Optical joint virtual energy storage model double-layer optimization scheduling method considering uncertainty

Abstract

The invention discloses an optical joint virtual energy storage model double-layer optimization scheduling method considering uncertainty, and the method considers the uncertainty of photovoltaic power generation, and builds a photovoltaic power generation GAN mathematical model through collecting historical data. Virtual energy storage is introduced, that is, electrochemical energy storage and virtual energy storage form an energy storage system, and double-layer optimization scheduling of a light storage combined system is realized by establishing a light storage combined double-layer optimization scheduling model. And finally, adding an improved strategy to the gravitational search algorithm, and solving the optimal configuration parameters of the optical storage operation model by using the improved gravitational search algorithm. The improved strategy is added on the basis of the gravitational search algorithm, the global search capability of the optimization algorithm is improved, the optimal configuration parameters of the optical storage model are obtained, and reference is conveniently provided for system operation.

Description

Technical Field

[0001] This invention relates to the field of power system dispatching technology, specifically to the field of photovoltaic-storage optimization dispatching technology, and particularly to a two-layer optimization dispatching method for a photovoltaic-storage joint virtual energy model that considers uncertainties. Background Technology

[0002] As the proportion of photovoltaic-storage integrated systems in new power systems increases, the scheduling of these systems faces challenges such as uncertainties in photovoltaic output, load fluctuations, and multi-timescale coupling. This invention focuses on photovoltaic-storage integrated systems, primarily investigating model building and optimized scheduling.

[0003] The establishment of photovoltaic-storage models is generally divided into traditional deterministic scheduling models, probabilistic modeling methods, and time series methods. Traditional deterministic scheduling models cannot handle prediction errors, stochastic probabilistic modeling methods rely on the assumption of accurate probability distribution, and time series methods are insufficient for handling non-stationary photovoltaic output sequences. Currently, the commonly used methods for handling photovoltaic uncertainties are robust optimization, chance constraints, and scenario analysis. The latter two both require describing the probability distribution of random variables.

[0004] The scheduling modes of multi-energy complementary systems can be mainly divided into the following two categories: one is to take electricity as the core, consider energy consumption, and build a multi-energy collaborative system that integrates electricity, gas, heat and cooling. Through the collaborative optimization and joint regulation between energy units, the system can achieve energy cascade utilization and efficiency improvement. The other is to give priority to the use of large-scale clean energy sources such as solar and wind energy, and secondly to give full play to the regulation role of energy storage systems and provide supporting energy storage equipment, thereby integrating the advantages of multiple power sources to stably transmit and efficiently absorb large-scale new energy sources. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a two-layer optimal scheduling method for a photovoltaic-integrated virtual energy storage model that considers uncertainties. This invention considers the uncertainties of photovoltaic power generation and establishes a photovoltaic power generation GAN mathematical model by collecting historical data. Virtual energy storage is introduced; that is, electrochemical energy storage and virtual energy storage constitute an energy storage system. By establishing a photovoltaic-energy storage integrated two-layer optimal scheduling model, the two-layer optimal scheduling of the photovoltaic-energy storage integrated system is achieved. Finally, an improved strategy is added to the gravity search algorithm, and the improved gravity search algorithm is used to solve for the optimal configuration parameters of the photovoltaic-energy storage operation model.

[0006] This invention is achieved through the following technical solution:

[0007] A two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties includes the following steps:

[0008] S1. Collect historical photovoltaic array data and meteorological data, and preprocess the collected data to train and improve the dataset;

[0009] S2. Considering the uncertainty of photovoltaic output, a photovoltaic power generation GAN model is established using the preprocessed dataset, and a multi-scenario photovoltaic output prediction sequence for the next 24 hours is output.

[0010] S3. Establish an electrochemical energy storage model and a virtual energy storage model, wherein the virtual energy storage model includes a transferable virtual energy storage model and a slashable virtual energy storage model;

[0011] S4. Using the photovoltaic power generation GAN model to generate a multi-scenario photovoltaic output prediction sequence for the next 24 hours, as well as the electrochemical energy storage model and virtual energy storage model, establish a photovoltaic-energy storage operation model and generate a preliminary joint scheduling strategy. Adopt a two-layer optimization scheduling strategy on the photovoltaic-energy storage operation model and correct the preliminary joint scheduling strategy based on the real-time collected data.

[0012] S5. Using a gravity search algorithm, incorporating chaotic operators, adaptive gravity decay factors, and population elite particles, the multivariable nonlinear global optimal solution of the photovoltaic-storage operation model is obtained, and the optimal configuration parameters are applied to the photovoltaic-storage system.

[0013] The specific details of step S1 are as follows:

[0014] S1.1 Data Acquisition: Collect real-time output power of photovoltaic panels and multi-source meteorological data, including light intensity, temperature, humidity and wind speed;

[0015] S1.2 Data Preprocessing: Combine deep learning algorithms to process outliers in the collected data. Specifically, select anomaly detection based on autoencoder to identify outliers in the data, and fill in missing data through bidirectional long short-term memory network interpolation.

[0016] S1.3 Constructing the correlation of multi-source data for temporal and spatial composite feature assessment, specifically: quantifying the spatial coupling relationship of preprocessed data using Pearson correlation coefficient, calculating the output correlation of photovoltaic panels in different regions, and extracting the rate of change of light intensity according to the time series;

[0017] Pearson correlation coefficient:

[0018] in These represent the power output in different regions at the i-th time period, where n is the length of the time series. , X and Y are the means of X and Y, respectively; X and Y represent different regions. These are the standard deviations of the two sequences;

[0019] S1.4. Adaptive standardization method is used to standardize the data obtained in S1.3, and the standardization parameters are dynamically adjusted according to the time period characteristics of the data.

[0020] The specific details of step S2 are as follows:

[0021] The photovoltaic power generation GAN model includes a generator and a discriminator;

[0022] The generator architecture is designed as follows:

[0023] The input layer fuses multimodal data, specifically random noise generated by the input layer, historical photovoltaic array data, meteorological data, and photovoltaic array topology information;

[0024] The spatiotemporal feature extraction network processes the standardized data as follows: Temporal feature extraction adopts a Transformer-CNN hybrid structure, using Transformer to capture long-term time dependence, i.e., the impact of seasonal changes on photovoltaic power output, and CNN to extract short-term high-frequency fluctuation features; Spatial feature extraction uses an improved U-Net++ network to process photovoltaic array topology information, using a combination of dense connections and attention mechanisms to identify spatial factors affecting photovoltaic power output and output spatial feature vectors.

