Source-load multi-subject layered collaborative optimization method based on fused niche

Abstract

The invention relates to a source-load multi-subject hierarchical collaborative optimization method based on a fused niche. According to the scheme, multi-subject modeling, a master-slave game mechanism, niche evolution optimization and an interlayer feedback coordination strategy are combined. Firstly, a unified mathematical model of various power supplies and loads in a source-load system is constructed, and optimization targets and constraints of the unified mathematical model are defined. Secondly, introducing a master-slave game (Stackelberg) mechanism, and simulating a dynamic game behavior of source first-onset and load response; thirdly, a niche evolution algorithm is adopted to improve the search diversity and the global optimal solution obtaining capability; and finally, through an interlayer feedback mechanism, realizing mutual guidance and correction of a game result and an evolution process, and outputting a global consistent scheduling scheme. According to the method, the optimization precision, the operation coordination and the scheduling robustness of the source-load system in a complex and changeable environment can be effectively improved, and the method is suitable for multi-source and multi-load collaborative optimization scenes such as a micro-grid and an integrated energy system.

Description

Technical Field

[0001] This invention relates to the field of power system optimization control, and in particular to a source-load multi-agent hierarchical collaborative optimization method based on integrated niches. Background Technology

[0002] With the rapid development of new energy technologies and the continuous improvement of user-side load response capabilities, modern power systems are gradually evolving into complex systems characterized by high distribution, strong uncertainty, and multi-stakeholder participation. In this system, the traditional "source follows load" dispatching concept is no longer adequate to meet the operational demands of the new situation. The large-scale integration of various distributed power sources (such as photovoltaic and wind power), controllable loads, electric vehicles, and energy storage devices has significantly increased the number of dispatchable objects within the system, making operating states more dynamic and varied, and exhibiting significant heterogeneity in optimization objectives and behavioral patterns among different stakeholders. Against this backdrop, traditional centralized optimization dispatching methods, due to their reliance on global information, high computational complexity, and poor real-time performance, are unable to meet the needs of multi-stakeholder autonomous decision-making, rapid response, and local optimization, thus hindering the improvement of collaborative dispatching capabilities in new power systems.

[0003] In recent years, multi-agent collaborative optimization has become a hot research topic in intelligent power system dispatching. Some studies have attempted to introduce game theory and multi-agent system modeling into the source-load optimization framework to improve the system's ability to characterize the heterogeneity of individual behaviors and their interactions. However, most current mainstream methods rely on centralized modeling and the assumption of complete information, failing to fully consider the incomplete rationality, behavioral coupling, and goal conflicts among individuals, making them difficult to adapt to the hierarchical structure and evolutionary characteristics of complex systems. At the optimization algorithm level, although research based on evolutionary computation methods such as genetic algorithms and particle swarm optimization is constantly emerging, traditional evolutionary algorithms generally suffer from the problems of easily getting trapped in local optima and having a single search path, making it difficult to effectively cope with complex optimization scenarios such as non-convex, nonlinear, and multi-peak. Niche optimization algorithms, as an extension mechanism of evolutionary algorithms, can effectively maintain population diversity and enhance the global exploration capability of complex search spaces. Therefore, it is necessary to propose a source-load multi-agent collaborative optimization method that combines hierarchical game modeling and niche evolutionary mechanisms to achieve intelligent collaboration between power sources and loads under their respective goal constraints. Summary of the Invention

[0004] Purpose of the invention: To overcome the difficulties in multi-agent optimization of source and load, this invention discloses a hierarchical collaborative optimization method for multi-agent source and load based on fusion niches. Through a master-slave game mechanism and niche evolution optimization, it effectively promotes collaborative optimization of multiple sources and loads.

[0005] Technical solution: This invention discloses a source-load multi-agent hierarchical collaborative optimization method based on fusion niches, comprising the following steps:

[0006] S1 constructs a mathematical modeling system for multi-source, multi-load optimization objects, clarifying the decision variables, optimization objectives, and constraints of various power sources (such as photovoltaic, wind power, and gas turbine units) and loads (such as industrial loads and residential loads), serving as the foundational model for subsequent hierarchical game theory and evolutionary optimization. The specific implementation steps are as follows:

[0007] S11 models the main body on the power supply side, clarifying its participating variables, optimization objectives, and operational constraints to support the setting and evolution optimization of the "leader" role in the master-slave game in S2;

[0008] S12 models the load-side entities, capturing their sensitivity and adjustment capabilities to electricity prices or dispatch signals, and supports their participation in decision-making, response, and optimization as "followers" in the game.

[0009] The S13 specification defines the decision variable structure, shared information variables, and objective function interface for multiple agents, enabling efficient integration of subsequent game theory and optimization algorithms.

[0010] S2 builds upon the multi-agent model constructed in S1, introducing a Stackelberg game mechanism to construct a hierarchical game structure with sources as the master and loads as the slaves. This simulates the dynamic interaction between power source preemptive strategies (such as electricity prices or output) and load response behavior, providing a hierarchical collaborative framework for evolutionary optimization. The specific implementation steps are as follows:

[0011] S21 constructs a master-slave game (Stackelberg) structure. It simulates a realistic scheduling mechanism with a master-slave decision-making order between power sources and loads. This allows power sources to specify electricity prices or output plans in advance, and loads to optimize their own response strategies after obtaining the power source's strategy, thus constructing a decision-making model with a clearly defined logical hierarchy. In the game structure, the power source set G = {G1, G2, ..., G...} N} is the leader, and the load set L = {L1, L2, ..., L} N} is a follower;

[0012] S22 constructs the objective function on the power source side. It clarifies the form of the power source's objective function and its driving effect on the game outcome, enabling it to influence load behavior and optimize its own interests through electricity price adjustments. Specifically, the power source adjusts electricity prices... Guided load response This creates a positive feedback loop between price and electricity consumption behavior;

[0013] S23 constructs an optimal response mechanism on the load side, enabling load entities to perform personalized adaptive optimization responses based on the electricity price strategy of the power source's location and their own preferences (comfort, cost, regulation capability, etc.).

