joint time series of wind speed and illumination as features for unsupervised scene classification. Several candidate cluster numbers are set, and the optimal cluster number is obtained by calculating the clustering evaluation index under each candidate cluster number. This divides the scene set into Each cluster is obtained, and the corresponding clusters are obtained. There are 1 cluster centers. Let be the 1st cluster center. The sample set in each cluster is Its cluster center is , where index To construct a set of typical scenarios, this embodiment selects the most representative real-world scenario from each cluster as the typical scenario for that cluster: specifically, the cluster is calculated... Each sample With cluster center Euclidean distance The sample with the smallest distance is denoted as And take this sample as the first A typical scenario for this class. Sequentially, for all... Each cluster performs the above operations to obtain the result from... Typical scenario set composed of typical scenarios :

[0055] Each of them All of these are complete wind and solar power output curves from real historical or generated scenarios, which can be directly used for subsequent scheduling calculations.

[0056] To enhance the model's robustness to extreme weather, an extreme scenario set is constructed from each cluster of scenario samples by taking the maxima and minima time-by-time. The construction of the... Maximum value scenario obtained by taking the maximum value time-by-time in the cluster sample :

[0057] Similarly, construct the first Minimum scenarios obtained by taking the minimum value time-by-time in the cluster sample Writing the Ultimate Extreme Scenario Collection :

[0058] Typical and extreme scenarios are combined to form the final scene set for any given season. : That is, a subset of the scenery and landscapes of each season. , and The seasonal landscape scene set consists of subsets of seasonal landscape scenes from the three seasons. .

[0059] To form a complete multi-timescale optimization input, this embodiment assigns a probability weight p to each scene. s p can be set according to the frequency ratio of each cluster of scene samples or in an equal probability manner. s This is used to construct scene partitioning and ensures it through normalization.

[0060] To ensure the seasonal adaptability of the scheduling scheme, this embodiment has independently generated and clustered corresponding seasonal scene sets for summer, transitional season, and winter during the scene construction phase. , and In actual operation, a subset of seasonal landscape scenes corresponding to the current season is selected from the three seasonal scene sets mentioned above, based on the season to which the scheduling day belongs, as the input to the scheduling model, thereby ensuring that the scheduling scheme is consistent with the actual seasonal situation.

[0061] In step S2, the wind speed sequence and irradiance sequence are mapped to the time-varying sequence of the upper limit of available wind power output and the upper limit of available photovoltaic output on the target scheduling day according to the output characteristics of wind power and photovoltaic units in the islanded microgrid. The time-varying sequence of the upper limit of available wind power output and the upper limit of available photovoltaic output, along with the load forecast, are input into the day-ahead optimization scheduling model. The day-ahead optimization scheduling model is solved with the goal of minimizing operating costs to obtain the day-ahead output plan of diesel generators, wind power, photovoltaics, and shared energy storage on the target scheduling day.

[0062] After constructing the set of typical and extreme landscape scenarios with seasonal characteristics, this embodiment enters the day-ahead scheduling phase. Day-ahead scheduling uses a 24-hour optimization cycle and a 15-minute time step, encompassing a total of... Each time period. The recent scheduling is based on scenario sets. ( , or With the goal of minimizing the weighted expected cost of all scenarios in the model, the output trajectories of various types of equipment, including diesel engines, wind turbines, photovoltaics, energy storage, and peak shaving and valley filling strategies, are optimized collaboratively over the next day.

[0063] The objective function for the current scheduling is:

[0064] The total number of time periods for the target scheduling day. The time period index for the target scheduling day is determined by the scenario set. The provided wind speed and sunlight time series only reflect external environmental conditions. To correlate the environmental scenario with the available power output of the unit, this embodiment uses the power characteristic curves of the wind turbine and photovoltaic module to correlate the scenario. Next, time wind speed and light intensity Mapped to the upper limits of available output for wind turbines and solar power, respectively. , denoted as:

[0065] in Let be the power curve function of the wind turbine. Let be the output power function of the photovoltaic module. In day-ahead and intraday scheduling optimization, the actual output of wind turbines and photovoltaics are determined as decision variables under the above upper limit constraint, satisfying:

[0066] This enables unified modeling from environmental scenarios to unit output: scenario sets The upper limit of available output at each moment is determined, and the scheduling optimization selects the optimal unit output trajectory within the range of this upper limit.

[0067] In the objective function of the current scheduling, each cost item is defined as follows: Diesel engine fuel costs for:

[0068] in, , , The fitting coefficient for the fuel cost of diesel generators. For the scene Next, time The active power output of a diesel generator.

[0069] Equipment operation and maintenance costs for:

[0070] in, , , These are the unit operation and maintenance cost coefficients for wind turbines, photovoltaic systems, and energy storage systems, respectively. , , Scenes Next, time The actual active power output of the wind turbine, the actual active power output of the photovoltaic array, and the net charging and discharging power of the energy storage system.

[0071] The penalties for wind and solar power curtailment and load reduction are respectively ,in For the scene Next, time The total power of wind and solar power curtailed. For the scene Next, time The amount of load reduction.

[0072] In the current scheduling constraints, the physical output of all types of equipment must meet upper and lower bound limits, as expressed in the following formula:

[0073] in, Representing a scene Next, time equipment Those who have made contributions A combination of wind turbines (WT), photovoltaic (PV), diesel generators (DG), and energy storage (B). and respectively equipment The minimum and maximum allowable output are used to ensure that various types of equipment operate within physically feasible limits.

[0074] The system power must be kept in real-time balance at every moment, and the relationship is as follows:

[0075] in, For the scene Next, time The actual output power of the fan This represents the actual output power of the photovoltaic array. This refers to the output power of the diesel engine. and For energy storage systems at time The discharge power and charging power, To meet system load requirements, This is the load reduction amount, and together these amounts ensure the supply and demand balance of the system in each time period.

