Microgrid group collaborative scheduling method based on master-slave strategy

Abstract

The present application relates to power system dispatching technical field, specifically, it relates to micro-grid group collaborative dispatching method based on master-slave strategy, including upper layer uncertainty propagation simulator and multiple lower layer local robust optimizer. The upper layer simulator collects real-time uncertainty source data and network topology, generates collaborative robust domain through simulating uncertainty propagation path, and issues to lower layer as probability constraint. The lower layer optimizer executes local cost minimization optimization under constraint, and reports operating point and uncertainty index. The upper layer calculates overall collaborative entropy, if lower than threshold value, expands robust domain through feedback mechanism to maximize benefit, if higher than threshold value, shrinks domain to reduce risk. Iterative process applies gradient algorithm to adjust domain boundary parameter until convergence or dispatching period ends, realizing micro-grid group robust collaborative dispatching. The present application realizes dynamically maximizing collaborative operation benefit and robust potential of micro-grid group under the premise of guaranteeing system global stability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system dispatching, in particular to a micro-grid group collaborative dispatching method based on a master-slave strategy. BACKGROUND

[0002] Distributed renewable energy represented by photovoltaic and wind power is connected to the power grid at an unprecedented scale. In order to effectively accommodate these intermittent and volatile power sources, and at the same time improve the power supply reliability and energy utilization efficiency of local power grids, micro-grids as an advanced regional autonomous energy system have emerged. With the further improvement of the penetration rate of distributed energy, the regulation capacity of a single micro-grid gradually reaches a bottleneck. By interconnecting multiple micro-grids that are geographically adjacent or functionally complementary, a micro-grid group is formed, which has become an important development direction to improve the flexibility and economy of regional power grids. Through internal energy interconnection and collaborative dispatching, a micro-grid group can suppress the output fluctuation of renewable energy on a larger scale, and realize the optimal allocation of resources and the sharing of operation benefits. However, this interconnected structure also brings new challenges: the uncertainties of sources and loads in each micro-grid will be transmitted, coupled and even amplified through the tie lines, forming a complex chain reaction, which threatens the safe and stable operation of the entire system. Therefore, how to accurately quantify and effectively manage the propagation effect of uncertainties in the micro-grid group, and develop a collaborative dispatching strategy that takes into account both economy and robustness, is a key technical problem that needs to be solved in the current field of power system research. At present, the micro-grid group dispatching method has obvious limitations in dealing with uncertainty propagation. On the one hand, although the traditional two-level optimization model can divide the master-slave dispatching hierarchy, the interaction between the upper and lower layers is usually limited to price signals or power plans, lacking quantification and transmission of uncertainty risks themselves, which makes the upper coordinator unable to accurately assess the global chain risk that a local disturbance may cause, and the decision-making has blind spots. On the other hand, although some methods using distributed algorithms improve the calculation efficiency and privacy protection, their iterative process usually relies on the exchange of information between adjacent individuals, making it difficult to form a macroscopic perception of the total amount of uncertainty in the entire system; in addition, the current method usually uses fixed robustness or randomness constraints to deal with uncertainty, and the static boundary setting cannot adapt to the real-time operating state of the system. When the overall uncertainty of the system is low, the overly conservative constraints will sacrifice economic efficiency, and when the uncertainty accumulates to a dangerous level, the fixed constraints may not be enough to ensure system safety, lacking dynamic adaptability and self-evolution ability. SUMMARY

[0003] The purpose of the present application is to provide a micro-grid group collaborative dispatching method based on a master-slave strategy, which quantifies the total uncertainty of the system through collaborative entropy, and drives the collaborative robust domain to expand or shrink, realizing the dynamic maximization of the collaborative operation benefits and robust potential of the micro-grid group under the premise of ensuring the global stability of the system.

[0004] The application is achieved by the following technical solutions:

[0005] The method comprises the following steps:

[0006] Step 1: a micro-grid group cooperative scheduling double-layer model is constructed, and the micro-grid group cooperative scheduling double-layer model comprises an upper-layer uncertainty propagation simulator and a lower-layer local robust optimizer;

[0007] Step 2: the upper-layer uncertainty propagation simulator collects real-time uncertainty source data and network topology of the micro-grid group, simulates a causal propagation path of the uncertainty in the micro-grid group system, and generates a cooperative robust domain, wherein the cooperative robust domain represents an operation set caused by quantized uncertainty propagation and stability risk;

[0008] Step 3: the cooperative robust domain is sent to the lower-layer local robust optimizer as a probability constraint condition;

[0009] Step 4: the lower-layer local robust optimizer performs independent local target optimization under the premise of meeting the cooperative robust domain as the probability constraint condition, and reports an optimized operating point and a local uncertainty index;