[0014] S24 solves for the master-slave game equilibrium between the power source and the load, ensuring that the power source and the load achieve their respective optimal and mutually adaptive strategy combinations in the game strategy space. First, the power source strategy variable X is fixed. s For each load body L j Solve for its optimal response. Then, summarize all load responses and substitute them into the power source objective function to form a constrained optimization model for the power source.

[0015] Building upon the game structure constructed in S2, S3 introduces a niche evolutionary optimization mechanism for each type of source load subject. Utilizing fitness sharing and a local ecosystem evolution algorithm, it maintains the diversity of the multi-solution space and global search capability, obtaining a set of locally optimal response strategies under multi-objective conflict. The specific implementation steps are as follows:

[0016] In the S31 load system, due to the inconsistent interests of various stakeholders, the complexity of the strategy space, and the existence of multiple local optima in the objective function, traditional optimization methods may fail to find the global optimum. Therefore, a niche strategy is introduced to construct a diverse basic population for the solution space, and multiple small ecological subpopulations are divided through feature clustering or spatial partitioning, allowing individuals in different regions to evolve in parallel.

[0017] To address the fitness inflation problem caused by dense clustering of individuals, S32 utilizes a fitness sharing mechanism to weaken the evaluation of locally highly concentrated individuals, enabling niches to be evenly distributed.

[0018] S33 independently conducts evolutionary search within each niche, achieving efficient optimization of the multi-objective function within local optima. For each niche, an evolutionary operator is used for individual updates.

[0019] To avoid the local convergence of evolution in S33, S34 utilizes the preservation of elite individuals and migration between microhabitats to retain superior individuals and prevent degeneration. At the same time, it strengthens information exchange between microhabitats and enhances the overall synergy and global exploration capabilities of the population.

[0020] Building upon the diverse strategy space established in S3, S4 introduces an inter-layer feedback mechanism, mutually guiding the game outcome and niche evolutionary solutions to form a dynamic closed-loop collaborative optimization process. It then integrates the global objective to achieve the final scheduling scheme output, improving the overall robustness and coordination of the system while ensuring local self-optimization. The specific implementation steps are as follows:

[0021] S41 injects the game equilibrium result obtained in step S2 as guiding knowledge into the niche evolutionary population in S3 to improve the quality of the optimization starting point and guide the search direction. First, the optimal power supply strategy X obtained in S2... s * and optimal load response Form a decision vector Xeq Next, during the initialization of the S3 population, X... eq As elite individuals are injected into various microhabitats, the resulting updated microhabitat N is obtained. k To strengthen information linkage between the game layer and the optimization layer and prevent strategy conflicts, the local optimum of the game is regarded as the navigation for individual optimization, thereby improving the initial quality of the search. Specifically, if an individual solution is close to the game solution, a reward term is added to adjust the shared fitness, promoting niche equilibrium distribution and individual optimization;

[0022] S42 back-projects the evolutionary optimization results to the modeling layer, updating or correcting key parameters in the source-load model, such as user preference coefficients and price response sensitivity, thus achieving closed-loop learning between modeling and optimization. First, it analyzes the evolutionary optimization result set and selects the optimal individual solution for each niche from S3. This includes electricity prices Output P i t Load response results Information such as this. Then, based on this data, multiple samples are constructed. Then, a regression model is used to learn the functional relationship between electricity price and load changes. The new parameters are then written back into the objective function of the load-side model in S2, and the long-term accuracy of the model is improved by updating the modeling parameters.

[0023] S43 integrates the optimization results of each layer into a globally consistent, constraint-satisfied, and implementable scheduling strategy, which is then deployed as the actual operation plan.

[0024] The beneficial effects of this invention are as follows: This invention innovatively addresses the problems of heterogeneous optimization objectives, coupled decision-making behaviors, and poor scheduling coordination among multiple entities (source and load). It effectively solves problems such as the lack of game theory mechanisms and the tendency of optimization algorithms to get trapped in local optima, achieving significant results. Specifically, to address the difficulty of traditional optimization methods in characterizing the interaction between sources and loads, this invention first proposes a hierarchical modeling method based on a master-slave game theory mechanism. This method clearly delineates the decision-making levels and behavioral logic of power sources and loads, constructing a game theory framework with sources as masters and loads as slaves, achieving hierarchical collaborative scheduling. Secondly, to address the high optimization difficulty of nonlinear, multi-objective scheduling problems, a niche evolutionary optimization mechanism is introduced. This mechanism maintains multiple stable families of locally optimal solutions within the search space, avoiding a single solution from getting trapped in local extrema. This better addresses non-convex, multi-peak optimization problems, improving the ability to maintain the diversity of the solution space and the global search capability. Furthermore, this invention constructs an inter-layer feedback coordination mechanism to achieve mutual guidance and closed-loop optimization between game theory solutions and evolutionary solutions. Finally, it outputs a scheduling strategy that balances global coordination and local responsiveness, significantly improving the system's intelligence, robustness, and collaborative operating efficiency. From a technical perspective, this invention organically integrates hierarchical game modeling, niche evolution optimization, and closed-loop feedback mechanisms, achieving autonomous optimization and global coordination among multiple agents without relying on perfect global information. It achieves a good balance between modeling accuracy, optimization efficiency, solution stability, and operational adaptability, significantly improving the intelligent collaborative scheduling capability of complex energy systems. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention;

[0026] Figure 2 This is a schematic diagram of the collaborative optimization hierarchical structure designed in this invention;

[0027] Figure 3 This is a schematic diagram of the source-load master-slave game interaction designed in this invention;

[0028] Figure 4 This is a schematic diagram of the niche evolution optimization mechanism designed in this invention. Detailed Implementation

[0029] The specific embodiments of the present invention will now be described in conjunction with the accompanying drawings to enable those skilled in the art to better understand the present invention.

