A multi-modal hierarchical multi-objective distributed optimization acceleration method for integrated energy systems

Abstract

The application provides a multi-modal hierarchical multi-objective distributed optimization acceleration method for a comprehensive energy system, which combines a multi-modal problem optimization method, a quantum and fuzzy logic full-connection layer embedded self-attention network acceleration method and a multi-layer multi-objective distributed consistency method; the multi-layer multi-objective distributed consistency method is applied as an iteration method and is suitable for optimization of a large-scale distributed multi-agent system; a small niche strategy and a binary tournament selection mechanism are introduced to endow a Pareto solution set with a multi-modal characteristic; and a quantum and fuzzy logic full-connection layer embedded self-attention network is added in the iteration process of the method to improve the speed of the method. The multi-modal hierarchical multi-objective distributed optimization acceleration method for the comprehensive energy system can solve a multi-objective optimization problem of the comprehensive energy system, realizes the function of quickly obtaining a multi-objective multi-modal Pareto solution set of the comprehensive energy system, reduces the required optimization time and improves the operation speed of the optimization method.

Description

Technical Field

[0001] This invention belongs to the field of multi-objective optimization of integrated energy systems, and involves analytical methods and artificial intelligence methods, which are applicable to the optimization control of integrated energy systems. Background Technology

[0002] Existing multi-objective hierarchical distributed optimization methods cannot provide more comprehensive solutions when solving problems with multiple objectives. For example, if one of the optimal solutions becomes unusable due to environmental changes or other factors, decision-makers cannot easily and quickly find another optimal solution that is not significantly different.

[0003] In addition, existing multi-objective, multi-modal, hierarchical, distributed optimization methods suffer from slow optimization speed.

[0004] Therefore, the introduction of niche strategies and binary tournament selection mechanisms endows the Pareto solution set with multimodal characteristics to address the problem that it cannot provide a more comprehensive solution strategy when applied to problems with multiple objectives; in the process of method iteration, a fully connected layer with quantum and fuzzy logic embedded self-focused network is added to accelerate the process and address the problem of slow optimization speed. Summary of the Invention

[0005] This invention proposes an accelerated method for multimodal hierarchical multi-objective distributed optimization of integrated energy systems. It combines multimodal optimization methods, a fully connected layer embedded self-concern network acceleration method based on quantum and fuzzy logic, and a multi-layer multi-objective distributed consensus method to solve multi-objective optimization problems in integrated energy systems. This method overcomes the shortcomings of slow computation and poor convergence of non-multi-layer distributed methods when applied to large-scale systems. Compared to centralized optimization methods, it offers higher information privacy. The acceleration process improves the overall convergence time and provides more diverse solution sets for multi-objective multimodal problems. The steps in its application are as follows:

[0006] Step (1): Set the current time period t = 0, and start the optimization solution for the t-th time period;

[0007] Step (2): Method initialization, determine the total number of system layers N layer The input is the Laplace matrix L, calculated from the topology diagram of the system being optimized. G The topology graph is an undirected, unweighted graph. The maximum number of iterations K for each layer is determined, and the number of iteration layers N is set. l =0, so let the number of iterations k in the current iteration process be 0;

[0008] Laplace matrix L G element l in ii With element l ij for:

[0009]

[0010] Among them, a ij Let A be the adjacency matrix. G The element in, a ij The possible values ​​are:

[0011]

[0012] Among them, v i and v j Two distinct points in the topology, v i ,v j ∈V G V G ={v1,v2,...,v M Let} be the set of nodes, and M be the number of nodes in the topology graph; v i v j An edge is formed by two distinct points in a topological structure. Let be the set of edges;

[0013] The objective function f(x) for the multimodal and multi-objective optimization problem of integrated energy systems is:

[0014]

[0015] Among them, f (e) The objective function representing the total cost objective, f (c) The objective function f represents the total carbon emission target. (ξ) The objective function represents the overall energy efficiency target of the entire system. The objective function f represents user satisfaction. (F) The objective function represents the flexibility of the system's regulation capability; x represents a multi-dimensional decision vector, including the active power output of the j-th power supply unit in time period t. The energy supply E of the power supply unit during time period t j,t The energy consumption W of the j-th power supply unit j The change in electricity, gas, and heat load in time period t after the user responds to demand. The flexibility indicators of the power supply units in adjusting the supply of electricity, heat, and gas upwards and downwards. min() is a function that selects the minimum value;

[0016] The objective function f of the total cost target (e) for:

[0017]

[0018] Where n is the number of power supply units. The cost of generating electricity for the j-th power unit during the t-th time period, t total a is the total number of segments that are evenly divided within the scheduling cycle. j b j and c j These are the cost coefficients for the secondary, primary, and constant terms of the j-th power supply unit, respectively.

