Microgrid real-time optimization operation method and device, computer equipment, storage medium and computer program product

Abstract

The invention relates to a micro-grid real-time optimization operation method and device, computer equipment, a storage medium and a computer program product. Relates to the technical field of power systems. The method comprises the following steps: constructing a sample set under various operation scenes of the micro-grid, and simultaneously extracting state space sampling points to obtain a scene state data set; processing the large coordinate set and the small landmark set in sequence based on a preset step-by-step Nystrom approximation algorithm, and outputting an optimal small landmark set; constructing an association relationship among the value function, the operation scene and the state variable based on a preset Wiener kernel regression algorithm, and constructing a value function formula corresponding to the state space of the microgrid based on a preset Loewner interpolation approximation algorithm; and obtaining a current operation scene of the micro-grid in real time, and processing the current operation scene and the optimal small landmark set based on a value function formula to obtain an optimal operation strategy of the micro-grid. By adopting the method, the accuracy and the efficiency of the real-time optimization operation method of the micro-grid can be improved.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method, apparatus, computer equipment, storage medium and computer program product for real-time optimization of microgrid operation. Background Technology

[0002] Currently, the park's microgrid is an aggregation of renewable energy, traditional energy, load, and energy storage. However, the integrated renewable energy sources such as photovoltaic and wind power have inherent intermittency and volatility, and face multiple uncertainties such as load demand and electricity price fluctuations, making it difficult to maintain system power balance and resulting in high operating costs. Therefore, achieving real-time optimized operation of the park's microgrid under complex and uncertain conditions has become a technical problem to be solved.

[0003] However, current methods for optimizing the operation of microgrids in industrial parks mainly rely on traditional algorithms or simple stochastic optimization strategies. For example, the value function for offline training in approximate dynamic programming algorithms is mainly formed based on historical data. When the operating conditions of the industrial park microgrid change, the accuracy of online optimization is easily reduced due to the difference between real-time operating conditions and historical conditions. Furthermore, the offline training process in approximate dynamic programming algorithms often requires a large number of samples to ensure the accuracy of the constructed value function, which affects the training efficiency in the offline stage.

[0004] Therefore, current methods for real-time optimization of microgrid operation suffer from low accuracy and low efficiency. Summary of the Invention

[0005] Therefore, it is necessary to address the technical problems of low accuracy and low efficiency of current microgrid real-time optimization operation methods by providing a microgrid real-time optimization operation method, device, computer equipment, computer-readable storage medium, and computer program product.

[0006] Firstly, this application provides a method for real-time optimized operation of a microgrid, including:

[0007] A sample set for various operating scenarios of microgrids is constructed, and the state space sampling points of the microgrid under each operating scenario are extracted to obtain a scenario state data set.

[0008] The large coordinate set and small landmark set in the scenario state data set are processed sequentially based on the preset step-by-step Nystrom approximation algorithm to output the optimal small landmark set;

[0009] The relationship between the value function, the operating scenario, and the state variables is constructed based on the preset Wiener kernel regression algorithm, and the value function formula corresponding to the state space of the microgrid is constructed based on the preset Loewner interpolation approximation algorithm.

[0010] The current operating scenario of the microgrid is acquired in real time, and the current operating scenario and the optimal small landmark set are processed based on the value function formula to obtain the optimal operating strategy of the microgrid.

[0011] In one embodiment, the construction of a sample set under multiple operating scenarios of the microgrid, and the extraction of state space sampling points of the microgrid under each operating scenario to obtain a scenario state data set, includes: randomly generating multiple microgrid operating scenarios; each operating scenario includes at least one parameter among renewable energy output, load, expected distribution of electricity price uncertainty, standard deviation of renewable energy output, standard deviation of load, and expected standard deviation of electricity price uncertainty distribution; collecting state space sampling points corresponding to multiple microgrid operating scenarios, and combining each microgrid operating scenario with its corresponding state space sampling points to obtain the scenario state data set.

[0012] In one embodiment, the step-by-step Nystrom approximation algorithm is used to process the large coordinate set and the small landmark set in the scenario state data set sequentially to output the optimal small landmark set. This includes: randomly extracting data from the scenario state data set as the large landmark set and obtaining the kernel matrix of the large coordinate set; obtaining the initial cluster centers of the large coordinate set based on a preset clustering algorithm and the kernel matrix, and determining the initial values ​​of the small landmark set based on the initial cluster centers, while setting a convergence threshold and a maximum number of iterations; updating the small landmark set using a preset gradient descent algorithm, and outputting the optimal small landmark set.

[0013] In one embodiment, the construction of the value function, the relationship between the running scenario and the state variables based on the Wiener kernel regression algorithm includes: constructing a kernel regression objective function containing noise terms based on a Gaussian distribution; processing the kernel space features of the scenario state data based on the kernel regression objective function to obtain corresponding random weights; processing the random weights based on a preset homogeneous chaotic algorithm to separate deterministic coefficients and the noise-related terms to obtain corresponding optimal weight coefficients; and constructing the relationship between the value function, the running scenario and the state variables based on the optimal weight coefficients.

[0014] In one embodiment, constructing the value function formula corresponding to the state space of the microgrid based on a preset Loewner interpolation approximation algorithm includes: splitting a pre-constructed sample set into a first sample set covering a region of drastic change in the value function and a second sample set covering a region of gradual change in the value function; constructing a Loewner matrix and a shifted Loewner matrix based on the first sample set and the second sample set, respectively; constructing a projection matrix based on the Loewner matrix and the shifted Loewner matrix; and stacking and processing the value functions of the first sample set and the second sample set based on the projection matrix to obtain the value function formula.

[0015] In one embodiment, the step of acquiring the current operating scenario of the microgrid in real time, processing the current operating scenario and the optimal small landmark set based on the value function formula to obtain the optimal operating strategy of the microgrid includes: acquiring the optimal small landmark set included in the current operating scenario of the microgrid, the combination of the operating scenario and the state space sampling points, the kernel matrix of the optimal small landmark set and the original data, and the binary membership matrix of the optimal small landmark set, to obtain the optimal operating strategy of the microgrid.

