Double-file super-multi-objective optimization method and system based on enhanced knowledge sharing

Abstract

The invention discloses a double-file super-multi-objective optimization method and system based on enhanced knowledge sharing. The method comprises the steps of generating an initial population based on integrated energy system data and storing the initial population into a convergence file and a diversity file; calculating a knowledge sharing factor based on a non-dominated solution set of a diversity file, and adaptively calculating a generalized displacement density index and a new displacement index according to the knowledge sharing factor; carrying out convergence evaluation and updating on the convergence archive individuals by utilizing the generalized displacement density index, and carrying out diversity evaluation and updating on the diversity archive individuals by utilizing the new displacement index; selecting a parent scheme from the updated double archives and generating a filial generation through genetic manipulation; and combining the parent and child generation schemes, updating the double archives, carrying out loop iteration until a termination condition is met, and finally outputting a non-dominated solution set in the diversity archives as an optimal scheduling scheme. According to the method, the convergence speed and the solution set distribution performance for solving the high-dimensional multi-objective optimization problem are remarkably improved through knowledge sharing and a double-file cooperation mechanism.

Description

Technical Field

[0001] This invention belongs to the field of integrated energy system operation optimization technology, and in particular relates to a dual-archive multi-objective optimization method and system based on enhanced knowledge sharing. Background Technology

[0002] In the energy and industrial sectors, practical problems such as workshop operation scheduling, sensor network configuration, and regional integrated energy system optimization are essentially engineering decision-making problems involving multiple conflicting objectives—that is, multi-objective optimization problems. The operation optimization of combined cooling, heating, and power (CCHP) integrated energy systems is a typical example. This system needs to simultaneously coordinate multiple physical devices such as gas turbines, energy storage devices, and refrigeration units. Its optimization objectives typically include minimizing system operating costs, minimizing primary energy consumption, and minimizing pollutant emissions. These objectives are mutually conflicting. Therefore, finding the optimal set of trade-offs among multiple conflicting objectives to improve the overall system operating efficiency is a core engineering challenge that urgently needs to be addressed in this field.

[0003] Traditional single-objective optimization methods often transform multi-objective optimization problems into single-objective optimization problems by weighting or other means when dealing with such problems. This method relies heavily on prior knowledge and is difficult to obtain a series of diverse trade-off solutions. As a result, the final decision solution may not truly reflect the complex preferences in engineering practice, thus affecting the actual operating efficiency and economy of the system.

[0004] To address this challenge, researchers have designed various multi-objective evolutionary algorithms. Existing methods can be broadly categorized into three types:

[0005] (1) Improve the selection pressure by improving the dominance relationship, but its performance is extremely sensitive to parameter settings. When dealing with high-dimensional, nonlinear target spaces such as combined cooling, heating and power integrated energy systems, excessive selection pressure can easily lead to premature convergence of the search process, making it impossible to explore a wide range of equipment scheduling schemes. The resulting solution set is not diverse enough, which is reflected in the actual system as a single operating strategy that cannot adapt to diverse load demands and operating conditions.

[0006] (2) Algorithms based on reference vectors. This type of method guides the search direction through a preset reference vector, which can improve the uniformity of the solution set distribution. However, when dealing with the complex and irregular Pareto front commonly found in combined cooling, heating and power integrated energy systems, its fixed reference vector system is difficult to adapt flexibly, resulting in insufficient convergence pressure of the algorithm and an inability to accurately approximate the true high-efficiency operating range of the system. The generated scheduling scheme may perform poorly in terms of either cost or emissions.

[0007] (3) Performance index-based algorithms. These algorithms use indices to guide the search, but a single fixed index can easily cause the optimization process to fall into local optima. In engineering terms, this means that the scheduling scheme falls into a certain fixed operating mode and cannot discover innovative operating strategies that can significantly improve the efficiency of comprehensive energy utilization. On the other hand, introducing multiple indices will significantly increase the computational overhead and make it difficult to meet the real-time requirements of large-scale combined cooling, heating and power integrated energy systems for optimization calculation.

[0008] Previous research has shown that dual-archive structure algorithms provide an effective framework for balancing convergence and diversity, and have been attempted to solve multi-objective optimization problems. However, most existing dual-archive algorithms primarily focus on improving the internal update strategy of a single archive, paying insufficient attention to how the two archives can achieve deep collaboration based on the characteristics of specific engineering problems. When dealing with complex practical engineering problems such as combined cooling, heating, and power (CCHP) systems, the lack of collaboration mechanisms between archives can easily lead to unbalanced optimization results: over-biasing towards the convergent archive will cause the obtained solution set to be concentrated in the low-cost region, ignoring environmentally friendly solutions; conversely, over-reliance on diverse archives may lead to slow algorithm convergence and low overall quality of the obtained solution set.

[0009] Therefore, in the specific industrial application context of combined cooling, heating and power integrated energy systems, how to design an optimization algorithm that can adapt to system operating characteristics, achieve deep collaborative convergence and diverse search capabilities, so that it can not only rapidly and stably approximate the real Pareto front, but also generate a series of executable equipment scheduling schemes that achieve a good balance between operating costs, energy efficiency and environmental performance, has become a key technical bottleneck for improving the intelligent operation level of integrated energy systems. Summary of the Invention

[0010] To address the aforementioned technical problems, this invention proposes a dual-archive multi-objective optimization method and system based on enhanced knowledge sharing, thereby resolving the issues present in the prior art.

