A micro-grid network online distributed optimization economic dispatch method and system

Abstract

The application discloses a kind of micro-grid network online distributed optimization economic scheduling method and system.The local cost function of each micro-grid node is modeled as strong pseudo-convex function in the application, which effectively solves the optimization problem of micro-grid network with strong pseudo-convex characteristic objective function under nonlinear power generation cost and power characteristics;By obtaining the random gradient containing heavy-tailed noise in each iteration and using the time-varying clipping threshold, nonlinear mapping is performed to suppress extreme outliers, overcoming the problem of complex dynamic environment caused by gradient explosion under heavy-tailed noise interference;Combined with the time-varying directed communication graph satisfying the periodic combination strong connectivity condition, the neighbor state weighted aggregation is performed, and the auxiliary optimization subproblem is constructed for projection update, thereby realizing the online distributed optimization scheduling scheme with high probability convergence and sublinear dynamic regret, and improving the scheduling reliability and long-term economy of micro-grid under severe noise and time-varying topology conditions.

Description

Technical Field

[0001] This invention relates to the field of distributed collaborative control technology for smart microgrids, and in particular to an online distributed optimized economic dispatch method and system for microgrid networks. Background Technology

[0002] With the large-scale integration of distributed renewable energy sources (such as wind and solar power), energy management and real-time scheduling of microgrid networks are becoming increasingly complex. Online distributed optimization scheduling of microgrids aims to minimize the overall network operating cost through local information interaction among multiple intelligent agents (such as distributed power sources and energy storage systems) during dynamic environmental changes.

[0003] However, in actual microgrid operation scenarios, existing distributed optimization scheduling technologies still face the following challenges: The non-convexity of the objective function. Traditional distributed optimization scheduling methods mostly assume the objective function is convex. However, in actual microgrid modeling, due to the diversity of generation costs, energy storage efficiency, and flexible load characteristics, local cost functions often exhibit strong pseudo-convexity. Although strongly pseudo-convex functions have better properties than general non-convex functions, conventional distributed gradient descent methods often struggle to guarantee the convergence and optimality of decisions when dealing with such functions.

[0004] Heavy-tailed characteristics of stochastic gradient noise. In microgrid environments, the gradient information acquired by agents is often interfered with by random noise. Existing techniques generally assume that the noise follows a light-tailed distribution with finite variance (such as a Gaussian distribution). However, in practical applications, due to extreme fluctuations in meteorological conditions (such as sudden strong winds or abrupt changes in cloud cover), occasional sensor failures, or severe jitter in communication links, random noise often exhibits a "heavy-tailed" characteristic, meaning that the probability of finding a maximal outlier is significantly higher than in a normal distribution. In this case, the second moment (variance) of the noise often does not exist, leading to gradient explosion or scheduling deviations in traditional stochastic gradient descent algorithms, seriously threatening the operational stability of the microgrid.

[0005] The dynamic nature of the environment. The load demand and resource output of microgrids fluctuate dramatically over time, requiring scheduling algorithms not only to perform static optimization but also to have low regret, i.e., to respond quickly to environmental changes.

[0006] Therefore, designing an online distributed optimization scheduling scheme with high probability of convergence and secondary line Regret for microgrid networks with strong pseudo-convex objective functions under complex dynamic environments with heavy-tailed noise interference has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] Therefore, it is necessary to provide an online distributed optimized economic dispatch method and system for microgrid networks to address the aforementioned technical problems.

[0008] The following technical solution is adopted in this specification: This specification provides an online distributed optimized economic dispatch method for microgrid networks, specifically including: A target microgrid network is obtained, comprising multiple microgrid nodes that interact with each other for local decision-making via a time-varying directed communication graph. An online distributed optimization scheduling model is constructed with the goal of minimizing the global operating cost of the target microgrid network at all operating times. At each time step, the local cost function of each microgrid node is a strongly pseudo-convex function on the microgrid network's operating constraint set. When scheduling the network using the online distributed optimization scheduling model, the following steps are executed iteratively: Obtain the stochastic gradient of the local cost function of each microgrid node, including heavy-tailed noise, and set the scheduling period.

[0009] A preset time-varying clipping threshold is used to perform a nonlinear mapping on the stochastic gradient containing heavy-tailed noise, transforming the stochastic gradient containing heavy-tailed noise into a clipped gradient that suppresses extreme outliers.