[0016] Secondly, this application also provides a microgrid real-time optimized operation device, comprising:

[0017] The data acquisition module is used to construct sample sets under various operating scenarios of the microgrid, and at the same time extract the state space sampling points of the microgrid under each operating scenario to obtain the scenario state data set.

[0018] The data processing module is used to process the large coordinate set and the small landmark set in the scenario state data set sequentially based on the preset step-by-step Nystrom approximation algorithm, and output the optimal small landmark set.

[0019] The data processing module is also used to construct the value function, the relationship between the operating scenario and the state variables based on the preset Wiener kernel regression algorithm, and to construct the value function formula corresponding to the state space of the microgrid based on the preset Loewner interpolation approximation algorithm.

[0020] The strategy generation module is used to acquire the current operating scenario of the microgrid in real time, process the current operating scenario and the optimal small landmark set based on the value function formula, and obtain the optimal operating strategy of the microgrid.

[0021] Thirdly, this application also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0022] A sample set for various operating scenarios of microgrids is constructed, and the state space sampling points of the microgrid under each operating scenario are extracted to obtain a scenario state data set.

[0023] The large coordinate set and small landmark set in the scenario state data set are processed sequentially based on the preset step-by-step Nystrom approximation algorithm to output the optimal small landmark set;

[0024] The relationship between the value function, the operating scenario, and the state variables is constructed based on the preset Wiener kernel regression algorithm, and the value function formula corresponding to the state space of the microgrid is constructed based on the preset Loewner interpolation approximation algorithm.

[0025] The current operating scenario of the microgrid is acquired in real time, and the current operating scenario and the optimal small landmark set are processed based on the value function formula to obtain the optimal operating strategy of the microgrid.

[0026] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0027] A sample set for various operating scenarios of microgrids is constructed, and the state space sampling points of the microgrid under each operating scenario are extracted to obtain a scenario state data set.

[0028] The large coordinate set and small landmark set in the scenario state data set are processed sequentially based on the preset step-by-step Nystrom approximation algorithm to output the optimal small landmark set;

[0029] The relationship between the value function, the operating scenario, and the state variables is constructed based on the preset Wiener kernel regression algorithm, and the value function formula corresponding to the state space of the microgrid is constructed based on the preset Loewner interpolation approximation algorithm.

[0030] The current operating scenario of the microgrid is acquired in real time, and the current operating scenario and the optimal small landmark set are processed based on the value function formula to obtain the optimal operating strategy of the microgrid.

[0031] Fifthly, this application also provides a computer program product, which includes a computer program that, when executed by a processor, performs the following steps:

[0032] A sample set for various operating scenarios of microgrids is constructed, and the state space sampling points of the microgrid under each operating scenario are extracted to obtain a scenario state data set.

[0033] The large coordinate set and small landmark set in the scenario state data set are processed sequentially based on the preset step-by-step Nystrom approximation algorithm to output the optimal small landmark set;

[0034] The relationship between the value function, the operating scenario, and the state variables is constructed based on the preset Wiener kernel regression algorithm, and the value function formula corresponding to the state space of the microgrid is constructed based on the preset Loewner interpolation approximation algorithm.

[0035] The current operating scenario of the microgrid is acquired in real time, and the current operating scenario and the optimal small landmark set are processed based on the value function formula to obtain the optimal operating strategy of the microgrid.

[0036] The aforementioned microgrid real-time optimization operation method, device, computer equipment, storage medium, and computer program product, in the process of real-time optimization operation of the microgrid, firstly constructs sample sets under various operating scenarios of the microgrid, and simultaneously extracts the state space sampling points of the microgrid under each operating scenario to obtain a scenario state data set; then, based on a preset step-by-step Nystrom approximation algorithm, processes the large coordinate set and small landmark set in the scenario state data set sequentially, and outputs the optimal small landmark set; next, based on a preset Wiener kernel regression algorithm, it constructs the correlation between value functions, operating scenarios, and state variables, and based on a preset Loewner interpolation approximation algorithm, it constructs the value function formula corresponding to the state space of the microgrid; finally, it obtains the current operating scenario of the microgrid in real time, processes the current operating scenario and the optimal small landmark set based on the value function formula, and obtains the optimal operating strategy of the microgrid. In the above configuration, a multi-scenario situational state dataset is constructed and merged, and a step-by-step Nystrom approximation algorithm is used to process high-dimensional data. A large coordinate set is used to initially represent the data distribution, and a small coordinate set is used for optimization and screening. This reduces computational complexity while ensuring approximation accuracy. Furthermore, by combining Wiener kernel regression and Loewner interpolation approximation algorithms, the correlation between value functions and operating scenarios and state variables is accurately constructed and transformed into explicit formulas. This enables rapid solution of value functions and real-time output of the optimal operating strategy adapted to the current scenario, effectively improving the accuracy and efficiency of the microgrid real-time optimization operation method. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating a method for real-time optimization of microgrid operation in one embodiment;

[0039] Figure 2 This is a schematic diagram of the process for acquiring scenario state data sets in a microgrid real-time optimized operation method in one embodiment;

[0040] Figure 3 This is a schematic diagram illustrating the process of obtaining the optimal small landmark set for the real-time optimization operation steps of a microgrid in one embodiment.

[0041] Figure 4 This is a flowchart illustrating the real-time optimization operation steps of a microgrid in one embodiment;

[0042] Figure 5 This is a flowchart illustrating the construction of the value function formula for the real-time optimization operation steps of a microgrid in one embodiment;

[0043] Figure 6 This is a flowchart illustrating the real-time optimization operation steps of a microgrid in another embodiment;

[0044] Figure 7 This is a structural block diagram of a microgrid real-time optimization operation device in one embodiment;

[0045] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] With the rapid development of distributed resources, campus microgrids, as a collection of renewable energy, traditional energy, load, and energy storage, play a crucial role in improving energy efficiency, reducing costs, and promoting low-carbon transformation. Therefore, achieving real-time optimized operation of campus microgrids and improving their adaptability to complex operating environments has become an important research direction. Real-time optimization of campus microgrids is essentially a complex decision-making problem involving multiple time periods. Approximate dynamic programming (APP) possesses efficient solution capabilities for this type of problem and is considered an effective algorithm for real-time optimization of campus microgrids. To improve the decision-making efficiency of APP in the real-time optimization process of campus microgrids, its application can be divided into two stages: offline training and online decision-making. The offline training stage uses given samples to learn and construct the value function corresponding to the state space of the campus microgrid. The online decision-making stage uses the value function obtained from the training and combines it with the current state of the campus microgrid to quickly obtain its approximate optimal operating strategy.

