Micro-grid multi-resource collaborative autonomous optimization method and system based on coupling constraint relaxation

Abstract

A microgrid multi-resource collaborative autonomous optimization method and system based on coupling constraint relaxation abstracts each physical unit with power regulation capability within the microgrid as an identical regulation agent. The minimum sum of the local objective functions of all regulation agents is used as the global objective function. Relaxation variables are introduced to correct the total power balance constraint satisfied by the global objective function. A global penalty term common to all regulation agents is constructed based on the relaxation variables and penalty coefficients, and this global penalty term is superimposed on the global objective function to obtain the collaborative autonomous optimization objective. Based on the collaborative autonomous optimization objective, a distributed gradient projection method is used to iteratively predict the output of each regulation agent. When the iterative convergence criterion is met, the iteration stops and the predicted output of each regulation agent is output as the result of the microgrid multi-resource collaborative autonomous optimization. This forms a microgrid internal collaborative optimization system that is decoupled from resource types, has no external dependencies, and possesses adaptive capabilities, realizing autonomous collaboration among multiple units under the constraint of coupling.

Description

Technical Field

[0001] This invention belongs to the field of microgrid dispatching technology, specifically, it relates to a microgrid multi-resource collaborative autonomous optimization method and system based on coupling constraint relaxation. Background Technology

[0002] The core of autonomous operation of microgrids lies in the collaborative optimization of multiple resources. Existing technologies mainly include: centralized optimization scheduling: establishing a global objective function encompassing all units and obtaining the optimal output through a solver (such as CPLEX). However, this requires a complete system model, incurs heavy communication burdens, and cannot handle dynamic changes in units; distributed control based on consensus algorithms: each unit achieves output consistency through neighbor communication. However, this usually assumes that all unit objective functions are isomorphic (e.g., all are quadratic forms) and handles coupling constraints (such as total power balance) simply by using Lagrange multiplier iterations, without considering constraint infeasibility cases; and Nash equilibrium methods based on game theory: treating each unit as a player in a game. However, the equilibrium point may not satisfy global constraints, and the computational complexity is high. The common drawbacks of the above methods are: treating coupling constraints as hard conditions that must be strictly satisfied, the optimization problem may become unsolvable once the system state changes abruptly (e.g., a unit exits); and the lack of a quantification mechanism for "tolerable deviations of constraints," leading to control rigidity.

[0003] In smart manufacturing microgrids, various local units with regulation capabilities exist, each with different operational objectives. However, these units share the same electrical or thermal coupling network and must satisfy global constraints such as total power balance and voltage / frequency stability. Existing autonomous control methods typically employ centralized optimization or fixed weight allocation, which suffers from the following problems: rigid handling of coupling constraints: traditional methods treat constraints such as power balance as hard equations, which can easily lead to optimization failures if local unit states change abruptly (e.g., equipment start-up or shutdown); lack of adaptive coordination mechanisms: each unit independently optimizes its local objectives, ignoring the impact of global coupling, resulting in regulation conflicts or resource waste; reliance on a central controller or pre-set model: all unit types and parameters need to be known in advance, making it difficult to adapt to dynamic resource access / exit scenarios; poor convergence performance: in manufacturing environments with frequent disturbances, fixed-step algorithms are prone to oscillations or slow convergence. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a microgrid multi-resource collaborative autonomous optimization method and system based on coupling constraint relaxation. This method forms a microgrid internal collaborative optimization that is decoupled from resource types, has no external dependencies, and possesses adaptive capabilities, enabling autonomous and efficient collaboration among multiple units while satisfying coupling constraints.

[0005] The present invention adopts the following technical solution.

[0006] This invention proposes a multi-resource collaborative autonomous optimization method for microgrids based on coupling constraint relaxation, comprising: Physical units with power regulation capabilities within the microgrid are abstracted as identical regulation agents. The basic attributes of each regulation agent include current output, feasible output range, and local objective function. The minimum sum of the local objective functions of all regulation agents is taken as the global objective function. Relaxation variables are introduced to correct the total power balance constraint satisfied by the global objective function. A global penalty term common to all regulation agents is constructed based on the relaxation variables and penalty coefficients, and the global penalty term is superimposed on the global objective function to obtain the collaborative autonomous optimization objective. Based on the collaborative autonomous optimization objective, a distributed gradient projection method is used to iteratively predict the output of each regulating agent. During the prediction process, the iteration step size is dynamically adjusted according to the change in the total power deviation of the microgrid in two consecutive iterations. Based on the contraction property of the distributed gradient projection method, a collaborative autonomous constraint is established based on the gradient of the local objective function, along with the penalty coefficient and the iteration step size. After updating the penalty coefficient, the dynamically adjusted iteration step size is updated based on the collaborative autonomous constraint. When the iteration convergence criterion is met, the iteration stops and the predicted output of each regulating agent is output as the result of the multi-resource collaborative autonomous optimization of the microgrid.

[0007] Preferably, the global objective function The minimum of the sum of the local objective functions of all regulating agents is shown in the following formula:

[0008] In the formula, To adjust the number of agents, For the current output set of all regulating agents, To regulate the agent The local objective function, To regulate the agent Current output; The total power balance constraint satisfied by the global objective function is shown in the following equation:

[0009] In the formula, For a moment The load demand.

