A distributed online energy management method for smart grid

Abstract

This invention discloses a distributed online energy management method for smart grids, relating to the field of smart grid energy management technology. It establishes a communication topology diagram based on the communication between each distributed generation (DG) in the grid system to determine initial data; it establishes an energy management model based on the actual constraints of generation balance and power output limitations in the smart grid; it constructs a gradient estimator to replace the true gradient information, proposing a Bandit-based distributed online energy management method to achieve optimal power output for each generator; this invention solves the energy management problem with energy storage devices under directed graphs, constructing a locally stochastic gradient-free estimator to replace the true gradient and guide the update of output power; addressing the difficulty of simultaneously sampling multiple points for online time-varying objective functions, it proposes an online distributed single-point gradient-free estimation method. Compared with two-point and multi-point gradient-free estimation methods, this method only requires sampling once at each time step, making it more convenient to operate.

Description

Technical Field

[0001] This invention relates to the field of smart grid energy management technology, and in particular to a distributed online energy management method for smart grids. Background Technology

[0002] Smart grids are one of the fundamental infrastructures of modern society, containing vast amounts of energy, energy storage facilities, and controllable loads, which leads to the complexity of smart grid energy management. Increasing research now recognizes that integrating and optimizing energy systems across multiple time scales can significantly improve economic efficiency and overall operational efficiency. Therefore, energy management has received considerable attention, emphasizing the inherent complementarity of various energy sources and developing a comprehensive energy management approach to effectively achieve optimal energy allocation.

[0003] In recent years, due to the rapid growth of renewable energy, an increasing number of distributed generation (DG) generators have been installed in energy management systems. In these distributed power systems, traditional centralized control methods are no longer suitable due to drawbacks such as high single-point failure rates, poor scalability, and heavy information transmission burdens. Furthermore, the intermittent and stochastic nature of DG power generation and the transient characteristics of most load changes make it difficult for predicted data to approximate true values, and discrete optimization strategies lack real-time capability. In contrast, online distributed optimization methods can adjust energy management control strategies in real time according to environmental changes. Therefore, distributed online energy management methods can solve the current problems.

[0004] A search of existing smart grid optimization and scheduling literature revealed that, among the existing technical literature, Ma Kai, Yu Yangqing, et al. published a paper titled "A Distributed Economic Dispatch Algorithm for Power Grids Based on Gradient Descent and Consensus" in *Science in China: Information Science* (2018, 48(10): 1364-1380), proposing a distributed optimization algorithm based on gradient descent and consensus strategies to minimize the total power generation cost. Liu Qingshan published a paper titled "A Distributed Optimization Algorithm Based on Multiagent Network for EconomicDispatch With Region Partitioning" in IEEE Transactions on Cybernetics (2019), proposing a distributed optimization algorithm to solve the economic dispatch problem with a connected graph. Comparing existing patent technology literature, Li Dongyuan's "A Distributed Smart Grid Energy Dispatch Method Based on Consensus Optimization Algorithm," Cai Defu's "A Distributed Economic Dispatch Method for Energy Storage Batteries Based on Differential Privacy Mechanism," and Li Chunhua's "An Energy Dispatch Optimization Method for Isolated Multi-Microgrids Based on Improved Multi-Agent Consensus Algorithm" are all power grid energy dispatch optimization methods designed based on consensus algorithms. The aforementioned literature assumes that cost function information is readily available. However, cost function information is usually incomplete in real-world systems. Furthermore, there is limited research on power grid energy management problems with multiple constraints and difficulty in obtaining gradient information, which fails to meet the needs of practical power grid energy management. Therefore, a distributed online energy management method for smart grids is proposed. Summary of the Invention

[0005] The purpose of this application is to provide a distributed online energy management method for smart grids to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this application provides the following technical solution: a distributed online energy management method for smart grids, comprising the following steps:

[0007] S1. Establish a communication topology diagram based on the communication between each DG in the power grid system and determine the initial data;

[0008] S2. Establish an energy management model based on the actual constraints of power generation balance and power output limits in the smart grid;

[0009] S3. Construct a gradient estimator to replace the real gradient information and propose a distributed online energy management method based on Bandit to achieve the optimal power output of each generator.

[0010] Preferably, the initial data includes total power demand and initial power for each DG.

[0011] Preferably, in step S2, the energy management model includes corresponding models for coal-fired generators, oil-fired generators, and energy storage systems, wherein:

[0012]

[0013]

[0014]

[0015] Where F t C represents the total revenue. t i P represents the total cost of electricity generation. D P represents the total load demand. S ρ represents the interface power connected to the main power grid. s Let N be the main grid electricity price, and N represent the number of generators. but:

[0016]

[0017] Where Ω represents a closed convex set, and P t ∈Ω indicates that all constraints are contained within the closed convex set.