[0076] The energy state of an energy storage system needs to be dynamically updated over time, and the update equation is as follows:

[0077] in, For the scene Next, time The state of charge of energy storage, This represents the state of charge at the previous moment. and These refer to the charging efficiency and discharging efficiency of energy storage, respectively. This is the scheduling step size, used to calculate energy increases or decreases.

[0078] To ensure that energy storage remains within a safe operating range, its state of charge further satisfies:

[0079] in, and These represent the minimum and maximum states of charge that energy storage can tolerate, used to prevent overcharging or over-discharging of energy storage.

[0080] By solving the above day-ahead scheduling model, we can obtain the results for each scenario. And every moment The optimal output trajectory of various generating units and the change trajectory of energy storage state of charge are used to form the baseline plan for the next day, which will serve as the reference input for intraday rolling scheduling.

[0081] In step S3, within the target scheduling day, the following steps are repeatedly executed in a rolling manner according to time sequence: the wind power available output forecast, photovoltaic available output forecast and load forecast for the current time period are input into the intraday rolling optimization model. With the goal of minimizing the output deviation and operating cost of the current time period relative to the corresponding time period of the previous day's output plan, the intraday rolling optimization model is solved to obtain the intraday output plan of diesel engines, wind power, photovoltaic and shared energy storage for the current time period.

[0082] Entering the intraday rolling scheduling phase aims to refine the plan based on real-time observation data and short-term forecasts while maintaining the overall trend of the daily plan. In this embodiment, a moving time window approach is used to refine the daily plan after entering the intraday rolling scheduling phase. Taking a 15-minute time step and a 4-hour window length as an example, the current rolling time is denoted as... The time period covered by the window is ,in Within this window, based on the latest wind turbine available power forecast, photovoltaic available power forecast, and load forecast data, as well as the planned output trajectory obtained from day-ahead scheduling, the unit output, energy storage charging and discharging power, wind and solar curtailment power, and load reduction amount for each time period within the window are used as decision variables. The constraints are consistent with day-ahead scheduling, and the following intraday optimization objective function is constructed:

[0083] in, This is an index for discrete time periods within a daily rolling window. , The set of time periods contained within the current scrolling window. For the first Operating costs for a given period of time As carbon cost weight, For the diesel engine to output power during this period, Weighting for plan deviation penalties The time period obtained by the day-ahead optimized scheduling model The recent work plan, To optimize the time period within the current scrolling window The adjusted output vector. The objective function, by taking into account operating costs, carbon emission costs, and deviations from the day-ahead planned trajectory, obtains the optimal output scheme for the entire segment within the window.

[0084] After finding the optimal solution for the current window, only the first time period of this window (i.e., the first time period) is executed. The output decision is used to guide actual operation. The system then advances by a time step to... The window will scroll forward accordingly. Discard the earliest time period in the original window, introduce a new prediction time period and its corresponding prediction data, and repeat the above optimization process on the new window to solve the problem again. This process is repeated until it covers the entire day's operation, achieving a seamless connection between intraday rolling optimization and real-time control.

[0085] The real-time control layer operates on a 5-minute cycle, performing a final power balance correction on the daily rolling scheduling plan provided by the upper layer based on high-precision real-time measurement data. This layer introduces a Q-Learning-based maximum power point tracking (MPPT) controller to perform real-time output optimization for the tidal power generation unit. The core objective of MPPT technology is to adjust the operating state of the power generation system in real time under different flow velocity conditions, ensuring it always operates near the point where it can output maximum power. Taking wind or tidal power generation as an example, the power output is related to factors such as the incoming flow velocity, generator speed, and blade pitch angle. For any fixed flow velocity, there exists an optimal combination of speed and pitch angle that maximizes energy capture efficiency; this combination point is the maximum power point. The task of the MPPT controller is to continuously find and track this operating point during online operation.

[0086] To achieve model-free self-learning control, this embodiment employs a Q-Learning-based MPPT controller to maximize power capture. Q-Learning is a classic model-free algorithm in reinforcement learning, learning the optimal control strategy through continuous interaction between the agent and the environment without relying on a precise mathematical model. Given a state, the agent selects a control action and applies it to the environment. The environment provides immediate feedback, rewarding the agent and initiating a new state. The algorithm iteratively updates the Q-value function to gradually approach the optimal strategy. The Q-Learning-based MPPT controller uses the Q-value function... Description in state Next action And the expected long-term cumulative reward that can be obtained by consistently taking the optimal action thereafter, its update rule is as follows:

[0087] in, It indicates the current operating status of the system and is used to describe the real-time environmental conditions and operating variables of the tidal power generation unit. Specifically, it may include quantitative measurement information such as current flow velocity, generator speed, pitch angle, and output power. Indicates the controller is in state The following adjustable actions can be performed, such as changing the blade pitch angle or adjusting the electromagnetic torque; To perform the action The system then transitions to a new state; The immediate reward for this action can be constructed based on the increase in power generation or the improvement in energy capture efficiency before and after the action, in order to encourage an increase in the output power of the tidal power generation unit. The learning rate has a range of values. This is used to control the extent to which new sampled information covers the original Q value; This is the discount factor, and its value range is... This is used to measure the importance of future rewards relative to current rewards. By iteratively learning using the above update formula within each 5-minute control cycle, the MPPT controller can gradually learn to select the optimal torque or pitch angle command under different flow rates and operating conditions, enabling the tidal power generation unit to operate in the high-efficiency energy capture range for a long time, thereby maximizing the tidal energy capture power at the real-time level.

[0088] While ensuring the tidal power generation unit operates as efficiently as possible, the real-time control layer also needs to address the power balance issues caused by frequency deviations in the isolated microgrid. When system frequency deviations are detected... When the frequency deviation exceeds a preset threshold, this embodiment invokes interruptible loads and other rapid demand response resources to adjust the load-side power according to the magnitude of the frequency deviation. The load adjustment power can be adjusted according to... The calculation yielded, where For the load power that needs to be cut off or restored, This is the frequency-load regulation coefficient. This represents the deviation of the current system frequency from the rated frequency. By combining Q-Learning-based power flow MPPT control with frequency deviation-based rapid load adjustment, the real-time control layer can finely revise the scheduling plan given by the upper layer on a minute-level time scale, further improving the utilization rate of renewable energy while ensuring system power balance and frequency stability.