[0019] The objective function f of the total carbon emission target (c) for:

[0020]

[0021] in, Let α be the carbon emissions of the j-th power supply unit during the t-th time period. j β j and γ j These are the carbon emission coefficients of the second-order, first-order, and constant terms of the j-th power supply unit, respectively.

[0022] The objective function f of the overall system energy efficiency target (ξ) for:

[0023]

[0024] Where ξ represents the overall energy efficiency over the complete system scheduling cycle. Let d be the load on the site corresponding to the j-th power supply unit after considering the comprehensive demand response in time period t. j η is the reference coefficient for standard coal equivalent of energy type j. s,j Let J be the energy efficiency of the j-th power supply unit. The energy efficiency of the j-th power supply unit is:

[0025]

[0026] Where E j Let J be the energy supply of the j-th power supply unit.

[0027] The objective function of user satisfaction for:

[0028]

[0029] User satisfaction is used express, The original electricity, gas, and heat loads; the larger the load response, the more user load adjustment behaviors there are, and the higher the user satisfaction adjustment value. The larger the value, the lower the user's satisfaction level with energy usage. max() is a function that selects the maximum value;

[0030] The objective function f of the system's adaptability flexibility(F) for:

[0031]

[0032] Where F load As an indicator for evaluating the flexibility of integrated energy systems; F E F H and F G These represent the comprehensive flexibility index of the electrical, thermal, and pneumatic links, respectively; ω1, ω2, and ω3 represent the weighting coefficients of the electrical, thermal, and pneumatic links, respectively.

[0033] The overall flexibility index F of the electrical link E for:

[0034]

[0035] in and These represent the flexibility indicators for adjusting the power supply capacity of the generating units upwards and downwards, respectively.

[0036] The overall flexibility index F of the thermal link H for:

[0037]

[0038] in and These represent the flexibility indicators for adjusting the thermal energy provided by the power supply unit upwards and downwards, respectively.

[0039] The overall flexibility index F of the gas process G for:

[0040]

[0041] in and These represent the flexibility indicators for adjusting the gas energy provided by the power supply unit upwards and downwards, respectively.

[0042] The power constraints of the generating unit include power balance constraints and upper and lower limit output constraints;

[0043] The power balance constraint is:

[0044] in Let t be the total system load demand for time period t; the upper and lower limits of the power supply unit output are constrained as follows: in The lower limit of the active power output of the j-th power supply unit in time period t. Let j be the upper limit of active power output of the j-th power supply unit in the t-th time period; Step (3): Divide the complete integrated energy system into regions, each region is regarded as an intelligent agent, and the initial intelligent agent of the current layer iteration is selected. Select a leader intelligent agent and the rest are follower intelligent agents. The leader intelligent agent is selected from the center of the topology; Step (4): Randomly generate the consistency variable value and iteration variable value of each intelligent agent;

[0045] Consistency variable elements, which are different from the elements of the same dimension corresponding to the decision variables, include the objective function of the cost target and the objective function of the carbon emission target of the j-th agent in time period t, and the active power output. The partial derivatives of the ... The partial derivatives are: Where is the total objective value of the j-th agent in time period t, obtained by linear weighting of the cost objective function and the carbon emission objective function. Let be the total objective value of the first agent in time period t, obtained by linear weighting of the cost objective function and the carbon emission objective function. Let be the total objective value of the nth agent in time period t, obtained by linearly weighting the cost objective function and the carbon emission objective function. The active power output of the first power supply unit in time period t. For the active power output of the nth power supply unit in time period t, For each element of the consistency variable that is not equal to the decision variable of the agent in time period t; the total objective value of the j-th agent in time period t after linear weighting of the cost objective function and the carbon emission objective function. for: Wherein, τ is the weighting factor, and its value range is 0≤τ≤1;

[0046] The initial decision variables derived from the consistency variables are: in, It is the decision variable element of agent j in the k-th iteration at time t that is not equal to the consistency variable element. Let be the consistency variable element of agent j in the k-th iteration at time t that is not equal to the decision variable element. For the χ-th layer The active power limit of the region is the first The sum of the active power limits of the lower-level sub-regions or power supply units of the region. Let be the total number of regions in layer χ. For decision variable elements that are the same as elements of the consistency variable, the decision variable elements that agent j has in time period t during the k-th iteration are equal to the consistency variable. for:

[0047]

[0048] in, Let be the consistency variable element that is equal to the decision variable in the k-th iteration at time t;

[0049] Step (5): Based on the topology diagram of the agents, update the consistency variables of the follower agents and the leader agents respectively, and update the decision variable elements. Then compare the decision variables with the corresponding objective space function values ​​to generate the Pareto optimal solution set and the Pareto front; based on the input Laplace matrix L G Find the k-th iteration element d in the row random matrix of the agent. ij (k):

[0050]

[0051] After iterative updates, the consistency variable of the follower agent is:

[0052]

[0053] After iterative updates, the consistency variable for the leader agent is:

[0054]

[0055] in, Let be the consistency variable of the i-th agent in the (k+1)-th iteration during the t-th time period; Let be the consensus variable of the i-th agent in time period t during the k-th iteration; ε is the power balance adjustment factor of the multi-level multi-objective distributed consensus method; ΔP t (k) is the power deviation in the k-th iteration during time period t;

[0056] Then update the decision variable elements. For decision variable elements that are different from the consistency variable elements, the decision variables derived from the consistency variables are:

[0057]

[0058] in It is the decision variable element of agent j in the (k+1)th iteration at time t that is not equal to the consistency variable element. Let be the consistency variable element that is not equal to the decision variable element in the (k+1)th time period of agent j;

[0059] For the decision variable element that is the same as the consistency variable element, the decision variable element of the agent at the t-th time period in the (k + 1)-th iteration that is equal to the consistency variable is:

[0060]

[0061] where is the consistency variable element of the agent at the t-th time period in the (k + 1)-th iteration that is equal to the decision variable;

[0062] Step (6): Determine whether the number of iterations in the current iteration satisfies k < n 10 or K - n 10 < k < K, where n 10 is the number of iterations before the acceleration stage and after the acceleration stage; if k < n 10 or K - n 10 < k < K is satisfied, then continue to use the multi-modal multi-layer multi-objective distributed consistency method for iteration; if k < n 10 or K - n 10 < k < K is not satisfied, then enter the fully connected layer embedded self-attention network of quantum and fuzzy logic to execute the acceleration process;

[0063] Step (7): If k < n 10 or K - n 10 < k < K is satisfied, then execute the iteration process, start the next iteration, and use the niche strategy and binary tournament selection mechanism to select the agents participating in the next iteration to solve the multi-modal problem; set both the niche size and the number of agents entering the next iteration to be N agent , and put the agents in the next iteration into the iteration set;

[0064] Step (8): Randomly select agents from the population to enter the candidate solution set until the niche is filled;

[0065] Step (9): Calculate the Euclidean distance D between the first agent in the candidate solution set as an individual a in the binary tournament and each agent other than this agent in the set, and select the agent with the smallest Euclidean distance from the agent as the other individual b in the binary tournament, judge the dominance relationship between the two individuals a and b in the binary tournament, and select one of the two individuals to enter the iteration set; the Euclidean distance D between the two individuals at the t-th time period t is:

[0066]

[0067] where and Let represent the i-th element of the decision variable for individual a and individual b in time period t, respectively, and m represent the dimension of the decision variable; and Let represent the first element of the decision variable for individual a and individual b in time period t, respectively; and Let m represent the m-th element of the decision variable for individual a and individual b in time period t, respectively.

[0068] The binary tournament selection mechanism is as follows: First, determine the dominance relationship between two individuals, a and b, in the binary tournament. If individual a dominates individual b, then individual a is added to the iteration set; otherwise, individual b is added to the iteration set. However, if the two individuals do not dominate each other, calculate the crowding distance of each individual in the decision space. And select individuals with larger crowding distances to add to the iteration set; if the crowding distance between two individuals is... They are equal, so individual a is placed into the iteration set; the solution selected by the binary tournament selection mechanism is the multimodal solution, ensuring the comprehensiveness of the solution in multiple cases, while the solutions not retained in the iteration set are those that the decision-maker will not use; crowding distance CD i,X for:

[0069]

[0070] in, This represents the crowding distance of the i-th agent in the decision space during time period t; 1≤i≤N agent ; Let m represent the value of the i-th agent in the j-th dimension during time period t, where 1 ≤ j ≤ m. This represents the value of the (i+1)th agent in the j-th dimension during time period t. This represents the value of the (i-1)th agent in the j-th dimension during time period t. This represents the value of the (i+1)th agent in time period t on the first dimension. This represents the value of the (i-1)th agent in time period t on the first dimension. This represents the value of the (i+1)th agent in the m-th dimension during time period t. This represents the value of the (i-1)th agent in the m-th dimension during time period t; This represents the maximum value in the j-th dimension during time period t. This represents the minimum value in the j-th dimension during time period t. This represents the maximum value in the first dimension during time period t. This represents the minimum value in the first dimension of the time interval t. This represents the maximum value in the m-th dimension during the t-th time period. This represents the minimum value in the m-th dimension during time period t;

[0071] Step (10): Repeat step (9) until the iteration set is filled;

[0072] Step (11): Combine the topological structure diagram of the agents to update the consistency variables of the follower agents and leader agents in the iteration set respectively, and update the decision variable elements. Then compare the decision variables with the corresponding values in the target space to generate the Pareto optimal solution set and the Pareto front;

[0073] Step (12): If it does not satisfy k < n 10 or K - n 10 < k < K, then enter the acceleration process of the fully connected layer embedded self-attention network accelerated by quantum and fuzzy logic, and input N agent m-dimensional decision variables and the corresponding objective function values;