[0018] Preferably, in step S3, the gradient is estimated based on the initial parameters.

[0019]

[0020] Preferably, each of the generators uses the following algorithm to update its output power:

[0021]

[0022]

[0023] Where Π represents the projection onto the closed convex set Ω, when P t i When the value exceeds the power generation limit, P is projected... t i Map back to the closed convex set, [A r ], [A c These are row random matrices and column random matrices, respectively. It is the auxiliary output power P t i Updated variable, α t It is the decay step size, and ε is a positive parameter.

[0024] In summary, the technical effects and advantages of this invention are as follows:

[0025] 1. This invention solves the energy management problem with energy storage devices in a directed graph, where the explicit cost function in the power grid is unknown and time-varying, which is more suitable for the actual generator set situation. A local stochastic gradient-free estimator is constructed to replace the real gradient and guide the update of output power.

[0026] 2. To address the problem of difficulty in simultaneously sampling multiple points for online time-varying objective functions, this invention proposes an online distributed single-point gradient-free estimation method. Compared with two-point and multi-point gradient-free estimation methods, this method only requires sampling once at each time step, making it more convenient to operate. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0028] Figure 1 This is a flowchart of the distributed online energy management method for smart grids in this embodiment;

[0029] Figure 2 This is a communication topology diagram of the three schedulable units in this embodiment;

[0030] Figure 3 This is a graph showing the relationship between output power and cost fitted based on function values ​​in this embodiment;

[0031] Figure 4 This is a graph showing the output power curves of each distributed power source in this embodiment;

[0032] Figure 5 This is a diagram showing the balance between total output power and total demand power in this embodiment;

[0033] Figure 6 This is a performance comparison chart of the algorithm (OPBF) in this embodiment with the RGF-DPGD and DPS algorithms;

[0034] Figure 7 This is a diagram showing the impact of dimension d on algorithm performance in this embodiment;

[0035] Figure 8 These are the cost function parameters and the upper and lower limits of the output power. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Example: Reference Figure 1 The method for distributed online energy management in a smart grid, as shown, includes the following steps:

[0038] S1. Establish a communication topology diagram based on the communication between each DG in the power grid system and determine the initial data;

[0039] S2. Establish an energy management model based on the actual constraints of power generation balance and power output limits in the smart grid;

[0040] S3. Construct a gradient estimator to replace the real gradient information and propose a distributed online energy management method based on Bandit to achieve the optimal power output of each generator.

[0041] In this embodiment, the power grid system consists of two distributed generation (DG) systems and one energy storage system (ESS). A corresponding communication topology is established based on the communication and status information of each distributed generation, such as... Figure 2 As shown. Considering the difficulty in obtaining gradient information for distributed power sources in practical applications, this application considers the case where the cost function form is unknown and time-varying, and only the function value is known. The cost function curve is fitted based on empirical data, as shown... Figure 3 As shown. By reviewing existing technical literature and comparing fitted curves, the cost functions of DG and ESS can be approximated as quadratic forms.

[0042] The cost function of DG can be expressed as:

[0043]

[0044] Where a i b i c i It is a cost parameter, N d It refers to the quantity of DG.

[0045] The power limit of DG can be expressed as:

[0046]

[0047] in These represent the lower and upper limits of the output of the i-th generator, respectively.

[0048] The cost function of ESS can be expressed as:

[0049]

[0050] in It is a penalty factor for rapid charging and discharging. It is a cost factor, N e This indicates the number of ESSs.

[0051] The power limit of an ESS is expressed as follows:

[0052]

[0053] in That is the maximum charging power. It is the maximum power of the discharge.

[0054] Considering both DG and ESS together, the total cost function can be expressed as:

[0055]

[0056] The corresponding new power output limit range is:

[0057]

[0058] Where N = N d +N e When the power generation unit is DG, When the power generation unit is ESS

[0059] Considering supply and demand balance constraints and power generation constraints, a smart grid energy management model is established:

[0060]

[0061]

[0062]

[0063]

[0064] Where F t Represents total revenue. P represents the total cost of electricity generation. D P represents the total load demand. S ρ represents the interface power connected to the main power grid. s It refers to the electricity price on the main power grid.

[0065] Update the output power using the following steps:

[0066] S1. Initialization: At iteration t=0, the initial power of generator i is... Let y be the average of the total demand, for all i∈V. i (0) = 0. The row random matrix A is obtained from the communication diagram. r Sum of column random matrices A c ε and the accuracy parameter η can be set to any small positive numbers. Setting the dimension d and the exploration parameter δ to 1... It is a zero-mean, unit random variable. The step size is set to...