[0048] While the above process shifts a significant portion of computational costs from the online decision-making stage to the offline training stage, improving the computational efficiency of real-time optimization of industrial microgrids, it also presents two problems: First, the value function trained offline is primarily based on historical data. When the operating conditions of the industrial microgrid change, the accuracy of online optimization may drop sharply due to the difference between real-time operating conditions and historical data. Second, the offline training process often requires extensive sampling to ensure the accuracy of the constructed value function, which to some extent affects the training efficiency of the offline stage. Therefore, current methods for real-time optimization of microgrid operation suffer from low accuracy and low efficiency.

[0049] To address the issues of low accuracy and low efficiency in the aforementioned microgrid real-time optimization operation methods, in one embodiment, such as... Figure 1 As shown, a method for real-time optimization of microgrid operation is provided. This embodiment illustrates the method applied to a terminal, but it is understood that the method can also be applied to a server, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0050] Step S102: Construct sample sets for various operating scenarios of the microgrid, and extract state space sampling points of the microgrid under each operating scenario to obtain scenario state data sets.

[0051] Among them, the microgrid operation scenario is a combination of the operation modes of the microgrid under different external conditions, internal configurations and load demands. It can include grid-connected operation scenario: interconnected with the main grid and capable of bidirectional power transmission; islanded operation scenario: independent power supply without the main grid; peak-valley load scenario: peak load, low-valley load, flat-section load; and renewable energy output fluctuation scenario: full generation, insufficient generation, zero generation of photovoltaic power, etc. The constraints and control objectives under different microgrid operation scenarios are significantly different.

[0052] Step S104: Based on the preset step-by-step Nystrom approximation algorithm, process the large coordinate set and small landmark set in the scenario state data set in sequence, and output the optimal small landmark set.

[0053] Among them, the scenario state data set is a multi-dimensional state parameter dataset obtained by sampling the state space of the microgrid under various operating scenarios. The multi-dimensional state parameters can typically include distributed power output, such as photovoltaic, wind power, and energy storage, as well as indicators such as load power, bus voltage, and line current. The step-by-step Nystrom approximation algorithm is an improved algorithm for low-rank approximation of large-scale matrices. It decomposes the calculation of the high-dimensional kernel matrix into two steps: sampling of a large coordinate set and optimization of a small coordinate set. It is suitable for dimensionality reduction and feature extraction of high-dimensional and massive datasets such as microgrid scenario state data sets.

[0054] More specifically, the large coordinate set and the small landmark set are two core data subsets in the stepwise Nystrom approximation algorithm. The large coordinate set is a large-scale sample subset randomly selected or drawn according to specific rules from the scenario state data set, used to initially characterize the distribution characteristics of the overall data. The small landmark set is a small-scale representative sample subset further filtered from the large coordinate set, which is the carrier used by the algorithm for the final kernel matrix approximation. Stepwise processing of the large coordinate set and the small landmark set can balance computational efficiency and improve accuracy. The optimal small landmark set is the small landmark set that best characterizes the overall distribution characteristics of the scenario state data set after optimization and filtering by the stepwise Nystrom approximation algorithm. Its optimality is reflected in the fact that when approximating the kernel matrix using the above-mentioned optimal small landmark set, the error is minimized and the computational efficiency is maximized.

[0055] Step S106: Construct the correlation between value functions, operating scenarios and state variables based on the preset Wiener kernel regression algorithm, and construct the value function formula corresponding to the state space of the microgrid based on the preset Loewner interpolation approximation algorithm.

[0056] Among them, the Wiener kernel regression algorithm is a nonparametric regression method based on kernel functions. It combines the noise reduction characteristics of Wiener filtering with the nonlinear fitting ability of kernel regression. By introducing kernel functions, it maps low-dimensional state variables to high-dimensional feature spaces, which can effectively construct the nonlinear relationship between value functions, operating scenarios, and state variables. Value functions are a core concept in reinforcement learning and optimal control. They are used to quantify the operating benefits or operating costs of microgrids under a certain state and scenario. The value of the value function can be used to reflect the merits of the operating strategy under the current state.

[0057] More specifically, the Loewner interpolation approximation algorithm is a function approximation method based on interpolation theory. It constructs a low-complexity approximation function to fit the original function using the function values ​​of known sample points. In this application, the Loewner interpolation approximation algorithm is used to transform the correlation obtained from Wiener kernel regression into explicit mathematical formulas for the microgrid state space and value function, enabling rapid calculation and invocation of the value function. The state space is a multi-dimensional parameter space describing the operating state of the microgrid, with each dimension corresponding to a state variable. Each point in the state space represents a specific operating state of the microgrid.

[0058] Step S108: Obtain the current operating scenario of the microgrid in real time, process the current operating scenario and the optimal small landmark set based on the value function formula, and obtain the optimal operating strategy of the microgrid.

[0059] Among them, the optimal operation strategy is the optimal control decision combination calculated by the value function formula and the optimal small local set under the current operation scenario of the microgrid. It can include distributed power generation output allocation strategy, energy storage charging and discharging strategy, grid-connected or off-grid switching strategy, load switching strategy, etc.