[0074] Step (13): According to the weighted total objective function values of N agent agents, divide the agents into four fuzzy logic subsets through the fuzzy logic strategy; each agent is classified into category z p by the fuzzy rules based on the weighted total objective function value; the weighted total objective function value f T,p of the p-th agent is:

[0075]

[0076] where ρ1, ρ2, ρ3, ρ4, ρ5 are the weighted parameter values corresponding to the objective functions f (e) , f (c) , f (ξ) , f (F) respectively; the category z p of the p-th agent obtained by the fuzzy rule classification is:

[0077]

[0078] where f Tmax and f Tmin are the maximum and minimum values of the weighted total objective function value respectively, and the subsets A, B, C, D are four fuzzy logic subsets, and the subsets A to D have H agent，A , H agent，B , H agent，C and H agent，D agents respectively, 0 < p ≤ N agent ;

[0079] Step (14): To prevent the data input to the fully connected layer embedded self-focused network from being over-synthesized and causing overfitting of the fully connected layer embedded self-focused network, 16% of the individuals in each subset are selected to enter the two-dimensional quantum universal gate; each selected individual enters one of the four fuzzy subsets through a two-dimensional quantum universal gate, and each universal quantum gate Q 1bit for:

[0080]

[0081] Where θ, λ and These are Felix Bloch parameters, where θ is the polar angle of the Bloch sphere. λ represents the relative phase of the Bloch sphere, and λ represents the global phase of the Bloch sphere; the two-position quantum universal gate Q 2bit It is the tensor product of two unit quantum universal gates, which is:

[0082]

[0083] Q 1,1bit and Q 2,1bit For two unit quantum universal gates, θ1, θ2, λ1, λ2, and Q 1,1bit Bloch's polar angle, Q 2,1bit Bloch's polar angle, Q 1,1bit Bloch sphere global phase, Q 2,1bit Bloch sphere global phase, Q 1,1bit The relative phase of the Bloch sphere and Q 2,1bit The relative phase of the Bloch sphere;

[0084] Step (15): Using four fuzzy subsets as datasets, train four fully connected layer embedded self-focused networks using momentum stochastic gradient descent. Input the randomly selected weighted total objective function value to obtain the corresponding decision variable value and provide input for the next iteration; the input and output are the weighted total objective function value and the decision variable, respectively.

[0085] Step (16): Restrict the predicted decision variables to the boundary. u represents the u-th element of the decision variable. and Let represent the lower and upper bounds of the u-th element of the decision variable in time period t, respectively; output the boundary-constrained decision variable values ​​as the initial values ​​of the decision variables in the next iteration of the multi-level multi-objective distributed consensus method.

[0086] Step (17): After the acceleration process is completed, continue the iterative process of the multi-layer multi-objective distributed consensus method. When k = K, end the iteration of this layer and let N l =Nl +1;

[0087] Step (18): Determine whether N is satisfied. l =N layer If not satisfied, proceed to the next iteration and output the Pareto optimal solution selected according to the decision-maker's preference. Each element of the output Pareto optimal solution is used as the sum of the elements of that dimension of the decision variables of all agents in each region in the next iteration, and is used as a constraint condition for the next iteration. If satisfied, output the Pareto solution set for that time period and let t = t + 1.

[0088] Step (19): When N is not satisfied l =N layer Then proceed to the next iteration, divide each region of the upper layer into sub-regions, and use each sub-region as the initial agent for that iteration. If it is the lowest iteration, each topological point is an agent, and select a leader agent in each region of the upper layer, and the others are followers.

[0089] Step (20): Perform a parallel iteration process for each partition, that is, execute steps (4) to (17) in parallel for each partition to complete the iteration process of this layer;

[0090] Step (21): If N is satisfied l =N layer If the Pareto solution set for that time period is output, the decision-maker can select the Pareto optimal solution according to the specific situation. Let t = t + 1, and proceed with the iteration for the next time period.

[0091] Step (22): Determine whether t≤t is satisfied. total If the condition is met, the iteration process continues for the next time period; otherwise, the method ends and the Pareto solution set for each time period is output.

[0092] The present invention has the following advantages and effects compared with the prior art:

[0093] (1) When the scale of the integrated energy system is very large, the multi-layer distributed multi-objective consensus method can overcome the disadvantages of slow calculation speed and poor convergence of the distributed multi-objective consensus method. Compared with the centralized multi-objective method, when the similar objective results are obtained, the information privacy is higher and the convergence time is shorter.

[0094] (2) Introduce the niche strategy and binary tournament selection mechanism to endow the Pareto solution set with multi-modal characteristics. For decision-makers, it is possible to obtain all the optimal solutions of the optimization problem, gain a deeper understanding of the problem, and play an important role in characterizing problem attributes, proposing improvement directions, and finding optimal solutions. At the same time, when some optimal solutions become unavailable due to factors such as environmental changes, decision-makers can conveniently and quickly find another optimal solution as a substitute.