[0011] To achieve the above objectives, this invention provides a dual-archive multi-objective optimization method based on enhanced knowledge sharing, comprising:

[0012] S1. Generate an initial population based on integrated energy system data, and store the initial population in the convergence file and the diversity file respectively;

[0013] S2. Calculate the knowledge sharing factor based on the non-dominated solution set in the aforementioned diversity archive; the knowledge sharing factor is used to adaptively calculate the generalized displacement density index and the new displacement index.

[0014] S3. Based on the generalized displacement density index, perform convergence evaluation on the individuals in the convergence archive, and update the convergence archive based on the evaluation results;

[0015] S4. Perform diversity assessment on individuals in the diversity profile based on the new displacement index, and update the diversity profile based on the assessment results;

[0016] S5. Select the parent population of the scheduling scheme from the updated convergence file and the updated diversity file through the tournament selection mechanism, and generate the offspring population of the scheduling scheme through genetic operations.

[0017] S6. Merge the child population of the scheduling scheme with the parent population of the scheduling scheme to obtain a merged population, and update the convergence file and the diversity file based on the merged population;

[0018] S7. Determine whether the preset termination condition has been met. If it has, output the non-dominated solution set in the diversity file as the optimal scheduling scheme for the combined cooling, heating and power integrated energy system. Otherwise, return to step S2.

[0019] Optionally, in step S1, an initial population is generated based on the combined cooling, heating and power integrated energy system model and operating data. The data includes equipment capacity constraints, energy prices, load demand, and upper and lower bounds of decision variables. The decision variables include at least one of gas turbine output, energy storage device charging and discharging power, and refrigeration unit power.

[0020] Optionally, in S2, the formula for calculating the knowledge-sharing factor is:

[0021] ;

[0022] Where M is the number of targets, Let be the average of all individuals on the i-th objective, and log() be the natural logarithm.

[0023] Optionally, in S3, the formula for calculating the generalized displacement density index is:

[0024] ;

[0025] in, For all that cannot be dominated by Pareto A collection of individuals For the target number, As a knowledge-sharing factor, For individuals In the Normalized values ​​on each objective For individuals The first one after mapping according to the shift rule One target value.

[0026] Optionally, in step S4, the process of performing diversity assessment on individuals in the diversity profile based on the new displacement index and updating the diversity profile includes:

[0027] The target vectors in the diversity archive are translated with the ideal point as the origin, and the most representative boundary individual is determined along the axial direction of each target. The determined boundary individuals are used as the initial elite subset. When the number of individuals in the diversity archive does not exceed the size limit, all individuals are directly used as the initial elite subset. The Minkowski distance from each individual not selected into the elite subset to the ideal point and the displacement difference distance of each individual relative to the current elite subset are calculated, and a new displacement index value is obtained based on the calculation results. A greedy selection strategy is adopted to select the individual with the highest new displacement index value from the remaining individuals at each step and incorporate it into the elite subset until the size of the elite subset reaches the size limit of the diversity archive, thus completing the update of the diversity archive.

[0028] Optionally, in S5, the parent population of the scheduling scheme is selected from the updated convergence file and the updated diversity file through a tournament selection mechanism, including:

[0029] The individuals in the updated convergence file and the updated diversity file are merged to form a merge pool; several individuals are randomly selected from the merge pool, and the individual with the best index value is selected from the several individuals according to the generalized displacement density index or the new displacement index to enter the mating pool; the above selection process is repeated until the size of the mating pool reaches a preset value, and the parent population of the scheduling scheme is obtained.

[0030] Optionally, in step S5, generating the offspring population of the scheduling scheme through genetic operations includes:

[0031] The parent individuals in the parent population of the scheduling scheme are paired up; for each pair of parent individuals, a simulated binary crossover operation is performed to generate two intermediate offspring individuals; for each intermediate offspring individual, a polynomial mutation operation is performed to perturb each of its decision variables with a preset mutation probability to obtain the offspring population of the scheduling scheme.

[0032] Optionally, in step S6, updating the convergence profile and the diversity profile based on the merged population includes:

[0033] The merged population and individuals in the current convergence archive are sorted using a non-dominated method, and a preset number of individuals are selected from them based on the generalized displacement density index to update the convergence archive. At the same time, the merged population and individuals in the current diversity archive are sorted using a non-dominated method to obtain a non-dominated frontier. If the number of individuals in the non-dominated frontier does not exceed the capacity of the diversity archive, all individuals are retained; otherwise, a preset number of individuals are selected from them based on the new displacement index to update the diversity archive.

[0034] This invention also provides a dual-archive multi-objective optimization system based on enhanced knowledge sharing, used to implement the above method, comprising:

[0035] An initialization module is used to generate an initial population based on integrated energy system data, and store the initial population in a convergence archive and a diversity archive, respectively.

[0036] The knowledge sharing factor calculation module is used to calculate the knowledge sharing factor based on the non-dominated solution set in the diversity archive.

[0037] The SDEp calculation and sorting module is used to evaluate and sort individuals in the converged archive based on the knowledge sharing factor and the generalized displacement density index.

[0038] The SBI calculation and selection module is used to evaluate and iteratively select individuals in the diversity archive based on the knowledge-sharing factor and the new displacement index.

[0039] The genetic operations module is used to select the parent population of the scheduling scheme from the updated convergence archive and diversity archive through a tournament selection mechanism, and to generate the offspring population of the scheduling scheme through genetic operations.