[0010] The decision state of each neighbor node in the set of neighbor nodes is obtained through the time-varying directed communication graph, and the communication weight matrix is ​​determined. The decision states of each neighbor node are weighted and aggregated according to the communication weight matrix to obtain local aggregated variables.

[0011] Based on the pruning gradient and local aggregation variables, construct an auxiliary optimization subproblem, solve it, obtain the decision state at the next time step, and project the decision state at the next time step into the microgrid network operation constraint set.

[0012] Determine whether the scheduling cycle has been reached. If so, combine the decision states of each microgrid node at each time point into a scheduling decision sequence, end the current scheduling, and output the scheduling decision sequence; otherwise, update the iteration count and continue iterating.

[0013] Furthermore, the time-varying directed communication graph satisfies the periodic combination strong connectivity condition, meaning that within any consecutive time interval, the parallel graphs of all communication graphs are strongly connected, thus supporting the dynamic access and disconnection of microgrid nodes.

[0014] Furthermore, for the local cost function of each microgrid node, there exists a positive number. This ensures that for any two decision states within the constraint set... All of them satisfy the strong pseudo-convexity property and are used to describe the nonlinear generation cost and power characteristics in microgrids.

[0015] Furthermore, a preset time-varying pruning threshold is used to perform a nonlinear mapping on the stochastic gradient containing heavy-tailed noise, transforming the stochastic gradient with heavy-tailed noise into a pruned gradient that suppresses extreme outliers. The expression is: ; in, For the first Each microgrid node in The stochastic gradient with heavy-tailed noise observed at each time step. The clipping threshold changes over time.

[0016] Furthermore, the preset time-varying pruning threshold satisfies the following condition: when the scheduling time approaches infinity, the growth rate of the pruning threshold is less than linear, and it satisfies an infinite series convergence condition related to the moment order of the heavy-tailed noise, thereby ensuring the convergence of the algorithm under high probability.

[0017] Furthermore, the communication weight matrix It satisfies the double random property, specifically: ; in, for The communication weight matrix at time step Line number Column elements, for The communication weight matrix at time step Line number The elements of the column.

[0018] Furthermore, an auxiliary optimization subproblem is constructed based on the pruning gradient and local aggregation variables, and solved to obtain the decision state at the next time step. Specifically: ; in, For the set of constraints for microgrid network operation, To optimize variables, For transpose calculation, It is a positive definite matrix. To learn step length, for Local aggregate variables at time points.

[0019] This specification provides an online distributed optimized economic dispatch system for microgrid networks, specifically including: The model building module is used to obtain a target microgrid network that includes multiple microgrid nodes that interact with local decision-making information through a time-varying directed communication graph, and to build an online distributed optimization scheduling model with the goal of minimizing the global operating cost of the microgrid network at all operating times.

[0020] The dynamic pruning module is used to obtain the stochastic gradient containing heavy-tailed noise of the local cost function of each microgrid node, and to use a preset pruning threshold that varies with time to transform the stochastic gradient containing heavy-tailed noise into a pruning gradient that suppresses extreme outliers.

[0021] The weighted aggregation module is used to weight and aggregate the decision states of each neighboring node according to the communication weight matrix to obtain local aggregated variables.

[0022] The decision state update module is used to construct an auxiliary optimization sub-problem based on the pruning gradient and local aggregation variables, solve it, obtain the decision state at the next time step, and project the decision state at the next time step into the microgrid network operation constraint set.

[0023] The output module determines whether the scheduling cycle has been reached. If so, it combines the decision states of each microgrid node at each time point into a scheduling decision sequence, ends the current scheduling, and outputs the scheduling decision sequence. Otherwise, it updates the iteration count and continues iterating.