[0010] Step 5: the upper-layer uncertainty propagation simulator starts a robust domain self-evolution protocol according to the operating point and the local uncertainty index reported by all lower-layer local robust optimizers: calculates a whole cooperative entropy of the micro-grid group system, if the cooperative entropy is lower than a preset threshold, evolves through a feedback mechanism of maximizing total robust benefit, and expands the shape and size of the cooperative robust domain; if the cooperative entropy is higher than the preset threshold, evolves through a feedback mechanism of minimizing the cooperative entropy, and shrinks the cooperative robust domain to reduce instability risk caused by uncertainty propagation;

[0011] Step 6: steps 4-5 are iteratively executed until the self-evolution of the cooperative robust domain converges or a maximum iteration number is reached, and a scheduling strategy of maximizing robust cooperative potential of the micro-grid group system is obtained.

[0012] Optionally, the objective function of the upper-layer uncertainty propagation simulator is specifically to minimize the uncertainty propagation entropy in the cooperative robust domain, and to maximize the total robust benefit of all lower-layer local robust optimizers in the cooperative robust domain, and the specific calculation formula is:

[0013]

[0014] wherein, is an upper-layer objective function value, is a whole cooperative entropy of the micro-grid group, is a benefit-entropy trade-off weight, is a total robust benefit of all lower-layer local robust optimizers in the cooperative robust domain. Robust benefits of a lower-level local robust optimizer For microgrid indexing, This represents the total number of microgrids.

[0015] Optionally, the objective function of the lower-level local robust optimizer is to minimize the local running cost, and its specific calculation formula is as follows:

[0016]

[0017] in, For the first The local operating cost of a lower-level local robust optimizer For the first Secondary generation cost coefficient of a microgrid For the first The primary generation cost coefficient of a microgrid For the first The fixed generation cost of a microgrid For the first microgrids Power generation during the period For the first microgrids Local loss cost during the time period For time period index, This represents the total number of time periods.

[0018] Optionally, the simulation of the causal propagation path of uncertainty in the microgrid cluster system specifically involves:

[0019] The probability distribution function is extracted from the real-time uncertainty source data of the microgrid group, and a dynamic causal graph model is constructed in combination with the network topology. In the dynamic causal graph model, the nodes represent each microgrid, and the edges represent the uncertainty propagation path.

[0020] Based on the probability distribution function, multiple uncertainty scenario samples are generated using the Monte Carlo chain method. Each scenario sample includes the initial uncertainty source and its possible propagation process.

[0021] For each scenario sample, all propagation paths are traversed sequentially in the dynamic causal graph model to calculate the uncertainty intensity, propagation amplification coefficient, and multi-source uncertainty superposition effect on each propagation path, thereby obtaining the uncertainty distribution of each propagation path in each scenario.

[0022] For the uncertainty distribution of each propagation path in various scenarios, risk indicators are predicted based on time-series dynamic simulation, including the probability of risk occurrence, frequency deviation and cumulative fluctuation amplitude of the propagation path, and the risk indicators of all propagation paths and all scenarios are statistically aggregated.

[0023] Based on the statistical aggregation result, a probability boundary including all risk scenarios greater than the threshold value is determined, and a multi-dimensional probability operation set of the cooperative robust domain is defined as a constraint condition.

[0024] Optionally, the respective independent local target optimization is performed under the premise that the cooperative robust domain is satisfied as a probability constraint condition, and the specific process is as follows.

[0025] The lower-layer local robust optimizer receives the cooperative robust domain issued by the upper-layer uncertainty propagation simulator, and converts the cooperative robust domain into a local stochastic inequality probability constraint;

[0026] Based on the converted constraint and the local real-time data, the objective function of the lower-layer local robust optimizer is initialized;

[0027] Based on Monte Carlo sampling, a plurality of scene samples are generated from the uncertainty indicators extracted from the local real-time data;

[0028] Under each scene sample, a genetic algorithm is applied to solve the initialized objective function, and the optimization variable is checked and adjusted scene by scene to satisfy the converted probability constraint;

[0029] The optimization results of all scenes are aggregated to obtain an optimized operating point and a local uncertainty indicator, and the aggregated optimization result is verified by a time series simulation test after the local operating cost is minimized within the cooperative robust domain, and is reported to the upper-layer uncertainty propagation simulator.

[0030] Optionally, the overall cooperative entropy of the micro-grid group system is calculated, and the specific process is as follows.

[0031] From the operating point and the local uncertainty indicator reported by all lower-layer local robust optimizers, the uncertainty probability distribution of each micro-grid is extracted, and the uncertainty propagation path is mapped in combination with the network topology;

[0032] For each uncertainty propagation path, the local uncertainty entropy is calculated based on the extracted uncertainty probability distribution of each micro-grid;

[0033] The local uncertainty entropies of all propagation paths are weighted and aggregated, and a propagation item is added to reflect the uncertainty accumulation, to obtain an aggregation result;

[0034] The numerical stability of the aggregation result is checked and output, and the aggregation result is represented as the overall cooperative entropy of the micro-grid group system.