[0089] The multi-time-scale optimization scheduling method for isolated microgrids provided in this embodiment further includes: using the rated power capacity and rated energy capacity of shared energy storage as capacity decision variables, taking the daily output plan for each time period within a given cycle as input, and aiming to minimize the overall life-cycle cost under the condition of satisfying the upper limit constraint of capacity engineering, using the improved gray wolf optimization algorithm to solve the capacity collaborative optimization model, and obtaining the optimal combination of shared energy storage power capacity and energy capacity.

[0090] To enhance the robust operation of isolated microgrid systems under various scenarios, this embodiment constructs a shared energy storage capacity collaborative optimization model during the capacity planning phase of shared energy storage, based on the aforementioned day-ahead / intra-day scheduling model. The model aims to minimize the total lifecycle cost and optimize the power capacity of the shared energy storage. and energy capacity Perform joint optimization. The objective function is:

[0091] in, For the one-time investment cost function of shared energy storage, For the maximum charge and discharge power capacity of the shared energy storage system, To share the energy capacity of the energy storage system; This represents the total number of scenarios constructed based on the aforementioned CGAN and clustering methods. For scene indexing, For the scene The probability of occurrence, For the scene The total number of corresponding typical intraday time periods; in this embodiment, a typical day uses a 15-minute resolution. The investment cost function is expressed as:

[0092] in, The investment coefficient per unit power capacity. The investment coefficient per unit energy capacity.

[0093] In the formula, , , , , , and cost function , , , The meaning is the same as in the aforementioned daytime scheduling objective function F1, both of which are given a capacity parameter. and scenes Under the given conditions, the solution is obtained through day-ahead and intraday scheduling models.

[0094] It should be noted that the output variable in the capacity collaborative optimization model , , , , and load reduction In each scene and each time period All of the above must satisfy the power balance constraints and shared energy storage operation constraints given in the aforementioned day-ahead scheduling objective function F1, including the power balance relationship between wind, solar, and diesel generator output and load, energy storage charging and discharging power, and the dynamic evolution equation and upper and lower bound constraints of the energy storage state of charge. Capacity variables By limiting the charging and discharging power and state of charge of energy storage, the output decision at different times is indirectly affected, thereby affecting the accumulated operating costs in different scenarios.

[0095] Regarding the scenario set, the scenario set used for capacity collaborative optimization in this embodiment is the same as the aforementioned sub-bar optimization scenario set. Consistent for Scene index in for Total number of scenes For the scene The probability of occurrence.

[0096] In the capacity collaborative optimization model, and As a variable for higher-level decision-making, it must also satisfy the constraints of engineering feasibility:

[0097]

[0098] in, The maximum buildable power capacity for shared energy storage is limited. The maximum achievable energy capacity limit for shared energy storage is determined in advance during the planning stage based on the site's internal capacity and equipment selection. To solve the above-mentioned... and To address the capacity coordination optimization problem, this embodiment employs an improved Grey Wolf optimization algorithm. As a continuous variable to be optimized, with the objective function As the fitness function, iterative search yields results that... The optimal capacity configuration is the one that takes the minimum value.

[0099] For each set of candidate capacity parameters Substituting this into the two-stage scheduling model of day-ahead and intraday scheduling, the corresponding operating costs are obtained under all scenarios, and then added to the investment cost to form the fitness value of F3. Since F3 is a high-dimensional, nonlinear optimization problem with multiple local extrema, this invention introduces an improved Grey Wolf Optimization algorithm (IGWO) that combines Grey Wolf Optimization (GWO) and Beetle Antenna Search (BAS) to efficiently search for the global optimum in the capacity variable space.

[0100] After establishing the capacity collaborative optimization model, in order to solve the objective function... To find the minimum shared energy storage capacity parameter, this embodiment uses an improved gray wolf-beetle antenna optimization algorithm. The rated power capacity of the shared energy storage is then considered. With rated energy capacity Combined into a two-dimensional vector As the position vector of an individual gray wolf, each gray wolf corresponds to a set of candidate capacity configuration schemes. Given a certain capacity vector... Under the condition of calling the aforementioned scenario set The day-ahead scheduling and intraday rolling scheduling models are used in various scenarios. and each time period The corresponding operating output is obtained from the above. and load reduction And substitute it into the capacity co-optimization objective function Calculate the target value of this capacity scheme, and then... As the fitness function of the gray wolf optimization algorithm The lower the adaptability, the better the capacity solution is in terms of the combined significance of investment cost and multi-scenario operation cost.

[0101] During the iterative search process, let the position vectors of the three gray wolves with the lowest fitness in the current population be respectively... The position vectors of the remaining gray wolves are uniformly denoted as... Introduce a shrinkage vector that decreases linearly with the number of iterations. and the element takes the value of random vectors in intervals Construct the coefficient vector , In each iteration, the individual gray wolves, guided by the three alpha wolves, update their intermediate position vectors:

[0102]

[0103] The average of the three vectors above is then used to obtain a new candidate capacity vector:

[0104] Each vector operation is performed component-by-component. To enhance local search capabilities, this embodiment introduces a beetle antennae search mechanism based on the standard gray wolf update: at the current position... Randomly generated unit direction vectors in the vicinity According to the current step size With antenna length Construct two sampling points, left and right. , Calculate separately and ,like Then let the next position follow fine-tuning of direction Otherwise, fine-tune in the opposite direction. , and according to The rule gradually decreases the step size, where This represents the step size decay coefficient, which is between 0 and 1. Through the iterative process combining global guidance and local search described above, the capacity variable is guaranteed to always satisfy... Under the constraints of engineering, the optimal individual in a gray wolf population Converging to make the objective function The minimum capacity combination is selected to obtain the optimal power capacity and energy capacity configuration of the shared energy storage system in this embodiment.