[0095] (3) Add a fully connected layer embedded self-attention network with quantum and fuzzy logic to accelerate the acceleration process during the iteration of the multi-layer multi-modal multi-objective distributed consensus method, and improve the computational speed of the method. Brief Description of the Drawings

[0096] Figure 1 is the complete flowchart of the method of the present invention. Detailed Embodiment

[0097] A multi-modal hierarchical multi-objective distributed optimization acceleration method for an integrated energy system proposed by the present invention is described in detail as follows in conjunction with the accompanying drawings: Figure 1 is the complete flowchart of the method of the present invention. First, let the initial time t = 0 and start the optimization calculation for the t-th period. Then, initialize the method and input the Laplacian matrix L calculated from the topological structure diagram of the system to be optimized G , determine the total number of layers N of the system layer , determine the maximum number of iteration steps K for each layer, and let the initial layer N l = 0. After that, partition the complete integrated energy system, regard each region as an initial agent. If it is not the iteration of the first layer, then partition each region of the upper layer respectively, select a leader agent in each sub-region, and the rest as follower agents. If it is the iteration process of the bottom layer, then each topological point is an agent. Then, start the iteration process, let the initial iteration step k = 0. After that, randomly generate the consensus variable values and iteration variable values of each agent. Then, use the topological structure diagram of the agent to update the consensus variables of the follower agents and leader agents respectively, and update the decision variable elements, compare the decision variables with the corresponding objective space function values, and generate the Pareto optimal solution set and the Pareto front. Then, judge whether the current iteration algebra satisfies k < n 10 or K - n 10 < k < K. If this condition is satisfied, continue to iterate using the multi-modal multi-layer multi-objective distributed consensus method. If this condition is not satisfied, enter the fully connected layer embedded self-attention network with quantum and fuzzy logic and execute the acceleration process. After that, if the acceleration process is executed, input N agentThe system generates m-dimensional decision variables and corresponding objective function values. Then, using a fuzzy logic strategy, the agent is divided into four fuzzy logic subsets, and 16% of the individuals in each subset are selected to enter a two-dimensional quantum universal gate to prevent overfitting in the fully connected layer embedded self-focused network. Next, using the four fuzzy subsets as datasets, four fully connected layer embedded self-focused networks are trained using momentum stochastic gradient descent. The system inputs randomly selected weighted overall objective function values ​​to predict the corresponding decision variable values, thus constraining them to the boundary values. This provides input for the next iteration; afterwards, let k = n 10 +1. Afterwards, if the iterative process is executed, the niche strategy and binary tournament selection mechanism are used to select agents to participate in the next iteration to solve the multimodal problem. Then, the consistency variables of the follower and leader agents are updated using the agent topology graph, and the decision variable elements are updated simultaneously. The decision variables are compared with the corresponding objective space function values ​​to generate the Pareto optimal solution set and the Pareto front. Then, let k = k + 1, and check if k = K is satisfied. If it is, the iteration of this layer ends; otherwise, the iterative process is executed again. Afterwards, if k = K is satisfied, the iteration of this layer ends, and N is set to N. l =N l +1. Then, determine if N is satisfied. l =N layer If not satisfied, proceed to the next iteration, using each element of the output Pareto optimal solution as the sum of that element in the decision variables of all agents in each region for the next iteration, as a constraint for the next iteration; if satisfied, output the Pareto solution set for that time period, and let t = t + 1. Finally, determine whether t ≤ t total If the condition is met, the iteration process continues for the next time period; otherwise, the method ends and the Pareto solution set for each time period is output.