[0035] Optionally, the overall cooperative entropy of the micro-grid group system is calculated, and the specific calculation formula is as follows.

[0036]

[0037]

[0038]

[0039] wherein, is the local uncertainty entropy of the i-th propagation path, is the total number of propagation paths, is the probability of the j-th uncertainty event on the i-th path, is the propagation path index, is the total number of propagation paths, is the uncertainty event index, is the total number of events for each propagation path, is the total local entropy after weighted aggregation, is the weight based on network topology, is the overall coordination entropy of the microgrid group, is the propagation aggregation item. Optionally, the feedback mechanism evolution by minimizing the coordination entropy, which is specifically:

[0040] The overall coordination entropy of the microgrid group system is obtained, and compared with a preset threshold to determine whether the coordination entropy is lower than the preset threshold;

[0041] If the coordination entropy is higher than the preset threshold, the coordination robust domain is contracted to reduce the instability risk caused by uncertainty propagation;

[0042] If the coordination entropy is lower than the threshold, the gradient contribution of each propagation path to the coordination entropy is calculated, the propagation path reaching the set contribution is identified, and the feedback mechanism parameters and the initial gradient direction are initialized;

[0043] The gradient descent algorithm is applied to iteratively adjust the domain boundary parameters, wherein the domain boundary parameters represent the multi-dimensional boundary control variables of the coordination robust domain, and the domain boundary parameters are updated based on the current coordination entropy and the initialized feedback mechanism parameters in each iteration until the coordination entropy is minimized or the total robust benefit is maximized;

[0044] Based on the minimized coordination entropy or the maximized total robust benefit, the multi-dimensional shape and size of the coordination robust domain are dynamically expanded.

[0045] Optionally, the gradient descent algorithm is applied to iteratively adjust the domain boundary parameters, and the specific calculation formula is:

[0046]

[0047]

[0048] wherein, is the domain boundary parameter of the i-th iteration, is the domain boundary parameter of the i-th iteration, ​​for the step iteration, a domain boundary parameter of the step iteration, for the learning rate, for the gradient of the collaborative entropy on the domain boundary parameter.

[0049] Optionally, the self-evolution of the collaborative robust domain converges or reaches a maximum number of iterations, wherein the self-evolution convergence condition is specifically that the change rate of the collaborative entropy is less than a preset threshold.

[0050] The technical scheme of the present application has at least the following advantages and beneficial effects:

[0051] The present application constructs an upper-layer uncertainty propagation simulator, which no longer only issues scheduling instructions, but generates a dynamic collaborative robust domain through simulation, and the collaborative robust domain accurately depicts the multi-dimensional safety boundary that the entire system must comply with to maintain stability, thereby providing a clear and globally collaborative constraint premise for the autonomous optimization of lower-layer microgrids. On the other hand, the present application establishes collaborative entropy as a core index for measuring the total amount of uncertainty of the entire microgrid group, so that the risk level of the system state is accurately quantified for the first time: when the collaborative entropy is low, the system automatically expands the robust domain and encourages each microgrid to explore a more optimal economic operating point; when the collaborative entropy is too high, the robust domain is actively contracted to ensure system safety. The feedback closed-loop mechanism based on gradient descent enables the system to continuously iterate and evolve itself, and dynamically find the optimal balance point between robustness and economy under the current state. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The overall flowchart of the microgrid group collaborative scheduling method based on the master-slave strategy provided by the present application is shown in the figure;

[0053] Figure 2 The execution logic diagram of the lower-layer local robust optimizer provided by the present application is shown in the figure;

[0054] Figure 3 The architecture diagram of the microgrid group collaborative scheduling system based on the master-slave strategy provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0056] As Figure 1As shown, this invention proposes a microgrid group cooperative scheduling method based on a master-slave strategy, aiming to solve the core technical challenge of balancing operational robustness and economy when microgrid groups with a high proportion of renewable energy are dealing with the propagation and accumulation effects of multi-source uncertainties. This method constructs an innovative two-layer scheduling architecture, introduces cooperative entropy to quantify system-level uncertainty risks, and designs a robust domain self-evolution protocol, enabling the system to dynamically and intelligently maximize economic benefits while ensuring safety.

[0057] The method of this invention can be implemented based on a collaborative scheduling system that executes the method. The physical and software architecture of the system can be deployed on the dispatch center of a regional power grid or a cloud server cluster. At the hardware level, the system includes high-performance computing servers, data storage arrays, and dedicated network equipment for communication with each microgrid. The communication network preferably employs a 5G private network or fiber optic network with high bandwidth and low latency to ensure reliable transmission of massive amounts of real-time data.