[0010] Preferably, slack variables are introduced to correct the total power balance constraint, as shown in the following equation:

[0011] In the formula, As slack variables, .

[0012] Preferably, a global penalty term common to all regulation agents is constructed based on slack variables and penalty coefficients. This global penalty term is added to the global objective function to form the reconstructed microgrid collaborative autonomous optimization objective, as shown in the following equation:

[0013]

[0014] In the formula, The penalty coefficient is... .

[0015] Preferably, based on the collaborative autonomous optimization objective, a distributed gradient projection method is used to iteratively predict the output of each regulating agent, as shown in the following equation:

[0016] In the formula, For iteration Central regulation agent The predicted output Feasible range for output The projection operator, The lower limit of the interval. The upper limit of the interval, For iteration Central regulation agent The current output, For iteration Step size, To regulate the agent Local objective function gradient, This represents the direction of the total power deviation of the microgrid. It is set to 0 when the total power is balanced, 1 when the output of all regulating agents exceeds the load demand, and -1 when the output of all regulating agents is less than the load demand. For iteration Total power deviation, .

[0017] Preferably, the iteration step size is dynamically adjusted according to the change in the total power deviation of the microgrid in two consecutive iterations, as shown in the following formula:

[0018] In the formula, As the reference step size, This is the sensitivity coefficient. .

[0019] Preferably, the convergence condition is established based on the shrinkage property of the distributed gradient projection method, as shown in the following equation:

[0020] In the formula, suprem is the supremum function. This is the upper bound of the Lipschitz constant for the gradient of the local objective function. ; The collaborative autonomous constraint based on the convergence condition, including the penalty coefficient and the step size of each iteration, is shown in the following equation:

[0021] In the formula,

[0022] When the adaptively adjusted step size satisfies the collaborative autonomy constraint, it is expressed in the following form:

[0023] In the formula, For iteration The latest step size.

[0024] Preferably, an iterative convergence criterion is established using the total power deviation accuracy and the output change threshold of each regulating agent, as shown in the following formula:

[0025]

[0026] In the formula, For iteration Predicted total power deviation , For iteration Medium load demand forecast For total power deviation accuracy, This is the threshold for output variation; The total power deviation prediction value is determined based on the predicted output of each regulating agent. When the total power deviation prediction value meets the total power deviation accuracy and the predicted output of each regulating agent meets the output change threshold of each regulating agent, and the iteration convergence criterion is met, the iteration stops and the predicted output of each regulating agent is output as the result of the microgrid multi-resource collaborative autonomous optimization.

[0027] In another aspect, this invention proposes a microgrid multi-resource collaborative autonomous optimization system based on coupling constraint relaxation, comprising: The regulation agent abstraction module is used to abstract the physical units with power regulation capabilities in the microgrid into the same regulation agent. The basic attributes of each regulation agent include the current output, the feasible range of output, and the local objective function. The collaborative autonomous optimization objective establishment module is used to set the minimum sum of the local objective functions of all regulating agents as the global objective function; introduces slack variables to modify the total power balance constraint satisfied by the global objective function; constructs a global penalty term common to all regulating agents based on slack variables and penalty coefficients, and superimposes the global penalty term onto the global objective function to obtain the collaborative autonomous optimization objective; The collaborative autonomous optimization module is used to iteratively predict the output of each regulating agent based on the collaborative autonomous optimization objective and employing the distributed gradient projection method. During the prediction process, the iteration step size is dynamically adjusted according to the change in the total power deviation of the microgrid in two consecutive iterations. Based on the contraction property of the distributed gradient projection method, a collaborative autonomous constraint is established based on the gradient of the local objective function, along with the penalty coefficient and the iteration step size. After updating the penalty coefficient, the dynamically adjusted iteration step size is updated based on the collaborative autonomous constraint. When the iteration convergence criterion is met, the iteration stops and the predicted output of each regulating agent is output as the result of the multi-resource collaborative autonomous optimization of the microgrid.

[0028] The present invention is also a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0029] The present invention is also a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0030] The beneficial effects of the present invention are as follows, compared with the prior art, at least the present invention proposes a coupling constraint softening mechanism based on slack variables, which transforms the hard physical constraints such as power balance that must be strictly satisfied in microgrids into a tolerable deviation form with a penalty term, thereby fundamentally solving the problem of the infeasibility of optimization due to resource response delay or sudden exit. This invention also constructs a universal proxy abstraction model that is completely decoupled from the physical type of resources. It only requires defining the current output, feasible adjustment range and local objective function triplet to support plug-and-play of heterogeneous units such as photovoltaics, flexible loads and backup power. This invention designs a fully distributed gradient projection coordination algorithm that relies only on local gradient calculation and global bias broadcasting, without the need for a central controller, internal device models, or point-to-point communication. An adaptive step size adjustment mechanism based on power deviation and cooperative autonomous constraints is embedded to achieve a balance between fast response under disturbances and fine convergence under steady state. Attached Figure Description