[0105] The multi-time-scale optimization scheduling method for isolated microgrids provided in this embodiment also includes: using the optimal capacity parameters of shared energy storage. and Using the intraday power output plan as input, a multi-entity energy trading model is constructed between the shared energy storage operator and multiple microgrid users. With the joint objective of minimizing the total operating cost of users and maximizing the revenue of the shared energy storage operator, an augmented Lagrangian function is constructed by introducing consistency constraints, and the Nash equilibrium solution between microgrid users and shared energy storage is obtained by using the alternating direction multiplier method.

[0106] After completing the construction of a seasonal wind-solar joint scenario, optimizing day-ahead and intraday multi-timescale scheduling, and obtaining the optimal capacity parameters for shared energy storage through a capacity collaborative optimization model and an improved gray wolf algorithm, this embodiment introduces multiple user nodes within the same isolated microgrid. Each user node can buy or sell electricity through shared energy storage, and different user nodes are game participants with each other. The isolated microgrid as a whole consists of only one node, but multiple nodes coexist within it. Each microgrid user node is denoted as a user. Each user node has its own electricity demand, distributed power sources, and the ability to interact with shared energy storage. The shared energy storage devices are uniformly managed by the operator, who is responsible for setting charging and discharging prices and allocating capacity resources. The solution results of the aforementioned capacity collaborative optimization model give the optimal power capacity and energy capacity configuration of the shared energy storage, denoted as […]. ,in This represents the upper limit of the absolute values ​​of the charging and discharging power of shared energy storage at any given time. This represents the maximum permissible energy storage capacity for shared energy storage. During energy trading and negotiation processes within the operational period, all charging and discharging power variables related to shared energy storage must meet the constraint of not exceeding these optimal capacity parameters, thereby ensuring that operational decisions are consistent with the capacity configuration during the planning period.

[0107] In the time dimension, the operating cycle is discretized into There are several scheduling periods, with period numbers as follows: For any given time period User records The charging power between the shared energy storage and the charging power is The discharge power is .like This indicates that the user is inputting electrical energy into the shared energy storage; if This indicates that the user is obtaining electrical energy from the shared energy storage. Shared energy storage at any given time... The state of charge is denoted as Charging efficiency is denoted as The discharge efficiency is denoted as At any given time period, the power and energy of the shared energy storage must meet the following constraints: for each user node and at each time point. ,have:

[0108] The overall state of charge evolution of shared energy storage is as follows:

[0109] And satisfy the energy constraint:

[0110] in and The minimum and maximum allowable charge ratios for shared energy storage are typically taken as a certain range of the rated capacity.

[0111] Shared energy storage operators at every moment Standardize charging and discharging service prices Fees are collected or paid to all user nodes. Specifically, when a user charges the shared energy storage, they need to pay according to the price. Fees are paid to the energy storage operator; when users discharge or draw energy from the shared energy storage, they receive compensation from the shared energy storage at the same price. The revenue function of shared energy storage is defined as the cumulative difference between the fees charged to each time period and each user node and the operation and maintenance costs, denoted as:

[0112] in For shared energy storage at any time Because for users The operation and maintenance cost function generated from providing charging and discharging services has the same form as the operation and maintenance cost function used in the aforementioned capacity collaborative optimization model, and can often be a function with a linear or quadratic relationship to the charging and discharging power. In this embodiment, the shared energy storage operator's decision is to select a price sequence. The goal is to maximize revenue while meeting the power and energy constraints of energy storage devices. .

[0113] For each user node within an isolated microgrid At any moment The local operating variables include the diesel engine output determined by the scheduling model. Actual output of wind power and photovoltaic power , Controllable load reduction Curtailed wind and solar power As well as the charging and discharging power involved in interacting with shared energy storage. , User node Operating costs include fuel costs, operation and maintenance costs, curtailment penalty costs, load shedding penalty costs, and fees paid to or received from shared energy storage. The total cost function for the entire operating cycle is defined as:

[0114] in The fuel cost function for diesel engines is typically expressed in terms of output. The quadratic function form; For user nodes All devices at any time Maintenance and operating costs; The penalty cost for curtailing wind and solar power is used to characterize the economic impact of unutilized renewable energy on users. The penalty cost for load reduction is used to reflect the negative effects of insufficient power supply. (Last item) This represents a user's expenditure or income during a transaction with shared energy storage. A positive value indicates payment when a user charges the energy storage, while a negative value indicates compensation received when a user draws electricity from the storage. User Node The behavioral objective is to optimize its own cost function while satisfying local power balance constraints, unit output constraints, energy storage power boundary constraints, and other operational safety constraints. Minimum.

[0115] To achieve global coordination while preserving the private operational information of each user node, this embodiment introduces a global charging and discharging power variable, treating the shared energy storage as the coordinator or "leader," and each user node as a "follower." At any given time... Define the global charging and discharging power provided by shared energy storage to all users as follows:

[0116] Since there is actually only one shared energy storage system, its charging and discharging power cannot be freely set by each user; instead, it must satisfy the consistency relationship between the global and local variables. This relationship can be expressed through a consistency constraint, which requires that the global variable equal the sum of the local charging and discharging powers of all users. To solve this problem using the alternating direction multiplier method, a consistency error is introduced:

[0117] Dual variables Lagrange multipliers used to characterize consistency constraints, penalty factors This is used to control the weight of the error term in the augmented Lagrangian function. Thus, the augmented Lagrangian function of the entire game system can be expressed as:

[0118] The first term is the sum of the operating costs of all user nodes, the second term is the revenue of the shared energy storage operator, and the third term is the dual term and the secondary penalty term corresponding to the consistency constraint. Through the analysis of... The distributed solution allows each user node and the shared energy storage operator to optimize their respective objective functions while maintaining global consistency, ultimately converging to a Nash equilibrium solution.