[0025] In the Transformer-CNN hybrid architecture, the Transformer module captures long-period dependencies, and the CNN module captures short-period dependencies. The final temporal feature vector of the Transformer-CNN hybrid architecture is:

[0026]

[0027] in For channel splicing, FC() is a fully connected layer. These are the long-term and short-term time series feature vectors, respectively.

[0028] The improved U-Net++ network replaces the traditional convolutions in the U-Net network with depthwise separable convolutions. A topology consistency penalty term is added to the loss function, resulting in the following overall loss function:

[0029]

[0030] Basic loss Using Dice loss, i.e. A and B represent the actual value and the predicted value, respectively. This represents the topology consistency penalty term. ,in To predict probabilities, =5; The background area is represented by h and w, which represent the height and width of the image, respectively.

[0031] Spatiotemporal feature fusion specifically involves introducing an adaptive fusion mechanism to dynamically adjust the fusion weights of spatiotemporal features according to different scenarios and output a joint feature vector.

[0032] The photovoltaic power output sequence prediction is as follows: the generator outputs a multi-scenario photovoltaic power output sequence for the next 24 hours based on the input layer data and joint feature vectors, with a 15-minute granularity.

[0033] The discriminator architecture is designed as follows:

[0034] The multi-scale discriminant structure is as follows: a pyramid discriminant network is used to judge the authenticity of the data generated by the generator from different levels. The shallow network quickly judges the overall distribution characteristics of the data, while the deep network analyzes the detailed features of the data in depth.

[0035] To guide the discriminator in accurately distinguishing between real and fake data and to conversely constrain the generator's optimization, a discriminator loss function is introduced, defined as follows:

[0036]

[0037] In the formula: Wasserstein distance loss is used to measure the difference between the generated data and the real data distribution. Loss of detailed features; , These are the weighting coefficients.

[0038] The specific details of step S3 are as follows:

[0039] The electrochemical energy storage model is as follows:

[0040]

[0041] In the formula: This represents the total energy stored in the battery at time t; , , These represent the battery's self-discharge efficiency, charging efficiency, and discharging efficiency, respectively. , Let represent the charging and discharging power of the battery at time t, respectively. Indicates a time step;

[0042] Virtual energy storage models include portable virtual energy storage models and slashable virtual energy storage models;

[0043] In transferable virtual energy storage, when the value of the load power after transfer is higher than the value before transfer, it is equivalent to the charging behavior of transferable virtual energy storage; otherwise, it is the discharging behavior. The transferable virtual energy storage model is as follows:

[0044]

[0045] In the formula: This represents the load value of transferable virtual energy storage participating in scheduling during time period t. , These represent the load value of transferable virtual energy storage before participating in scheduling and the change in the amount of energy participating in scheduling, respectively. This indicates the transfer status of increased load during time period t. This indicates the transfer status of load reduction during time period t. A value of 0 indicates no transfer, while a value of 1 indicates that a transfer has been performed. , These represent the increase or decrease in load during time period t, respectively. , These represent the upper and lower limits of the change, respectively, and T is the scheduling period;

[0046] In reducible virtual energy storage, the power of load reduction is equivalent to the discharge power of virtual energy storage. The reducible virtual energy storage model is as follows:

[0047]

[0048] In the formula: This represents the amount of virtual energy storage load that can be reduced and participates in scheduling during time period t. , These represent the load that can be reduced from virtual energy storage before it participates in scheduling and the change in the amount of virtual energy storage participating in scheduling, respectively. This indicates the upper limit of the amount of virtual energy storage change that can be reduced.

[0049] The specific details of step S4 are as follows:

[0050] Every night, a preliminary joint scheduling strategy is generated by using the photovoltaic power generation GAN model to generate a multi-scenario photovoltaic output prediction sequence for the next 24 hours, as well as an electrochemical energy storage model and a virtual energy storage model. On the same day, photovoltaic output and energy storage status data are collected in real time, and the current operating status is evaluated at a 15-minute interval. The preliminary joint scheduling strategy is then corrected using a two-layer optimization scheduling strategy.

[0051] The described two-layer optimization scheduling strategy consists of two layers: the upper layer aims to minimize the system operating cost, and the lower layer aims to minimize the net load variance. The objective function of the upper layer is the operating cost. Minimum:

[0052]

[0053] In the formula: , , These are the virtual energy storage dispatch cost, electrochemical energy storage dispatch cost, and photovoltaic dispatch cost, respectively, and their mathematical expressions are as follows:

[0054]

[0055]

[0056]

[0057] In the formula: , These represent the unit dispatch cost coefficients for transferable virtual energy storage and reducible virtual energy storage, respectively. This represents the unit dispatch cost coefficient for electrochemical energy storage. Indicates the light-wasting penalty coefficient; Indicates the output power of photovoltaics. This indicates the upper limit of the photovoltaic output power; , Let t represent the charging power and discharging power of the battery at time t, respectively, and T represent the scheduling period;

[0058] The lower-level objective function is the net load variance. Minimum:

[0059]

[0060] In the formula: , Let represent the net load value during time period t and the average net load value within the scheduling period, respectively. Their mathematical expressions are as follows:

[0061]

[0062]

[0063] In the formula: This represents the net power output of the photovoltaic system during time period t. , This represents the net power of the energy storage system during charging and discharging. This represents the electricity demand during time period t;

[0064] Constraints include upper-level constraints and lower-level constraints, specifically:

[0065] The lower-level constraints include photovoltaic output constraints, energy storage charge and discharge constraints, and energy storage system capacity constraints, among which:

[0066] Photovoltaic output constraints for:

[0067]

[0068]

[0069] In the formula: This represents the active power delivered by the photovoltaic system to the energy storage system during time period t; This indicates the maximum active power of photovoltaic power. This represents the rated power of the photovoltaic system at time t;

[0070] Energy storage charging and discharging constraints are:

[0071]

[0072]

[0073]

[0074]

[0075]

[0076] In the formula: This indicates the energy storage status during time period t; , These represent the charging and discharging efficiencies of the energy storage system, respectively. , These represent the maximum charging and discharging power of the energy storage system, respectively. , These represent the maximum charging and discharging power at time t, respectively. , These represent the maximum and minimum energy storage states of the energy storage system, respectively.