[0058] like Figure 3 As shown, at the software level, the system mainly consists of four core functional modules that work together to realize the scheduling method of this invention.

[0059] Upper-layer Uncertainty Propagation Simulator Module: This module is the "brain" of the entire system, deployed on the main server. It is responsible for grasping the operational status of the entire microgrid cluster from a macroscopic perspective. Its main functions include: real-time collection and integration of operational data from all microgrids via data interfaces, such as output, load demand, energy storage status, and tie-line power of each distributed power source (for example); and access to external data sources, such as weather forecast APIs, to obtain forecast information like solar intensity and wind speed. This module incorporates an advanced causal propagation simulation engine, capable of constructing dynamic causal graph models based on network topology and historical data to simulate how uncertainties (such as a sudden drop in photovoltaic output of a microgrid) are propagated and amplified among units within the cluster. Its final output is a dynamically changing "cooperative robust domain," a multi-dimensional set of probabilistic operations that defines the safety boundaries that all microgrids must adhere to during joint operation to ensure the stability of the entire system.

[0060] Lower-layer local robust optimizer module: This module is usually in the form of a software agent, deployed on each microgrid's respective local controller or edge computing node. It serves as the "limbs" of the system, responsible for executing specific, local optimization tasks. Each optimizer receives the collaborative robust domain issued by the upper-layer simulator and parses it into locally executable probabilistic constraints. Under the premise of meeting these global constraints, the module aims to minimize local operating costs or maximize local benefits, using built-in optimization algorithms such as genetic algorithms, particle swarm optimization, etc. to solve the locally optimal scheduling strategy, such as adjusting local diesel generator output, energy storage charging and discharging plans, etc. After completing the optimization, it reports the calculated optimal operating point and key indicators reflecting the local uncertainty level to the upper-layer simulator.

[0061] Robust domain self-evolution module: This is the core innovative module of the invention, closely integrated with the upper-layer simulator. It acts as a dynamic balance regulator, responsible for implementing the intelligent evolution of the system. The module receives all the operating points and uncertainty indicators reported by the lower-layer optimizers and calculates a key macro-state variable - the overall collaborative entropy of the microgrid group system. Collaborative entropy accurately quantifies the current uncertainty level and potential risks of the entire system. The module has a built-in feedback control logic: when the calculated collaborative entropy is below the pre-set safety threshold, it means that the system is currently very stable and has the potential to pursue higher economic benefits. At this time, the module will trigger the expansion mechanism to actively and smoothly relax the boundaries of the collaborative robust domain through gradient descent algorithms; conversely, when the collaborative entropy is higher than the threshold, it means that the system risks are accumulating, and the module will trigger the "contraction" mechanism to tighten the robust domain boundaries, forcing all lower-layer microgrids to adopt more conservative operating strategies at the expense of some economic benefits for the sake of system safety.

[0062] Iterative control module: This module serves as the commander of the entire scheduling process, responsible for managing and coordinating the interaction and iteration of the above three modules. It sets the scheduling period, such as a rolling optimization every 15 minutes, and the termination condition of iteration. Within a scheduling period, it starts one or more rounds of closed-loop iteration process of "upper-layer issuance - lower-layer optimization - upper-layer evaluation and evolution". The termination condition of iteration can be the convergence of the self-evolution process of the collaborative robust domain, i.e. the change rate of collaborative entropy is less than a small pre-set value, or the pre-set maximum iteration number is reached, or the end of the scheduling period. When the iteration terminates, the module is responsible for outputting the final scheduling strategy that has been verified to maximize the robust collaborative potential under the current state, and issuing it to each microgrid for execution.

[0063] In the specific application of the invention, assume that a microgrid group consists of three microgrids (MG1, MG2, MG3), which are connected to each other through tie lines.

[0064] MG1: mainly photovoltaic power generation, equipped with energy storage system, the load is mainly residential electricity.

[0065] MG2: mainly wind power generation, equipped with diesel generator as backup, the load is mainly small industrial users.

[0066] MG3: no local power supply, pure load type microgrid, equipped with large energy storage system, mainly relying on purchasing electricity from MG1 and MG2.

[0067] The dispatching period is set to the future 24 hours, and the time resolution is 1 hour.

[0068] Step 1: Construct a microgrid group collaborative scheduling two-level model.

[0069] At the beginning of scheduling, the system first initializes the two-level model. The upper layer uncertainty propagation simulator loads the network topology data of the entire microgrid group, including the tie line parameters (impedance, maximum transmission capacity, etc.) between MG1, MG2, and MG3. The lower layer is three independent local robust optimizers, corresponding to MG1, MG2, and MG3 respectively.