[0024] This specification provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0025] The beneficial effects of this invention are as follows: Compared with the prior art, this invention effectively solves the optimization problem of microgrid networks with strong pseudo-convex objective functions under nonlinear generation cost and power characteristics by modeling the local cost function of each microgrid node as a strong pseudo-convex function; by obtaining stochastic gradients containing heavy-tailed noise in each iteration and using a time-varying pruning threshold to perform nonlinear mapping to suppress extreme outliers, it overcomes the complex dynamic environment problem of gradient explosion leading to scheduling jumps under heavy-tailed noise interference; on this basis, by combining a time-varying directed communication graph that satisfies the periodic combination strong connectivity condition to perform neighbor state weighted aggregation and constructing an auxiliary optimization subproblem for projection update, a high-probability convergent online distributed optimization scheduling scheme with sublinear dynamic Regret is realized, which significantly improves the scheduling reliability and long-term economic efficiency of microgrids under harsh noise and time-varying topology conditions. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0027] Figure 1 This is a flowchart illustrating an embodiment of the present invention; Figure 2 This is the time-varying directed communication diagram of the present invention; Figure 3 The output power of the microgrid node in this embodiment of the invention A schematic diagram of the trajectory; Figure 4 This is a schematic diagram illustrating the average values ​​of the stochastic gradient and the clipping gradient in an embodiment of the present invention. Figure 5 This is a consistency error diagram of an embodiment of the present invention; Figure 6 This is a dynamic regret graph from an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this invention.

[0029] This invention provides an online distributed optimization and economic scheduling method for microgrid networks, comprising six steps: constructing a distributed optimization scheduling model based on a time-varying directed communication graph; obtaining stochastic gradients; transforming stochastic gradients into pruning gradients based on a pruning threshold; weighted aggregation of the decision states of neighboring nodes; solving for the decision state at the next time step; and determining whether the process has ended and outputting the scheduling result. By modeling the local cost function of each microgrid node as a strongly pseudo-convex function, the optimization problem of microgrid networks with strongly pseudo-convex objective functions under nonlinear generation cost and power characteristics is effectively solved. By obtaining stochastic gradients containing heavy-tailed noise in each iteration and using a time-varying pruning threshold to perform nonlinear mapping to suppress extreme outliers, the complex dynamic environment problem of gradient explosion leading to scheduling jumps under heavy-tailed noise interference is overcome. Based on this, a time-varying directed communication graph satisfying the periodic combination strong connectivity condition is combined with neighbor state weighted aggregation, and an auxiliary optimization subproblem is constructed for projection update, thereby realizing an online distributed optimization scheduling scheme with high probability convergence and sublinear dynamic Regret, significantly improving the scheduling reliability and long-term economic efficiency of microgrids under harsh noise and time-varying topology conditions.

[0030] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0031] A method for online distributed optimal economic dispatch of microgrid networks specifically includes the following steps: A target microgrid network is obtained, comprising multiple microgrid nodes that interact locally through a time-varying directed communication graph. An online distributed optimization scheduling model is constructed with the optimization objective of minimizing the global operating cost of the target microgrid network at all operating times. At each time step, the local cost function of each microgrid node is a strongly pseudo-convex function on the microgrid network's operating constraint set. When scheduling the network using the online distributed optimization scheduling model, the following steps are executed iteratively: Step 1: Obtain the stochastic gradient of the local cost function of each microgrid node, including heavy-tailed noise, and set the scheduling period.

[0032] Specifically, the time-varying directed communication graph satisfies the periodic combination strong connectivity condition, meaning that all communication graphs are strongly connected within any continuous time interval, thus supporting the dynamic access and disconnection of microgrid nodes.

[0033] Furthermore, for the local cost function of each microgrid node, there exists a positive number. This ensures that for any two decision states within the constraint set... All of them satisfy the strong pseudo-convexity property and are used to describe the nonlinear generation cost and power characteristics in microgrids.

[0034] Step 2: Using a preset time-varying clipping threshold, perform a nonlinear mapping on the stochastic gradient containing heavy-tailed noise, transforming the stochastic gradient containing heavy-tailed noise into a clipped gradient that suppresses extreme outliers.

[0035] Specifically, a preset time-varying pruning threshold is used to perform a nonlinear mapping on the stochastic gradient containing heavy-tailed noise, transforming the stochastic gradient with heavy-tailed noise into a pruned gradient that suppresses extreme outliers. The expression is: ; in, For the first Each microgrid node in The stochastic gradient with heavy-tailed noise observed at each time step. The clipping threshold changes over time.

[0036] Preferably, the preset time-varying pruning threshold satisfies the following conditions: when the scheduling time approaches infinity, the growth rate of the pruning threshold is less than linear, and it satisfies an infinite series convergence condition related to the moment order of the heavy-tailed noise, thereby ensuring the convergence of the algorithm under high probability.