[0077] Energy storage system capacity constraints:

[0078]

[0079] In the formula: This indicates the state of charge of the energy storage system during time period t; This indicates the maximum state of charge of the energy storage system;

[0080] Substituting the power load obtained from the optimization model into the lower-level optimization model, we obtain the power balance constraint of the upper-level constraint, as follows:

[0081]

[0082] In the formula: This represents the electricity demand during time period t.

[0083] The specific details of step S5 are as follows:

[0084] (5a) Based on the gravity search algorithm, the following three improvement strategies are introduced: First, a chaotic operator is added to increase population diversity and search randomness; second, an adaptive gravity decay factor is introduced; finally, a formula for the update rate of elite particles in the population is introduced to increase the population's memory capacity and information exchange capacity, as detailed below:

[0085] The formula for chaotic operators is The particle velocity update formula after adding the chaotic operator is: ;in It is a randomly generated d-dimensional random vector and k represents the number of individuals. It is a factor that controls the range of chaos. It is the current d-dimensional velocity vector. It is the current d-dimensional acceleration vector; The particle velocity after adding the chaotic operator; It is a random number;

[0086] The formula for the adaptive gravitational decay factor is: ;in This is the initial value of the gravitational decay factor. and Scaling factor , , The shift factor is T, the maximum number of iterations is T, and the current number of iterations is t.

[0087] The velocity update formula after introducing elite particles from the population is as follows: ;

[0088] in , It is an adaptive learning factor that changes with the number of iterations. and represents the individual optimal position and the global optimal position in the d-th dimension, respectively, and rand represents a random number;

[0089] (5b) Update the parameters of the gravity search algorithm, generate the configuration parameters of the system's optical storage operation model, and determine whether the constraints are met and calculate the individual fitness value;

[0090] (5c) Perform system optimization to obtain the output P of each device and the system operating cost F. 1t The obtained F 1t Compared to the F obtained last time 1t-1 Compare the results. If the result is lower than the previous iteration value, proceed to step (5d). If the upper limit of the running process has not been reached, proceed to step (5a) and output the result of the previous run.

[0091] (5d) Update the optimal cost F1 and determine whether the number of iterations has reached the upper limit. If the maximum number of iterations has been reached, output the running results and parameters. The output results are the optimal configuration parameters of the optical storage running model. If not, return to step (5a).

[0092] The specific details of step (5b) are as follows:

[0093] 1) Randomly initialize the population and set parameters, including population size N, maximum number of iterations T, and location boundaries. Initial value of universal gravitation and initial value of gravitational decay factor Randomly initialize the positions of individuals in the population;

[0094] 2): Determine if the updated individual position is out of bounds; if so, randomly assign a new position.

[0095] 3): Calculate the fitness value of all individuals in the population;

[0096] 4): Update the global optimal position and individual optimal position Find the optimal objective function value for the population. and corresponding individual locations ,if < ,but = , = , This represents the fitness value after t cycles. The optimal fitness value is obtained after t iterations. This is the globally optimal fitness value;

[0097] 5): Calculate individual mass According to the formula G(t) = G0 × exp(- Calculate the gravitational parameter G(t) using the formula (×t / T). and Calculate the gravitational force acting on you , where R ij (t) = ||X i (t), X j (t)||2, finally calculate the acceleration ;in It is the gravitational parameter at the t-th iteration. Let i be the Euclidean distance between individuals i and j. , and It is a constant. , ,and ; Let represent the gravitational component of the j-th individual on the i-th individual in the d-th dimension at the t-th iteration; This represents the quality of the i-th individual in the t-th iteration; This represents the quality of the j-th individual in the t-th iteration; This represents the position coordinates of the i-th individual in the d-th dimension during the t-th iteration. Let X represent the position coordinates of the j-th individual in the d-th dimension during the t-th iteration. i (t) is the d-dimensional position vector of the i-th individual at the t-th iteration, X j (t) is the d-dimensional position vector of the j-th individual at the t-th iteration. This represents the set of the K individuals with the highest fitness in the current iteration. Let N be a random number, and N be the population size.

[0098] 6): Calculate the adaptive learning factor based on universal gravitation, and the formula... and Update speed and location, V i d (t) represents the movement speed of the i-th individual in the d-th dimension of the decision variable at the t-th iteration;

[0099] 7): Number of iterations ,like (If the condition is met, proceed to step 2); otherwise, end the loop and obtain the optimal objective function value. and optimal position vector .

[0100] A two-layer optimized scheduling system for a combined optical and virtual energy storage model considering uncertainties includes:

[0101] Data acquisition and processing module: Collects historical photovoltaic array data and meteorological data, and preprocesses the collected data to train and improve the dataset;

[0102] A photovoltaic power generation GAN model construction module is used to build a photovoltaic power generation GAN model based on the uncertainty of photovoltaic output and output a multi-scenario photovoltaic output prediction sequence for the next 24 hours.

[0103] The energy storage model construction module establishes an electrochemical energy storage model and a virtual energy storage model, wherein the virtual energy storage model includes a transferable virtual energy storage model and a slashable virtual energy storage model.

[0104] The photovoltaic-storage operation model construction module, using the photovoltaic power generation GAN model to generate a multi-scenario photovoltaic output prediction sequence for the next 24 hours, as well as an electrochemical energy storage model and a virtual energy storage model, establishes a photovoltaic-storage operation model and generates a preliminary joint scheduling strategy. A two-layer optimization scheduling strategy is adopted for the photovoltaic-storage operation model, and the preliminary joint scheduling strategy is corrected based on the real-time collected data.

[0105] Solution module: Employing a gravity search algorithm, incorporating chaotic operators, adaptive gravity decay factors, and population elite particles, the module solves for the multivariable nonlinear global optimal solution of the photovoltaic-storage operation model, obtaining the optimal configuration parameters and applying them to the photovoltaic-storage system.

[0106] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of a two-layer optimal scheduling method for a combined optical virtual energy storage model that considers uncertainties.

[0107] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the two-layer optimization scheduling method for the optical joint virtual energy storage model considering uncertainties.