[0031] Figure 1 This is a flowchart of a microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation proposed in this invention; Figure 2This is the optimized allocation result of multiple types of active power resource adjustment instructions in the embodiments of the present invention; Figure 3 This is the effect of coordinated control of multiple types of active resources in the embodiments of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0033] This invention proposes a multi-resource collaborative autonomous optimization method for microgrids based on coupling constraint relaxation. Physical units within the microgrid with power regulation capabilities are abstracted as identical regulation agents. Each regulation agent's basic attributes include current output, feasible output range, and local objective function. Relaxation variables are introduced to correct the total power balance constraint. A global penalty term common to all regulation agents is constructed based on the relaxation variables and penalty coefficients, and this global penalty term is superimposed on the local objective function of each regulation agent. The summation of the local objective functions of the superimposed global penalty terms for all regulation agents is used, with minimizing this sum as the microgrid collaborative autonomous optimization objective. The penalty coefficient is determined under the corrected total power balance constraint. A local penalty term common to all regulation agents is constructed using the penalty coefficients and the direction of the microgrid's total power deviation. A distributed gradient projection coordination method is adopted, and each regulation agent iteratively updates its output based on the gradient of its local objective function and the local penalty term. The iteration step size is adaptively adjusted according to the microgrid's total power deviation.

[0034] like Figure 1 As shown, it includes the following steps: Step 1: Abstract all physical units with power regulation capabilities within the microgrid into the same regulation agent to construct a multi-agent abstract model of the microgrid; where each regulation agent contains three basic attributes: current output, feasible output range, and local objective function.

[0035] Power system interaction demands typically arise within a given timescale. The power system needs to maintain real-time power balance, necessitating the regulation of a portion of flexible and controllable power sources or resources to track power changes in real time. This demand for active power regulation constitutes the system's interaction demand. Grid interaction demands primarily stem from factors causing power fluctuations in the power system, including the random volatility of loads and renewable energy sources, as well as prediction errors. All units within a microgrid with power regulation capabilities (regardless of their physical form as generation, consumption, or energy storage devices) are abstracted as "Regulation Agents." Each agent... Only three basic attributes need to be defined: current output Feasible range of output and a locally differentiable local objective function. .in, It can represent any optimization intention such as minimum energy consumption, lowest operating cost, and power fluctuation suppression. Its specific form is determined by the agent itself and does not need to be disclosed or standardized to other parts of the system.

[0036] To illustrate the feasibility of this modeling approach, we will use two typical power consumption-side resources and one typical power generation-side resources as examples for specific explanation.

[0037] First, the attribute definition of the electricity consumption agent is explained using a representative example of a constant-temperature flexible load cluster with continuous power regulation capability (such as a large central air conditioning system): A1) Current output This refers to the actual active power consumption of the load at the current moment, which can be read in real time from a smart meter or energy management system. A2) Feasible output range: This is determined by both the rated capacity of the equipment and the operating boundaries set by the user. For example, if an air conditioning system is allowed to adjust within the range of 300–600kW to maintain the room temperature within the comfortable range, then this range is its feasible domain. A3), Local Objective Function : Set as a penalty item for deviating from the current execution point, such as This reflects the operational preference of "fewer adjustments are better than more adjustments".

[0038] Secondly, taking grid-connected photovoltaic inverter clusters as an example, the attribute definition of the generation-side agent is explained: B1) Current output This refers to the actual active power output of the photovoltaic array, which can be directly obtained through the inverter communication interface or SCADA system. B2), Feasible output range: including the upper limit The maximum available power output (i.e., "usable output") under current illumination conditions is determined and can be estimated through short-term irradiance prediction or measured MPPT points. The lower limit is usually set to 0 (allowing complete curtailment), but it can also be set to a positive value based on reactive power support requirements (e.g., reserving 10% capacity for voltage regulation). For example, if the predicted maximum available power output at a certain moment is 800kW, then the feasible range can be set to [0, 800]kW. B3), Local Objective Function If the goal is to maximize the utilization of new energy sources, a penalty system for wasted solar power can be implemented. , >0, The weighting factor is equivalent to encouraging maximum power generation; if inverter losses are considered, a quadratic model can be used. ,in , These are the fitting parameters.

[0039] Both of the above examples do not rely on production processes, internal equipment states, or superior scheduling instructions; modeling can be completed solely based on locally observable electrical quantities (power, capacity limits) and the operational intentions set by the user / operation and maintenance party. Similarly, other resources such as diesel generators, fuel cells, and charging pile clusters can also have their triple attributes defined using the same logic. Thus, the entire microgrid is transformed into a set of generic agents carrying only information on "current state—adjustment boundary—optimization tendency," providing a unified interface for subsequent undifferentiated, decentralized collaborative optimization.

[0040] This modeling approach completely eliminates reliance on prior information such as equipment type, process flow, and internal structure, basing its model solely on observable power boundaries and local optimization objectives, thus laying the foundation for subsequent generalized collaboration. The key to this step lies in "de-identification"—the system only cares about how much the adjustable unit can be adjusted and to what extent.