[0119] In the ADMM iteration process, the solution process is divided into three alternately updated sub-problems. First, in the... In the next iteration, given the global charge / discharge power of the previous round... Dual variables and price series Under these conditions, each user node Solve its local optimization problem independently, that is, minimize the cost function plus the local consistency penalty while satisfying all operational constraints locally:

[0120] In this optimization problem, The form is the same as described above, except that the charging and discharging power variable is treated as the decision variable for the current iteration; the quadratic term in parentheses is used to penalize the deviation between the local total charging and discharging power of the user node and the global charging and discharging power and dual variable of the previous round, thereby promoting the local decision to gradually converge with the global variable. Since each user node's local problem only depends on its own operating parameters and global signals... Therefore, it can be solved in parallel, ensuring the scalability of the algorithm and privacy protection.

[0121] After all user nodes have completed their partial updates, the uploads from each user are aggregated. and Calculate the new global charge / discharge power:

[0122] Based on this, shared energy storage operators reassess their revenue function and update their pricing decisions for the next round. To balance revenue maximization and consistency constraints, the operator's update process can be formulated as the following optimization problem:

[0123] In this expression, the first term is the energy trading revenue obtained by shared energy storage based on the overall charging and discharging demand of users in the current iteration; the second term is the corresponding operation and maintenance cost; and the third term is the consistency penalty term, whose form is consistent with the quadratic error term in the augmented Lagrangian function. By solving the above maximization problem, a new price can be obtained. This price will be used for local decision-making by user nodes in the next round.

[0124] After updating the price and global power, the dual variables are updated according to the standard form of ADMM to gradually converge the consistency error. The update formula for the dual variables is as follows:

[0125] This formula can be understood as superimposing the total charging and discharging power of the current round onto the original dual variables, gradually amplifying the impact of consistency deviation on the Lagrangian function, thereby guiding local and global variables to converge in subsequent iterations. This is achieved by repeatedly executing the three steps of "user node local update—shared energy storage global update and price adjustment—dual variable update" until the convergence condition is met, for example...

[0126] in This is a pre-set convergence threshold. When the above conditions are met, the ADMM iterative process is considered to have converged, and the decisions of all user nodes and shared energy storage operators no longer change significantly. At this point, the... That is, it is constituted in the capacity parameter The Nash equilibrium solution for energy trading between user nodes and shared energy storage within a sub-islanded microgrid is presented. This equilibrium solution simultaneously considers user-side operating costs and shared energy storage revenue. Under the premise of ensuring energy storage power and energy constraints, state of charge constraints, and grid security constraints, it achieves fair, economical, and privacy-friendly utilization of shared energy storage resources among multiple users. The result can further serve as a reference power trajectory and price signal for the real-time control layer, providing a coordination benchmark for 5-minute MPPT control and local power correction.

[0127] Through the coordinated operation of multiple modules, including scenario generation, typical and extreme scenario construction, day-ahead scheduling, intraday rolling scheduling, real-time control, shared energy storage capacity optimization, and energy storage trading game theory, this embodiment forms a complete multi-timescale closed-loop energy management system. Starting from data-driven wind and solar scenario construction, this system uses optimized scheduling and real-time control at different time levels as its framework, and extends in depth with energy storage capacity optimization and trading game mechanisms. This enables isolated microgrids to achieve stable, economical, and robust operation under uncertain wind and solar conditions, extreme weather events, load changes, and multi-entity collaborative environments.

[0128] The logical flow of the entire system from top to bottom is as follows: The implementation process begins with historical wind speed and irradiance data, constructing historical scenarios for combined wind and solar power output of wind turbines and photovoltaic modules. These historical scenarios are normalized to a uniform range, enabling them to be recognized and learned by the neural network. To reflect seasonal differences, each historical scenario is assigned a seasonal label, serving as a key input to the conditional generation network, allowing the generator to learn the corresponding wind and solar resource distribution characteristics in different seasons. This forms the foundation of the entire scenario generation process.

[0129] Building upon this foundation, a conditional generative adversarial network (CGAN) is trained, allowing the generator to learn realistic patterns of wind and solar combined output under constraints of random noise and seasonal labels. The large-scale scene set obtained from CGAN is denormalized to restore wind speed and illumination data to physical quantities. Then, K-means clustering is performed, and the optimal number of clusters is determined based on silhouette coefficients. Typical scenes near each cluster center are extracted. These typical scenes reflect common wind and solar curves under different operating modes, summarizing ordinary weather conditions. Maximum and minimum value curves are extracted from each sample class at each time step to form extreme scenes, representing extreme situations under sudden severe weather. The typical and extreme scenes are merged to form the final scene set. Each scenario is assigned a probability weight, making it a unified source of uncertainty input for subsequent scheduling models.

[0130] Once the scenario set is built, it enters the most forward-looking day-ahead scheduling phase. Day-ahead scheduling operates on a 24-hour cycle within the scenario set. and scenario probability Driven by this approach, a multi-scenario optimization model is solved, encompassing diesel engines, wind turbines, photovoltaics, energy storage, and load shedding strategies. This model aims to minimize the expected total cost, balancing fuel costs, operation and maintenance costs, wind and solar curtailment penalties, and load shedding costs, while satisfying a series of physical conditions such as equipment operation constraints, energy storage status constraints, renewable energy availability limits, and power balance constraints. The solution yields the baseline output trajectory P_plan and energy storage status trajectory SOC_plan for each device the following day, providing a global reference for intraday rolling scheduling.

[0131] Once the day begins, based on the latest real-time monitoring data and ultra-short-term forecasts, intraday rolling scheduling updates unit output using fixed-length rolling windows. Each rolling window uses the previous day's plan as a baseline, adjusting the plan through local optimization to adapt to the latest forecasts without deviating too much from the previous day's plan. This process can handle real-time updates of wind speed, irradiance, and load forecasts, thereby enhancing the system's adaptability to forecast errors. After the intraday scheduling solution is completed, the window moves forward one step, and the solution is repeated until the entire daily operating cycle is covered.