[0108] The advantages of this invention are as follows: This invention addresses the scheduling optimization of photovoltaic (PV) and energy storage (ESS) systems in power systems. To reduce the impact of PV grid connection uncertainties on system scheduling, it proposes a mathematical model for establishing a PV power generation GAN (Gas-Oriented Animation) system and introduces virtual energy storage to obtain a PV-ESS operation model. It then outputs a multi-scenario PV output sequence for the next 24 hours to provide a reference for joint PV-ESS scheduling. Based on this, a two-layer optimization scheduling strategy is adopted, using an improved gravity search algorithm to obtain the optimal capacity configuration. Specific advantages are manifested in the following aspects:

[0109] First, it takes into account the uncertainty of renewable energy, enhances the accuracy of output modeling of photovoltaic and energy storage operation models, and makes photovoltaic output prediction based on the model, providing a reference for the joint dispatch of photovoltaic and energy storage.

[0110] Secondly, virtual energy storage is introduced on the basis of electrochemical energy storage. Virtual energy storage adopts an energy peak-shifting management mode, which not only improves the operational reliability and economic benefits of the entire system, but also reduces the dependence on physical energy storage devices such as electrochemical energy storage batteries in the process, thereby reducing the negative impact on the ecological environment.

[0111] Third, a two-level scheduling optimization is introduced, which considers different objective functions at the upper and lower levels during the optimization process, thereby reducing the system's operating cost and decreasing the system's net load variance.

[0112] Fourth, an improved strategy was added to the gravity search algorithm to enhance the global search capability of the optimization algorithm and obtain the optimal configuration parameters of the optical storage model, which can provide a reference for system operation. Attached Figure Description

[0113] Figure 1 This is a structural diagram of a photovoltaic-storage combined power generation system;

[0114] Figure 2 Flowchart for improving the gravity search algorithm. Detailed Implementation

[0115] like Figure 1 As shown, the main steps of the optical-storage joint two-layer optimization scheduling method considering uncertainties adopted in this invention are as follows:

[0116] (1) Collect photovoltaic array data, meteorological data, etc., and preprocess the data to train and improve the dataset;

[0117] (2) Considering the uncertainty of photovoltaic output, a photovoltaic power generation GAN mathematical model is established using the preprocessed data and the photovoltaic output sequence for the next 24 hours is output.

[0118] (3) Energy storage includes electrochemical energy storage and virtual energy storage systems. Virtual energy storage is divided into fixed load and flexible load. Flexible load includes transferable load and reduceable load. Virtual energy storage models and electrochemical energy storage models for the two types of flexible loads are established respectively.

[0119] (4) Based on the photovoltaic power generation model and energy storage model, establish a photovoltaic-energy storage operation model, and generate a preliminary joint dispatch strategy by combining the generated photovoltaic output prediction and energy storage status. A two-layer optimized dispatch strategy is adopted for the photovoltaic-energy storage operation model. The preliminary dispatch strategy is revised based on real-time data to form more refined dispatch instructions, with the upper layer based on the overall system operating cost. The objective is to minimize the net load variance, while the lower level uses the minimum net load variance as its objective function.

[0120] (5) The improved gravity search algorithm is used to solve the multivariate nonlinear global optimal solution of the photovoltaic storage operation model, obtain the optimal configuration parameters, and reduce the average operating cost of the system.

[0121] Based on the above-described optical-storage joint two-layer optimization scheduling method (prediction) that considers uncertainties, the specific content of step (1) is as follows:

[0122] Data acquisition: High-density radiation and temperature sensors are used to collect real-time power output of photovoltaic panels, and multi-source meteorological data, including light intensity, temperature, humidity and wind speed, are collected.

[0123] Data preprocessing: Outliers are processed using deep learning algorithms, specifically by selecting anomaly detection based on autoencoders (AE) to identify outliers in complex data, and filling in missing data through interpolation using a bidirectional long short-term memory network (Bi-LSTM).

[0124] To construct a composite feature assessment of "time + space" for multi-source data correlation, specifically, the spatial coupling relationship is quantified by using the Pearson correlation coefficient, the output correlation of photovoltaic panels in different regions is calculated (such as the power deviation coefficient of adjacent photovoltaic strings), and the rate of change of light intensity is extracted according to the time series.

[0125] An adaptive standardization method is used to standardize the data, and the standardization parameters are dynamically adjusted according to the time period characteristics of the data (specifically, the "regular and repetitive change characteristics" of core data such as photovoltaic output, meteorological conditions (sunlight, temperature, etc.), and power load demand over time).

[0126] The specific content of step (2) is as follows:

[0127] The photovoltaic power generation GAN model includes a generator and a discriminator;

[0128] The generator architecture is designed as follows:

[0129] The input layer fuses multimodal data, specifically random noise (256 dimensions) generated by the input layer, historical photovoltaic array data (last 48 hours, 15-minute granularity), meteorological data (sunlight, temperature, etc. for the next 24 hours), and photovoltaic array topology information (through graph embedding encoding).

[0130] The spatiotemporal feature extraction network specifically employs a Transformer-CNN hybrid structure for temporal feature extraction. The Transformer captures long-term time dependencies, i.e., the impact of seasonal changes on photovoltaic power output, while the CNN extracts short-term high-frequency fluctuation features, such as power changes caused by cloud cover and wind speed variations. For spatial feature extraction, an improved U-Net++ network is used to process photovoltaic array topology information. By combining dense connections and attention mechanisms, it identifies spatial factors affecting photovoltaic power output and outputs spatial feature vectors.

[0131] In the Transformer-CNN hybrid architecture, the Transformer module captures long-period dependencies, and the CNN module captures short-period dependencies. The final temporal feature vector of the Transformer-CNN hybrid architecture is:

[0132]

[0133] in For channel splicing, FC() is a fully connected layer. These are the long-term and short-term time series feature vectors, respectively.

[0134] The improved U-Net++ network replaces the traditional convolutions in the U-Net network with depthwise separable convolutions. A topology consistency penalty term is added to the loss function, resulting in the following overall loss function:

[0135]

[0136] Basic loss Using Dice loss, i.e. A and B represent the actual value and the predicted value, respectively. This represents the topology consistency penalty term. ,in To predict probabilities, =5; Let h and w represent the background area, and h and w represent the height and width of the image; for example, if pixel (h, w) is located in the photovoltaic module area, then... =1, if the background is blank, then =0;

[0137] Spatiotemporal feature fusion specifically involves introducing an adaptive fusion mechanism that dynamically adjusts the fusion weights of spatiotemporal features according to different scenarios and outputs a joint feature vector.