[0060] In the aforementioned microgrid real-time optimization operation method, firstly, a sample set under various operating scenarios of the microgrid is constructed, and the state space sampling points of the microgrid under each operating scenario are extracted to obtain a scenario state data set; then, based on a preset step-by-step Nystrom approximation algorithm, the large coordinate set and the small landmark set in the scenario state data set are processed sequentially to output the optimal small landmark set; next, based on a preset Wiener kernel regression algorithm, the correlation between the value function, the operating scenario, and the state variables is constructed, and based on a preset Loewner interpolation approximation algorithm, the value function formula corresponding to the state space of the microgrid is constructed; finally, the current operating scenario of the microgrid is obtained in real time, and the current operating scenario and the optimal small landmark set are processed based on the value function formula to obtain the optimal operating strategy of the microgrid. In the above configuration, a multi-scenario situational state dataset is constructed and merged, and a step-by-step Nystrom approximation algorithm is used to process high-dimensional data. A large coordinate set is used to initially represent the data distribution, and a small coordinate set is used for optimization and screening. This reduces computational complexity while ensuring approximation accuracy. Furthermore, by combining Wiener kernel regression and Loewner interpolation approximation algorithms, the correlation between value functions and operating scenarios and state variables is accurately constructed and transformed into explicit formulas. This enables rapid solution of value functions and real-time output of the optimal operating strategy adapted to the current scenario, effectively improving the accuracy and efficiency of the microgrid real-time optimization operation method.

[0061] Furthermore, in an exemplary embodiment, the specific process of the microgrid online optimization method based on approximate dynamic programming is described in detail:

[0062] For the real-time optimization problem of industrial microgrids, approximate dynamic programming algorithms are typically based on Markov decision processes, which are expressed as multi-stage sequential decision-making forms for solution. A Markov decision process can be described by basic elements such as state variables, decision variables, exogenous information, state transition equations, and cost functions.

[0063] First, determine the state variables. The state variables for the real-time optimization problem of microgrids can be defined as the state of charge of energy storage. That is, formula (1): Then, the decision variables are determined; the decision variables for the real-time optimization problem of microgrids are... This includes the active / reactive power output of distributed generation, the charging and discharging power of energy storage systems, the load reduction power, the renewable energy reduction power, the power purchased and sold by the microgrid, and the reactive power exchanged between the microgrid and the upper-level grid, i.e., formula (2): In the formula, Represents a node During the period Distributed power output; , Representing nodes respectively During the period The energy storage charging and discharging power; , Representing nodes respectively During the period Load reduction and renewable energy reduction; , They represent time periods respectively. Power purchased and sold by the microgrid; Represents a node During the period The reactive power output of distributed power sources; Indicates time period Reactive power exchanged with the upper-level power grid.

[0064] Furthermore, the above decision variables must satisfy the constraints of formulas (3)-(14), where formulas (3)-(6) are the branch power flow model after the second-order cone relaxation of the microgrid, formula (7) is the upper and lower limit constraint of voltage amplitude, formula (8) is the upper and lower limit constraint of power flow, formulas (9)-(10) are the upper and lower limit constraints of load and renewable energy reduction, formula (11) is the upper and lower limit constraint of distributed power output, formula (12) is the distributed power ramping constraint, and formulas (13)-(14) are the energy storage system operation constraints. The specific expressions of formulas (3)-(14) are as follows:

[0065] Formula (3):

[0066] Formula (4):

[0067] Formula (5):

[0068] Formula (6):

[0069] Formula (7):

[0070] Formula (8):

[0071] Formula (9):

[0072] Formula (10):

[0073] Formula (12):

[0074] Formula (13):

[0075] Formula (13):

[0076] Formula (14):

[0077] In the formula, Represents the set of branches; Representing nodes respectively Parent and child nodes; Indicates the line Complex impedance; Represents a node During the period The square of the voltage amplitude; Represents a node During the period Net injected complex power; Indicates time period From node The square of the branch current amplitude to the parent node; Indicates the time period node Branch power flow to the parent node; Represents a node During the period Renewable energy output; Represents a node During the period The reactive power demand of the load; , Representing nodes respectively The upper and lower limits of voltage; , These represent the upper and lower limits of active power flow, respectively. , Representing nodes respectively The upper and lower limits of distributed power generation output; Represents a node The maximum ramp power of the distributed power supply; Represents a node Maximum charge / discharge power of energy storage; Represents a node During the period Energy storage capacity; , Representing nodes respectively The upper and lower limits of energy storage capacity.

[0078] More specifically, identify exogenous information, where exogenous information Including time period The renewable energy output, load, and electricity price information obtained from real-time observations are given by formula (15): In the formula, They represent time periods respectively. The renewable energy output, load, and electricity price information obtained from real-time observation are used to determine the state transition equation. The state transition equation is used to characterize the transition process from making a decision in a certain state to the next state. In the real-time optimization problem of the park microgrid, it can be defined as the transition of the energy storage charging state, i.e., formula (16): In the formula, Represents a node The charge / discharge coefficient of energy storage.

[0079] Finally, the cost function is determined. The cost function is used to represent the specific cost of making a certain decision in a certain state at a certain time period. In the real-time optimization problem of microgrids, it can be expressed as the following formula (17):

[0080]

[0081] In the formula, , , These represent the sets of nodes for energy storage, distributed power sources, and renewable energy, respectively. Represents the set of microgrid nodes; , Representing nodes respectively Cost coefficient of distributed power generation and energy storage systems; , These represent the load reduction penalty factor and the renewable energy reduction penalty factor, respectively. , These represent the purchase price and the selling price of electricity, respectively.

[0082] Furthermore, based on the Bellman equations, the value function and optimal decision of the microgrid can be obtained by solving the following formulas (18)-(19):

[0083]

[0084]

[0085] To improve the computational efficiency of online optimization, the approximate dynamic programming algorithm usually divides the problem into two stages: offline training and online decision-making. In the offline training stage, the given samples are used to learn and construct the value function corresponding to the state space of the microgrid in the park. In the online decision-making stage, the value function corresponding to the current state is substituted into formula (19) to solve and obtain the optimal operation strategy of the microgrid.