[0040] The dual-file update module is used to merge the offspring population and the parent population of the scheduling scheme to obtain a merged population, and update the convergence file and the diversity file respectively based on the merged population;

[0041] The termination judgment module is used to determine whether a preset termination condition has been met, and when the condition is met, it outputs the non-dominated solution set in the diversity file as an optimized scheduling scheme.

[0042] Compared with the prior art, the present invention has the following advantages and technical effects:

[0043] This invention closely integrates a multi-objective evolutionary algorithm with specific engineering problems of combined cooling, heating and power (CCHP) integrated energy systems. By strengthening the knowledge sharing mechanism and the dual-archive collaborative update strategy, it not only improves the convergence speed and solution quality of the algorithm, but more importantly, its output directly corresponds to a series of executable and selectable system scheduling instructions, which can be directly applied to physical devices, significantly improving the intelligent operation level and overall energy efficiency of CCHP integrated energy systems. Attached Figure Description

[0044] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0045] Figure 1This is a flowchart illustrating high-dimensional multi-objective optimization based on a reinforced knowledge-sharing-assisted dual-archive evolutionary algorithm, according to an embodiment of the present invention.

[0046] Figure 2 This is a network architecture diagram of a combined cooling, heating and power integrated energy system according to an embodiment of the present invention;

[0047] Figure 3 Different embodiments of the present invention A schematic diagram of the unit Minkowski distance;

[0048] Figure 4 The distribution diagram of the solutions with three objectives obtained by DEA-GNG, LMPFE, MOEAD-DQN, and MaOEAIH on the MaF4 test problem according to the embodiments of the present invention;

[0049] Figure 5 The distribution diagram of the solutions with three objectives obtained by RVEA-DWC, hpaEA, and RKS-TAEA in the MaF4 test problem according to the embodiments of the present invention is shown.

[0050] Figure 6 The distribution diagram shows the solutions with 8 objectives obtained by DEA-GNG, LMPFE, MOEAD-DQN, and MaOEAIH on the MaF13 test problem according to embodiments of the present invention.

[0051] Figure 7 The distribution diagram of the solutions with 8 objectives obtained by RVEA-DWC, hpaEA, and RKS-TAEA in the MaF13 test problem according to the embodiments of the present invention;

[0052] Figure 8 The graph shows the trend of HV values ​​for the seven algorithms of this invention under different population iterations in the WFG9 test problem.

[0053] Figure 9 This is a 24-hour thermal network operation diagram according to an embodiment of the present invention;

[0054] Figure 10 This is a diagram of the 24-hour cold network operation according to an embodiment of the present invention;

[0055] Figure 11 This is a 24-hour electrical network operation diagram according to an embodiment of the present invention;

[0056] Figure 12 The figures show the minimum normalization target and curves for different iteration numbers in this embodiment of the invention. Detailed Implementation

[0057] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0058] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0059] Example 1

[0060] This invention deeply couples the algorithm search process with the physical operating characteristics of a combined cooling, heating, and power (CCHP) integrated energy system. Through an adaptive knowledge-sharing mechanism, it dynamically balances the simultaneous search for low-operating-cost scheduling schemes and low-pollutant-emission scheduling schemes, thereby achieving precise optimization and control of the system's operating state. Figure 1 As shown, this embodiment provides a dual-archive multi-objective optimization method based on reinforced knowledge sharing, including:

[0061] Step S1: Based on the combined cooling, heating and power integrated energy system model and operation data, generate an initial scheduling scheme population and initialize parameters.

[0062] Based on the actual operating parameters and constraints of each physical device in a combined cooling, heating, and power (CCHP) integrated energy system, a population of initial system scheduling schemes is randomly generated. Each scheme is encoded to represent a complete and feasible set of system operation instructions. The initial populations are stored separately in a convergence archive focused on searching for the economically optimal scheme and a diversity archive focused on maintaining diversity.

[0063] Specifically, the inputs include upper and lower limits of equipment capacity, gas price, grid electricity purchase price, population size, objective function (operating cost, primary energy consumption, carbon dioxide emissions), decision variables (gas turbine output, absorption chiller operating degree, electric cooling ratio, energy storage charging and discharging power, etc.), problem dimension, maximum number of generations, file size, upper and lower bounds of variables, and constraints (power balance, equipment ramp-up, energy storage capacity). First, the combined cooling, heating, and power (CCHP) dataset is called and constructed into a multidimensional array. Then, N individuals satisfying the constraints are randomly generated and stored in the convergence file CA and the diversity file DA, respectively, completing the initialization.

[0064] Specific details of a combined cooling, heating and power (CCHP) integrated energy system:

[0065] like Figure 2As shown, the combined cooling, heating, and power (CCHP) integrated energy system consists of a power grid, a natural gas supply side, a heating network, a cooling network, and energy storage and conversion equipment such as electricity storage, thermal storage, and cold storage. Each energy flow is coupled through power balance constraints and operational constraints such as equipment capacity, ramp-up, and efficiency. This provides a system structure basis for the subsequent coding, constraint verification, and target calculation of the 24-hour scheduling scheme.

[0066] This system aims to minimize the economic cost and optimize the environmental performance of an integrated energy system. The system's economic cost mainly consists of the unit investment cost. Electricity purchase cost and the cost of purchasing natural gas .