[0138] The photovoltaic power output sequence prediction is specifically that the generator outputs a multi-scenario photovoltaic power output sequence for the next 24 hours based on the input layer data and feature vectors, with a 15-minute granularity.

[0139] The discriminator architecture is designed as follows:

[0140] The multi-scale discriminant structure specifically utilizes a pyramid-shaped discriminant network to judge the authenticity of generated data from different levels. The shallow network quickly judges the overall distribution characteristics of the data, while the deep network deeply analyzes the detailed characteristics of the data.

[0141] To guide the discriminator in accurately distinguishing between real and fake data and to inversely constrain the generator's optimization, the discriminator loss function is defined as follows:

[0142]

[0143] In the formula: Wasserstein distance loss is used to measure the difference between the generated data and the real data distribution. For detailed feature loss, such as high frequency, it provides differences in fluctuation characteristics; , These are the weighting coefficients.

[0144] The specific content of step (3) is as follows:

[0145] The mathematical model for electrochemical energy storage is as follows:

[0146]

[0147] In the formula: This represents the total energy stored in the battery at time t; , , These represent the battery's self-discharge efficiency, charging efficiency, and discharging efficiency, respectively. , Let represent the charging and discharging power of the battery at time t, respectively. It indicates a time step.

[0148] Virtual energy storage models include transferable virtual energy storage mathematical models and slashable virtual energy storage mathematical models;

[0149] In transferable virtual energy storage, when the load power after transfer is higher than the value before transfer, it is equivalent to charging behavior of the transferable virtual energy storage; otherwise, it is discharging behavior. The mathematical model of transferable virtual energy storage is as follows:

[0150]

[0151] In the formula: This represents the load value of transferable virtual energy storage participating in scheduling during time period t. , These represent the load value of transferable virtual energy storage before participating in scheduling and the change in the amount of energy participating in scheduling, respectively. This indicates the transfer status of increased load during time period t. This indicates the transfer status of load reduction during time period t; a value of 1 indicates that a transfer has occurred. , These represent the increase or decrease in load during time period t, respectively. , These represent the upper and lower limits of the change, respectively, and T is the scheduling period.

[0152] The power reduction from load cuts in a reducible virtual energy storage system is equivalent to the discharge power of the virtual energy storage. The mathematical model for reducible virtual energy storage is as follows:

[0153]

[0154] In the formula: This represents the amount of virtual energy storage load that can be reduced and participates in scheduling during time period t. , These represent the load that can be reduced from virtual energy storage before it participates in scheduling and the change in the amount of virtual energy storage participating in scheduling, respectively. This indicates the upper limit of the amount of virtual energy storage change that can be reduced.

[0155] The specific content of step (4) is as follows:

[0156] Every night, a preliminary joint scheduling strategy is generated using the photovoltaic GAN model to predict photovoltaic output for the next 24 hours and the energy storage model. On the same day, real-time data on photovoltaic output and energy storage status are collected, and the current operating status is evaluated at 15-minute intervals. A two-layer optimization scheduling approach is then used to refine the preliminary scheduling strategy.

[0157] The described two-layer optimization scheduling strategy consists of two layers: the upper layer aims to minimize system operating cost, and the lower layer aims to minimize net load variance. The objective function of the upper layer is to minimize operating cost:

[0158]

[0159] In the formula: , , These are the virtual energy storage dispatch cost, electrochemical energy storage dispatch cost, and photovoltaic dispatch cost, respectively, and their mathematical expressions are as follows:

[0160]

[0161]

[0162]

[0163] In the formula: , These represent the unit dispatch cost coefficients for transferable virtual energy storage and reducible virtual energy storage, respectively. This represents the unit dispatch cost coefficient for electrochemical energy storage. Indicates the light-wasting penalty coefficient; This indicates the upper limit of photovoltaic power output. , Let t represent the charging power and discharging power of the battery at time t, respectively, and T represent the scheduling period.

[0164] The lower-level objective function is to minimize the net load variance:

[0165]

[0166] In the formula: , Let represent the net load value during time period t and the average net load value within the scheduling cycle, respectively. Their mathematical expressions are:

[0167]

[0168]

[0169] In the formula: This represents the net power output of the photovoltaic system during time period t. , This indicates the net power of the energy storage system during charging and discharging.

[0170] The constraints include upper-level constraints and lower-level constraints, specifically:

[0171] The lower-level constraints include photovoltaic output constraints, energy storage charge and discharge constraints, and energy storage system capacity constraints, among which:

[0172] Photovoltaic output constraints are:

[0173]

[0174]

[0175] In the formula: This represents the active power delivered by the photovoltaic system to the energy storage system during time period t; Indicates the maximum active power of photovoltaic power. This represents the rated power of the photovoltaic system at time t.

[0176] Energy storage charging and discharging constraints are:

[0177]

[0178]

[0179]

[0180]

[0181]

[0182] In the formula: This indicates the energy storage status during time period t; , These represent the charging and discharging efficiencies of the energy storage system, respectively. , These represent the maximum charging and discharging power of the energy storage system, respectively. , These represent the maximum charging and discharging power at time t, respectively. , These represent the maximum and minimum energy storage states of the energy storage system, respectively.

[0183] Energy storage system capacity constraints:

[0184]

[0185] In the formula: This indicates the state of charge of the energy storage system during time period t; This indicates the maximum state of charge of the energy storage system.

[0186] Substituting the power load obtained from the optimization model into the lower-level optimization model, we obtain the power balance constraint of the upper-level constraint, as follows:

[0187]

[0188] In the formula: This represents the electricity demand during time period t.

[0189] The specific content of step (5) is as follows:

[0190] (5a) Based on the standard gravitational search algorithm, the following three improvement strategies are introduced: First, a chaotic operator is added to increase population diversity and search randomness; second, an adaptive gravitational decay factor is introduced to improve the variation law of the gravitational constant in the algorithm; finally, a formula for the update rate of elite particles in the population is introduced to increase the collective memory capacity and information exchange capacity, resulting in the improved gravitational search algorithm, the specific algorithm of which is as follows (e.g. Figure 2 (as shown)

[0191] The formula for chaotic operators is The particle velocity update formula after adding the chaotic operator is: ;in It is a randomly generated d-dimensional random vector and k represents the number of individuals. It is a factor that controls the range of chaos. It is the current d-dimensional velocity vector. It is the current d-dimensional acceleration vector; The particle velocity after adding the chaotic operator; It is a random number;

[0192] The formula for the adaptive gravitational decay factor is: ;in This is the initial value of the gravitational decay factor. and Scaling factor , , The shift factor is T, the maximum number of iterations is T, and the current number of iterations is t.