[0086] In one exemplary embodiment, this application constructs a correlation between the offline training value function and the microgrid operating scenario based on an improved approximate dynamic programming algorithm using stepwise Nystrom approximation and Wiener kernel regression, thereby improving the real-time optimization performance of the microgrid in complex and variable operating environments. The specific steps are as follows:

[0087] First, let's introduce the problem scenario. The original Markov decision process is improved to map the operating scenario to the decision space, switching costs, and optimal strategy. In the microgrid optimization problem, the problem scenario... It can be described in the form of formula (20): In the formula, These represent the expected distributions of renewable energy output, load, and electricity price uncertainty in the microgrid under the current scenario; These represent the standard deviations of the uncertainty distributions of renewable energy output, load, and electricity price in the microgrid under the current scenario; based on the problem scenario The value function formula of the approximate dynamic programming algorithm will become the form of formula (21): In the formula, The scenario value function will be calculated to have different values ​​under different microgrid operating environments. Based on formula (21), the scenario value function can be obtained from the scenario. With system status A joint decision, when When determined, a unique value can be obtained, namely formula (22): .

[0088] The aforementioned mapping relationship is implicit in Markov decision processes. To address this issue, this application employs a pre-defined distribution Nystrom approximation method to map the scene and state variables to the kernel space, and uses a pre-defined Wiener kernel regression method to explicitly describe the relationship between the scene value function and the scene and state variables. Specifically, this includes: Sample set construction: randomly generated... Each microgrid faces uncertain scenarios, with different loads, renewable energy outputs, and electricity price curves, as shown in formula (23): For each scenario The state space of its value function can be described as a simplex, i.e., formula (24): In the formula, Indicates system state The dimension; Representing a scene Next The first affine independent state, that is, the first simple state. One vertex; Representing a scene Next The coefficients of each state; This represents the coefficient vector.

[0089] Based on the above theory, by extracting all vertices after simplex segmentation of each scene, we can obtain the set of scene and state variables, i.e., formula (25): For each The corresponding value function can be calculated using formula (21), and the output set can be constructed as shown in formula (26): .

[0090] Therefore, the above formulas (20)-(26) correspond to Figure 2 In step S202, multiple microgrid operation scenarios are randomly generated; each operation scenario includes at least one parameter among renewable energy output, load, expected value of electricity price uncertainty distribution, standard deviation of renewable energy output, standard deviation of load, and expected standard deviation of electricity price uncertainty distribution; in step S204, state space sampling points corresponding to multiple microgrid operation scenarios are collected, and each microgrid operation scenario is combined with the corresponding state space sampling points to obtain a scenario state data set.

[0091] Furthermore, the kernel space construction based on the preset step-by-step Nystrom approximation includes: the scene and state variable set constructed by formula (25) has a high dimension, making it difficult to directly describe the relationship between it and the scene value function. Therefore, this application adopts the step-by-step Nystrom approximation method to map it to a low-dimensional kernel space to reduce the solution complexity. The specific steps include: sampling the local set of landmarks to maintain approximate accuracy: from the scene and state variable set Randomly selected from Data as a large-scale landmark set And its kernel matrix is ​​calculated based on the Gaussian kernel formula, as shown in formula (27): In the formula, For large coordinate sets The sample data in the set; based on the Nystrom approximation theory, the set kernel matrix Can be derived from large coordinate sets kernel matrix Approximate representation, i.e., formulas (28)-(30): , , In the formula, express The false reversal; This represents the kernel matrix between the entire dataset and the landmark set; Representing a large coordinate set kernel matrix eigenvectors; Representing a large coordinate set kernel matrix eigenvalues.

[0092] Furthermore, using the above formulas (28)-(30), for high-dimensional datasets... Its kernel space features can be approximated by the eigenvectors of the kernel matrix of the large-scale landmark set, that is, the original high-dimensional dataset can be mapped to the low-dimensional kernel space through the large-scale landmark set.

[0093] In the Nystrom approximation, both accuracy and computation time are determined by the size of the landmark set. Landmark sets often require a large amount of data to ensure approximation accuracy, leading to high computational costs. To address this, this application further proposes a method using a small landmark set. To equivalently replace the large landmark set constructed earlier To improve overall computational efficiency while ensuring approximate accuracy, the formula is constructed as shown in formula (31): In the formula, Let Z be the Firobenius norm; Z is the desired set of small landmarks; The kernel matrix represents the scene, state data, and small landmark set; F is a binary matrix representing... The membership relationship, where each column contains only one unit value to indicate the membership. and The correspondence can be obtained by... Each data point in the matrix is ​​connected to the nearest data point in the Z matrix.

[0094] Furthermore, considering both computational accuracy and solution efficiency, this application employs the gradient descent algorithm to solve for the small landmark set. The specific steps include:

[0095] ① Using clustering methods such as K-means (K-means clustering algorithm) to obtain Major landmarks The cluster centers were used as the initial set of small landmarks. ;

[0096] ② Set the convergence threshold With maximum number of iterations ;

[0097] ③while( )and ( do:

[0098] ④ Calculation ;

[0099] ⑤ Calculate the kernel matrix gradient of the objective function using the following formulas (32)-(33);

[0100]

[0101]

[0102] ⑥ Calculate the gradient of the objective function with respect to each matrix element using formulas (34)-(35);

[0103]

[0104]

[0105] In the formula, Represents the trace of a matrix; Indicates the first element in the original input data. The first input sample Dimensional features; Indicates the first The input sample and the first Kernel function values ​​between small landmark samples.

[0106] ⑦ Update the small landmark set using the following formulas (36)-(37) With convergence tolerance;

[0107]

[0108]

[0109] In the formula, The set coefficient.

[0110] ⑧ ;

[0111] ⑨ Output the final optimal set of small landmarks obtained from the solution. .

[0112] In summary, the kernel matrix of the scene and state dataset shown in formula (25) can be approximately represented in the form of formula (38): In the formula, The kernel matrix represents the optimal set of small landmarks and the original data; This represents the binary membership matrix of the optimal small landmark set.