[0041] Step 2: Establish an economic dispatch optimization model for the microgrid. The global objective function of the economic dispatch optimization model is the minimum of the sum of the local objective functions of all regulation agents. The global objective function satisfies the total power balance constraint of the microgrid.

[0042] After completing the abstract model of the multi-regulatory agent in a microgrid, it is important to understand that a microgrid is essentially a strongly coupled energy system. All regulators share the same bus voltage and the same distribution network, and their power behaviors are mutually influential and inseparable. If each regulator only pursues local optima, such as maximizing photovoltaic power generation and minimizing air conditioning regulation, the total output of the microgrid may far exceed or fall below load demand, leading to voltage exceeding limits, frequency instability, and even protection tripping. Therefore, based on the regulator abstraction, this invention establishes a coordination mechanism that enables each regulator to jointly satisfy the global physical constraints of the system while pursuing its own goals. The coordination mechanism established in this invention clarifies the system-level coupling relationships, including identifying the physical quantities that all regulators interact with, such as total net power, critical node voltage, and trunk line power flow. Among these, the total power balance constraint is the most fundamental and universal coupling relationship, applicable to both grid-connected and islanded operation modes, and relies only on active power data without requiring topological details; therefore, it is selected as the core coupling constraint. Simultaneously, the total power balance constraint defines the objective paradigm for collaborative optimization. Since the objective functions of each regulating agent are different, some minimize cost and some minimize regulation amount, it is necessary to unify the objective functions of each regulating agent into an additivity global objective framework. Therefore, this invention adopts the idea of ​​maximizing social welfare and defines the overall system performance as the weighted sum of the local utilities of all regulating agents, thereby forming an optimization structure that can be decomposed but requires collaborative solution. Then, a distributed algorithm is designed to achieve decentralized collaboration.

[0043] Specifically, the global objective function of the economic scheduling optimization model The minimum of the sum of the local objective functions of all regulating agents is shown in the following formula:

[0044] In the formula, To adjust the number of agents, For the current output set of all regulating agents, To regulate the agent The local objective function, To regulate the agent The current output.

[0045] global objective function It serves as the basis and foundation for all distributed iterative updates of regulating agents. The core logic of the distributed gradient projection algorithm is to guide each regulating agent along the global objective function. The gradient direction updates its output; the final convergence condition of the entire distributed iteration is to make the global objective function... The minimum value is achieved while satisfying the system's global physical constraints.

[0046] The constraints that the global objective function must satisfy fall into two categories: 1) Microgrid total power balance constraint, i.e., the sum of the net output of all regulating agents must be equal to the value at time t. load demand As shown in the following formula:

[0047] 2) Several shared physical constraints, such as bus voltage limits, line thermal stability limits, and thermodynamic balance equations, are collectively expressed as the following formula: ,

[0048] In the formula, For the first A shared physical constraint function expression represents a certain type of physical operating state quantity (such as bus voltage, line power flow, branch temperature rise, etc.) jointly determined by the output of all regulating agents. It is used to quantify whether the current physical operating state of the system exceeds the constraint. The total number of shared physical constraints; These constraints are inherently coupled; any change in the output of a regulating agent will affect the feasibility of the entire system. However, in traditional frameworks, such constraints are treated as rigid conditions, meaning they must be strictly satisfied. If an agent fails to act as expected due to equipment failure, communication delays, or response saturation, the constraint equations may instantly become invalid, rendering the optimization problem unsolvable and forcing the system to degrade. Therefore, a mechanism is urgently needed to alleviate this rigidity.

[0049] Step 3: Introduce slack variables to correct the total power balance constraint. Based on the slack variables and penalty coefficients, construct a global penalty term that is common to all regulation agents. Add the global penalty term to the global objective function to form the reconstructed microgrid collaborative autonomous optimization objective.

[0050] To address the infeasibility issue caused by overly rigid coupling constraints in traditional optimization models, this invention introduces slack variables to soften rigid coupling constraints, particularly the total power balance constraint of the microgrid. Slack variables are a classic technique in mathematical programming used to transform rigid equality or inequality constraints that must be strictly satisfied into soft constraints that allow for certain deviations. The magnitude of these deviations is controlled by adding a penalty term to the global objective function.

[0051] In engineering practice, slack variables have been widely applied in various fields. For example, in power system unit commitment, to avoid unsolvable scheduling schemes due to load forecasting errors, slack variables are often introduced into power balance constraints, and high-cost penalties are imposed to simulate the cost of load shedding or starting / stopping standby units. In robot motion planning, to handle the conflict between obstacle avoidance constraints and path smoothness, slack variables are also introduced into obstacle boundaries, allowing the planner to generate "approximately feasible" trajectories even in extreme cases. In machine learning support vector machines (SVM), soft-margin classification uses slack variables to allow for some misclassification, thereby improving the model's generalization ability under noisy data. The common logic in these cases is: it is better to accept a controllable small deviation than to fall into a completely infeasible deadlock.

[0052] Specifically, this invention softens the total power balance constraint by introducing relaxation variables, as shown in the following equation:

[0053] In the formula, As slack variables, This represents the maximum allowable power imbalance margin of the system.