[0067] ;

[0068] ;

[0069] in, For the rate of return on investment; , , These are the rated electrical power of the gas turbine, gas boiler, and natural gas internal combustion engine unit, respectively. , These are the rated cooling capacities of the electric chiller and the absorption chiller, respectively. , , , , These are the operating costs of the rated power internal combustion engine, gas boiler, natural gas internal combustion engine unit, electric chiller, and absorption chiller, respectively.

[0070] ;

[0071] in, One system cycle; for Time-of-use electricity purchase price for Power purchased by the system during a specific time period.

[0072] ;

[0073] in, for The unit price of natural gas purchased through the time-slot system; for Gas purchase capacity of the time-period system.

[0074] During system operation, the combustion of natural gas produces carbon dioxide gas. Therefore, the emission of carbon dioxide during system operation incurs environmental costs. The objective function is as follows:

[0075] ;

[0076] in, For emissions Unit price; This represents the carbon dioxide emission coefficient per unit of electrical power. This is the carbon dioxide emission factor per unit volume of natural gas.

[0077] System overall optimization function as follows:

[0078] ;

[0079] The system can generate electricity by burning natural gas using natural gas internal combustion engines and gas turbines. It can also use the heat generated from these devices and gas boilers for heating and cooling, thus integrating power supply, heating, cooling, and gas supply. This strengthens the coupling between the various functional systems and improves energy utilization efficiency. The mathematical model and constraints of the gas turbine are as follows:

[0080] ;

[0081] ;

[0082] ;

[0083] in, For the power generation of gas turbines; For heat generation power; For power generation efficiency; For heat production efficiency; This refers to the power consumption of gas. , These represent the upper and lower limits of the gas turbine's electrical power.

[0084] The mathematical model and constraints of the gas-fired boiler are as follows:

[0085] ;

[0086] ;

[0087] in The heat output of the gas-fired boiler; This refers to the power consumption of gas. For heat production efficiency; , These represent the upper and lower limits of the thermal power of a gas-fired boiler.

[0088] The mathematical model of a natural gas internal combustion engine is as follows:

[0089] ;

[0090] ;

[0091] in, Power generation for natural gas internal combustion engines; For heat generation power; For power generation efficiency; For heat production efficiency; This refers to the gas consumption power.

[0092] The mathematical model of the electric chiller is as follows:

[0093] ;

[0094] in, It is the cooling power output of the electric chiller; It is the coefficient of performance (COP) of an electric refrigeration system; Input electrical power.

[0095] The mathematical model of an absorption chiller is as follows:

[0096] ;

[0097] in, To output cooling power for absorption chillers; Coefficient of performance (COP); This is the input thermal power.

[0098] Mathematical models of energy storage, heat and cold equipment , , as follows:

[0099] ;

[0100] ;

[0101] in, The self-discharge rate of the energy storage device; for( -1) Energy consumed during a specific charging / discharging period; , Power for charging and discharging energy storage devices; , To improve the charging and discharging efficiency of energy storage devices; , These are the upper and lower limits for energy storage in energy storage devices.

[0102] ;

[0103] ;

[0104] in, This refers to the thermal energy storage loss rate. for( -1) Time-limited energy storage heat; , Input and output thermal power for thermal storage equipment; , The input and output conversion efficiency of thermal storage equipment; , These are the upper and lower limits for heat storage in thermal storage equipment.

[0105] ;

[0106] ;

[0107] in, This refers to the cold energy storage loss rate. ( -1) Time-limited cold energy storage; , Input and output cooling power for cold storage equipment; , The input and output conversion efficiency of cold storage equipment; , These are the upper and lower limits for cold storage equipment.

[0108] The power balance constraints of the system for electrical energy, natural gas, heat energy, and cold energy are as follows:

[0109] ;

[0110] ;

[0111] ;

[0112] in, The electrical load delivered by the system to the user side; This refers to the amount of heat load delivered by the system to the user side. The amount of gas load delivered by the system to the user side; The amount of cooling load delivered by the system to the user side.

[0113] Step S2: Based on the set of non-dominated scheduling schemes in the current diversity archive, calculate the knowledge sharing factor according to the formula. This factor can adaptively fit the shape of the Pareto front in the target space of operating cost and emission cost for current combined cooling, heating and power (CCHP) integrated energy systems, and dynamically determine the parameters of the generalized Minkowski distance in subsequent evaluation indicators based on this.

[0114] The formula for calculating the knowledge sharing factor is:

[0115] ;

[0116] in, To achieve three objectives—operating cost, primary energy consumption, and CO2 emissions—this [method / technology] was used. Value adaptive determination and The Minkowski distance parameter in the text, For all individuals in the first The average value across all targets It is the natural logarithm. Figure 3 It is a diagram showing different Let's set the unit Minkowski distance from any point in a two-dimensional Cartesian coordinate system to the origin, in order to... and To flexibly adapt to multi-objective problems with different Pareto geometries, this invention derives the solution by fitting the Pareto nondominated solution in the DA and utilizing the learned geometric information. value.

[0117] In this embodiment of the combined cooling, heating and power integrated energy system, the factor The significance lies in its ability to learn the Pareto front geometry from currently found high-quality scheduling schemes. If the schemes are mostly concentrated in the low-cost region, then... The value will adaptively guide subsequent searches to focus on diversity; and vice versa. This achieves dynamic adaptation of algorithm parameters to the specific operating characteristics of energy systems, freeing the search process from blind mathematical calculations and closely aligning it with practical engineering problems.