[0132] At minute-level timescales, maximum power point tracking (MPPT) regulation is performed through a real-time control layer. Specifically, through Q-Learning's reinforcement learning mechanism, the optimal control strategy can be learned in real-time without precise mechanical or electrical models. This adjusts the wind turbine pitch angle or motor torque, ensuring that the wind turbine and photovoltaic modules operate as close as possible to their maximum available power point. The state-transition-update structure of Q-Learning ensures the model can adapt to rapidly changing natural conditions. Furthermore, when system frequency fluctuates, the real-time layer automatically triggers interruptible loads, proportionally reducing load power according to the frequency deviation to achieve instantaneous stability.

[0133] Meanwhile, to ensure the matching of energy storage scale with microgrid operation needs at the system level, a shared energy storage capacity collaborative optimization model is introduced. Based on scenario sets... The full probability operation results are obtained by searching for energy storage power capacity. With energy capacity The optimal combination is found to maximize the sum of investment cost and multi-scenario costs during the operating period. The nonlinearity and multimodality of this problem are addressed by the improved Grey Wolf Optimization Algorithm (IGWO), where the α, β, and δ mechanisms of the Grey Wolf Algorithm guarantee global search capability, while the BAS local optimization mechanism enhances local approximation capability, ultimately converging to the optimal solution for the capacity parameter.

[0134] Building upon this foundation, this embodiment also considers the usage of shared energy storage among multiple users, establishing a multi-entity energy storage trading game mechanism through the Alternating Directional Multiplier Method (ADMM). Each user possesses its own charging and discharging strategy and cost function. Through alternating coordination of local and global decisions, the energy storage behavior of all users consistently meets the overall operational rules of shared energy storage. ADMM achieves convergence through three iterative steps: local variable update, global consistency variable update, and dual variable update, ensuring that energy storage trading achieves economic rationality and operational consistency among multiple users.

[0135] All the above modules form a closed-loop system from data-driven to real-time control, with the specific chain as follows: historical landscape data → normalization → CGAN training → large-scale seasonal landscape scenes → clustering and dimensionality reduction → typical and extreme scene sets. → Input to the day-ahead scheduling model → obtain the baseline output plan P_plan → intraday rolling scheduling adjusts the output based on the latest forecast → real-time layer executes fast MPPT and frequency control → obtain the actual operating trajectory for the day → the operating trajectory is used as input to participate in the multi-scenario operating cost calculation of capacity optimization F3 → optimize capacity through IGWO → multi-entity energy storage trading ADMM conducts game theory under fixed capacity → obtain the final shared energy storage operation strategy.

[0136] Through the aforementioned complete chain, this embodiment achieves comprehensive energy management across multiple time scales in isolated microgrids: CGAN ensures the realism and diversity of scenarios; K-means and extreme value analysis guarantee the representativeness of scenarios; F1 day-ahead scheduling provides globally optimal plans; F2 intraday rolling adjustment provides a buffer mechanism for prediction errors; Q-Learning real-time adjustment ensures efficient operation of equipment in rapidly changing environments; F3 energy storage capacity optimization ensures resource allocation matches needs; and ADMM energy storage trading game ensures multi-user cooperation and economic viability. The overall solution possesses significant systemic advantages and engineering feasibility.

[0137] Example 2 like Figure 2 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a multi-time-scale optimized scheduling device for isolated microgrids, comprising: The scene determination module is used to select a subset of seasonal scenery scenes corresponding to the season from a pre-built set of seasonal scenery scenes based on the season to which the target scheduling day belongs. The subset of seasonal scenery scenes includes typical scenes and extreme scenes that represent the changing characteristics of the scenery resources in that season, and each scene contains a wind speed sequence that varies with time and an irradiance sequence that varies with time. The day-ahead scheduling module is used to map the wind speed sequence and irradiance sequence to the output characteristics of wind power and photovoltaic units in the islanded microgrid as a time-varying sequence of the upper limit of available wind power output and the upper limit of available photovoltaic output on the target scheduling day. The time-varying sequence of the upper limit of available wind power output and the upper limit of available photovoltaic output, along with the load forecast, is input into the day-ahead optimization scheduling model. With the goal of minimizing operating costs, the day-ahead optimization scheduling model is solved to obtain the day-ahead output plan of diesel generators, wind power, photovoltaics, and shared energy storage on the target scheduling day. The intraday scheduling module is used to repeatedly execute the following steps in a rolling manner according to time sequence within the target scheduling day: input the wind power available output forecast, photovoltaic available output forecast, and load forecast for the current time period into the intraday rolling optimization model, and solve the intraday rolling optimization model with the goal of minimizing the output deviation and operating cost of the current time period relative to the corresponding time period of the previous day's output plan, to obtain the intraday output plan for diesel engines, wind power, photovoltaics, and shared energy storage for the current time period.