[0193] The velocity update formula after introducing elite particles from the population is as follows: ;

[0194] in , It is an adaptive learning factor that changes with the number of iterations. and represents the individual optimal position and the global optimal position in the d-th dimension, respectively, and rand represents a random number;

[0195] (5b) Update and improve the parameters of the gravity search algorithm, generate the system optical storage configuration parameters, determine whether the constraints are met, and calculate the individual fitness value. The specific algorithm steps are as follows:

[0196] Step 1: Randomly initialize the population and set parameters. Parameters include population size N, maximum number of iterations T, and location boundaries. Initial value of universal gravitation and initial value of gravitational decay factor Etc. Randomly initialize the positions of individuals in the population.

[0197] Step 2: Determine if the updated individual position is out of bounds. If it is, randomly assign a new position.

[0198] Step 3: Calculate the fitness value of all individuals in the population.

[0199] Step 4: Update the global optimal position and individual optimal position Find the optimal objective function value for the population. and corresponding individual locations (The loop will compare the result with the previous one, and after a certain number of iterations, it will output the best result.) If < ,but = , = , This represents the fitness value after t cycles. The optimal fitness value is obtained after t iterations. This is the globally optimal fitness value;

[0200] Step 5: Calculate individual mass According to the formula G(t) = G0 × exp(- Calculate the gravitational parameter G(t) using the formula (×t / T). and Calculate the gravitational force acting on you , where R ij (t) = ||X i (t), X j (t)||2, finally calculate the acceleration ;in It is the gravitational parameter at the t-th iteration. Let i be the Euclidean distance between individuals i and j. , and It is a constant. , ,and ; Let represent the gravitational component of the j-th individual on the i-th individual in the d-th dimension at the t-th iteration; This represents the quality of the i-th individual in the t-th iteration; This represents the quality of the j-th individual in the t-th iteration; This represents the position coordinates of the i-th individual in the d-th dimension during the t-th iteration. Let X represent the position coordinates of the j-th individual in the d-th dimension during the t-th iteration. i (t) is the d-dimensional position vector of the i-th individual at the t-th iteration, X j (t) is the d-dimensional position vector of the j-th individual at the t-th iteration. This represents the set of the K individuals with the highest fitness in the current iteration. Let N be a random number, and N be the population size.

[0201] Step 6: Calculate the adaptive learning factor based on universal gravitation, and the formula... and Update speed and location, V i d (t) represents the movement speed of the i-th individual in the d-th dimension of the decision variable at the t-th iteration;

[0202] Step 7: Number of iterations ,like Proceed to step 2; otherwise, end the loop and obtain the optimal objective function value. and optimal position vector .

[0203] (5c) Perform system optimization to obtain the output P of each device and the system operating cost F. 1t The obtained F 1t Compared to the F obtained last time 1t-1 Compare the results. If the result is lower than the previous iteration value, proceed to step (5d). If the upper limit of the running process has not been reached, proceed to step (5a) and output the result of the previous run.

[0204] (5d) Update the optimal cost F1 and determine whether the number of iterations has reached the upper limit. If the maximum number of iterations has been reached, output the running results and parameters. The output results are the optimal configuration parameters for the operation of the photovoltaic energy storage system. If not, return to step (5a).

[0205] After obtaining the optimized parameters, the optimized system's best configuration parameters are applied to the photovoltaic energy storage system to reduce the system's operating costs and achieve system efficiency optimization.

[0206] For example, in multi-source joint scheduling operation, the photovoltaic-storage system optimization scheduling model established in this invention can be used to obtain a better parameter configuration for the photovoltaic-storage model. At the same time, the two-layer optimization scheduling can take into account the load variance while considering the operating cost. The load variance corresponds to the load curve. The smaller the value, the flatter the curve, and the better the peak-shaving effect of the photovoltaic-storage joint power generation system.

Claims

1. A two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties, characterized in that, Includes the following steps: S1. Collect historical photovoltaic array data and meteorological data, and preprocess the collected data to train and improve the dataset; S2. Considering the uncertainty of photovoltaic output, a photovoltaic power generation GAN model is established using the preprocessed dataset, and a multi-scenario photovoltaic output prediction sequence for the next 24 hours is output. S3. Establish an electrochemical energy storage model and a virtual energy storage model, wherein the virtual energy storage model includes a transferable virtual energy storage model and a slashable virtual energy storage model; S4. Using the photovoltaic power generation GAN model to generate a multi-scenario photovoltaic output prediction sequence for the next 24 hours, as well as the electrochemical energy storage model and virtual energy storage model, establish a photovoltaic-energy storage operation model and generate a preliminary joint scheduling strategy. Adopt a two-layer optimization scheduling strategy on the photovoltaic-energy storage operation model and correct the preliminary joint scheduling strategy based on the real-time collected data. S5. Using a gravity search algorithm, incorporating chaotic operators, adaptive gravity decay factors, and population elite particles, the multivariable nonlinear global optimal solution of the photovoltaic-storage operation model is obtained, and the optimal configuration parameters are applied to the photovoltaic-storage system.

2. The two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties according to claim 1, characterized in that, The specific details of step S1 are as follows: S1.1 Data Acquisition: Collect real-time output power of photovoltaic panels and multi-source meteorological data, including light intensity, temperature, humidity and wind speed; S1.2 Data Preprocessing: Combine deep learning algorithms to process outliers in the collected data. Specifically, select anomaly detection based on autoencoder to identify outliers in the data, and fill in missing data through bidirectional long short-term memory network interpolation. S1.3 Constructing the correlation of multi-source data for temporal and spatial composite feature assessment, specifically: quantifying the spatial coupling relationship of preprocessed data using Pearson correlation coefficient, calculating the output correlation of photovoltaic panels in different regions, and extracting the rate of change of light intensity according to the time series; Pearson correlation coefficient: , in These represent the power output in different regions at the i-th time period, where n is the length of the time series. , X and Y are the means of X and Y, respectively; X and Y represent different regions. These are the standard deviations of the two sequences; S1.