[0118] Step S3: In the convergence archive, the generalized displacement density index is used to screen out the more economical scheduling schemes. This index can identify and retain those scheduling schemes with the greatest potential to reduce the total operating cost of the system, thereby accelerating the convergence of the algorithm to the high-economic region. Its technical effect is to quickly eliminate high-cost operating modes and focus on the most cost-effective scheduling strategy.

[0119] The generalized shift density index is as follows:

[0120] ;

[0121] Among them, all those belonging to the target space individuals and The distances are all measured using the Minkowski distance. According to... The current definition makes it difficult to distinguish The displacement-based density difference of all dominated individuals is considered because their displacement-based density values ​​are all equal to zero. Therefore, the following modification is made:

[0122] ;

[0123] in, As a knowledge-sharing factor, Includes all in The middle cannot control the current individual The solution, and if If =∅, then it satisfies .once It is a non-dominant individual, in the two formulas The result is the same. For individuals In the Normalized values ​​on each objective For individuals The first one after mapping according to the shift rule One target value. According to the revised definition, Each individual in the system can have a distinguishable density value based on displacement.

[0124] In the optimization of combined cooling, heating, and power (CCHP) integrated energy systems, this metric physically measures the minimum distance between a scheduling scheme and other non-dominant schemes in the database within the target space. Its technical effect is to accurately identify and retain scheduling schemes with the greatest potential for reducing total system operating costs, thereby quickly eliminating high-cost, low-efficiency operating modes. This concentrates search resources on the most economically efficient system operation strategies, significantly improving the economic convergence speed of the optimization process.

[0125] Step S4: In the diversity archive, identify boundary scheduling schemes based on the new displacement index and construct an elite subset;

[0126] New displacement indexes are adopted in the diversity archive. Each scheduling scheme is evaluated using this metric. This metric comprehensively considers the scheme's convergence performance and its contribution to enriching decision options. Iterative selection is achieved through a greedy strategy. The solution with the highest value is added to the elite subset. The technical effect is to ensure that the final set of scheduling solutions not only has superior performance, but also covers various technical paths and trade-offs ranging from purely pursuing economic efficiency to the ultimate pursuit of environmental protection.

[0127] The process of conducting diversity assessments of individuals in the diversity profile based on the new displacement index and updating the diversity profile includes:

[0128] The target vectors in the diversity archive are translated with the ideal point as the origin, and the most representative boundary individual is determined along the axial direction of each target. The determined boundary individuals are used as the initial elite subset. When the number of individuals in the diversity archive does not exceed the size limit, all individuals are directly used as the initial elite subset. The Minkowski distance from each individual not selected into the elite subset to the ideal point and the displacement difference distance of each individual relative to the current elite subset are calculated, and a new displacement index value is obtained based on the calculation results. A greedy selection strategy is adopted to select the individual with the highest new displacement index value from the remaining individuals in each step and add it to the elite subset until the size of the elite subset reaches the size limit of the diversity archive, and the update is completed.

[0129] A new displacement-based index is as follows: Indicators can effectively measure The convergence of each individual is determined based on a displacement strategy, where each individual should perform the following in the worst case: Sub-individual displacement. Inspired by displacement strategies, this embodiment proposes a new displacement-based index, referred to as... This is used to study the impact of an individual joining another elite population. The performance in the middle. It consists of two parts, namely and Each measure the individual's participation in Post-convergence and diversity.

[0130] New displacement index The calculation and acquisition of the diversity candidate solution set are as follows:

[0131] (1) For each individual x in the diversity archive DA, calculate the Minkowski distance to the ideal point.

[0132]

[0133] in, For all individuals in the DA archive in the first Minimum value on each objective for Normalized One target value, As a knowledge-sharing factor, The target number.

[0134] (2) Calculate the nearest shift distance to the current elite subset EP.

[0135]

[0136] in, A collection of elite individuals selected for the next generation. To The first after using shift mapping One target value.

[0137] (3) According to the formula: The comprehensive evaluation value will be obtained. The individuals with the highest values ​​are selected into the DA in turn until the maximum file size is reached.

[0138] Specifically Instead of using the common Euclidean distance, the Minkowski distance between the individual and the ideal point is calculated. The Euclidean distance cannot accurately measure the distance from the Pareto front to the origin under different geometries. Therefore, When applied to complex multi-objective optimization problems, it more accurately reflects the convergence of individuals in the population. It calculates the Minkowski distance between an individual in the population and its corresponding point of displacement in the elite set. It determines the population density by considering the elite individuals in the elite set. The computational complexity is reduced by simply shifting the individuals in the elite group.

[0139] Step S5: Intelligent integration and exploration of scheduling schemes;

[0140] A tournament selection mechanism is used to select parent scheduling schemes from a mating pool constructed from convergent and diverse archives. Genetic operations such as simulated binary crossover and polynomial mutation are then employed to intelligently fuse and fine-tune the equipment output parameters of different schemes, thereby generating new offspring scheduling schemes. In the context of optimizing a combined cooling, heating, and power (CCHP) integrated energy system, the technical essence of this process is to mix and fine-tune the equipment output parameters of different parent scheduling schemes at specific times. For example, combining the gas turbine output of scheme A at midday with the chiller power of scheme B at midday may produce a new scheme with lower costs or emissions. The technical effect of this step is that, through simulating the recombination and innovative exploration of excellent operating strategies, potential more efficient equipment collaborative operation modes within the system can be discovered, breaking through local optima.