[0030] Example: Figure 1 As shown, a multi-agent hierarchical collaborative optimization method for source and load based on fusion niches includes the following steps:

[0031] S1 constructs a mathematical modeling system for multi-source, multi-load optimization objects, clarifying the decision variables, optimization objectives, and constraints of various power sources (such as photovoltaic, wind power, and gas turbine units) and loads (such as industrial loads and residential loads), serving as the foundational model for subsequent hierarchical game theory and evolutionary optimization. The specific implementation steps are as follows:

[0032] S11 models the power supply entities, clarifying their participating variables, optimization objectives, and operational constraints to support the "leader" role setting and evolutionary optimization in the master-slave game in S2. Power supply entities include, but are not limited to, photovoltaic power generation units, wind turbine generators, gas turbine generators, and energy storage devices. Each type of power supply entity is denoted as set G:

[0033] G = {G1, G2, ..., G} N}

[0034] Where N represents the number of power supply units, G i Let represent the i-th power source. The decision variables for each power source include: the output power P of the i-th power source at time t. i t The operating cost of the i-th power supply The carbon emission factor e of the i-th power source i The electricity price set by the power source at time t

[0035] The objective function for each power source is to maximize its net benefit:

[0036]

[0037] Where T represents the total duration of the scheduling cycle. C i (P i t The function denoted by ) represents the output cost function of the i-th power source at time t. It is typically a convex function, such as a quadratic cost function.

[0038] C i (P i t ) = a i (P i t ) 2 +b i P i t +c i

[0039] Among them, a i b i c i These represent cost coefficients.

[0040] The output of each power source is limited by operational constraints:

[0041] P i min ≤P i t ≤P i max

[0042] |P i t -P i t-1 |≤R i

[0043] Among them, P i min P i max These represent the minimum and maximum allowable output power of the power source, respectively. P i t P represents the output power of the i-th power source at time t. i t-1 R represents the output power of the i-th power source at time t-1. i This indicates the power output ramp-up rate limit.

[0044] S12 models the load-side entities, capturing their sensitivity and adaptability to electricity prices or dispatch signals, supporting their participation as "followers" in decision-making and optimization within the game. Load-side entities include industrial loads, commercial loads, residential electricity loads, and controllable loads (such as electric vehicles, heating and cooling loads). All load entities are denoted as set L:

[0045] L = {L1, L2, ..., L} N}

[0046] Where M represents the number of load cells, L i Let represent the i-th load subject. The decision variables for each load subject include: the power consumption of the j-th load at time t. Adjustable amount relative to the reference load Adjusting state variables A value of 1 indicates that the response is enabled, and a value of 0 indicates that the response is disabled; comfort deviation (such as room temperature, brightness, etc.). Received electricity price signal (from the power supply side) λ t .

[0047] The objective function for load, taking into account cost, comfort, and adjustment range, can be expressed as:

[0048]

[0049] Where T represents the total duration of the scheduling cycle. α j β jThis represents the comfort penalty weight and the response penalty factor. This represents the squared penalty for deviations from the user-defined value, reflecting the loss of comfort. This represents the frequency penalty for the excitation load response, used to suppress frequent start-stop cycles.

[0050] The load's response capability is limited by constraints:

[0051]

[0052] in, These represent the upper and lower limits of load regulation, respectively. δ j This indicates the maximum adjustment range of the load.

[0053] The S13 specification defines the decision variable structure, shared information variables, and objective function interface for multiple agents, enabling efficient integration of subsequent game theory and optimization algorithms.

[0054] First, a unified set of interaction strategies between source and load is defined, mainly including a power supply decision set. and load decision set

[0055] Next, we define shared information variables, including the electricity price signal (transmitted from source to load) λ. t And load response feedback (i.e., the adjustable amount relative to the reference load, transmitted from the load to the source).

[0056] Finally, the information structure between power sources and loads should be clearly defined: the power source, as the "leader," controls its own output and revenue model and can issue electricity prices; the load, as the "follower," makes its own decisions after receiving the electricity price and does not share comfort parameters.

[0057] S2 builds upon the multi-agent model constructed in S1, introducing a Stackelberg game mechanism to construct a hierarchical game structure with sources as the master and loads as the slaves. This simulates the dynamic interaction between power source preemptive strategies (such as electricity prices or output) and load response behavior, providing a hierarchical collaborative framework for evolutionary optimization. The specific implementation steps are as follows:

[0058] S21 constructs a master-slave game (Stackelberg) structure. It simulates a realistic scheduling mechanism with a master-slave decision-making order between power sources and loads. This allows power sources to specify electricity prices or output plans in advance, and loads to optimize their own response strategies after obtaining the power source's strategy, thus constructing a decision-making model with a clearly defined logical hierarchy. In the game structure, the power source set G = {G1, G2, ..., G...} N} is the leader, and the load set L = {L1, L2, ..., L} N} is a follower. Using time t as the scheduling period unit, the following decision sequence is established:

[0059] Power Supply G i The variables for deciding its electricity pricing or power output strategy are:

[0060] • Load L j Receive electricity price signal λ t Then, response optimization is performed, with the variable being...

[0061] The load side optimizes its objective function U based on the power supply distribution strategy. f Constructing the response function, also known as the optimal response mapping:

[0062]

[0063] Then, based on the load response results, the power supply further optimizes its objective function U. s Solve for the game equilibrium:

[0064]

[0065] Among them, U f The objective function representing the load typically considers cost, comfort, and adjustment costs. s The objective function of a power source is usually expressed as profit or revenue. Given a power supply strategy X s The optimal load strategy at that time. s * This represents the optimal power supply strategy obtained through game equilibrium solution.