[0070] Specifically, the objective of the upper layer model is to achieve a balance between global robustness and economy, and its objective function is set to minimize the uncertainty propagation entropy within the collaborative robust domain, while maximizing the total robust benefit of all lower layer microgrids. Its specific calculation formula is:

[0071]

[0072] wherein, is the value of the upper layer objective function, is the overall collaborative entropy of the microgrid group, with units of bits, is the benefit-entropy trade-off weight, used to balance the safety (low entropy) and economy (high benefit) two goals. In this embodiment, the initial value is set to 0.5, indicating equal attention to both, is the robust benefit of the th lower layer local robust optimizer, usually referring to its operating income under the robustness constraint, with units of yuan, is the microgrid index, is the total number of microgrids.

[0073] The objective of each local robust optimizer (taking MG2 as an example) in the lower layer is to minimize its local operating cost. Its objective function is specifically in the form of:

[0074]

[0075] wherein, is the total operating cost of MG2 within 24 hours, Its diesel generator is in Power generation during a given time period. , , This is the cost coefficient of a diesel generator, used as an example. =0.005 yuan / kW², =20 yuan / kW, =10 yuan. This refers to local network loss costs.

[0076] Step 2: The upper-level simulator generates the initial cooperative robust domain.

[0077] The upper-level simulator begins operation. It first collects real-time and forecast data for all sources of uncertainty. For example, it obtains the wind speed forecast curve for the next 24 hours from the meteorological service and combines it with the historical output characteristics of the MG2 wind turbine to generate a probability distribution function for wind power output. Assuming that at 3 PM, the predicted wind speed has a 70% probability of being 10 m / s and a 30% probability of being 8 m / s, this constitutes one source of uncertainty.

[0078] The upper-level simulator constructs a dynamic cause-effect graph model. The nodes are MG1, MG2, MG3, and the tie line. Edges represent the propagation path of uncertainty. As an example, an edge from the wind turbine node of MG2 to the MG2-MG3 tie line node indicates that wind power fluctuations directly affect the power transmitted to MG3.

[0079] Simulations were performed using the Monte Carlo chain method. The system generated 10,000 uncertainty scenario samples. Each sample represents one possible scenario for the next 24 hours. For example, the photovoltaic output of MG1 is 10% lower than predicted, while the wind power output of MG2 is 5% higher than predicted.

[0080] In each scenario, the system traverses all propagation paths and calculates risk indicators. As an example, in a certain scenario, the probability of the power flow exceeding the limit on the MG2-MG3 interconnect line due to a significant increase in the output of the MG2 wind power was calculated to be 5%. The system analyzed the results of all 10,000 scenarios and found that a total of 800 scenarios had risk events such as frequency deviation exceeding 0.2Hz or voltage exceeding the limit by more than 5%.

[0081] A probability boundary is defined, which precisely encompasses all 9200 safe scenarios where no risk events have occurred. The probability boundary consists of a set of operations defined by multidimensional probability inequalities, such as MG1 photovoltaic power output. The variance of the fluctuation is less than And the MG2 wind power output The variance of the fluctuation is less than In addition to fluctuation constraints of other random variables, an initial cooperative robust domain is formed.

[0082] Step 3: The cooperative robust region is issued as a probabilistic constraint.

[0083] The upper simulator issues the cooperative robust region to the three lower optimizers. As an example, the constraint issued to MG1 is converted to a specific local stochastic inequality probabilistic constraint, which is calculated as:

[0084]

[0085] where, is the probability function, is the actual PV output of microgrid 1 at time, is the minimum value of the PV output of microgrid 1 at time, is the maximum value of the PV output of microgrid 1 at time.

[0086] This constraint means that MG1 must ensure that its PV output falls within the safety interval specified by the upper layer at a confidence level of 95%. The confidence level of 0.95 here is converted from the probabilistic boundary of the cooperative robust region.

[0087] Step 4: The lower layer performs local robust optimization and reports the results.

[0088] As shown in Figure 2 , after receiving the constraints, the three lower optimizers start working in parallel. Taking MG2 as an example, it initializes its cost minimization objective function based on the converted probabilistic constraint and local real-time data, such as the current state of the diesel generator and the measured value of the wind speed.

[0089] Using Monte Carlo sampling, multiple local uncertainty scenarios are generated, such as random fluctuations in local industrial loads. In each scenario, it applies a genetic algorithm to solve its cost minimization problem. During the solving process, each individual of the genetic algorithm, representing a scheduling scheme, must be tested to ensure that it meets the probabilistic constraint issued by the upper layer. Individuals that do not meet the constraint are eliminated or subjected to a high penalty.