4. Adaptive standardization method is used to standardize the data obtained in S1.3, and the standardization parameters are dynamically adjusted according to the time period characteristics of the data.

3. The two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties according to claim 1, characterized in that, The specific details of step S2 are as follows: The photovoltaic power generation GAN model includes a generator and a discriminator; The generator architecture is designed as follows: The input layer fuses multimodal data, specifically random noise generated by the input layer, historical photovoltaic array data, meteorological data, and photovoltaic array topology information; The spatiotemporal feature extraction network is as follows: Temporal feature extraction adopts a Transformer-CNN hybrid structure, using Transformer to capture long-period time dependence, i.e. the impact of seasonal changes on photovoltaic power output, and CNN to extract short-term high-frequency fluctuation features; Spatial feature extraction adopts an improved U-Net++ network to process photovoltaic array topology information, using a combination of dense connections and attention mechanisms to identify spatial factors affecting photovoltaic power output and output spatial feature vectors. In the Transformer-CNN hybrid architecture, the Transformer module captures long-period dependencies, and the CNN module captures short-period dependencies. The final temporal feature vector of the Transformer-CNN hybrid architecture is: ， in For channel splicing, FC() is a fully connected layer. These are the long-term and short-term time series feature vectors, respectively. The improved U-Net++ network incorporates a topology consistency penalty term into the loss function, resulting in the following total loss function: ， Basic loss Using Dice loss, i.e. A and B represent the actual value and the predicted value, respectively. This represents the topology consistency penalty term. ,in To predict probabilities, =5; The background area is represented by h and w, which represent the height and width of the image, respectively. Spatiotemporal feature fusion specifically involves introducing an adaptive fusion mechanism to dynamically adjust the fusion weights of spatiotemporal features according to different scenarios and output a joint feature vector. The photovoltaic power output sequence prediction is as follows: the generator outputs a multi-scenario photovoltaic power output sequence for the next 24 hours based on the input layer data and joint feature vectors, with a 15-minute granularity. The discriminator architecture is designed as follows: The multi-scale discriminant structure is as follows: a pyramid discriminant network is used to judge the authenticity of the data generated by the generator from different levels. The shallow network quickly judges the overall distribution characteristics of the data, while the deep network analyzes the detailed features of the data in depth. To guide the discriminator to accurately distinguish between real and fake data and to conversely constrain the generator's optimization, a discriminator loss function is introduced. The formula is defined as follows: ， In the formula: Wasserstein distance loss is used to measure the difference between the generated data and the real data distribution. Loss of detailed features; , These are the weighting coefficients.

4. The two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties according to claim 1, characterized in that, The specific details of step S3 are as follows: The electrochemical energy storage model is as follows: ， In the formula: This represents the total energy stored in the battery at time t; , , These represent the battery's self-discharge efficiency, charging efficiency, and discharging efficiency, respectively. , Let represent the charging power and discharging power of the battery at time t, respectively. Indicates a time step; Virtual energy storage models include portable virtual energy storage models and slashable virtual energy storage models; In transferable virtual energy storage, when the value of the load power after transfer is higher than the value before transfer, it is equivalent to the charging behavior of transferable virtual energy storage; otherwise, it is the discharging behavior. The transferable virtual energy storage model is as follows: ， In the formula: This represents the load value of transferable virtual energy storage participating in scheduling during time period t. , These represent the load value of transferable virtual energy storage before participating in scheduling and the change in the amount of energy participating in scheduling, respectively. This indicates the transfer status of increased load during time period t. This indicates the transfer status of load reduction during time period t. A value of 0 indicates no transfer, while a value of 1 indicates that a transfer has been performed. , These represent the increases and decreases in load during time period t, respectively. , These represent the upper and lower limits of the change, respectively, and T is the scheduling period; In reducible virtual energy storage, the power of load reduction is equivalent to the discharge power of virtual energy storage. The reducible virtual energy storage model is as follows: ， In the formula: This represents the amount of virtual energy storage load that can be reduced and participates in scheduling during time period t. , These represent the load that can be reduced from virtual energy storage before it participates in scheduling and the change in the amount of virtual energy storage participating in scheduling, respectively. This indicates the upper limit of the amount of virtual energy storage change that can be reduced.

5. A two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties, as described in claim 4, is characterized in that... The specific details of step S4 are as follows: Every night, a preliminary joint scheduling strategy is generated by using the photovoltaic power generation GAN model to generate a multi-scenario photovoltaic output prediction sequence for the next 24 hours, as well as an electrochemical energy storage model and a virtual energy storage model. On the same day, photovoltaic output and energy storage status data are collected in real time, and the current operating status is evaluated at a 15-minute interval. The preliminary joint scheduling strategy is then corrected using a two-layer optimization scheduling strategy. The described two-layer optimization scheduling strategy consists of two layers: the upper layer aims to minimize the system operating cost, and the lower layer aims to minimize the net load variance. The objective function of the upper layer is the operating cost. Minimum: ， In the formula: , , These are the virtual energy storage dispatch cost, electrochemical energy storage dispatch cost, and photovoltaic dispatch cost, respectively, and their mathematical expressions are as follows: ， ， ， In the formula: , These represent the unit dispatch cost coefficients for transferable virtual energy storage and reducible virtual energy storage, respectively. This represents the unit dispatch cost coefficient for electrochemical energy storage. Indicates the light-wasting penalty coefficient; Indicates the output power of photovoltaics. This indicates the upper limit of the photovoltaic output power; , Let t represent the charging power and discharging power of the battery at time t, respectively, and T represent the scheduling period; The lower-level objective function is the net load variance. Minimum: ， In the formula: , Let represent the net load value during time period t and the average net load value within the scheduling period, respectively. Their mathematical expressions are as follows: ， ， In the formula: This represents the net power output of the photovoltaic system during time period t. , This represents the net power of the energy storage system during charging and discharging. This represents the electricity demand during time period t; Constraints include upper-level constraints and lower-level constraints, specifically: The lower-level constraints include photovoltaic output constraints, energy storage charge and discharge constraints, and energy storage system capacity constraints, among which: Photovoltaic output constraints for: ， ， In the formula: This represents the active power delivered by the photovoltaic system to the energy storage system during time period t; This indicates the maximum active power of photovoltaic power. Indicates the rated power of the photovoltaic system; Energy storage charging and discharging constraints are: ， ， ， ， ， In the formula: This indicates the energy storage status during time period t; , These represent the charging and discharging efficiencies of the energy storage system, respectively. , These represent the maximum charging and discharging power of the energy storage system, respectively. , These represent the maximum charging and discharging power at time t, respectively. , These represent the maximum and minimum energy storage states of the energy storage system, respectively. Energy storage system capacity constraints: ， In the formula: This indicates the state of charge of the energy storage system during time period t; This indicates the maximum state of charge of the energy storage system; Substituting the power load obtained from the optimization model into the lower-level optimization model, we obtain the power balance constraint of the upper-level constraint, as follows: ， In the formula: This represents the electricity demand during time period t.