[0138] Example 3 like Figure 3 As shown, the present invention also provides an electronic device 100 for implementing a multi-time-scale optimized scheduling method for isolated microgrids; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0139] The memory 101 can be used to store the computer program 103. The processor 102 implements the multi-time-scale optimization scheduling method for isolated microgrids in Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0140] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0141] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0142] The memory 101 in the electronic device 100 stores multiple instructions to implement a multi-time-scale optimized scheduling method for an islanded microgrid, and the processor 102 can execute multiple instructions to achieve the following: Based on the season of the target scheduling day, a subset of seasonal landscape scenes corresponding to the season is selected from the pre-constructed set of seasonal landscape scenes. The subset of seasonal landscape scenes includes typical scenes and extreme scenes that represent the changing characteristics of landscape resources in that season. Each scene contains a wind speed sequence that varies with time and an irradiance sequence that varies with time. The wind speed and irradiance sequences are mapped to the time-varying sequences of the wind power and photovoltaic power generation caps on the target scheduling day according to the output characteristics of wind power and photovoltaic power generation units in the isolated microgrid. The time-varying sequences of the wind power and photovoltaic power generation caps, along with the load forecast, are input into the day-ahead optimization scheduling model. The day-ahead optimization scheduling model is solved with the goal of minimizing operating costs to obtain the day-ahead plan. The day-ahead plan includes the output plans of diesel generators, wind power, photovoltaic power, and shared energy storage on the target scheduling day, as well as the energy storage status of charge. Within the target scheduling day, the following steps are repeated in a rolling manner according to time sequence: Input the wind power available output forecast, photovoltaic available output forecast, and load forecast for the current time period into the intraday rolling optimization model. Solve the intraday rolling optimization model with the goal of minimizing the output deviation and operating cost of the current time period relative to the corresponding time period of the previous day's plan, and obtain the output plan for the current time period. The output plan for the current time period includes the output plan and energy storage status of diesel engines, wind power, photovoltaics, and shared energy storage for the current time period.

[0143] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0144] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0145] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0146] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0147] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0148] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A multi-time-scale optimization scheduling method for isolated microgrids, characterized in that, include: Based on the season of the target scheduling day, a subset of scenes corresponding to the season is selected from a pre-constructed set of seasonal landscape scenes. The subset of scenes includes: typical scenes and extreme scenes representing the changing characteristics of seasonal landscape resources. Each scene includes: a wind speed sequence that varies with time and an irradiance sequence that varies with time. Based on the output characteristics of wind and solar power units within the isolated microgrid, the wind speed sequence and irradiance sequence are mapped to obtain the time-varying sequences of the upper limits of available wind and solar power output on the target scheduling day. The time-varying sequences of the upper limits of available wind and solar power output, along with load forecasts, are input into the day-ahead optimization scheduling model. The day-ahead optimization scheduling model is solved with the goal of minimizing operating costs to obtain the day-ahead output plans for diesel generators, wind power, solar power, and shared energy storage on the target scheduling day. Within the target scheduling day, the following steps are repeated in a rolling manner according to time sequence: the wind power available output forecast, photovoltaic available output forecast and load forecast for the current time period are input into the intraday rolling optimization model. With the goal of minimizing the output deviation and operating cost relative to the previous day's output plan, the intraday rolling optimization model is solved to obtain the intraday output plan for diesel engines, wind power, photovoltaics and shared energy storage for the current time period.

2. The multi-time-scale optimization scheduling method for isolated microgrids according to claim 1, characterized in that, The steps for constructing the seasonal landscape scene set include: Based on historical wind speed and irradiance data, multiple historical scene sequences were obtained by dividing the data into preset time windows, forming a set of historical scene sequences. ,in Indicates the first A sequence of historical scenes; Historical scene sequence The wind speed and irradiance at each time point were normalized to obtain the corresponding normalized feature sequences. ; Based on each historical scene sequence Constructing seasonal tag vectors from date information ,in, The value is an integer code representing summer, transitional season, or winter; Normalized feature sequences and their corresponding seasonal tags As input to the conditional generative adversarial network, the adversarial objective function is: in, For generator, Let E be the discriminator and E be the expectation. As a joint sample of real landscapes, For the true sample distribution, For random noise samples, This is a vector of seasonal labels; By alternating training of the generator With discriminator Until the given seasonal label vector Under these conditions, the distribution of fake samples generated by the generator approximates the real wind-sun joint output distribution, resulting in a large-scale wind-sun joint candidate scene set generated under the drive of random noise and seasonal labels. The candidate scene set is divided into three seasonal subsets: summer, transitional season, and winter, according to seasonal labels. For each seasonal subset, several representative scene curves are automatically selected from the candidate scenes of that season based on a preset clustering algorithm, and the representative scenes are used as typical scenes of that season. Within each seasonal subset, the maximum and minimum values ​​of wind speed and irradiance for all candidate scenes in that season are taken at each moment to obtain the upper and lower envelope curves of wind speed and irradiance. Based on the upper and lower envelope curves, extreme scenes reflecting maximum and minimum power output are constructed. By combining all typical scenes obtained in the same season with the extreme scenes, a subset of seasonal landscape scenes for that season is obtained.

3. The multi-time-scale optimization scheduling method for isolated microgrids according to claim 1, characterized in that, The steps for solving the day-ahead optimal scheduling model include: Using the time-varying sequences of wind power available output upper limits and photovoltaic available output upper limits corresponding to each scenario in the seasonal wind and solar scene subset, along with the load forecast for the corresponding time period, as input, the objective function of the day-ahead optimization scheduling model is solved to obtain the day-ahead output plan. The objective function of the day-ahead optimization scheduling model is: in, For scene indexing, A collection of seasonal landscape scenes. For the scene The probability of occurrence, The total number of time periods for the target scheduling day. For the time period index within the target scheduling day, , , , Scenes Time period The actual output of diesel engines, wind power, photovoltaics, and shared energy storage; For the scene Time period The power of wind and solar power curtailment For the scene Time period The load reduction amount, For diesel engine fuel cost function, The operation and maintenance cost function for wind power, solar power, and shared energy storage. The cost function for the penalty of curtailing wind and solar power. The load reduction penalty cost function; The constraints of the day-ahead optimization scheduling model are: in, For device indexing, , For the scene Time period Lower device of efforts, and respectively equipment The minimum and maximum allowable output, and Scenes Time period The charging and discharging power of shared energy storage Time period The system load demand, For the scene Time period State of charge of shared energy storage and These represent the minimum and maximum states of charge allowed for shared energy storage, respectively. and These represent the charging efficiency and discharging efficiency of shared energy storage, respectively.