[0090] After hundreds of generations of evolution, the genetic algorithm of MG2 converges, obtaining a scheduling scheme that minimizes its own operating cost while meeting the global robustness constraint. The scheduling scheme includes the output curve of the diesel generator for the next 24 hours, planned exchange power with MG1 and MG3, etc. This is the optimized operating point of MG2. At the same time, MG2 also calculates and reports its local uncertainty indicators, such as the variance of its net load.

[0091] MG1 and MG3 also complete similar processes and report their respective operating points and uncertainty indicators to the upper layer.

[0092] Step 5: Start robust domain self-evolution protocol.

[0093] Robust domain self-evolution module of upper simulator is activated. It collects all reported data from MG1, MG2, MG3.

[0094] First, calculate the overall coordination entropy of microgrid group system:

[0095] According to the reported data, the module extracts the uncertainty probability distribution of the entire system at the current joint operating point. As an example, analyze the photovoltaic output plan and its uncertainty index reported by MG1, combined with the wind power plan and uncertainty index of MG2, map out the propagation path of uncertainty on the MG1-MG2 tie line.

[0096] For each propagation path, calculate its local uncertainty entropy. As an example, for the MG1-MG2 tie line path , there are possible states of uncertainty events (such as reverse power flow), and each state occurs with a probability of . The local entropy calculation formula is:

[0097]

[0098] Suppose the calculation shows that the local entropy of the MG1-MG2 path is 0.8 bits, and the local entropy of the MG2-MG3 path is 1.1 bits.

[0099] Then perform weighted aggregation. The weight is determined based on the importance of the network topology, such as the tie line connected to the main grid having a higher weight. Suppose the MG1-MG2 weight is 0.4 and the MG2-MG3 weight is 0.6. The aggregated entropy is 0.95 bits.

[0100] Then add the propagation aggregation item . The propagation aggregation item is calculated by analyzing the superposition effect of multi-source uncertainty. As an example, when the photovoltaic output of MG1 decreases and the wind power output of MG2 decreases at the same time, the joint impact on MG3 is much greater than the sum of the two independent impacts. This superposition effect increases the system entropy. Suppose after calculation, 0.52 bits.

[0101] Finally, the overall coordination entropy of the microgrid group system is obtained:

[0102]

[0103] Next, execute the feedback mechanism evolution.

[0104] The system pre-sets a safety threshold for the synergy entropy, which is set based on historical operation data and dispatchers' experience. In this embodiment, the safety threshold is 1.8 bits.

[0105] The currently calculated synergy entropy is 1.50 bits, which is lower than the pre-set threshold of 1.8 bits. This indicates that the current operation state of the system is very safe, and the robustness is sufficient, so that more profitable operation modes can be explored.

[0106] At this time, the robust domain self-evolution module starts the expansion mechanism. It evolves by minimizing the negative gradient of the synergy entropy, that is, it finds the direction that can slightly increase the synergy entropy (within the safety range) while maximizing the economic benefit. Specifically, it adjusts the boundary parameters of the synergy robust domain by applying the gradient descent algorithm. The iteration formula of the gradient descent is:

[0107]

[0108] The goal here is to maximize the total benefit So it is gradient ascent. is the boundary parameter of the robust domain, such as , , etc. , , respectively, as examples, which represent the photovoltaic and wind power fluctuation variances of microgrid 1 and microgrid 2, respectively. The larger the variance, the higher the uncertainty, and the robust domain needs to be expanded accordingly. is the learning rate, which is set to a small value such as 0.01 to ensure the smoothness of the evolution process. The gradient represents the sensitivity of the total benefit to the domain boundary parameter, which can be calculated by the numerical perturbation method.

[0109] After several iterations, the system finds a new synergy robust domain with a relaxed boundary. As an example, the original requirement that the photovoltaic fluctuation variance of MG1 be less than is now adjusted to be less than .

[0110] As an example, if the synergy entropy calculated in the previous step is 2.1 bits, which is higher than the threshold of 1.8 bits. Then the system will start the contraction mechanism. At this time, the goal of the gradient descent is to minimize the synergy entropy, and the iteration formula becomes:

[0111]

[0112] The system will iteratively tighten the boundary of the robust domain until the synergy entropy is reduced below the safety threshold.

[0113] Step 6: Iterate until convergence.

[0114] The steps 4 to 5 above constitute a complete closed-loop iteration. The new, evolved, synergistically robust domain is fed back to the lower-level optimizers, i.e. back to step 4, which again perform local optimization within their new active space, and report the results back to the upper level, which again evaluates the synergy entropy and performs evolution. This process is repeated continuously. In this example, the value of synergy entropy can undergo the following changes:

[0115] Iteration 1: Bit, at threshold, expand robust domain.