[0054] To prevent Unrestricted expansion can compromise system stability. A global penalty term, based on slack variables and penalty coefficients, is constructed that is applicable to all regulating agents. This term is then applied to the global objective function. Add a global penalty item The reconstructed microgrid collaborative autonomous optimization objective is formed as shown in the following formula:

[0055]

[0056] In the formula, The penalty coefficient is... It is used to weigh the benefits of local optimization against the costs of global balancing.

[0057] This refactoring ensures that a feasible solution always exists for the problem, for example, let It just needs to be large enough to fundamentally avoid unsolvable downtime. At the same time, The parameters can be designed as time-varying parameters, taking larger values ​​in steady state to pursue high-precision balance, and smaller values ​​in the early stages of disturbance to prioritize the continuity of adjustment. The penalty term has a dual function: it is used not only for constraint relaxation but also to accelerate the convergence of distributed gradient projection or to guarantee privacy protection boundaries between agents. Therefore, this invention proposes that when the cooperative autonomous optimization objective satisfies the modified total power balance constraint, the penalty coefficient obtained through iterative solution is used as the initial value in the improved distributed gradient projection coordination method.

[0058] After adopting the reconfigured microgrid collaborative autonomous optimization objective, the system can ensure that the problem is always feasible. Even if some regulating agents are temporarily unable to respond, such as being in a regulating dead zone or fault state, the remaining regulating agents can still introduce small imbalances (i.e., Under the premise of [missing information], continue collaborative optimization to avoid interruptions in the entire coordination process; achieve a dynamic balance between safety and economy. Through adjustment The size allows for flexible control over the tolerance for imbalance, and the high... It is better to make more adjustments than to lose balance, suitable for steady-state high-precision scenarios; low Prioritizing continuity while allowing for temporary imbalances is suitable for the initial stages of disturbances or periods of resource scarcity; and adjustments... The size of the algorithm can also improve its robustness and convergence stability; since the optimization problem no longer collapses due to instantaneous overshoot, the distributed iterative process can continue, and eventually automatically returns to a high-precision equilibrium state after the disturbance subsides.

[0059] This invention transforms the rigid coupling constraints of a microgrid into soft constraints by introducing relaxation variables and penalty coefficients. This solves the problems of infeasibility in optimization and system collapse under traditional hard constraints. Simultaneously, by adjusting the penalty coefficient, a dynamic trade-off is achieved between steady-state high accuracy and continuity during disturbances, providing feasible space and robustness guarantees for distributed iteration. This soft constraint system and objective function are the core foundation of the distributed gradient projection algorithm: the total power deviation is directly taken from the relaxation amount of the soft constraints, the adaptive step size is adjusted in conjunction with the penalty coefficient, and each agent iteratively updates along the gradient direction of the soft constraint objective function, with global optimum as the convergence target, forming a complete logical closed loop from constraint softening and parameter trade-offs to distributed solution.

[0060] Step 4: Based on the collaborative autonomous optimization objective, the distributed gradient projection method is used to iteratively predict the output of each regulating agent. During the prediction process, the iteration step size of each round is dynamically adjusted according to the change in the total power deviation of the microgrid in two consecutive rounds of iteration. Based on the shrinkage property of the distributed gradient projection method, a collaborative autonomous constraint of the penalty coefficient and the iteration step size of each round is established according to the gradient of the local objective function. After updating the penalty coefficient, the dynamically adjusted iteration step size of each round is updated based on the collaborative autonomous constraint.

[0061] After softening the coupling constraints, a collaborative optimization algorithm needs to be designed that can be autonomously executed by each regulating agent under conditions without a central controller. While a centralized solution, such as calling a commercial optimizer, can obtain the global optimum, it relies on a complete system model, incurs high communication overhead, and cannot adapt to dynamic resource access / exit, making it difficult to meet the high autonomy and fast response requirements of microgrids. Therefore, this invention designs a distributed coordination algorithm based on gradient projection to achieve consensus optimization among multiple regulating agents.

[0062] The core of this algorithm is that each regulating agent only needs to use local information (the gradient of the local objective function) and lightweight global feedback (the direction of the total power deviation of the microgrid) to iteratively adjust the output and gradually approach the global optimum.

[0063] Specifically, step 4 includes: Step 4.1: Based on the collaborative autonomous optimization objective, the distributed gradient projection method is used to iteratively predict the output of each regulating agent; Specifically, in iteration In the middle, based on the goal of collaborative autonomy optimization, the agent is regulated. Perform the following operations: Obtain the total power deviation of the microgrid. It can be obtained from the communication bus via periodic broadcasts (e.g., once every 5 seconds), without the need for a point-to-point connection; the gradient of the local objective function can be calculated. and gradient correction direction based on penalty coefficient This comprehensively reflects the improvement trend of the current output direction on the local objective function; Using the time scale as the iteration scale, in iteration In this study, based on the collaborative autonomous optimization objective, a distributed gradient projection method is used to calculate the iterative... The predicted output of each regulating agent is shown in the following formula:

[0064] In the formula, For iteration Central regulation agent The predicted output Feasible range for output The projection operator, The lower limit of the interval. The upper limit of the interval, For iteration Central regulation agent The current output, For iteration Step size, To regulate the agent Local objective function gradient, The penalty coefficient is... This represents the direction of the total power deviation of the microgrid. It takes a value of 0 when the total power is balanced, a value of 1 when the output of all regulating agents is greater than the load demand, and a value of -1 when the output of all regulating agents is less than the load demand. The indicator system determines whether there is "excess power generation" or "insufficient power consumption," guiding all regulating agents to correct in the same direction. It is a projection operator that ensures that the output always falls within the physically feasible range.