[0037] Step 3: Obtain the decision state of each neighbor node in the neighbor node set through the time-varying directed communication graph, and determine the communication weight matrix; weight and aggregate the decision states of each neighbor node according to the communication weight matrix to obtain local aggregate variables.

[0038] Furthermore, the communication weight matrix It satisfies the double random property, specifically: ; ; in, for The communication weight matrix at time step Line number Column elements, for The communication weight matrix at time step Line number The elements of the column.

[0039] Step 4: Construct an auxiliary optimization subproblem based on the pruning gradient and local aggregation variables, solve it to obtain the decision state at the next time step, and project the decision state at the next time step into the microgrid network operation constraint set.

[0040] Preferably, an auxiliary optimization subproblem is constructed based on the pruning gradient and local aggregation variables, and then solved to obtain the decision state at the next time step. Specifically: ; in, For the set of constraints for microgrid network operation, To optimize variables, For transpose calculation, It is a positive definite matrix. To learn step length, for Local aggregate variables at time points.

[0041] Step 5: Determine whether the scheduling cycle has been reached. If so, combine the decision states of each microgrid node at each time point into a scheduling decision sequence, end the current scheduling, and output the scheduling decision sequence; otherwise, update the iteration count and continue iterating.

[0042] Example This implementation example Figure 1 As shown, the specific steps include the following: S1 initialization steps: Construct a microgrid distributed network with multiple nodes, define the strongly pseudo-convex local cost function and operating constraint set for each node; initialize the symmetric positive definite matrix. Learning Step Size and dynamic gradient clipping threshold S2 stage iterative steps: Each microgrid node obtains the local cost stochastic gradient affected by heavy-tailed noise, and uses the pruning threshold... By performing a nonlinear mapping on it and restricting the gradient magnitude within a threshold range, the clipped gradient can be obtained. In the gradient clipping step, the gradient is clipped. The calculation formula is: in, For the first The stochastic gradient containing heavy-tailed noise observed at each microgrid node at time t. The pruning parameters vary over time. Neighborhood information interaction and consensus aggregation are implemented. Each node obtains the scheduling decision state of its neighbors through a time-varying directed communication graph, and uses a double-random weight matrix for weighted aggregation to generate auxiliary variables reflecting the regional collaborative state. State updates are based on the auxiliary strategy. Combining the pruning gradient and auxiliary variables, an auxiliary optimization subproblem for a strongly pseudo-convex function is constructed to solve and update the power scheduling decision at the current time, which is then projected onto the microgrid operating constraint set. ;in, This is a weighted aggregate value representing the scheduling status of neighboring microgrid nodes. To ensure communication weights conform to the characteristics of a double random matrix, To learn step length, It is a symmetric positive definite matrix. This represents the set of constraints for microgrid operation. It involves command issuance and dynamic iteration. The system outputs commands for each power source and energy storage charging / discharging commands, and automatically enters the next optimization cycle based on feedback from the scheduling environment.

[0043] Based on the above Figure 1 The illustrated embodiment, specifically, This embodiment uses real power operation data from five typical regions in Italy in 2018 for simulation verification. A distributed network consisting of six microgrid nodes is constructed, and a set of... ={1,2,3,4,5,6} represents a microgrid. Each node represents an independent microgrid region, integrating wind power generation devices, photovoltaic power generation devices, energy storage systems, and distribution facilities that interact with the main grid. Each transformer node... The decision variables are defined as follows: ,in This represents the adjustment of active power output of distributed power sources (wind / solar) at this node. This represents the charging and discharging power of the battery energy storage system (BESS) at that node. Each microgrid node can only access local load and generation data for its own region and cannot obtain global information about the entire grid. Microgrid nodes exchange information through time-varying directed communication graphs, such as... Figure 2As shown. This embodiment considers four possible communication topologies. Specify the switching order as The weight of each edge is set to... ,in For nodes The number of neighbors (including itself). This topological sequence satisfies the periodic combination strong connectivity condition, ensuring consistent propagation of distributed information. The goal of the distributed economic scheduling problem is to determine the output power of each microgrid node in real time, minimizing the total cumulative operating cost of the system when facing environmental uncertainties. Considering the nonlinear characteristics of microgrid operation, this embodiment uses a local cost function... It is defined as a strongly pseudoconvex function. Its specific mathematical description is as follows:

[0044] ; in, Indicates the first Each microgrid node (such as distributed power sources, energy storage) at time . output power. This represents the local cost function of the node. This function is modeled as a strongly pseudo-convex function, capable of describing converter losses, energy storage degradation costs, and nonlinear generation characteristics more accurately than traditional quadratic functions. The decision variables of the grid nodes must satisfy a set of physical constraints. , , This indicates a lower bound constraint. This represents an upper limit constraint, ensuring that the output of distributed wind turbines, photovoltaic systems, or conventional units remains within the rated range. Further, the online distributed optimization economic dispatch method model for microgrid networks based on heavy-tailed noise gradient information specifically includes:

[0045] I. Initialization Steps: Construct a microgrid distributed network with multiple nodes, define the strongly pseudo-convex local cost function and operating constraint set for each node; initialize the symmetric positive definite matrix. Learning Step Size and dynamic gradient clipping threshold ; II. Stage Cyclic Steps: Each microgrid node obtains the local cost stochastic gradient affected by heavy-tailed noise, and uses a pruning threshold. By performing a nonlinear mapping on it and restricting the gradient magnitude within a threshold range, the clipped gradient can be obtained. ; In the gradient clipping step, the gradient is clipped. The calculation formula is: ; in, For the first Each microgrid node in The stochastic gradient with heavy-tailed noise observed at each time step. These are the cutting parameters that change over time.

[0046] Neighborhood information exchange and consensus aggregation. Each node obtains the scheduling decision status of its neighbors through a time-varying directed communication graph, and uses a double random weight matrix for weighted aggregation to generate auxiliary variables reflecting the regional collaborative state;

[0047] State updates based on auxiliary strategies. Combining the aforementioned pruning gradient and auxiliary variables, an auxiliary optimization subproblem for strongly pseudo-convex functions is constructed to solve and update the power scheduling decision at the current time step, which is then projected onto the microgrid operation constraint set.

[0048] Command issuance and dynamic iteration. Output commands for each power source and energy storage charging / discharging, and automatically enter the next cycle of optimization process based on feedback from the scheduling environment.

[0049] Furthermore, the time-varying directed communication graph is a periodic strongly connected birandom equilibrium graph, which characterizes the topological structure of information interaction between scheduling nodes of the microgrid, satisfying local information characteristics, communication topology characteristics, equilibrium graph characteristics, and periodic strong connectivity characteristics. The local information characteristics include: each microgrid node can only access its own local strong pseudo-convex cost information, local constraints, and online historical data affected by heavy tail noise; the local constraints include nonlinear power constraints and convex set constraints, wherein the nonlinear power constraints include the dynamic gradient pruning threshold constraints as described in claim 2, and the convex set constraints correspond to the physical operating range of the microgrid node output power and the state of charge (SoC) of the energy storage system.

[0050] The communication topology characteristics include: given a time-varying directed graph ,in Represents a set of microgrid nodes. Represents the communication edge set, a nonnegative matrix. It is the adjacency weight matrix of the communication link. If Then the weight ,otherwise If the node Information can be transmitted to nodes Then it is called a node. For nodes The neighbors represent the asymmetric communication in a microgrid due to geographical distribution.

[0051] The characteristics of the balance diagram include: weight matrix. It satisfies the double random property, that is, for all nodes Satisfy rows and and column sum All of these conditions are met; this characteristic indicates that the information sent and received by each microgrid node during the information exchange process has equal weight, which is used to ensure the conservation of global energy information when dealing with heavy-tailed noise with unbounded variance, and to prevent the dispatch center from generating random biases.

[0052] The periodic strongly connected property includes: definition Laplace matrix at time If for any time There exist positive integers This makes it possible to achieve the following within the time interval: Union of all directed graphs in the inner region There exists from any node To the node If there are directed paths, then the graph sequence is said to be periodically strongly connected; this characteristic indicates that any two microgrid nodes can achieve cross-time domain information exchange through multi-hop communication within a finite scheduling period, ensuring that the effective gradient information after pruning can be fully diffused in the dynamically evolving network.

[0053] To solve the economic dispatch problem in microgrid networks, for An online distributed optimization scheduling method based on heavy-tailed noise gradient information is designed for time-varying directed graphs as follows: Initialization steps: Construct a microgrid distributed network containing multiple nodes, set the strong pseudo-convex local cost function and operating constraint set for each node; initialize the symmetric positive definite matrix, learning step size, and dynamic gradient pruning threshold.