[0116] Iteration 2: The lower level optimizers optimize within a wider domain, achieving higher economic benefit, but also increasing system uncertainty, Rise to 1.75 bits, still below threshold, continue to expand.

[0117] Iteration 3: Rise to 1.85 bits, above threshold, trigger contraction mechanism.

[0118] Iteration 4: Fall to 1.78 bits, below threshold, expand slightly again.

[0119] The termination condition of the iteration is that the rate of change of synergy entropy is less than a pre-set convergence threshold. As an example, set the convergence threshold = 0.01 bits. When, in a certain iteration, the system is considered to have reached a dynamic, robust and economic balance, and the self-evolution process converges, where, is the synergy entropy of the microgrid group as a whole in the first iteration, is the synergy entropy of the microgrid group as a whole in the first iteration.

[0120] At this point, the iteration control module terminates the iteration, and the joint dispatching strategy reported by the lower-level optimizers and verified by the upper level in the last iteration is taken as the final robust synergistically potential maximized dispatching strategy. The strategy contains the detailed output plan and exchange power plan of all controllable units (generators, energy storage, etc.) in each microgrid in the next 24 hours. This strategy is officially issued, and each microgrid executes accordingly.

[0121] At this point, a complete dispatching cycle is completed. The system will wait for the next dispatching time, such as 15 minutes, and then start a new rolling optimization process.

[0122] The above merely describes a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any modification or replacement within the technical range disclosed by the present application can be easily conceived by those skilled in the art, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0123] The above merely describes a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any modification or replacement within the technical range disclosed by the present application can be easily conceived by those skilled in the art, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A microgrid group collaborative scheduling method based on a master-slave strategy, characterized in that, The steps of the method comprise: Step 1: constructing a micro-grid group collaborative scheduling double-layer model, the micro-grid group collaborative scheduling double-layer model comprising an upper-layer uncertainty propagation simulator and a lower-layer local robust optimizer; Step 2: the upper-layer uncertainty propagation simulator, collecting real-time uncertainty source data and network topology of the micro-grid group, and generating a collaborative robust domain by simulating a causal propagation path of uncertainty in the micro-grid group system, the collaborative robust domain representing an operation set quantifying stability risk caused by uncertainty propagation; Step 3: issuing the collaborative robust domain as a probability constraint condition to the lower-layer local robust optimizer; Step 4: the lower-layer local robust optimizer, under the premise of meeting the collaborative robust domain as a probability constraint condition, performing independent local target optimization, and reporting an optimized operating point and a local uncertainty index; Step 5: the upper-layer uncertainty propagation simulator, according to the operating point and the local uncertainty index reported by all lower-layer local robust optimizers, starting a robust domain self-evolution protocol: calculating a collaborative entropy of the micro-grid group system as a whole, if the collaborative entropy is lower than a preset threshold, evolving through a feedback mechanism of maximizing total robust benefit, and expanding the shape and size of the collaborative robust domain; if the collaborative entropy is higher than the preset threshold, evolving through a feedback mechanism of minimizing the collaborative entropy, and shrinking the collaborative robust domain to reduce instability risk caused by uncertainty propagation; Step 6: iteratively performing steps 4-5 until the self-evolution of the collaborative robust domain converges or reaches a maximum iteration number, obtaining a scheduling strategy maximizing robust collaborative potential of the micro-grid group system. 2.The microgrid cluster collaborative scheduling method based on master-slave strategy of claim 1, wherein, The objective function of the upper-layer uncertainty propagation simulator is specifically to minimize the uncertainty propagation entropy in the collaborative robust domain, while maximizing the total robust benefit of all lower-layer local robust optimizers in the collaborative robust domain, and the specific calculation formula is: wherein, is the upper layer objective function value, is the microgrid group overall coordination entropy, is the benefit-entropy trade-off weight, is the robust benefit of the th lower layer local robust optimizer, is the microgrid index, is the total number of microgrids. 3.The microgrid cluster coordinated dispatching method based on master-slave strategy of claim 2, wherein, The objective function of the lower-layer local robust optimizer is specifically to minimize the local operation cost, and the specific calculation formula is: wherein, is the local running cost of the th lower-layer local robust optimizer, is the quadratic generation cost coefficient of the th microgrid, is the linear generation cost coefficient of the th microgrid, is the fixed generation cost of the th microgrid, is the generation power of the th microgrid in the th time period, is the local loss cost of the th microgrid in the th time period, is the time period index, is the total number of time periods.