[0098] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for accelerating multimodal hierarchical multi-objective distributed optimization of integrated energy systems, characterized in that: This method combines multimodal problem optimization methods, quantum and fuzzy logic fully connected layer embedded self-concern network acceleration methods, and multi-layer multi-objective distributed consensus methods to solve multi-objective optimization problems in integrated energy systems. It overcomes the shortcomings of slow computation speed and poor convergence of non-multi-layer distributed methods when used in large-scale systems. Compared with centralized optimization methods, it has higher information privacy. The acceleration process improves the convergence time of the entire method and provides more diverse solution sets for multi-objective multimodal problems. The steps during use are as follows: Step (1): Set the current time period t = 0, and start the optimization solution for the t-th time period; Step (2): Method initialization, determine the total number of system layers N layer The input is the Laplace matrix L, calculated from the topology diagram of the system being optimized. G The topology graph is an undirected, unweighted graph. The maximum number of iterations K for each layer is determined, and the number of iteration layers N is set. l =0, so let the number of iterations k in the current iteration process be 0; Laplace matrix L G element l in ii With element l ij for: Among them, a ij Let A be the adjacency matrix. G The element in, a ij The value can be: Among them, v i and v j Two distinct points in the topology, v i ,v j ∈V G V G ={v1,v2,...,v M Let} be the set of nodes, and M be the number of nodes in the topology graph; v i v j An edge is formed by two distinct points in a topological structure. Let be the set of edges; The objective function f(x) for the multimodal and multi-objective optimization problem of integrated energy systems is: Among them, f (e) The objective function representing the total cost objective, f (c) The objective function f represents the total carbon emission target. (ξ) The objective function represents the overall energy efficiency target of the entire system. The objective function f represents user satisfaction. (F) The objective function represents the flexibility of the system's regulation capability; x represents a multi-dimensional decision vector, including the active power output of the j-th power supply unit in time period t. The energy supply E of the power supply unit during time period t j,t The energy consumption W of the j-th power supply unit j The change in electricity, gas, and heat load ΔW in time period t after the user responds to demand. t e ΔW t g ΔW t h The flexibility index of the power supply units in adjusting the supply of electricity, heat, and gas upwards and downwards. min() is a function that selects the minimum value; The objective function f of the total cost target (e) for: Where n is the number of power supply units. The cost of generating electricity for the j-th power unit during the t-th time period, t total a is the total number of segments that are evenly divided within the scheduling cycle. j b j and c j These are the cost coefficients for the secondary, primary, and constant terms of the j-th power supply unit, respectively. The objective function f of the total carbon emission target (c) for: in, Let α be the carbon emissions of the j-th power supply unit during the t-th time period. j β j and γ j These are the carbon emission coefficients of the second-order, first-order, and constant terms of the j-th power supply unit, respectively. The objective function f of the overall system energy efficiency target (ξ) for: Where ξ represents the overall energy efficiency over the complete system scheduling cycle. Let d be the load on the site corresponding to the j-th power supply unit after considering the comprehensive demand response in time period t. j η is the reference coefficient for standard coal equivalent of energy type j. s,j Let J be the energy efficiency of the j-th power supply unit. The energy efficiency of the j-th power supply unit is: Where E j Let J be the energy supply of the j-th power supply unit. The objective function of user satisfaction for: User satisfaction is used W indicates t e W t g W t h The original electricity, gas, and heat loads are considered; the larger the load response, the more load adjustments users make, and the higher the user satisfaction adjustment value. The larger the value, the lower the user's satisfaction level with energy usage. max() is a function that selects the maximum value; The objective function f of the system's adaptability flexibility (F) for: minf (F) =min(-F load )=-(ω1·F E +ω2·F H +ω3·F G ) (9) Where F load As an indicator for evaluating the flexibility of integrated energy systems; F E F H and F G These represent the comprehensive flexibility index of the electrical, thermal, and pneumatic links, respectively; ω1, ω2, and ω3 represent the weighting coefficients of the electrical, thermal, and pneumatic links, respectively. The overall flexibility index F of the electrical link E for: in and These represent the flexibility indicators for adjusting the power supply capacity of the generating units upwards and downwards, respectively. The overall flexibility index F of the thermal link H for: in and These represent the flexibility indicators for adjusting the thermal energy provided by the power supply unit upwards and downwards, respectively. The overall flexibility index F of the gas process G for: in and These represent the flexibility indicators for adjusting the gas energy provided by the power supply unit upwards and downwards, respectively. The power constraints of the generating unit include power balance constraints and upper and lower limit output constraints; The power balance constraint is: in Let t be the total system load demand in time period t; The upper and lower limits of the power supply unit's output are constrained as follows: in The lower limit of the active power output of the j-th power supply unit in time period t. The maximum active power output of the j-th power supply unit during the t-th time period; Step (3): Divide the complete integrated energy system into regions, treat each region as an agent, and use it as the initial agent for the current layer iteration. Select one leader agent and the rest as follower agents. The leader agent is selected from the center of the topology. Step (4): Randomly generate the consistency variable value and iterative variable value for each agent; Consistency variables, which correspond to elements of the same dimension but are different from the decision variables, include