4. The multi-time-scale optimization scheduling method for isolated microgrids according to claim 3, characterized in that, The solution steps for the intraday rolling optimization model include: Using the day-ahead plan, wind power available output forecast, photovoltaic available output forecast, and load forecast for the current time period as inputs, the objective function of the intraday rolling optimization model is solved to obtain the output plan for the current time period. The constraints of the intraday rolling optimization model are the same as those of the day-ahead optimization scheduling model. The objective function of the intraday rolling optimization model is: in, The set of time periods contained within the current scrolling window. For the time period index within the scrolling window, Time period Operating costs As a carbon emission cost weight, Time period The output of the diesel engine; Weighting for plan deviation penalties The time period obtained by the day-ahead optimized scheduling model The recent work plan, To optimize the time period within the current scrolling window The adjusted output vector.

5. The multi-time-scale optimization scheduling method for isolated microgrids according to claim 4, characterized in that, Also includes: Using the rated power capacity and rated energy capacity of shared energy storage as capacity decision variables, and the daily output plan for each time period within a given cycle as input, with the goal of minimizing the overall life-cycle cost while meeting the upper limit constraints of capacity engineering, the improved gray wolf optimization algorithm is used to solve the capacity collaborative optimization model to obtain the optimal combination of shared energy storage power capacity and energy capacity. The objective function of the capacity collaborative optimization model is: in, To share the power capacity of energy storage, To share the energy storage capacity, , and These are the cost coefficients per unit power and per unit capacity, respectively. The upper limit constraint for capacity engineering under this objective function is: in, and These refer to the maximum power capacity and maximum energy capacity that can be constructed for shared energy storage, respectively. The power capacity and energy capacity of shared energy storage constitute a capacity vector: ,by The fitness function is minimized using an improved gray wolf optimization algorithm to solve the objective function of the capacity collaborative optimization model. The optimal capacity vector is obtained through iterative updates of the positions of the alpha wolf, sub-optimal individuals, and second-best individuals. ,in, and These represent the optimal rated power capacity and optimal rated energy capacity of the shared energy storage, respectively.

6. The multi-time-scale optimization scheduling method for isolated microgrids according to claim 5, characterized in that, Also includes: Using the optimal capacity parameters of shared energy storage and Using the intraday power output plan as input, a multi-entity energy trading model is constructed between the shared energy storage operator and multiple microgrid users. With the joint objective of minimizing the total operating cost of users and maximizing the revenue of the shared energy storage operator, an augmented Lagrangian function is constructed by introducing consistency constraints, and the Nash equilibrium solution between microgrid users and shared energy storage is obtained by using the alternating direction multiplier method. In the game step described above, let the user... In time period The charging power and discharging power are respectively ; The global charging power and discharging power of shared energy storage are as follows: Define the consistency error as: Shared energy storage operators during the time period The payoff function is: in Time period set for operators The price of charging and discharging services; user The operating cost function is: 。 7. The multi-time-scale optimization scheduling method for isolated microgrids according to claim 6, characterized in that, Also includes: Q-learning-based real-time control steps: State vectors are constructed using real-time measurement information from wind power and photovoltaic power generation units as well as critical loads. The state vector includes at least one or more of the following: current system frequency, wind power output, photovoltaic power output, and shared energy storage charge state. The blade pitch angle, electromagnetic torque, shared energy storage charging and discharging power regulation, and interruptible load connection status are taken as the action. An instant reward is constructed based on the increase in power generation, the reduction in power point tracking error, and the improvement in frequency deviation. The Q-learning method is used to establish the state-action value function. And iteratively update according to the following formula: Within each real-time control cycle, based on the updated Q-value function in the current state... Select below The largest action is taken as the optimal control action, and the operating points of the wind power and photovoltaic power generation units are adjusted so that they achieve maximum power point tracking near the planned output for the current time period. When a system frequency deviation is detected Exceeding the preset threshold At that time, the adjustable power of the interruptible load is calculated according to the following formula: And within the set range, the corresponding interruptible load can be cut off or restored; in, For learning rate, As a discount factor, and The first Second and third The Q-value function at the next iteration In the state Next action The new state that is obtained later For candidate actions in the new state, For the first System frequency deviation per control cycle For frequency deviation threshold, This is the frequency-load regulation coefficient. This refers to the interruptible load power that needs to be adjusted within the current control cycle.

8. A multi-time-scale optimized scheduling device for isolated microgrids, characterized in that, include: The scene determination module is used to select a subset of seasonal scenery scenes corresponding to the season from a pre-built set of seasonal scenery scenes based on the season to which the target scheduling day belongs. The subset of seasonal scenery scenes includes typical scenes and extreme scenes that represent the changing characteristics of the scenery resources in that season, and each scene contains a wind speed sequence that varies with time and an irradiance sequence that varies with time. The day-ahead scheduling module is used to map the wind speed sequence and irradiance sequence to the output characteristics of wind power and photovoltaic units in the islanded microgrid as a time-varying sequence of the upper limit of available wind power output and the upper limit of available photovoltaic output on the target scheduling day. The time-varying sequence of the upper limit of available wind power output and the upper limit of available photovoltaic output, along with the load forecast, is input into the day-ahead optimization scheduling model. With the goal of minimizing operating costs, the day-ahead optimization scheduling model is solved to obtain the day-ahead output plan of diesel generators, wind power, photovoltaics, and shared energy storage on the target scheduling day. The intraday scheduling module is used to repeatedly execute the following steps in a rolling manner according to time sequence within the target scheduling day: input the wind power available output forecast, photovoltaic available output forecast, and load forecast for the current time period into the intraday rolling optimization model, and solve the intraday rolling optimization model with the goal of minimizing the output deviation and operating cost of the current time period relative to the corresponding time period of the previous day's output plan, to obtain the intraday output plan for diesel engines, wind power, photovoltaics, and shared energy storage for the current time period.

9. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement the multi-time-scale optimized scheduling method for islanded microgrids as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the multi-time-scale optimized scheduling method for isolated microgrids as described in any one of claims 1 to 7.

Images