[0066] S22 constructs the objective function on the power source side. It clarifies the form of the power source's objective function and its driving effect on the game outcome, enabling it to influence load behavior and optimize its own interests through electricity price adjustments. Specifically, the power source adjusts electricity prices... Guided load response This creates a positive feedback loop between price and electricity consumption behavior. Therefore, based on the power supply model defined in S1, the power supply entity G... i Objective function U s It can be represented as:

[0067]

[0068] in, Let represent the electricity price of the i-th power source at time t. This represents the total electrical energy consumed by all loads in response to the power supply, from X. f In-process aggregation. C i (P i t Let represent the output cost function of the i-th power source. The optimization constraints of this objective function are:

[0069] P i min ≤P i t ≤P i max ,|P i t -P i t-1 |≤R i

[0070] Among them, P i min P i max These represent the minimum and maximum allowable output power of the power source, respectively. P i t P represents the output power of the i-th power source at time t. i t-1 R represents the output power of the i-th power source at time t-1. i This indicates the power output ramp-up rate limit.

[0071] S23 constructs an optimal response mechanism on the load side, enabling load entities to perform personalized adaptive optimization responses based on the electricity pricing strategy of the power source's location, combined with their own preferences (comfort, cost, adjustment capability, etc.). Each load entity L j The objective function is expressed as follows:

[0072]

[0073] Where T represents the total duration of the scheduling cycle. λ t This indicates the electricity price signal emitted by the power supply side. This represents the electricity consumption of the load at time t. α j β j This represents the comfort penalty weight and the response penalty factor. This represents the squared penalty for deviations from the user-defined value, reflecting the loss of comfort. The frequency penalty for the excitation load response is used to suppress frequent start-stop cycles. The optimization constraints of this objective function are:

[0074]

[0075] in, These represent the upper and lower limits of load regulation, respectively. δ j This indicates the maximum adjustment range of the load.

[0076] S24 solves for the master-slave game equilibrium between the power source and the load, ensuring that the power source and the load achieve their respective optimal and mutually adaptive strategy combinations in the game strategy space. First, the power source strategy variable X is fixed.s For each load body L j Solve for its optimal response:

[0077]

[0078] Next, after summarizing all load responses, they are substituted into the power supply objective function to form a constrained optimization model for the power supply:

[0079]

[0080] Building upon the game structure constructed in S2, S3 introduces a niche evolutionary optimization mechanism for each type of source load subject. Utilizing fitness sharing and a local ecosystem evolution algorithm, it maintains the diversity of the multi-solution space and global search capability, obtaining a set of locally optimal response strategies under multi-objective conflict. The specific implementation steps are as follows:

[0081] In the S31 source-load system, due to the inconsistent interests of various stakeholders, the complexity of the strategy space, and the existence of multiple local optima in the objective function, traditional optimization methods may fail to find the global optimum. Therefore, a niche strategy is introduced to construct a diverse basic population for the solution space, and multiple small ecological subpopulations are divided through feature clustering or spatial partitioning, allowing individuals in different regions to evolve in parallel.

[0082] For each entity in the power supply and load, its decision variables, such as electricity price, are... Output P i t Load response results Load response status Each of these is encoded as an individual solution x in vector form. i :

[0083]

[0084] Where, x i This represents the solution for the i-th individual, which includes the scheduling strategies for all source loads.

[0085] Based on the above individual solutions, a random number of size N is generated. pop The initial population P0:

[0086]

[0087] The population is divided into K subhabitats using neighborhood partitioning based on Euclidean distance:

[0088] N k ={x i ∈P0|dist(x i ,μ k )≤ε},k=1,…,K

[0089] Where, N k This represents the k-th niche. μ k Let represent the center of the k-th niche, and ε represent the neighborhood radius. dist(·) represents the Euclidean distance.

[0090] To address the fitness inflation problem caused by dense clustering of individuals, S32 utilizes a fitness sharing mechanism to weaken the evaluation of locally highly concentrated individuals, enabling a more balanced distribution of niches. First, the initial fitness f(x) of each individual is calculated from the objective function value (such as total system cost, carbon emissions, load balancing deviation, etc.). i The calculation formula is as follows:

[0091] f(x i )=ω1·C total (x i )+ω2·E total (x i )+ω3·Δ balance (x i )

[0092] Among them, C total (x i ) represents the system operating cost under the current strategy. E total (x i The ω represents total carbon emissions. ω1, ω2, and ω3 represent multi-objective weighting coefficients. Δ balance (x i ) indicates the source-load power imbalance deviation.

[0093] Next, define the shared function sh(d) ij ):

[0094]

[0095] Where, d ij Represents individual x i With x j The policy distance between them. σ share Indicates the shared radius.

[0096] Ultimately, individual x i Shared fitness f i shared for:

[0097]

[0098] S33 independently conducts evolutionary search within each niche, achieving efficient optimization of the multi-objective function within local optima. For each niche, the following evolutionary operators are used for individual updates:

[0099] (1) Selection

[0100] Using selection methods such as roulette wheel betting and tournaments, based on shared fitness f i shared Select parent individuals.

[0101] (2) Cross

[0102] The decision variables of two parent individuals are combined using either uniform crossover or single-point crossover. For example, using single-point crossover yields the combined individual x. offspring :

[0103]

[0104] Where d represents the dimension and c represents the intersection point.

[0105] (3) Variation

[0106] By introducing small-amplitude perturbations or Gaussian noise, we obtain the mutated individual x. mutated :

[0107] x mutated =x+δ,δ~N(0,σ) 2 )

[0108] Where δ represents the following: N(0,σ) 2 Gaussian noise with a distribution of )

[0109] (4) Boundary Repair Mechanism

[0110] All new individual solutions obtained through the evolutionary operator must satisfy the physical constraints of the decision variables (such as non-negative electricity prices, load within allowable limits, etc.):

[0111] x i ←min(max(x i ,x min ),x max )

[0112] To avoid the local convergence of evolution in S33, S34 utilizes the preservation of elite individuals and migration between microhabitats to retain superior individuals and prevent degeneration. At the same time, it strengthens information exchange between microhabitats and enhances the overall synergy and global exploration capabilities of the population.