the objective functions of the j-th agent for cost targets and carbon emission targets for active power output in time period t. The partial derivatives of the ... The partial derivatives are: Where is the total objective value of the j-th agent in time period t, obtained by linear weighting of the cost objective function and the carbon emission objective function. Let be the total objective value of the first agent in time period t, obtained by linear weighting of the cost objective function and the carbon emission objective function. Let be the total objective value of the nth agent in time period t, obtained by linearly weighting the cost objective function and the carbon emission objective function. The active power output of the first power supply unit in time period t. For the active power output of the nth power supply unit in time period t, For each element of the consistency variable that is not equal to the decision variable of the agent in time period t; the total objective value of the j-th agent in time period t after linear weighting of the cost objective function and the carbon emission objective function. for: Wherein, τ is the weighting factor, and its value range is 0≤τ≤1; The initial decision variables derived from the consistency variables are: in, It is the decision variable element of agent j in the k-th iteration at time t that is not equal to the consistency variable element. Let be the consistency variable element of agent j in the k-th iteration at time t that is not equal to the decision variable element. For the χ-th layer The active power limit of the region is the first The sum of the active power limits of the lower-level sub-regions or power supply units of the region. Let be the total number of regions in layer χ. For decision variable elements that are the same as elements of the consistency variable, the decision variable elements that agent j has in time period t during the k-th iteration are equal to the consistency variable. for: in, Let be the consistency variable element that is equal to the decision variable in the k-th iteration at time t; Step (5): Based on the topology diagram of the agents, update the consistency variables of the follower agents and the leader agents respectively, and update the decision variable elements. Then compare the decision variables with the corresponding objective space function values ​​to generate the Pareto optimal solution set and the Pareto front; based on the input Laplace matrix L G Find the k-th iteration element d in the row random matrix of the agent. ij (k): After iterative updates, the consistency variable of the follower agent is: After iterative updates, the consistency variable for the leader agent is: in, Let be the consistency variable of the i-th agent in the (k+1)-th iteration during the t-th time period; Let be the consensus variable of the i-th agent in time period t during the k-th iteration; ε is the power balance adjustment factor of the multi-level multi-objective distributed consensus method; ΔP t (k) is the power deviation in the k-th iteration during time period t; Then update the decision variable elements. For decision variable elements that are different from the consistency variable elements, the decision variables derived from the consistency variables are: in It is the decision variable element of agent j in the (k+1)th iteration at time t that is not equal to the consistency variable element. Let be the consistency variable element that is not equal to the decision variable element in the (k+1)th time period of agent j; For decision variable elements that are identical to the consistency variable elements, the decision variable elements of the agent in time period t during the (k+1)th iteration that are equal to the consistency variable are... for: in These are the consistency variables that are equal to the decision variables in the agent during time period t of the (k+1)th iteration; Step (6): Determine whether the number of generations of the current iteration satisfies \(k < n\) 10 or \(K - n\) 10 where \(<k < K, n\) 10 are the number of iterations before and after the acceleration phase; if \(k < n\) 10 or \(K - n\) 10 where \(<k < K\), continue to iterate using the multi-modal multi-layer multi-objective distributed consensus method; if \(k < n\) is not satisfied 10 or \(K - n\) 10 where \(<k < K\), enter the fully connected layer embedded self-attention network of quantum and fuzzy logic to perform the acceleration process; Step (7): If k < n 10 or K - n 10 < k < K, then execute the iterative process, start the next iteration, and use the niche strategy and binary tournament selection mechanism to select the agents participating in the next iteration to solve the multi-modal problem; set both the niche size and the number of agents entering the next iteration to be N agent , and put the agents of the next iteration into the iteration set; Step (8): Randomly select agents from the population to enter the candidate solution set until the niche is filled; Step (9): Calculate the Euclidean distance D between the first agent in the candidate solution set as individual a in the binary tournament and every agent in the set except for that agent, and select the agent with the smallest Euclidean distance as the other individual b in the binary tournament. Determine the dominance relationship between the two individuals a and b in the binary tournament, and select one of the two individuals to enter the iteration set; the Euclidean distance D between the two individuals in time t. t for: in and Let represent the i-th element of the decision variable for individual a and individual b in time period t, respectively, and m represent the dimension of the decision variable; and Let represent the first element of the decision variable for individual a and individual b in time period t, respectively; and Let m represent the m-th element of the decision variable for individual a and individual b in time period t, respectively. The binary tournament selection mechanism is as follows: First, determine the dominance relationship between two individuals, a and b, in the binary tournament. If individual a dominates individual b, then individual a is added to the iteration set; otherwise, individual b is added to the iteration set. However, if the two individuals do not dominate each other, calculate the crowding distance of each individual in the decision space. And select individuals with larger crowding distances to add to the iteration set; if the crowding distance between two individuals is... They are equal, so individual a is placed into the iteration set; the solution selected by the binary tournament selection mechanism is the multimodal solution, ensuring the comprehensiveness