[0113] Therefore, the above formulas (27)-(37) can be represented as follows: Figure 3In step S302, data is randomly extracted from the scenario state data set as the large coordinate set, and the kernel matrix of the large coordinate set is obtained; in step S304, the initial cluster center of the large coordinate set is obtained based on the preset clustering algorithm and the kernel matrix, and the initial value of the small coordinate set is determined based on the initial cluster center, while setting the convergence threshold and the maximum number of iterations; in step S306, the small coordinate set is updated through the preset gradient descent algorithm, and the optimal small coordinate set is output.

[0114] Through the above steps, the originally high-dimensional scene and state data are successfully transformed into low-dimensional kernel spatial features using small landmark sets, reducing the computational complexity from... Transform into Due to the number of small landmark sets Much smaller than the original dataset size This will significantly improve computational efficiency.

[0115] More importantly, the mapping expression is constructed based on Wiener kernel regression: In the process of constructing the value function in the above scenarios, there are generally problematic scenarios. Observation error, system state The observation error, kernel approximation error and other noises. Therefore, this application uses the preset Wiener kernel regression algorithm to realize the linear mapping from the low-rank kernel space features of the scene and state variables to the scene value function, and transforms the deterministic weights in the original kernel regression into random weights to capture noise information. The optimal weights under noise-free conditions are separated by polynomial chaotic expansion technology, eliminating the influence of noise on the construction of the scene value function and further improving the fitting accuracy. The specific steps include: constructing the objective function considering noise information: considering the influence of noise information in the construction of the scene value function, the fitting relationship between the value function and the scene and state variables can be expressed as formula (39): In the formula, This represents the elements of the output set, i.e., the corresponding value functions obtained by solving the problem; This represents information about the input set, namely the scenario and state variables; This is noise error; based on prior knowledge of noise distribution, noise... It can be assumed to conform to a Gaussian distribution, i.e., formulas (40)-(41): , Based on the above, the kernel regression objective function can be expressed as the following formula (42):

[0116]

[0117] In the formula, The kernel space features representing the scene and state data set can be obtained by solving equation (38); To account for the random weights of noise information, they exist in the probability space. Among them This represents the sample space corresponding to the scene and state data. express Algebra This represents a probability measure, specifically the probability that a noise value falls within a certain interval. Represents the set of real numbers; Expressing expectations; Indicates the number of scenarios and state datasets; The noise standard deviation can be estimated through simulation experiments or historical data. Solving formula (42) yields the optimal random weights. .

[0118] Among them, the separation of noise-free optimal weights: based on the theory of "homogeneous chaos", the optimal random weights It can be expressed as a series of deterministic coefficients plus noise-related terms, i.e., formula (43): In the formula, Indicates a weighting coefficient that is independent of noise; The weighting coefficients related to noise are the core weighting coefficients calculated based on formula (43). The relationship between the scene value function and the scene and state variables can be explicitly described, i.e., formula (44): Formula (44) successfully eliminates the influence of various noises in the calculation and measurement process, and more accurately describes the relationship between the value function and the scene and state variables.

[0119] Therefore, the above formulas (39)-(44) can correspond to as follows: Figure 4 Step S402: Construct a kernel regression objective function containing noise terms based on Gaussian distribution; Step S404: Process the kernel space features of the scenario state data based on the kernel regression objective function to obtain the corresponding random weights; Step S406: Process the random weights based on the preset homogeneous chaotic algorithm to separate the deterministic coefficients and noise-related terms to obtain the corresponding optimal weight coefficients; Step S408: Construct the correlation between the value function, the running scenario, and the state variables based on the optimal weight coefficients.

[0120] In summary, based on the above content, for new microgrid scenarios... The value function corresponding to each sampling point The following formulas (45)-(47) can be substituted directly to obtain the approximate dynamic programming algorithm value function, which improves the adaptability of the approximate dynamic programming algorithm value function to the complex operating environment of microgrids: , , In the formula, This represents the combination of the new operating scenario and the state space sampling points.

[0121] In one embodiment, a Loewner interpolation approximation algorithm is further proposed to improve the construction process of the value function. The Loewner interpolation approximation algorithm greedily selects points based on the local features of the value function to improve the utilization rate of sample points, thereby enabling the fitting of nonlinear functions with fewer sample points. The specific steps include:

[0122] First, initial sample points are selected. Due to factors such as operational constraints, the value function may change significantly at the state space boundary. Therefore, this application first defines the energy storage capacity boundary as the key region for state space sampling, selects a large number of sample points within the region, and a small number of sample points outside the region, constructing a set. And calculate its corresponding value function using formula (21). The initial sample set is represented by formula (48): In the formula, , This represents the state dimension. Then, further analysis of the above sample set... We split the sample into two disjoint sample sets, as shown in formulas (49)-(51): , , In the formula, Covering regions where the value function changes drastically. Covering regions where the value function changes gradually. and A reasonable decomposition can capture the global characteristics of the microgrid value function.

[0123] Secondly, construct the value function model of Loewner interpolation: based on the above sample set. and The Loewner matrix and the shifted Loewner matrix are constructed using the formulas shown in (52)-(53): , In the formula, if N S >1, For element-wise subtraction, matrix division is implemented using the Moore-Penrose pseudo-inverse (Moore-Penrose Generalized Inverse Matrix); if N S =1, and formulas (52)-(53) degenerate into scalar division. Furthermore, based on formulas (52)-(53), we can obtain the following formulas (54)-(55):

[0124]

[0125]

[0126] In the formula, For one The block matrix, For the position located at the The identity matrix of each block position, formula (55) is the projection matrix of formula (54), and based on the projection matrix constructed by formula (55), the following formulas (56)-(57) can be further derived:

[0127]

[0128]

[0129] In the formula, Represents the sample set Value function column stacked matrix; Represents the sample set The value function is a stacked matrix of rows.

[0130] Therefore, based on the above, the explicit value function formula corresponding to the state space of the microgrid can be derived, namely formula (58): .

[0131] Therefore, the above formulas (48)-(58) can correspond to as follows: Figure 5 In step S502, the pre-constructed sample set is split into a first sample set covering the region of drastic change in the value function and a second sample set covering the region of gradual change in the value function; in step S504, a Loewner matrix and a shifted Loewner matrix are constructed based on the first sample set and the second sample set, respectively; in step S506, a projection matrix is ​​constructed based on the Loewner matrix and the shifted Loewner matrix, and the value functions of the first sample set and the second sample set are stacked based on the projection matrix to obtain the value function formula.