[0141] Step S6: Dual-archive co-evolution and dynamic regulation;

[0142] The newly generated offspring scheduling scheme is merged with the parent scheme, and the convergence and diversity files are updated respectively. Simultaneously, the knowledge-sharing factor is recalculated based on the distribution of the new generation of elite schemes. This step constitutes a closed-loop feedback system, the technical effect of which is to achieve the synergistic evolution and dynamic precise control of the algorithm search strategy and the multi-objective optimization process of the combined cooling, heating and power integrated energy system.

[0143] Step S7: Optimize the output and application of the scheduling scheme.

[0144] The algorithm terminates when the evolutionary process meets the preset termination condition. The final output is the set of all non-dominated solutions in the diversity archive. This set of solutions constitutes a complete set of optimized scheduling plans that can be directly used to guide the actual operation of the combined cooling, heating, and power (CCHP) system. Thus, this algorithm completes the full technical loop from mathematical optimization to physical system control, achieving comprehensive optimization and direct control of the operating costs and environmental benefits of the integrated energy system.

[0145] like Figures 9-11 As shown, mapping the output non-dominated scheduling scheme to the 24-hour hot network, cold network, and power network yields the supply-demand balance and equipment output trajectory for each time period; among which... Figure 9 This reflects the relationship between heating supply and heat storage regulation in the heating network at different times. Figure 10 This reflects the relationship between the refrigeration supply and cold storage regulation of the cold network. Figure 11 This reflects the balance between power generation, energy storage, and power purchase on the grid side, thereby verifying the feasibility of the output scheme at the physical system level.

[0146] This embodiment also provides a dual-archive multi-objective optimization system based on enhanced knowledge sharing to implement the above method, including:

[0147] An initialization module is used to generate an initial population based on integrated energy system data, and store the initial population in a convergence archive and a diversity archive, respectively.

[0148] The knowledge sharing factor calculation module is used to calculate the knowledge sharing factor based on the non-dominated solution set in the diversity archive.

[0149] The SDEp calculation and sorting module is used to evaluate and sort individuals in the converged archive based on the knowledge sharing factor and the generalized displacement density index.

[0150] The SBI calculation and selection module is used to evaluate and iteratively select individuals in the diversity archive based on the knowledge-sharing factor and the new displacement index.

[0151] The genetic operations module is used to select the parent population of the scheduling scheme from the updated convergence archive and diversity archive through a tournament selection mechanism, and to generate the offspring population of the scheduling scheme through genetic operations.

[0152] The dual-file update module is used to merge the offspring population and the parent population of the scheduling scheme to obtain a merged population, and update the convergence file and the diversity file respectively based on the merged population;

[0153] The termination judgment module is used to determine whether a preset termination condition has been met, and when the condition is met, it outputs the non-dominated solution set in the diversity file as an optimized scheduling scheme.

[0154] To verify the effectiveness of this invention, it was compared with several state-of-the-art multi-objective evolutionary algorithms. These algorithms included DEA-GNG, LMPFE, MaOEAIH, MOEA / DDQN, RVEA-DWC, hpaE, and Two_Arch2. Experiments were conducted on the evolutionary multi-objective optimization platform (PlatEMO).

[0155] For the tested problems, 33 benchmark problems were selected, with a total of 133 test instances. These problems included the MaF, WFG, DTLZ, and IDTLZ test suites. For the MaF problem, the decision variables... The settings are as follows: For MaF1-MaF9, set to... ,in It is the target quantity. Fixed at 10; for MaF10-MaF13, set to ,in This is the number of decision variables; for MaF14 and MaF15, it is set to... The WFG problem Set as ,in It is a distance-related variable, fixed at 20. According to the DTLZ problem... Set as Among them, DTLZ1 DTLZ2-DTLZ6 10, DTLZ7 20. IDTLZ1 and IDTLZ2 Same as DTLZ1 and DTLZ2.

[0156] The selected metrics are the hypervolume metric (HV) and the inverse generation distance (IGDP). HV is a widely used evaluation metric that provides a comprehensive assessment of the population without requiring knowledge of the true Pareto front of the problem. A higher HV value indicates better convergence and diversity of solutions. IGDP, an improvement on the inverse generation distance (IGD), also provides a comprehensive assessment of the population.

[0157] The target number is set to 3, 5, 8, 10; the population size is set to 91, 210, 156, 275; and the maximum number of iterations is set to... Experimental method: For each test problem, all algorithms were run independently 30 times. The results were analyzed using the Wilcoxon rank-sum test at a significance level of 0.05.

[0158] Experiments show that all strategies and strategy combinations proposed in this embodiment are effective. The dual-archive multi-objective optimization algorithm based on reinforced knowledge sharing first utilizes Minkowski distance to transform the original displacement-based index (…). (Promoted to the new version) This invention proposes a novel offset-based index that effectively preserves the convergence property of the original index in estimating the population. For updating DA, this invention suggests a new offset-based index (...). This invention estimates the diversity and convergence of candidate solutions by moving them into their neighborhoods and utilizing the Minkowski distances from the moved solutions to the ideal point and from the original solutions to the Minkowski distances to the ideal point. Furthermore, the invention proposes an enhanced knowledge-sharing mechanism. This mechanism extracts knowledge factors learned from fitting the geometry of the power factor (PF) for computation. and The value. Through and With two updated archives, this algorithm can effectively solve complex multi-objective problems with different PF geometries.