[0065] This design has three major advantages: first, it is decentralized, eliminating the risk of single points of failure; second, it has low communication overhead, requiring only the broadcast of a single scalar (total power or deviation); and third, it features implicit duality coordination with a penalty term. It essentially simulates the effect of Lagrange multipliers, avoiding the complexity of explicitly solving for dual variables.

[0066] Real-world testing shows that in a 6-node microgrid, this algorithm can reduce power imbalance to less than 1% within 20 iterations, fully meeting the requirements for real-time control.

[0067] Step 4.2: Dynamically adjust the iteration step size for each iteration based on the change in the total power deviation of the microgrid in two consecutive iterations.

[0068] Gradient algorithms with fixed step sizes perform poorly when faced with dynamic disturbances: too large a step size can easily lead to oscillations, while too small a step size results in slow convergence. This problem is particularly prominent in microgrids in manufacturing parks—loads often change abruptly due to production line start-ups and shutdowns, and photovoltaic output fluctuates drastically due to cloud cover. To improve the robustness of the algorithm, this invention introduces an adaptive step size mechanism, whose design draws on the error-driven approach from adaptive filtering (such as the LMS algorithm).

[0069] Specifically, step size It is no longer a constant, but is dynamically adjusted based on the change in the total power deviation of the microgrid in two consecutive iterations, as shown in the following formula:

[0070] In the formula, The reference step size is 0.01 in this example; This is the sensitivity coefficient. In this example, the value is 0.1.

[0071] When the system is suddenly disturbed (such as a sudden increase in load of 100 kW), the deviation changes drastically. As the step size increases, the step size automatically decreases to suppress overshoot; when the system stabilizes and the deviation changes more slowly, the step size gradually recovers, accelerating fine convergence. This mechanism requires no prediction or parameter identification, relying solely on the deviation observations of the previous two steps, making it extremely simple to implement. Simulations show that under the same perturbation, the algorithm using an adaptive step size converges 45% faster than the fixed step size scheme, with no significant oscillations, significantly improving the reliability of practical deployments.

[0072] Step 4.3: Based on the shrinkage property of the distributed gradient projection method, establish a collaborative autonomous constraint of penalty coefficient and iteration step size for each round according to the gradient of the local objective function.

[0073] The convergence condition is established based on the shrinkage property of the distributed gradient projection method, as shown in the following equation:

[0074] In the formula, suprem is the supremum function. This is the upper bound of the Lipschitz constant for the gradient of the local objective function. .

[0075] When the step size is large and the penalty coefficient is large, the current output of each regulating agent may cause microgrid oscillations; while when the step size is small and the penalty coefficient is small, the process of determining the current output of each regulating agent converges slowly and cannot meet the real-time requirements of scheduling. Therefore, this invention establishes a collaborative autonomous constraint based on the convergence condition, using the penalty coefficient and the step size of each iteration, as shown in the following equation:

[0076] In the formula, In the embodiments, The value is 0.9; Step 4.4: After updating the penalty coefficient under the intraday scheduling scale, update the dynamically adjusted iteration step size under the real-time scheduling scale based on the collaborative autonomy constraint.

[0077] Through collaborative autonomy constraints, the penalty coefficient is made... Iteration under fixed conditions step size Having a fixed upper bound also makes iteration possible. step size Penalty coefficient under relatively fixed conditions Since there is a fixed upper bound, when the adaptively adjusted step size satisfies the cooperative autonomy constraint, it can be expressed in the following form:

[0078] In the formula, For iteration The latest step size.

[0079] Furthermore, the penalty coefficient will be... Treating it as a slow-time-scale variable, such as minute-level adjustments, the penalty coefficient is updated based on the intraday scheduling scale; simultaneously, Treating it as a fast-time-scale variable, such as adjusting at the second level, after updating the penalty coefficient at each slow-time-scale, it is recalculated based on the real-time scheduling scale. The upper boundary.

[0080] Step 5: When the iterative convergence criterion is met, stop the iteration and output the predicted output of each regulating agent as the result of the microgrid multi-resource collaborative autonomous optimization.

[0081] By utilizing the total power deviation accuracy and the output change threshold of each regulating agent, a dual criterion for iterative convergence is established, as shown in the following equation:

[0082]

[0083] In the formula, For iteration Predicted total power deviation , For iteration Medium load demand forecast For total power deviation accuracy, The value is 0.01 pu. The threshold for output variation Values ; The total power deviation prediction value is determined based on the predicted output of each regulating agent. When the total power deviation prediction value meets the total power deviation accuracy and the predicted output of each regulating agent meets the output change threshold of each regulating agent, that is, when the dual criteria for iterative convergence are met, the iteration stops and the predicted output of each regulating agent is output as the result of multi-resource collaborative autonomous optimization of the microgrid.