[0132] In summary, the complete calculation process of the real-time optimized operation method for industrial park microgrids based on the improved approximate dynamic programming algorithm proposed in this application is as follows:

[0133] Offline phase: ① Randomly generated ① For each uncertain operation scenario of a microgrid, state space sampling points are extracted based on rational interpolation theory; ② Construct the scenario state set as shown in formula (25), and use the Nystrom approximation algorithm to calculate its corresponding large coordinate set. That is, formula (27)-(30); ③ Further calculate the large coordinate set using formula (32)-(37). The corresponding optimal small landmark set , reduce computational complexity; map scene state data to low-dimensional kernel space according to formula (38) to obtain its corresponding kernel space features; ⑤ consider the influence of noise in the value function construction process, construct Wiener kernel regression mapping formula to capture noise information, i.e. formula (42); ⑥ further use the "homogeneous chaos" theory to separate the weight coefficients that are not related to noise, i.e. formula (43); ⑦ finally construct the correlation formula as shown in formula (45)-(47) to explicitly characterize the correlation between scene value function and scene and state variables.

[0134] Online Phase: ① Determine the current operating scenario of the park's microgrid. Combined with state-space sampling points, an input is constructed. ; ② Use formulas (45)-(47) to calculate the corresponding value functions. ; ③ will and Construct a disjoint sample set as shown in formulas (48)-(51), and use the sample set to update the explicit value function formula of the microgrid constructed by the Loewner interpolation approximation algorithm, i.e., formula (58); ④ Input the current system state of the microgrid into formula (58) to obtain the corresponding value function; ⑤ Finally, substitute the value function into formula (19) to obtain the optimal operation strategy of the microgrid.

[0135] Therefore, the above formulas (1)-(58) can be used to obtain the optimal small landmark set, the combination of the operating scenario and the state space sampling points, the kernel matrix of the optimal small landmark set and the original data, and the binary membership matrix of the optimal small landmark set, thus obtaining the optimal operating strategy of the microgrid.

[0136] Through the above embodiments, an improved approximate dynamic programming algorithm based on stepwise Nystrom approximation and Wiener kernel regression is proposed. This algorithm can explicitly characterize the correlation between the value function and the microgrid operation scenario, improving the adaptability of the approximate dynamic programming algorithm to the complex operating environment of the park microgrid. Furthermore, the proposed improved approximate dynamic programming algorithm can help the microgrid make rapid and accurate online decisions in complex operating environments, which is conducive to the real-time and efficient operation of the park microgrid.

[0137] More often, in one embodiment, such as Figure 6As shown, firstly, the microgrid operating scenario and system state are mapped to the kernel space based on the pre-defined distributed Nystrom approximation algorithm. Then, the pre-defined Wiener kernel regression algorithm is used to describe the relationship between the value function formed by the approximate dynamic programming algorithm and the kernel space variables. The corresponding value function is directly obtained through the microgrid operating scenario and state variables, which improves the adaptability of the value function formed by the offline training of the approximate dynamic programming algorithm to complex operating environments. Secondly, based on rational interpolation theory, key sampling points in the state space are selected as inputs, and the Loewner interpolation approximation algorithm is used to explicitly construct the value function formula corresponding to the microgrid state space. This effectively reduces the number of samples in the microgrid state space during the value function construction process and improves the training efficiency of the value function in the offline stage of the approximate dynamic programming algorithm.

[0138] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0139] Based on the same inventive concept, this application also provides a microgrid real-time optimization operation device for implementing the microgrid real-time optimization operation method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the microgrid real-time optimization operation device provided below can be found in the limitations of the microgrid real-time optimization operation method described above, and will not be repeated here.

[0140] In one exemplary embodiment, such as Figure 7 As shown, a microgrid real-time optimization operation device is provided, comprising: a data acquisition module 701, a data processing module 702, and a strategy generation module 703, wherein:

[0141] The data acquisition module 701 is used to construct sample sets under various operating scenarios of the microgrid, and at the same time extract the state space sampling points of the microgrid under each operating scenario to obtain the scenario state data set.

[0142] The data processing module 702 is used to process the large coordinate set and the small landmark set in the scenario state data set sequentially based on the preset step-by-step Nystrom approximation algorithm, and output the optimal small landmark set.

[0143] The data processing module 702 is also used to construct the correlation between value functions, operating scenarios and state variables based on the preset Wiener kernel regression algorithm, and to construct the value function formula corresponding to the state space of the microgrid based on the preset Loewner interpolation approximation algorithm.

[0144] The strategy generation module 703 is used to obtain the current operating scenario of the microgrid in real time, process the current operating scenario and the optimal small landmark set based on the value function formula, and obtain the optimal operating strategy of the microgrid.

[0145] Furthermore, in one embodiment, the data acquisition module 701 is also used to randomly generate multiple microgrid operation scenarios; each operation scenario includes at least one parameter among renewable energy output, load, expected distribution of electricity price uncertainty, standard deviation of renewable energy output, standard deviation of load, and expected standard deviation of electricity price uncertainty distribution; collect state space sampling points corresponding to multiple microgrid operation scenarios, and combine each microgrid operation scenario with the corresponding state space sampling points to obtain a scenario state data set.

[0146] Furthermore, in one embodiment, the data acquisition module 701 is also used to randomly extract data from the scenario state data set as a large coordinate set and obtain the kernel matrix of the large coordinate set; obtain the initial cluster center of the large coordinate set based on the preset clustering algorithm and the kernel matrix, and determine the initial value of the small coordinate set based on the initial cluster center, while setting the convergence threshold and the maximum number of iterations; update the small coordinate set through the gradient descent algorithm and output the optimal small coordinate set.