[0054] Each microgrid node acquires the local cost stochastic gradient affected by heavy-tailed noise, and performs a nonlinear mapping on it using a pruning threshold to limit the gradient magnitude within the threshold range, thus obtaining the pruned gradient. The formula for calculating the pruned gradient in the gradient pruning step is as follows: ; Neighborhood information exchange and consensus aggregation. Each node obtains the scheduling decision status of its neighbors through a time-varying directed communication graph, and uses a double random weight matrix for weighted aggregation to generate auxiliary variables reflecting the regional collaborative state;

[0055] ; State update based on auxiliary strategy. Combining the pruning gradient and auxiliary variables, an auxiliary optimization subproblem for strongly pseudo-convex functions is constructed to solve and update the power scheduling decision at the current time, and then projected onto the microgrid operation constraint set;

[0056] ; get Decision-making at all times , , which serves as the initial value for scheduling in the next time step.

[0057] Given an intelligent agent The local cost function is as follows: ; ; Wherein, the constraint set is set as The parameter values ​​of the objective function are set as follows: The algorithm's parameters are set as follows: number of iterations. , matrix; initial state , , , , Set step size Clipping threshold The stochastic gradient follows a symmetric zero-mean property. stable distribution Among them, the characteristic index skewness parameter , scale parameters Position parameters .

[0058] Depend on Figure 3 It can be seen that the states of all agents have reached a consensus. Figure 4 and Figure 5 The mean value and consistency error of the stochastic gradient and the clipped gradient were plotted separately. As can be seen from the figure, the clipping strategy effectively eliminates the extreme values ​​in the stochastic gradient. Figure 6 All intelligent agents were drawn. The trajectory can be seen from this diagram. Decay to 0, i.e. It exhibits sublinear growth. Therefore, the method proposed in this invention is effective.

[0059] This specification provides an online distributed optimized economic dispatch system for microgrid networks, including: The model building module is used to obtain the target microgrid network, which includes multiple microgrid nodes that interact with local decision-making information through a time-varying directed communication graph, and to build an online distributed optimization scheduling model with the goal of minimizing the global operating cost of the microgrid network at all operating times.

[0060] The dynamic pruning module is used to obtain the stochastic gradient containing heavy-tailed noise of the local cost function of each microgrid node, and to use a preset pruning threshold that varies with time to transform the stochastic gradient containing heavy-tailed noise into a pruning gradient that suppresses extreme outliers.

[0061] The weighted aggregation module is used to weight and aggregate the decision states of each neighboring node according to the communication weight matrix to obtain local aggregated variables.

[0062] The decision state update module is used to construct auxiliary optimization sub-problems based on the pruning gradient and local aggregation variables, solve them, obtain the decision state at the next time step, and project the decision state at the next time step into the microgrid network operation constraint set.

[0063] The output module determines whether the scheduling cycle has been reached. If so, it combines the decision states of each microgrid node at each time point into a scheduling decision sequence, ends the current scheduling, and outputs the scheduling decision sequence. Otherwise, it updates the iteration count and continues iterating.

[0064] Specific limitations regarding the online distributed optimization economic dispatch system for microgrid networks can be found in the limitations of the online distributed optimization economic dispatch method for microgrid networks described above, and will not be repeated here. Each module in the aforementioned online distributed optimization economic dispatch system for microgrid networks 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 operations corresponding to each module.

[0065] This specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0066] This specification provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0067] Those skilled in the art will understand that all or part of the processes in the methods of 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, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention 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, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0068] 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.