[0084] After establishing the update mechanism for each iteration step size, the system enters a closed-loop operation phase. Each agent repeatedly executes the above steps to continuously optimize output allocation. To determine when to stop iteration, a dual convergence criterion is set; once both are satisfied, autonomous equilibrium is considered achieved, and each regulating agent locks the current output obtained from the latest iteration and enters steady-state operation. If a significant disturbance is subsequently detected, such as... If the cycle continues for more than 10 seconds, a new round of optimization will be automatically triggered. The entire process requires no external scheduling commands, no manual intervention, and no prior knowledge of resource types, truly achieving a microgrid autonomous closed loop integrating "perception-decision-execution".

[0085] In another aspect, this invention proposes a microgrid multi-resource collaborative autonomous optimization system based on coupling constraint relaxation, comprising: The regulation agent abstraction module is used to abstract the physical units with power regulation capabilities in the microgrid into the same regulation agent. The basic attributes of each regulation agent include the current output, the feasible range of output, and the local objective function. The collaborative autonomous optimization objective establishment module is used to set the minimum sum of the local objective functions of all regulating agents as the global objective function; introduces slack variables to modify the total power balance constraint satisfied by the global objective function; constructs a global penalty term common to all regulating agents based on slack variables and penalty coefficients, and superimposes the global penalty term onto the global objective function to obtain the collaborative autonomous optimization objective; The collaborative autonomous optimization module is used to iteratively predict the output of each regulating agent based on the collaborative autonomous optimization objective and employing the distributed gradient projection method. During the prediction process, the iteration step size is dynamically adjusted according to the change in the total power deviation of the microgrid in two consecutive iterations. Based on the contraction property of the distributed gradient projection method, a collaborative autonomous constraint is established based on the gradient of the local objective function, along with the penalty coefficient and the iteration step size. After updating the penalty coefficient, the dynamically adjusted iteration step size is updated based on the collaborative autonomous constraint. When the iteration convergence criterion is met, the iteration stops and the predicted output of each regulating agent is output as the result of the multi-resource collaborative autonomous optimization of the microgrid.

[0086] Through simulation verification of a 6-node microgrid, the optimized allocation results of various types of active power resource regulation commands obtained by the method proposed in this invention are as follows: Figure 2 As shown, Figure 2 The vertical axis represents the proportion of adjustment commands for each type of active power resource, and the effect of coordinated control of multiple types of active power resources is as follows: Figure 3 As shown, Figure 3 The vertical axis represents the power of the lower grid point. Case 1 is the controlled scheme obtained by using the method proposed in this invention, and Case 2 is the uncontrolled scheme obtained without using the method proposed in this invention. The dashed line represents the power target command. It can be seen that, without relying on external commands and without pre-setting resource types, a power balance accuracy of over 98% can be achieved, and the convergence speed is improved by 45% compared with the traditional gradient method, providing theoretical support for the autonomous operation of intelligent manufacturing microgrids.

[0087] Because this invention introduces slack variables to soften coupling constraints, the optimization problem still has feasible solutions when the system encounters partial unit failure or regulation saturation. This avoids coordination interruptions or forced load shedding caused by "unsatisfiable hard constraints" in traditional methods, thus significantly improving the system's operational resilience and continuity. Furthermore, since the surrogate model only relies on power boundaries and local targets, without involving equipment process details or type identification, new resources do not need to be remodeled or their parameters tuned when they are added. They can be automatically integrated into the cooperative network simply by publishing their triple attributes, thus achieving true plug-and-play functionality and architectural scalability. Moreover, the designed distributed algorithm only needs to broadcast a scalar deviation information, resulting in extremely low communication overhead. The update rule only contains local gradients and sign functions, with linear computational complexity, making it suitable for resource-constrained edge controllers. Finally, the adaptive step size mechanism automatically adjusts the learning rate according to the system dynamics, suppressing oscillations during load mutations and accelerating convergence in steady state. The overall convergence performance is improved by more than 40% compared to the fixed step size scheme. In summary, this invention outperforms existing technologies in four dimensions: feasibility, robustness, scalability, and economy, providing a universal, reliable, and lightweight solution for the highly autonomous operation of intelligent manufacturing microgrids.

[0088] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0089] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0090] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0091] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation, characterized in that, include: The physical units with power regulation capabilities within the microgrid are abstracted into identical regulation agents. The basic attributes of each regulation agent include the current output, the feasible output range, and the local objective function. The minimum sum of the local objective functions of all regulation agents is taken as the global objective function. Relaxation variables are introduced to modify the total power balance constraint satisfied by the global objective function. A global penalty term common to all regulatory agents is constructed based on slack variables and penalty coefficients, and the global penalty term is superimposed on the global objective function to obtain the cooperative autonomous optimization objective; Based on the collaborative autonomous optimization objective, a distributed gradient projection method is used to iteratively predict the output of each regulating agent. During the prediction process, the iteration step size is dynamically adjusted according to the change in the total power deviation of the microgrid in two consecutive iterations. Based on the contraction property of the distributed gradient projection method, a collaborative autonomous constraint is established based on the gradient of the local objective function, along with the penalty coefficient and the iteration step size. After updating the penalty coefficient, the dynamically adjusted iteration step size is updated based on the collaborative autonomous constraint. When the iteration convergence criterion is met, the iteration stops and the predicted output of each regulating agent is output as the result of the multi-resource collaborative autonomous optimization of the microgrid.