[0159] Table 1

[0160]

[0161] Table 1 (continued)

[0162]

[0163] Table 1 (continued from 2)

[0164]

[0165] Table 1 shows the HV values ​​obtained by the proposed algorithm (RKS-TAEA) and six other state-of-the-art algorithms for 3, 5, 8, and 10 objectives in the MaF1 to MaF15 test problems. Overall, DEA-GNG outperforms the other algorithms on two test functions, while LMPFE and MOEA / D-DQN achieve the best HV values ​​on six test functions respectively. MaOEAIH and hpaEA only outperform the other algorithms on three and two test functions, respectively, while RVAE-DWC performs exceptionally well on six test functions. The proposed algorithm achieves 35 best scores across 60 test functions, outperforming all other algorithms. This indicates that the proposed algorithm achieves the best overall performance in the MaF test suite. Specifically, in the MaF2, MaF4, and MaF12 test problems, the proposed algorithm outperforms the other six algorithms on all objectives. In the MaF3, MaF7, MaF13, and MaF15 test problems, the proposed algorithm performs exceptionally well on a total of three objectives. RVAE-DWC performs best on all objectives of the MaF10 test problem, while LMPFE achieves best results on objectives 5, 8, and 10 of the MaF11 test problem. For other test problems (MaF1, MaF5, MaF6, MaF8, MaF9, and MaF14), although no algorithm has a significant overall advantage, the proposed algorithm, MaOEAIH, and MOEA / D-DQN consistently outperform other algorithms when considered comprehensively. For complex test problems such as MaF2-4, MaF7, MaF12-13, and MaF15 (with multiple peaks, large scale, and irregular fronts), the proposed algorithm significantly outperforms other comparative algorithms in terms of HV values. Although algorithms like DEA-GNG and LMPFE have strategies for identifying Pareto front geometry, their results are still inferior to the proposed algorithm when facing these complex problems, demonstrating the effectiveness of the enhanced knowledge-sharing mechanism proposed in this invention in solving problems with complex Pareto fronts.

[0166] Table 2

[0167]

[0168] Table 2 (continued)

[0169]

[0170] Table 2 shows the HV values ​​for all algorithms. It is clear from the table that the proposed algorithm outperforms the others on most test functions. Specifically, the proposed algorithm performs excellently on all objectives of DTLZ3, while MaOEAIH and LMPFE have advantages on one and two test functions, respectively. DEA-GNG, RVEA-DWC, and hpaEA perform similarly on a similar number of test functions, while MOEA / DD-DQN achieves the best performance on eight test functions. Notably, the proposed algorithm performs best on DTLZ7 and IDTLZ2, while MOEA / DD-DQN performs best on DTLZ4, DTLZ5, and DTLZ6. In the remaining DTLZ1 and IDTLZ1 test problems, hpaEA and RVEA-DWC perform significantly better. Overall, the proposed algorithm performs strongly on DTLZ test problems. Detailed analysis of the HV values ​​shows that the proposed algorithm significantly outperforms other algorithms on DTLZ3, DTLZ7, and IDTLZ2, and also exhibits suboptimal results on most dimensions of DTLZ1, DTLZ2, and DTLZ4. The test problems include common Pareto fronts, such as convex, concave, and linear problems. Through the interaction of three proposed strategies, the proposed algorithm outperforms other comparative algorithms in solving these generalized problems.

[0171] To fully demonstrate the universality and effectiveness of the proposed algorithm in solving multi-objective optimization problems, performance metrics were used to rank the algorithms and conduct a comprehensive comparison. Figure 8 The graph shows the HV value trends of seven algorithms under different population iterations on the WFG9 test problem. As can be seen from the graph, the proposed algorithm (RKS-TAEA) quickly gained an advantage in the early iterations of the population and has maintained its lead since then.

[0172] In addition, such as Figures 4-5 As shown, in the MaF4 (3 targets) test problem, the solution set obtained by this invention can form a more uniform distribution on the front edge and cover the boundary region; as Figures 6-7 As shown, on the MaF13 (8 targets) test problem, this invention can still maintain good frontier coverage and distribution balance in the high-dimensional target space, demonstrating the effect of diverse archive update strategies on solution set expansion. Furthermore, as... Figure 12 As shown, with the increase of the number of iterations, the minimum normalization objective and the overall result show a decreasing trend and gradually stabilize, indicating that the algorithm maintains stable convergence behavior while continuously improving the quality of the solution set.

[0173] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dual-archive multi-objective optimization method based on reinforced knowledge sharing, characterized in that, Includes the following steps: S1. Generate an initial population based on integrated energy system data, and store the initial population in the convergence file and the diversity file respectively; S2. Calculate the knowledge sharing factor based on the non-dominated solution set in the aforementioned diversity archive; the knowledge sharing factor is used to adaptively calculate the generalized displacement density index and the new displacement index. S3. Based on the generalized displacement density index, perform convergence evaluation on the individuals in the convergence archive, and update the convergence archive based on the evaluation results; S4. Perform diversity assessment on individuals in the diversity profile based on the new displacement index, and update the diversity profile based on the assessment results; S5. Select the parent population of the scheduling scheme from the updated convergence file and the updated diversity file through the tournament selection mechanism, and generate the offspring population of the scheduling scheme through genetic operations. S6. Merge the child population of the scheduling scheme with the parent population of the scheduling scheme to obtain a merged population, and update the convergence file and the diversity file based on the merged population; S7. Determine whether the preset termination condition has been met. If it has, output the non-dominated solution set in the diversity file as the optimal scheduling scheme for the combined cooling, heating and power integrated energy system. Otherwise, return to step S2.