6. The two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties according to claim 1, characterized in that, The specific details of step S5 are as follows: (5a) Based on the gravity search algorithm, the following three improvement strategies are introduced: First, a chaotic operator is added to increase population diversity and search randomness; second, an adaptive gravity decay factor is introduced; finally, a formula for the update rate of elite particles in the population is introduced to increase the population's memory capacity and information exchange capacity, as detailed below: The formula for chaotic operators is , The particle velocity update formula after adding the chaotic operator is: ;in It is a randomly generated d-dimensional random vector and k represents the number of individuals. It is a factor that controls the range of chaos. It is the current d-dimensional velocity vector. It is the current d-dimensional acceleration vector; The particle velocity after adding the chaotic operator; It is a random number; The formula for the adaptive gravitational decay factor is: ;in This is the initial value of the gravitational decay factor. and All are scaling factors. , , The shift factor is T, the maximum number of iterations is T, and the current number of iterations is t. The velocity update formula after introducing elite particles from the population is as follows: ; in , These are all adaptive learning factors that change with the number of iterations. and represents the individual optimal position and the global optimal position in the d-th dimension, respectively, and rand represents a random number; (5b) Update the parameters of the gravity search algorithm, generate the configuration parameters of the system's optical storage operation model, and determine whether the constraints are met and calculate the individual fitness value; (5c) Perform system optimization to obtain the output P of each device and the system operating cost F. 1t The obtained F 1t Compared to the F obtained last time 1t-1 Compare the results. If the result is lower than the previous iteration value, proceed to step (5d). If the upper limit of the running process has not been reached, proceed to step (5a) and output the result of the previous run. (5d) Update the optimal cost F1 and determine whether the number of iterations has reached the upper limit. If the maximum number of iterations has been reached, output the running results and parameters. The output results are the optimal configuration parameters of the optical storage running model. If not, return to step (5a).

7. The two-layer optimization scheduling method for a combined optical and virtual energy storage model considering uncertainties according to claim 6, characterized in that, The specific details of step (5b) are as follows: 1) Randomly initialize the population and set parameters, including population size N, maximum number of iterations T, and location boundaries. Initial value of universal gravitation and initial value of gravitational decay factor Randomly initialize the positions of individuals in the population; 2): Determine if the updated individual position is out of bounds; if so, randomly assign a new position. 3): Calculate the fitness value of all individuals in the population; 4): Update the global optimal position and individual optimal position Find the optimal objective function value for the population. and corresponding individual locations ,if < ,but = , = , This represents the fitness value after t cycles. The optimal fitness value is obtained after t iterations. This is the globally optimal fitness value; 5): Calculate individual mass According to the formula G(t) = G0 × exp(- Calculate the gravitational parameter G(t) using the formula (×t / T). and Calculate the gravitational force acting on you , where R ij (t) = ||X i (t), X j (t)||2, finally calculate the acceleration ;in It is the gravitational parameter at the t-th iteration. Let i be the Euclidean distance between individuals i and j. , and It is a constant. , ,and ; Let represent the gravitational component of the j-th individual on the i-th individual in the d-th dimension at the t-th iteration; This represents the quality of the i-th individual in the t-th iteration; This represents the quality of the j-th individual in the t-th iteration; This represents the position coordinates of the i-th individual in the d-th dimension during the t-th iteration. Let X represent the position coordinates of the j-th individual in the d-th dimension during the t-th iteration. i (t) is the d-dimensional position vector of the i-th individual at the t-th iteration, X j (t) is the d-dimensional position vector of the j-th individual at the t-th iteration. This represents the set of the K individuals with the highest fitness in the current iteration. Let N be a random number, and N be the population size. 6): Calculate the adaptive learning factor based on universal gravitation, and the formula... and Update speed and location, V i d (t) represents the movement speed of the i-th individual in the d-th dimension of the decision variable at the t-th iteration; 7): Number of iterations ,like (If the condition is met, proceed to step 2); otherwise, end the loop and obtain the optimal objective function value. and optimal position vector .

8. A two-layer optimized scheduling system for a combined optical and virtual energy storage model considering uncertainties, characterized in that, Including: Data acquisition and processing module: Collects historical photovoltaic array data and meteorological data, and preprocesses the collected data to train and improve the dataset; A photovoltaic power generation GAN model construction module is used to build a photovoltaic power generation GAN model based on the uncertainty of photovoltaic output and output a multi-scenario photovoltaic output prediction sequence for the next 24 hours. The energy storage model construction module establishes an electrochemical energy storage model and a virtual energy storage model, wherein the virtual energy storage model includes a transferable virtual energy storage model and a slashable virtual energy storage model. The photovoltaic-storage operation model construction module, using the photovoltaic power generation GAN model to generate a multi-scenario photovoltaic output prediction sequence for the next 24 hours, as well as an electrochemical energy storage model and a virtual energy storage model, establishes a photovoltaic-storage operation model and generates a preliminary joint scheduling strategy. A two-layer optimization scheduling strategy is adopted for the photovoltaic-storage operation model, and the preliminary joint scheduling strategy is corrected based on the real-time collected data. Solution module: Employing a gravity search algorithm, incorporating chaotic operators, adaptive gravity decay factors, and population elite particles, the module solves for the multivariable nonlinear global optimal solution of the photovoltaic-storage operation model, obtaining the optimal configuration parameters and applying them to the photovoltaic-storage system.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the two-layer optimization scheduling method for the optical joint virtual energy storage model considering uncertainty as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the two-layer optimization scheduling method for the optical joint virtual energy storage model considering uncertainty as described in any one of claims 1-7.

Images