[0067] S2. Calculate the estimated gradient: Estimate the gradient based on the initial parameters and the gradient calculation method below.

[0068]

[0069] S3, Global Constraint Projection: At iteration t≥1, each generator j collects its generated P t i and weight matrix information Then send the information to its neighbor. Then, generator i uses this information to update its power according to the following iterative rules. The actual constraints of each generator are contained in the projection set Ω.

[0070]

[0071]

[0072] S4. Return the updated estimate: Compare the updated estimate in S3. Compared to the previous P t i The difference, if Then let t = t + 1, and proceed to the first step. If Then find the optimal output power P for each generator. i * It outputs optimal power.

[0073] The parameter settings related to the algorithm in this embodiment are as follows:

[0074] ε=0.01, d=1, δ=1, P D =105MW, η=0.01,

[0075] The cost function parameters and upper and lower limits of output power are shown in [reference]. Figure 8 .

[0076] This embodiment was tested on an IEEE-6 bus system, which consists of one ESS, two DGs (one coal-fired generator and one oil-fired generator), three loads, and eleven transmission lines. Under the given conditions, the method proposed in this invention was simulated and the simulation results were analyzed. The simulation results are attached. Figure 4 - Appendix Figure 7 As shown. Among them, the appendix Figure 4 The optimal output power of DG1 and ESS is clearly shown. After 10 iterations, the optimal output powers of DG1, DG2, and ESS are P1 = 44.19 MW, P2 = 43.36 MW, and P3 = 17.5 MW, respectively. Simultaneously, the optimal solutions satisfy their respective power output constraints. Then, the three power outputs generated in each iteration are summed and compared with the total power demand; the results are shown in the appendix. Figure 5 As shown in the figure, it is clear that the total output power always equals the total demand, indicating that the supply and demand balance is met.

[0077] Then, the advantages of the present invention are highlighted by comparing it with methods in other technical literature. The stochastic gradient-free distributed projective gradient descent (RGFDPGD) algorithm and the directional distributed projective subgradient (D-DPS) algorithm are respectively applied to the model established in this invention, and the average regret curves of the three algorithms are shown in the attached figure. Figure 6 As shown, the D-DPS algorithm exhibits the best convergence performance, followed by RGF-DPGD, while the algorithm proposed in this invention (OPBF) has the worst convergence performance. Although its convergence performance is poor, this invention only requires one point of function value information, while the RGF-DPGD algorithm requires two points, and the D-DPS algorithm requires knowledge of the function's gradient. Obviously, in practical electromagnetic analysis, it is not guaranteed that two points of function value or gradient information can be found every time. The algorithm proposed in this invention (OPBF) has a wider applicability, therefore, a slight loss in convergence performance is acceptable.

[0078] Finally, the impact of dimension d on algorithm performance was analyzed, and specific results were given as an example. Figure 7 As shown, d=1 is the parameter selected in this invention. It can be seen that the convergence performance of the algorithm deteriorates as d increases. The smaller the value of d, the greater the variability of the algorithm. Therefore, the parameter selected in this invention is optimal.

[0079] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A distributed online energy management method for a smart grid, characterized by, Includes the following steps: S1. Establish a communication topology diagram based on the communication between each DG in the power grid system and determine the initial data; S2. Establish an energy management model based on the actual constraints of power generation balance and power output limits in the smart grid; S3. Construct a gradient estimator to replace the real gradient information and propose a distributed online energy management method based on Bandit to achieve the optimal power output of each generator; In S2, the energy management model includes the corresponding model of coal-fired generators, oil-fired generators and energy storage systems, wherein: wherein represents the total income, represents the total power generation cost, represents the total load demand, represents the interface power connected to the main grid, is the main grid electricity price, N represents the number of generators, and let then: in Describes a closed convex set. This indicates that all constraints are contained within the closed convex set; In step S3, the gradient is estimated based on the initial parameters. : 。 2. The distributed online energy management method for smart grids according to claim 1, characterized in that: The initial data includes the total power demand and the initial power of each DG.

3. A distributed online energy management method for smart grids according to claim 2, characterized in that: Each generator uses the following algorithm to update its output power: in, Indicates a closed convex set The projection on the surface, when When the value exceeds the power generation limit, it will be projected... Map back to the closed convex set. , These are row random matrices and column random matrices, respectively. It is the auxiliary output power Updated variables, It is the decay step size. It is a positive parameter.

Images