2. The dual-archive multi-objective optimization method based on reinforced knowledge sharing according to claim 1, characterized in that, In step S1, an initial population is generated based on the combined cooling, heating and power integrated energy system model and operating data. The data includes equipment capacity constraints, energy prices, load demand, and upper and lower bounds of decision variables. The decision variables include at least one of gas turbine output, energy storage device charging and discharging power, and refrigeration unit power.

3. The dual-archive multi-objective optimization method based on reinforced knowledge sharing according to claim 1, characterized in that, In S2, the formula for calculating the knowledge sharing factor is: ； Where M is the number of targets, Let be the average of all individuals on the i-th objective, and log() be the natural logarithm.

4. The dual-archive multi-objective optimization method based on reinforced knowledge sharing according to claim 1, characterized in that, In S3, the formula for calculating the generalized displacement density index is: ； in, For all that cannot be dominated by Pareto A collection of individuals For the target number, As a knowledge-sharing factor, For individuals In the Normalized values ​​on each objective For individuals The first one after mapping according to the shift rule One target value.

5. The dual-archive multi-objective optimization method based on reinforced knowledge sharing according to claim 1, characterized in that, In step S4, the process of assessing the diversity of individuals in the diversity profile based on the new displacement index and updating the diversity profile includes: The target vectors in the diversity archive are translated with the ideal point as the origin, and the most representative boundary individual is determined along the axial direction of each target. The determined boundary individuals are used as the initial elite subset. When the number of individuals in the diversity archive does not exceed the size limit, all individuals are directly used as the initial elite subset. The Minkowski distance from each individual not selected into the elite subset to the ideal point and the displacement difference distance of each individual relative to the current elite subset are calculated, and a new displacement index value is obtained based on the calculation results. A greedy selection strategy is adopted to select the individual with the highest new displacement index value from the remaining individuals at each step and incorporate it into the elite subset until the size of the elite subset reaches the size limit of the diversity archive, thus completing the update of the diversity archive.

6. The dual-archive multi-objective optimization method based on reinforced knowledge sharing according to claim 1, characterized in that, In S5, a tournament selection mechanism is used to select the parent population of the scheduling scheme from the updated convergence file and the updated diversity file, including: The individuals in the updated convergence file and the updated diversity file are merged to form a merge pool; several individuals are randomly selected from the merge pool, and the individual with the best index value is selected from the several individuals according to the generalized displacement density index or the new displacement index to enter the mating pool; the above selection process is repeated until the size of the mating pool reaches a preset value, and the parent population of the scheduling scheme is obtained.

7. The dual-archive multi-objective optimization method based on reinforced knowledge sharing according to claim 6, characterized in that, In step S5, the generation of the scheduling scheme's offspring population is performed through genetic operations, including: The parent individuals in the parent population of the scheduling scheme are paired up; for each pair of parent individuals, a simulated binary crossover operation is performed to generate two intermediate offspring individuals; for each intermediate offspring individual, a polynomial mutation operation is performed to perturb each of its decision variables with a preset mutation probability to obtain the offspring population of the scheduling scheme.

8. The dual-archive multi-objective optimization method based on reinforced knowledge sharing according to claim 1, characterized in that, In step S6, updating the convergence profile and the diversity profile based on the merged population includes: The merged population and individuals in the current convergence archive are sorted using a non-dominated method, and a preset number of individuals are selected from them based on the generalized displacement density index to update the convergence archive. At the same time, the merged population and individuals in the current diversity archive are sorted using a non-dominated method to obtain a non-dominated frontier. If the number of individuals in the non-dominated frontier does not exceed the capacity of the diversity archive, all individuals are retained; otherwise, a preset number of individuals are selected from them based on the new displacement index to update the diversity archive.

9. A dual-archive multi-objective optimization system based on reinforced knowledge sharing, used to implement the method described in any one of claims 1-8, characterized in that, include: An initialization module is used to generate an initial population based on integrated energy system data, and store the initial population in a convergence archive and a diversity archive, respectively. The knowledge sharing factor calculation module is used to calculate the knowledge sharing factor based on the non-dominated solution set in the diversity archive. The SDEp calculation and sorting module is used to evaluate and sort individuals in the converged archive based on the knowledge sharing factor and the generalized displacement density index. The SBI calculation and selection module is used to evaluate and iteratively select individuals in the diversity archive based on the knowledge-sharing factor and the new displacement index. The genetic operations module is used to select the parent population of the scheduling scheme from the updated convergence archive and diversity archive through a tournament selection mechanism, and to generate the offspring population of the scheduling scheme through genetic operations. The dual-file update module is used to merge the offspring population and the parent population of the scheduling scheme to obtain a merged population, and update the convergence file and the diversity file respectively based on the merged population; The termination judgment module is used to determine whether a preset termination condition has been met, and when the condition is met, it outputs the non-dominated solution set in the diversity file as an optimized scheduling scheme.

Images