[0113] First, elite preservation refers to the retention of the most fit individuals from each niche in each generation. Ensure that high-quality solutions are not eliminated.

[0114] Next, we introduce the migration probability P. migrate Inter-microhabitat migration is controlled, allowing individuals to move between different microhabitats. Specifically, for migrating individuals, migration occurs from their original microhabitat N. kRandomly select another niche N k' If the individual can increase N k' If the individual has the worst fitness, it is accepted for migration; otherwise, it is regressed or replaced.

[0115] Finally, local updates and global convergence determination are performed. Every few generations, the globally optimal rate of change δ is determined. f If the value is less than α (α is a pre-set threshold), then the global re-initialization mechanism is triggered.

[0116] Building upon the diverse strategy space established in S3, S4 introduces an inter-layer feedback mechanism, mutually guiding the game outcome and niche evolutionary solutions to form a dynamic closed-loop collaborative optimization process. It then integrates the global objective to achieve the final scheduling scheme output, improving the overall robustness and coordination of the system while ensuring local self-optimization. The specific implementation steps are as follows:

[0117] S41 injects the game equilibrium result obtained in step S2 as guiding knowledge into the niche evolutionary population in S3 to improve the quality of the optimization starting point and guide the search direction. First, the optimal power supply strategy X obtained in S2... s * and optimal load response Form a decision vector X eq :

[0118]

[0119] Next, during the initialization of the S3 population, X... eq As elite individuals are injected into various microhabitats, the resulting updated microhabitat N is obtained. k ':

[0120]

[0121] To enhance information exchange between the game theory layer and the optimization layer and prevent policy conflicts, the local optimum of the game is viewed as a guide for individual optimization, thereby improving the initial quality of the search. Specifically, if an individual solution is close to the game solution, a reward term is added to adjust the shared fitness, promoting niche equilibrium distribution and individual optimization.

[0122] f i adjusted =f i shared -γ·dist(x i ,X eq )

[0123] in, It is the shared fitness calculated in S3. This is the adjusted shared fitness. γ represents the reward factor. dist(·) calculates the Euclidean distance between decision vectors.

[0124] S42 back-projects the evolutionary optimization results to the modeling layer, updating or correcting key parameters in the source-load model, such as user preference coefficients and price response sensitivity, thus achieving closed-loop learning between modeling and optimization. First, it analyzes the evolutionary optimization result set and selects the optimal individual solution for each niche from S3. This includes electricity prices Output P i t Load response results Information such as this. Then, based on this data, multiple samples are constructed. Then, a regression model is used to learn the functional relationship between electricity price and load change:

[0125]

[0126] Where, λ t Indicates electricity price f j (·) represents the response function of the j-th load subject to price. j This represents the price response coefficient. j This is the baseline load term. ε j This represents the residual term. It is the load response result, representing the optimal response of the load subject under this electricity price.

[0127] Then the new parameter a j b j The objective function of the load-side model in S2 is written back to, and the long-term accuracy of the model is improved by updating the modeling parameters. The updated load-side objective function is as follows:

[0128]

[0129] S43 integrates the optimization results from each layer into a consistent, constraint-satisfied, and implementable global scheduling strategy, which is then deployed as the actual operational plan. First, the candidate decision set X is formed by combining the game equilibrium solution and the niche optimal solution. all :

[0130]

[0131] Next, the comprehensive scoring function is calculated for each candidate solution:

[0132] S(x)=ω1·C(x)+ω2·E(x)+ω3·Q(x)

[0133] Where C(x) represents the normalized system cost. E(x) represents the normalized carbon emissions. Q(x) represents the load response satisfaction score. ω1, ω2, and ω3 are weighting factors, and satisfy ω1 + ω2 + ω3 = 1.

[0134] The solution with the highest overall score is selected as the final scheduling scheme based on the evaluation. It also outputs scheduling instructions at various times, mainly including power output P. i t Electricity price Optimal load response Load response status Information such as...

[0135] Further explanation is needed:

[0136] The implementation method of this invention involves executing the various steps described herein by controlling hardware through a computer program. Specifically, all or part of the processes in the above methods can be implemented by controlling related hardware through computer program instructions. This computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can complete the corresponding operations according to the processes of the above method embodiments. Any references to memory, storage, database, or other media involved in the embodiments provided in this application may include non-volatile memory and / or volatile memory.

[0137] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the above-described method of dividing functional units or modules is used as an example. However, in practical applications, the above-described functional units or modules can be divided into different functional units or modules as needed to accomplish all or part of the functions in the method of this invention.

[0138] The embodiments described above are only for illustrating the technical solutions of the present invention and are not intended to limit them. Although we have described the specific embodiments of the present invention in detail, those skilled in the art should understand that they can modify the technical solutions in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not depart from the core spirit and scope embodied in the embodiments of the present invention and should be included within the protection scope of the present invention.