[0147] Furthermore, in one embodiment, the data processing module 702 is also used to construct a kernel regression objective function containing noise terms based on a Gaussian distribution; process the kernel space features of the scenario state data based on the kernel regression objective function to obtain the corresponding random weights; process the random weights based on a preset homogeneous chaotic algorithm to separate deterministic coefficients and noise-related terms to obtain the corresponding optimal weight coefficients; and construct the correlation between the value function, the running scenario, and the state variables based on the optimal weight coefficients.

[0148] Furthermore, in one embodiment, the data processing module 702 is also used to split the pre-constructed sample set into a first sample set covering the region of drastic change in the value function and a second sample set covering the region of gradual change in the value function; construct a Loewner matrix and a shifted Loewner matrix based on the first sample set and the second sample set respectively; construct a projection matrix based on the Loewner matrix and the shifted Loewner matrix; and stack the value functions of the first sample set and the second sample set based on the projection matrix to obtain the value function formula.

[0149] Furthermore, in one embodiment, the strategy generation module 703 is also used to obtain the optimal small landmark set, the combination of the operating scenario and the state space sampling points included in the current operating scenario of the microgrid, the kernel matrix of the optimal small landmark set and the original data, and the binary membership matrix of the optimal small landmark set, so as to obtain the optimal operating strategy of the microgrid.

[0150] Each module in the aforementioned microgrid real-time optimization operation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0151] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores real-time optimized operation data for the microgrid. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a real-time optimized operation method for the microgrid.

[0152] Those skilled in the art will understand that Figure 8The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0153] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0154] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0155] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0156] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0157] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0158] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0159] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for real-time optimized operation of a microgrid, characterized in that, The method includes: A sample set is constructed for various operating scenarios of microgrids, and the state space sampling points of the microgrids under each operating scenario are extracted to obtain a scenario state data set. The large coordinate set and small landmark set in the scenario state data set are processed sequentially based on the preset step-by-step Nystrom approximation algorithm to output the optimal small landmark set; The relationship between the value function, the operating scenario, and the state variables is constructed based on the preset Wiener kernel regression algorithm, and the value function formula corresponding to the state space of the microgrid is constructed based on the preset Loewner interpolation approximation algorithm. The current operating scenario of the microgrid is acquired in real time, and the current operating scenario and the optimal small landmark set are processed based on the value function formula to obtain the optimal operating strategy of the microgrid.

2. The method according to claim 1, characterized in that, The process involves constructing sample sets for various operating scenarios of the microgrid, and simultaneously extracting state-space sampling points of the microgrid under each operating scenario to obtain a scenario state data set, including: Multiple microgrid operation scenarios are randomly generated; each operation scenario includes at least one of the following parameters: renewable energy output, load, expected value of electricity price uncertainty distribution, standard deviation of renewable energy output, standard deviation of load, and expected standard deviation of electricity price uncertainty distribution; Multiple state space sampling points corresponding to the microgrid operation scenarios are collected, and each microgrid operation scenario is combined with its corresponding state space sampling points to obtain the scenario state data set.

3. The method according to claim 1, characterized in that, The method based on the preset step-by-step Nystrom approximation algorithm processes the large coordinate set and the small landmark set in the scenario state data set sequentially, and outputs the optimal small landmark set, including: Randomly extract data from the scenario state data set as a large coordinate set, and obtain the kernel matrix of the large coordinate set; The initial cluster centers of the large coordinate set are obtained based on the preset clustering algorithm and the kernel matrix, and the initial values ​​of the small coordinate set are determined based on the initial cluster centers. At the same time, the convergence threshold and the maximum number of iterations are set. The set of small landmarks is updated using a preset gradient descent algorithm, and the optimal set of small landmarks is output.

4. The method according to claim 1, characterized in that, The relationship between the value function constructed based on the Wiener kernel regression algorithm, the operating scenario, and the state variables includes: Construct a kernel regression objective function containing a noise term based on the Gaussian distribution; Based on the kernel regression objective function, the kernel space features of the scenario state data are processed to obtain the corresponding random weights; The random weights are processed using a pre-defined homogeneous chaotic algorithm to separate the deterministic coefficients and the noise-related terms, thereby obtaining the corresponding optimal weight coefficients. The value function, the operating scenario, and the state variables are associated based on the optimal weight coefficients.

5. The method according to claim 1, characterized in that, The method for constructing the value function formula corresponding to the state space of the microgrid based on the preset Loewner interpolation approximation algorithm includes: The pre-constructed sample set is split into a first sample set covering the region of drastic change in the value function and a second sample set covering the region of gradual change in the value function; Construct the Loewner matrix and the shifted Loewner matrix based on the first sample set and the second sample set, respectively; A projection matrix is ​​constructed based on the Loewner matrix and the shifted Loewner matrix. The value functions of the first sample set and the second sample set are stacked and processed based on the projection matrix to obtain the value function formula.

6. The method according to claim 1, characterized in that, The process of acquiring the current operating scenario of the microgrid in real time, processing the current operating scenario and the optimal small landmark set based on the value function formula, and obtaining the optimal operating strategy of the microgrid includes: The optimal operating strategy of the microgrid is obtained by acquiring the optimal set of local landmarks, the combination of the operating scenario and the state space sampling points, the kernel matrix of the optimal set of local landmarks and the original data, and the binary membership matrix of the optimal set of local landmarks.

7. A microgrid real-time optimized operation device, characterized in that, The device includes: The data acquisition module is used to construct sample sets under various operating scenarios of the microgrid, and at the same time extract the state space sampling points of the microgrid under each operating scenario to obtain the scenario state data set; The data processing module is used to process the large coordinate set and the small landmark set in the scenario state data set sequentially based on the preset step-by-step Nystrom approximation algorithm, and output the optimal small landmark set; The data processing module is also used to construct the correlation between the value function, the operating scenario and the state variables based on the preset Wiener kernel regression algorithm, and to construct the value function formula corresponding to the state space of the microgrid based on the preset Loewner interpolation approximation algorithm. The strategy generation module is used to acquire the current operating scenario of the microgrid in real time, process the current operating scenario and the optimal small landmark set based on the value function formula, and obtain the optimal operating strategy of the microgrid.

8. 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 method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images