Claims

1. A method for online distributed optimal economic dispatch of microgrid networks, characterized in that, Specifically, it includes: A target microgrid network is obtained, comprising multiple microgrid nodes that interact with each other for local decision-making via a time-varying directed communication graph. An online distributed optimization scheduling model is constructed with the goal of minimizing the global operating cost of the target microgrid network at all operating times. At each time step, the local cost function of each microgrid node is a strongly pseudo-convex function on the microgrid network's operating constraint set. When scheduling the network using the online distributed optimization scheduling model, the following steps are executed iteratively: Obtain the stochastic gradient of the local cost function of each microgrid node, including heavy-tailed noise, and set the scheduling period; A preset time-varying clipping threshold is used to perform a nonlinear mapping on the stochastic gradient containing heavy-tailed noise, transforming the stochastic gradient containing heavy-tailed noise into a clipped gradient that suppresses extreme outliers. The decision state of each neighbor node in the set of neighbor nodes is obtained through the time-varying directed communication graph, and the communication weight matrix is ​​determined; the decision state of each neighbor node is weighted and aggregated according to the communication weight matrix to obtain local aggregated variables; Based on the pruning gradient and local aggregation variables, construct an auxiliary optimization sub-problem, solve it, obtain the decision state at the next time step, and project the decision state at the next time step into the microgrid network operation constraint set; Determine whether the scheduling cycle has been reached. If so, combine the decision states of each microgrid node at each time point into a scheduling decision sequence, end the current scheduling, and output the scheduling decision sequence; otherwise, update the iteration count and continue iterating.

2. The online distributed optimal economic dispatch method for microgrid networks as described in claim 1, characterized in that: The time-varying directed communication graph satisfies the periodic combination strong connectivity condition. Within any continuous time interval, the parallel graphs of all communication graphs are strongly connected to support the dynamic access and disconnection of microgrid nodes.

3. The online distributed optimal economic dispatch method for microgrid networks as described in claim 1, characterized in that: For the local cost function of each microgrid node, there exists a positive number. This ensures that for any two decision states within the constraint set... All of them satisfy the strong pseudo-convexity property and are used to describe the nonlinear generation cost and power characteristics in microgrids.

4. The online distributed optimal economic dispatch method for microgrid networks as described in claim 1, characterized in that, The method employs a preset time-varying pruning threshold to perform a nonlinear mapping on the stochastic gradient containing heavy-tailed noise, transforming the stochastic gradient with heavy-tailed noise into a pruned gradient that suppresses extreme outliers. The expression is: ； in, For the first Each microgrid node in The stochastic gradient with heavy-tailed noise observed at each time step. The clipping threshold changes over time.

5. The online distributed optimal economic dispatch method for microgrid networks as described in claim 1, characterized in that, The preset time-varying pruning threshold satisfies the following conditions: when the scheduling time approaches infinity, the growth rate of the pruning threshold is less than linear, and it satisfies an infinite series convergence condition related to the moment order of the heavy-tailed noise, thereby ensuring the convergence of the algorithm under high probability.

6. The online distributed optimal economic dispatch method for microgrid networks as described in claim 1, characterized in that, The communication weight matrix It satisfies the double random property, specifically: ； in, for The communication weight matrix at time step Line 1 Column elements, for The communication weight matrix at time step Line 1 The elements of the column.

7. The online distributed optimal economic dispatch method for microgrid networks as described in claim 1, characterized in that, Based on the pruning gradient and local aggregation variables, construct an auxiliary optimization subproblem, solve it, and obtain the decision state at the next time step. Specifically: ； in, For the set of constraints for microgrid network operation, To optimize variables, For transpose calculation, It is a positive definite matrix. To learn step length, for Local aggregate variables at time points.

8. A microgrid network online distributed optimized economic dispatch system, characterized in that, include: The model building module is used to obtain a target microgrid network including multiple microgrid nodes that interact with local decision-making information through a time-varying directed communication graph, and to build an online distributed optimization scheduling model with the optimization objective of minimizing the global operating cost of the microgrid network at all operating times. The dynamic pruning module is used to obtain the stochastic gradient containing heavy-tailed noise of the local cost function of each microgrid node, and to use a preset pruning threshold that varies with time to transform the stochastic gradient containing heavy-tailed noise into a pruning gradient that suppresses extreme outliers. The weighted aggregation module is used to weight and aggregate the decision states of each neighboring node according to the communication weight matrix to obtain local aggregated variables; The decision state update module is used to construct an auxiliary optimization sub-problem based on the pruning gradient and local aggregation variables, solve it, obtain the decision state at the next time step, and project the decision state at the next time step into the microgrid network operation constraint set. The output module determines whether the scheduling cycle has been reached. If so, it combines the decision states of each microgrid node at each time point into a scheduling decision sequence, ends the current scheduling, and outputs the scheduling decision sequence. Otherwise, it updates the iteration count and continues iterating.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 7.

10. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.

Images