of the solution in multiple cases, while the solutions not retained in the iteration set are those that the decision-maker will not use; crowding distance CD i,X for: in, This represents the crowding distance of the i-th agent in the decision space during time period t; 1≤i≤N agent ; Let m represent the value of the i-th agent in the j-th dimension during time period t, where 1 ≤ j ≤ m. This represents the value of the (i+1)th agent in the j-th dimension during time period t. This represents the value of the (i-1)th agent in the j-th dimension during time period t. This represents the value of the (i+1)th agent in time period t on the first dimension. This represents the value of the (i-1)th agent in time period t on the first dimension. This represents the value of the (i+1)th agent in the m-th dimension during time period t. This represents the value of the (i-1)th agent in the m-th dimension during time period t; This represents the maximum value in the j-th dimension during time period t. This represents the minimum value in the j-th dimension during time period t. This represents the maximum value in the first dimension during time period t. This represents the minimum value in the first dimension of the time interval t. This represents the maximum value in the m-th dimension during the t-th time period. This represents the minimum value in the m-th dimension during time period t; Step (10): Repeat step (9) until the iteration set is filled; Step (11): Update the consistency variables of the follower agents and leader agents in the iteration set respectively in combination with the topological structure diagram of the agents, and update the decision variable elements. Then compare the decision variables with the corresponding values in the target space to generate the Pareto optimal solution set and the Pareto front; Step (12): If it does not satisfy k < n 10 or K - n 10 < k < K, then enter the acceleration process accelerated by the fully connected layer embedded self-attention network of quantum and fuzzy logic, and input N agent m-dimensional decision variables and the corresponding objective function values; Step (13): According to N agent The weighted total objective function value of each agent is used to divide the agents into four fuzzy logic subsets using a fuzzy logic strategy; each agent is then classified into category z by fuzzy rules based on the weighted total objective function value. p In the middle; the weighted total objective function value f of the p-th agent. T,p for: Where ρ1, ρ2, ρ3, ρ4, and ρ5 are the objective functions f (e) f (c) f (ξ) , f (F) The corresponding weighted parameter values; the category z of the p-th agent obtained by fuzzy rule classification. p for: Where f Tmax and f Tmin These are the maximum and minimum values ​​of the weighted overall objective function, respectively. Subsets A, B, C, and D are four fuzzy logic subsets, and subsets A to D respectively have H... agent，A H agent，B H agent，C and H agent，D One intelligent agent, 0 <p≤N agent ; Step (14): To prevent the data input to the fully connected layer embedded self-focused network from being over-synthesized and causing overfitting of the fully connected layer embedded self-focused network, 16% of the individuals in each subset are selected to enter the two-dimensional quantum universal gate; each selected individual enters one of the four fuzzy subsets through a two-dimensional quantum universal gate, and each universal quantum gate Q 1bit for: Where θ, λ and These are Felix Bloch parameters, where θ is the polar angle of the Bloch sphere. λ represents the relative phase of the Bloch sphere, and λ represents the global phase of the Bloch sphere; the two-position quantum universal gate Q 2bit It is the tensor product of two unit quantum universal gates, which is: Q 1,1bit and Q 2,1bit For two unit quantum universal gates, θ1, θ2, λ1, λ2, and Q 1,1bit Bloch's polar angle, Q 2,1bit Bloch's polar angle, Q 1,1bit Bloch sphere global phase, Q 2,1bit Bloch sphere global phase, Q 1,1bit The relative phase of the Bloch sphere and Q 2,1bit The relative phase of the Bloch sphere; Step (15): Using four fuzzy subsets as datasets, train four fully connected layer embedded self-focused networks using momentum stochastic gradient descent. Input the randomly selected weighted total objective function value to obtain the corresponding decision variable value and provide input for the next iteration; the input and output are the weighted total objective function value and the decision variable, respectively. Step (16): Restrict the predicted decision variables to the boundary. u represents the u-th element of the decision variable. and Let represent the lower and upper bounds of the u-th element of the decision variable in time period t, respectively; output the boundary-constrained decision variable values ​​as the initial values ​​of the decision variables in the next iteration of the multi-level multi-objective distributed consensus method. Step (17): After the acceleration process is completed, continue the iterative process of the multi-layer multi-objective distributed consensus method. When k = K, end the iteration of this layer and let N l =N l +1; Step (18): Determine whether N is satisfied. l =N layer If not satisfied, proceed to the next iteration and output the Pareto optimal solution selected according to the decision-maker's preference. Each element of the output Pareto optimal solution is used as the sum of the elements of that dimension of the decision variables of all agents in each region in the next iteration, and is used as a constraint condition for the next iteration. If satisfied, output the Pareto solution set for that time period and let t = t + 1. Step (19): When N is not satisfied l =N layer Then proceed to the next iteration, divide each region of the upper layer into sub-regions, and use each sub-region as the initial agent for that iteration. If it is the lowest iteration, each topological point is an agent, and select a leader agent in each region of the upper layer, and the others are followers. Step (20): Perform a parallel iteration process for each partition, that is, execute steps (4) to (17) in parallel for each partition to complete the iteration process of this layer; Step (21): If N is satisfied l =N layer If the Pareto solution set for that time period is output, the decision-maker can select the Pareto optimal solution according to the specific situation. Let t = t + 1, and proceed with the iteration for the next time period. Step (22): Determine whether t≤t is satisfied. total If the condition is met, the iteration process continues for the next time period; otherwise, the method ends and the Pareto solution set for each time period is output.

Images