4. The microgrid cluster coordinated dispatching method based on master-slave strategy according to claim 3, characterized in that, The simulation of the causal propagation path of uncertainty in the micro-grid group system specifically comprises: extracting a probability distribution function from real-time uncertainty source data of the micro-grid group, and constructing a dynamic causal graph model combining network topology, wherein nodes in the dynamic causal graph model represent micro-grids, and edges represent uncertainty propagation paths; generating a plurality of uncertainty scenario samples based on the probability distribution function through a Monte Carlo chain method, each scenario sample comprising an initial uncertainty source and a possible propagation process thereof; for each scenario sample, sequentially traversing all propagation paths in the dynamic causal graph model, calculating uncertainty intensity, propagation amplification coefficient and multi-source uncertainty superposition effect of the uncertainty propagated on each propagation path, and obtaining an uncertainty distribution of each propagation path under each scenario; based on time-series dynamic simulation, predicting risk indexes of each propagation path under each scenario, including a risk occurrence probability, a frequency deviation and a cumulative fluctuation amplitude of the propagation path, and statistically aggregating the risk indexes of all propagation paths under all scenarios; and Based on the statistical aggregation result, a probability boundary including all risk scenarios greater than a threshold value is determined, and a multi-dimensional probability operation set of the cooperative robust domain is defined as a constraint condition of the probability boundary.

5. The microgrid cluster coordinated dispatching method based on master-slave strategy according to claim 4, characterized in that, The cooperative robust domain is used as a probability constraint condition to perform independent local target optimization. The lower-layer local robust optimizer receives the cooperative robust domain from the upper-layer uncertainty propagation simulator and converts it into a local stochastic inequality probability constraint. The lower-layer local robust optimizer initializes a target function based on the converted constraint and local real-time data. Based on Monte Carlo sampling, a plurality of scenario samples are generated from the uncertainty indicators extracted from the local real-time data. In each scenario sample, a genetic algorithm is applied to solve the initialized target function, and the optimization variables are checked and adjusted scene by scene to meet the converted probability constraint. The optimization results of all scenarios are aggregated to obtain an optimized operating point and local uncertainty indicator, and the aggregated optimization results are verified by a time series simulation test. 6.The microgrid cluster coordinated dispatching method based on master-slave strategy of claim 5, wherein, The overall cooperative entropy of the microgrid group system is calculated. The uncertainty probability distribution of each microgrid is extracted from the operating point and local uncertainty indicator reported by all lower-layer local robust optimizers, and the uncertainty propagation path is mapped based on the network topology. For each uncertainty propagation path, the local uncertainty entropy is calculated based on the uncertainty probability distribution of each microgrid. The local uncertainty entropies of all propagation paths are weighted and aggregated, and a propagation item is added to reflect the accumulation of uncertainty to obtain an aggregation result. The numerical stability of the aggregation result is checked and output, and the aggregation result represents the overall cooperative entropy of the microgrid group system. 7.The microgrid cluster coordinated dispatching method based on master-slave strategy according to claim 6, characterized in that, The overall cooperative entropy of the microgrid group system is calculated. wherein, is the local uncertainty entropy of the i-th propagation path, is the probability of the j-th uncertainty event on the i-th path, is the propagation path index, is the total number of propagation paths, is the uncertainty event index, is the total number of events for each propagation path, is the total sum of local entropy after weighted aggregation, is the weight based on network topology, is the overall coordination entropy of the microgrid group, is the propagation aggregation item.​​​ 8.The microgrid cluster coordinated dispatching method based on master-slave strategy of claim 7, wherein, The feedback mechanism is evolved by minimizing the cooperative entropy. The overall cooperative entropy of the microgrid group system is obtained and compared with a preset threshold to determine whether the cooperative entropy is lower than the preset threshold. If the cooperative entropy is higher than the preset threshold, the cooperative robust domain is shrunk to reduce the instability risk caused by uncertainty propagation. If the cooperative entropy is lower than the threshold, the gradient contribution of each propagation path to the cooperative entropy is calculated, the propagation path reaching the set contribution is identified, and the feedback mechanism parameters and initial gradient direction are initialized. The gradient descent algorithm is applied to iteratively adjust the domain boundary parameters, which represent the multi-dimensional boundary control variables of the cooperative robust domain. Based on the minimized cooperative entropy or the maximized total robust benefit, the multi-dimensional shape and size of the cooperative robust domain are dynamically expanded. 9.The microgrid cluster coordinated dispatching method based on master-slave strategy of claim 8, wherein, The gradient descent algorithm is applied to iteratively adjust the domain boundary parameters. in, For the first Domain boundary parameters for each iteration, For the first Domain boundary parameters for each iteration, For learning rate, This is the gradient of the cooperative entropy with respect to the domain boundary parameters.

10. The microgrid cluster coordinated scheduling method based on master-slave strategy according to any one of claims 1-9, characterized in that, The self-evolution of the cooperative robust domain converges or reaches the maximum number of iterations. The self-evolution converges when the change rate of the cooperative entropy is less than a preset threshold.

Images