Claims

1. A source-load multi-agent hierarchical collaborative optimization method based on fusion niches, characterized in that, Promoting collaborative optimization of multiple sources and multiple loads through master-slave game theory and niche evolution optimization includes the following steps: S1 constructs a mathematical modeling system for multi-source and multi-load optimization objects, clarifies the decision variables, optimization objectives and constraints of various power sources and loads, and serves as the basic model support for subsequent hierarchical game and evolutionary optimization; S2 is based on the multi-agent model built in S1. It introduces the master-slave game (Stackelberg) mechanism to construct a hierarchical game structure with the source as the master and the load as the slave. It simulates the dynamic interaction between the power supply's first-mover strategy and the load's response behavior, and provides a hierarchical collaborative framework for evolutionary optimization. Based on the game structure constructed in S2, S3 introduces a niche evolution optimization mechanism for each type of source load subject. By utilizing fitness sharing and local ecosystem evolution algorithms, it maintains the diversity of the multi-solution space and global search capability, and obtains a set of locally optimal response strategies under multi-objective conflict. Based on the diversified strategy space formed by S3, S4 introduces an inter-layer feedback mechanism, which uses the game results and niche evolution solutions as mutual guides to form a dynamic closed-loop collaborative optimization process. It also integrates the global goal to achieve the final scheduling scheme output, thereby improving the overall robustness and coordination of the system while ensuring local self-optimization.

2. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, characterized in that, Step S1 includes the following steps: S11 models the main body on the power supply side, clarifying its participating variables, optimization objectives, and operational constraints to support the setting and evolution optimization of the "leader" role in the master-slave game in S2; S12 models the load-side entities, capturing their sensitivity and adjustment capabilities to electricity prices or dispatch signals, and supports their participation in decision-making, response, and optimization as "followers" in the game. The S13 specification defines the decision variable structure, shared information variables, and objective function interface for multiple agents, enabling efficient integration of subsequent game theory and optimization algorithms.

3. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, wherein step S2 includes the following steps: S21 constructs a master-slave game (Stackelberg) structure to simulate a real scheduling mechanism with a master-slave decision-making order between power sources and loads. This allows power sources to specify electricity prices or output plans in advance, and loads to optimize their own response strategies after receiving the power source's strategy. A decision-making model with a clear logical hierarchy is constructed. In the game structure, the power source set G = {G1, G2, ..., G...} N } is the leader, and the load set L = {L1, L2, ..., L} N } is a follower; S22 constructs the objective function on the power source side, clarifying the form of the power source's objective function and its driving effect on the game outcome, enabling it to influence load behavior and optimize its own interests through electricity price adjustments. Specifically, the power source adjusts the electricity price... Guided load response This creates a positive feedback loop between price and electricity consumption behavior; S23 constructs an optimal response mechanism on the load side, enabling load entities to perform personalized adaptive optimization responses based on the electricity price strategy of the power source's location and their own preferences (comfort, cost, regulation capability, etc.). S24 solves for the master-slave game equilibrium between the power source and the load, ensuring that the power source and the load achieve their respective optimal and mutually adaptive strategy combinations in the game strategy space. First, the power source strategy variable X is fixed. s For each load body L j The optimal response is then determined. Subsequently, all load responses are summarized and substituted into the power supply objective function to form a constrained optimization model for the power supply.

4. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, characterized in that, Step S3 includes the following steps: In the S31 system, due to the inconsistent interests of various subjects, the complexity of the strategy space, and the existence of multiple local optima in the objective function, a niche strategy is introduced to construct a diverse basic population in the solution space. Multiple small ecological sub-populations are then divided through feature clustering or spatial partitioning, allowing individuals in different regions to evolve in parallel. To address the fitness inflation problem caused by dense clustering of individuals, S32 utilizes a fitness sharing mechanism to weaken the evaluation of locally highly concentrated individuals, enabling niches to be evenly distributed. S33 independently conducts evolutionary search within each niche, achieving efficient optimization of the multi-objective function within the local optimum region. For each niche, an evolutionary operator is used for individual updates. To avoid the local convergence of evolution in S33, S34 utilizes the preservation of elite individuals and migration between microhabitats to retain superior individuals and prevent degeneration. At the same time, it strengthens information exchange between microhabitats and enhances the overall synergy and global exploration capabilities of the population.

5. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, characterized in that, Step S4 includes the following steps: S41 uses the game equilibrium result obtained in step S2 as guiding knowledge to inject into the niche evolutionary population in S3, in order to improve the quality of the optimization starting point and guide the search direction. First, the power-optimal strategy obtained in S2 will be used. and optimal load response Form a decision vector X eq Then, during the initialization of the S3 population, X... eq As elite individuals, they are injected into various microenvironments N k To strengthen the information linkage between the game layer and the optimization layer and prevent strategy conflicts, the local optimal solution of the game is regarded as the navigation of individual optimization to improve the initial quality of the search. Specifically, if an individual solution is close to the game solution, a reward term is added to adjust the shared fitness, which promotes the equilibrium distribution of niches and individual optimization. S42 back-projects the evolutionary optimization results to the modeling layer, updating or correcting key parameters in the source-load model, such as user preference coefficients and price response sensitivity, thus achieving closed-loop learning between modeling and optimization. First, it analyzes the evolutionary optimization result set and selects the optimal individual solution for each niche from S3. This includes electricity prices contribute Load response results Information, and then, based on this data, construct multiple samples. Then, the regression model is used to learn the functional relationship between electricity price and load change, and the new parameters are written back into the objective function of the load-side model in S2. By updating the modeling parameters, the long-term accuracy of the model is improved. S43 integrates the optimization results of each layer into a globally consistent, constraint-satisfied, and implementable scheduling strategy, which is then deployed as the actual operation plan.

6. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, characterized in that, Step S1 includes the following steps: S11 models the power supply entities, clarifying their participating variables, optimization objectives, and operational constraints to support the "leader" role setting and evolutionary optimization in the master-slave game in S2. The power supply entities include photovoltaic power generation units, wind turbine generators, gas turbine generators, and energy storage devices. Each type of power supply entity is denoted as set G. G={G1,G2,…,G N } Where N represents the number of power supply units, G i Let i represent the i-th power source. The decision variables for each power source include: the output power of the i-th power source at time t. Operating cost of the i-th power source The carbon emission factor e of the i-th power source i The electricity price set by the power source at time t The objective function for each power source is to maximize its net benefit: Where T represents the total duration of the scheduling cycle, The output cost function of the i-th power source at time t is typically a convex function, a quadratic cost function: Among them, a i b i c i These represent cost coefficients, The output of each power source is limited by operational constraints: in, These represent the minimum and maximum allowable output power of the power supply, respectively. This represents the output power of the i-th power source at time t. R represents the output power of the i-th power source at time t-1. i This indicates the power output ramp-up rate limit. S12 models the load-side entities, capturing their sensitivity and adjustment capabilities to electricity prices or dispatch signals, supporting their participation in decision-making and optimization as "followers" in the game. The load-side entities include industrial loads, commercial loads, residential electricity loads, and controllable loads, and all load entities are denoted as set L: L={L1,L2,…,L N } Where M represents the number of load cells, L i Let i represent the i-th load subject. The decision variables for each load subject include: the power consumption of the j-th load at time t. Adjustable amount relative to the reference load Adjusting state variables A value of 1 indicates that the response is enabled, and a value of 0 indicates that the response is disabled; comfort deviation amount Received electricity price signal λ t , The objective function for load, taking into account cost, comfort, and adjustment range, is expressed as: Where T represents the total duration of the scheduling cycle, α j β j This represents the comfort penalty weight and the response penalty factor. This represents the squared penalty for deviations from the user-defined value, reflecting the loss of comfort. This represents a frequency penalty for the excitation load response, used to suppress frequent start-stop cycles. The load's response capability is limited by constraints: in, These represent the upper and lower limits of load regulation, δ j Indicates the maximum adjustment range of the load; The S13 specification defines the decision variable structure, shared information variables, and objective function interface for multiple agents, enabling efficient integration of subsequent game theory and optimization algorithms. First, a unified set of interaction strategies between source and load is defined, mainly including a power supply decision set. and load decision set Next, we define shared information variables, including the electricity price signal (transmitted from source to load) λ. t And load response feedback (i.e., the adjustable amount relative to the reference load, transmitted from the load to the source). Finally, the information structure between power sources and loads should be clearly defined: the power source, as the "leader," controls its own output and revenue model and can issue electricity prices; the load, as the "follower," makes its own decisions after receiving the electricity price and does not share comfort parameters.

7. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, characterized in that, Step S2 includes the following steps: S21 constructs a master-slave game (Stackelberg) structure to simulate a real scheduling mechanism with a master-slave decision-making order between power sources and loads. This allows power sources to specify electricity prices or output plans in advance, and loads to optimize their own response strategies after receiving the power source's strategy. A decision-making model with a clear logical hierarchy is constructed. In the game structure, the power source set G = {G1, G2, ..., G...} N } is the leader, and the load set L = {L1, L2, ..., L} N As a follower, with time t as the scheduling period unit, establish the following decision sequence: Power Supply G i The variables for deciding its electricity pricing or power output strategy are: • Load L j Receive electricity price signal λ t Then, response optimization is performed, with the variables being... The load side optimizes its objective function U based on the power supply distribution strategy. f Constructing the response function, i.e., the optimal response mapping: Then, based on the load response results, the power supply further optimizes its objective function U. s Solve for the game equilibrium: Among them, U f The objective function representing the load typically considers cost, comfort, and adjustment costs. U s The objective function of a power source is typically expressed as profit or revenue. Given a power supply strategy X s The optimal load strategy at that time This represents the optimal power supply strategy obtained through game equilibrium solution. S22 constructs the objective function on the power source side, clarifying the form of the power source's objective function and its driving effect on the game outcome, enabling it to influence load behavior and optimize its own interests through electricity price adjustments. Specifically, the power source adjusts the electricity price... Guided load response This creates a positive feedback loop between price and electricity consumption behavior, based on the power supply model defined in S1, with the power supply entity G... i Objective function U s Represented as: in, This represents the electricity price of the i-th power source at time t. This represents the total electrical energy consumed by all loads in response to the power supply, from X. f In-process aggregation, Let represent the output cost function of the i-th power source. The optimization constraints of this objective function are: in, These represent the minimum and maximum allowable output power of the power supply, respectively. This represents the output power of the i-th power source at time t. R represents the output power of the i-th power source at time t-1. i This indicates the power output ramp-up rate limit. S23 constructs an optimal response mechanism on the load side, enabling load entities to perform personalized adaptive optimization responses based on the electricity price policy of the power source's issuing location and their own preferences. Each load entity L j The objective function is expressed as follows: Where T represents the total duration of the scheduling cycle, λ t This indicates the electricity price signal emitted by the power supply side. α represents the electricity consumption of the load at time t. j β j This represents the comfort penalty weight and the response penalty factor. This represents the squared penalty for deviations from the user-defined value, reflecting the loss of comfort. The frequency penalty representing the excitation load response is used to suppress frequent starts and stops. The optimization constraints of this objective function are: in, These represent the upper and lower limits of load regulation, δ j Indicates the maximum adjustment range of the load; S24 solves for the master-slave game equilibrium between the power source and the load, ensuring that the power source and the load achieve their respective optimal and mutually adaptive strategy combinations in the game strategy space. First, the power source strategy variable X is fixed. s For each load body L j Solve for its optimal response: Next, after summarizing all load responses, they are substituted into the power supply objective function to form a constrained optimization model for the power supply:

8. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, characterized in that, Step S3 includes the following steps: In the S31 source-load system, due to the inconsistent interests of various stakeholders, the complexity of the strategy space, and the existence of multiple local optima in the objective function, traditional optimization methods may fail to find the global optimum. Therefore, a niche strategy is introduced to construct a diverse basic population for the solution space. Multiple small ecological subpopulations are then divided through feature clustering or spatial partitioning, allowing individuals in different regions to evolve in parallel. For each entity in the power supply and load, its decision variables, such as electricity price, are... contribute Load response results Load response status Each of these is encoded as an individual solution x in vector form. i : Where, x i Let represent the solution for the i-th individual, which includes the scheduling strategies for all source loads. Based on the above individual solutions, a random number of size N is generated. pop The initial population P0: The population is divided into K subhabitats using neighborhood partitioning based on Euclidean distance: N k ={x i ∈P0|dist(x i ,μ k )≤ε},k=1,…,K Where, N k Let μ represent the k-th niche. k Let ε denote the center of the k-th niche, ε denote the neighborhood radius, and dist(·) denote the Euclidean distance. To address the fitness inflation problem caused by dense clustering of individuals, S32 utilizes a fitness sharing mechanism to weaken the evaluations of locally highly concentrated individuals, enabling a more balanced distribution of niches. First, the initial fitness f(x) of each individual is calculated from the objective function value. i The calculation formula is as follows: f(x i )=ω1·C total (x i )+ω2·E total (x i )+ω3·D balance (x i ) Among them, C total (x i E represents the system operating cost under the current strategy. total (x i ) represents total carbon emissions, ω1, ω2, and ω3 represent multi-objective weighting coefficients, and Δ balance (x i ) indicates the source-load power imbalance deviation. Next, define the shared function sh(d) ij ): Where, d ij Represents individual x i With x j The policy distance between them, σ share Indicates the shared radius. Ultimately, individual x i Shared fitness f i shared for: S33 independently conducts evolutionary search within each niche to achieve efficient optimization of the multi-objective function within local optima. For each niche, the following evolutionary operators are used for individual updates: (1) Selection Using selection methods such as roulette wheel betting and tournaments, based on shared fitness f i shared Select parent individuals, (2) Cross By using either uniform crossover or single-point crossover, the decision variables of two parent individuals can be combined. Using single-point crossover yields the combined individual x. offspring : Where d represents the dimension and c represents the intersection point. (3) Variation By introducing small-amplitude perturbations or Gaussian noise, we obtain the mutated individual x. mutated : x mutated =x+δ,δ~N(0,σ 2 ) Where δ represents the following: N(0,σ) 2 Gaussian noise with a distribution of ) (4) Boundary Repair Mechanism All new individual solutions obtained through the evolutionary operator must satisfy the physical constraints of the decision variables: x i ←min(max(x i ,x min ),x max ) To avoid the local convergence of evolution seen in S33, S34 utilizes the preservation of elite individuals and migration between microhabitats to retain superior individuals and prevent degeneration. Simultaneously, it enhances information exchange between microhabitats, improving overall population synergy and global exploration capabilities. First, elite preservation refers to the retention of the most fit individuals from each niche in each generation. To ensure that high-quality solutions are not eliminated. Next, we introduce the migration probability P. migrate To facilitate inter-niche migration, the movement of individuals between different niches is controlled. For migrating individuals, from their original niche N... k Randomly select another niche N k' If the individual can increase N k' If the individual has the worst fitness, then the migration is accepted; otherwise, the individual is regressed or replaced. Finally, local updates and global convergence determination are performed, and the globally optimal rate of change δ is determined every few generations. f If the value is less than α (α is a pre-set threshold), then the global re-initialization mechanism is triggered.

9. The source-load multi-agent hierarchical collaborative optimization method based on fusion niches according to claim 1, characterized in that, Step S4 includes the following steps: S41 uses the game equilibrium result obtained in step S2 as guiding knowledge to inject into the niche evolutionary population in S3, in order to improve the quality of the optimization starting point and guide the search direction. First, the power-optimal strategy obtained in S2 will be used. and optimal load response Form a decision vector X eq : Next, during the initialization of the S3 population, X... eq As elite individuals are injected into various microhabitats, the resulting updated microhabitat N is obtained. k ': If an individual's solution is close to the game's solution, an additional reward term is added to adjust the shared fitness, promoting niche equilibrium and individual optimization. f i adjusted =f i shared -γ·dist(x i ,X eq ) Among them, f k shared This is the shared fitness calculated in S3, f k adjusted It is the adjusted shared fitness, γ represents the reward factor, and dist(·) represents the calculation of the Euclidean distance between decision vectors; S42 back-projects the evolutionary optimization results to the modeling layer, updating or correcting key parameters in the source-load model, such as user preference coefficients and price response sensitivity, thus achieving closed-loop learning between modeling and optimization. First, it analyzes the evolutionary optimization result set and selects the optimal individual solution for each niche from S3. This includes electricity prices contribute Load response results Information, and then, based on this data, construct multiple samples. Then, a regression model is used to learn the functional relationship between electricity price and load change: Where, λ t Indicates electricity price f j (·) represents the response function of the j-th load subject to price, a j b represents the price response coefficient. j It is the baseline load term, ε j Represents the residual term. This is the load response result, representing the optimal response of the load subject under this electricity price. Then the new parameter a j b j The objective function of the load-side model in S2 is written back to, and the long-term accuracy of the model is improved by updating the modeling parameters. The updated objective function of the load-side model is as follows: S43 integrates the optimization results from each layer into a consistent, constraint-satisfied, and implementable global scheduling strategy, which is then deployed as the actual operational plan. First, it combines the game equilibrium solution and the niche optimal solution to form a candidate decision set X. all : Next, the comprehensive scoring function is calculated for each candidate solution: S(x)=ω1·C(x)+ω2·E(x)+ω3·Q(x) Where C(x) represents the normalized system cost, E(x) represents the normalized carbon emissions, Q(x) represents the load response satisfaction score, and ω1, ω2, and ω3 are weighting factors that satisfy ω1 + ω2 + ω3 = 1. The solution with the highest overall score is selected as the final scheduling scheme based on the evaluation. It also outputs scheduling instructions at various times, mainly including power output. Electricity price Optimal load response Load response status information.

Images