2. The microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation according to claim 1, characterized in that, global objective function The minimum of the sum of the local objective functions of all regulating agents is shown in the following formula: In the formula, To adjust the number of agents, For the current output set of all regulating agents, To regulate the agent The local objective function, To regulate the agent Current output; The total power balance constraint satisfied by the global objective function is shown in the following equation: In the formula, For a moment The load demand.

3. The microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation according to claim 2, characterized in that, The total power balance constraint is modified by introducing slack variables, as shown in the following equation: In the formula, As slack variables, .

4. The microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation according to claim 3, characterized in that, A global penalty term, common to all regulation agents, is constructed based on slack variables and penalty coefficients. This global penalty term is then added to the global objective function to form the reconstructed microgrid collaborative autonomous optimization objective, as shown in the following equation: In the formula, The penalty coefficient is... .

5. The microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation according to claim 4, characterized in that, Based on the collaborative autonomous optimization objective, the distributed gradient projection method is used to iteratively predict the output of each regulating agent, as shown in the following equation: In the formula, For iteration Central regulation agent The predicted output Feasible range for output The projection operator, The lower limit of the interval. The upper limit of the interval, For iteration Central regulation agent The current output, For iteration Step size, To regulate the agent Local objective function gradient, This represents the direction of the total power deviation of the microgrid. It is set to 0 when the total power is balanced, 1 when the output of all regulating agents exceeds the load demand, and -1 when the output of all regulating agents is less than the load demand. For iteration Total power deviation, .

6. The microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation according to claim 5, characterized in that, The iteration step size is dynamically adjusted based on the change in the total power deviation of the microgrid in two consecutive iterations, as shown in the following formula: In the formula, As the reference step size, This is the sensitivity coefficient. .

7. The microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation according to claim 6, characterized in that, The convergence condition is established based on the shrinkage property of the distributed gradient projection method, as shown in the following equation: In the formula, suprem is the supremum function. This is the upper bound of the Lipschitz constant for the gradient of the local objective function. ; The collaborative autonomous constraint based on the convergence condition, including the penalty coefficient and the step size of each iteration, is shown in the following equation: In the formula, When the adaptively adjusted step size satisfies the collaborative autonomy constraint, it is expressed in the following form: In the formula, For iteration The latest step size.

8. The microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation according to claim 5, characterized in that, Using the total power deviation accuracy and the output change threshold of each regulating agent, an iterative convergence criterion is established, as shown in the following formula: In the formula, For iteration Predicted total power deviation , For iteration Medium load demand forecast For total power deviation accuracy, This is the threshold for output variation; The total power deviation prediction value is determined based on the predicted output of each regulating agent. When the total power deviation prediction value meets the total power deviation accuracy and the predicted output of each regulating agent meets the output change threshold of each regulating agent, and the iteration convergence criterion is met, the iteration stops and the predicted output of each regulating agent is output as the result of the microgrid multi-resource collaborative autonomous optimization.

9. A microgrid multi-resource collaborative autonomous optimization system based on coupling constraint relaxation, used to implement the microgrid multi-resource collaborative autonomous optimization method based on coupling constraint relaxation as described in any one of claims 1 to 8, characterized in that, include: The regulation agent abstraction module is used to abstract the physical units with power regulation capabilities in the microgrid into the same regulation agent. The basic attributes of each regulation agent include the current output, the feasible range of output, and the local objective function. The collaborative autonomous optimization objective establishment module is used to set the minimum sum of the local objective functions of all regulating agents as the global objective function; slack variables are introduced to modify the total power balance constraint satisfied by the global objective function; A global penalty term common to all regulatory agents is constructed based on slack variables and penalty coefficients, and the global penalty term is superimposed on the global objective function to obtain the cooperative autonomous optimization objective; The collaborative autonomous optimization module is used to iteratively predict the output of each regulating agent based on the collaborative autonomous optimization objective and employing the distributed gradient projection method. During the prediction process, the iteration step size is dynamically adjusted according to the change in the total power deviation of the microgrid in two consecutive iterations. Based on the contraction property of the distributed gradient projection method, a collaborative autonomous constraint is established based on the gradient of the local objective function, along with the penalty coefficient and the iteration step size. After updating the penalty coefficient, the dynamically adjusted iteration step size is updated based on the collaborative autonomous constraint. When the iteration convergence criterion is met, the iteration stops and the predicted output of each regulating agent is output as the result of the multi-resource collaborative autonomous optimization of the microgrid.

10. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-8.

Images