Intelligent generation method of building energy-saving scheme based on two-stage agent-assisted evolution

Abstract

The application discloses a kind of building energy-saving scheme intelligent generation method based on two-stage agent-assisted evolution, belong to building energy-saving field, according to resident building model decision variable and range, generate uniform feasible solution with rejection sampling method, and the real target value obtained by EnergyPlus simulation is stored in sample set DB;Uniformly construct reference vector in target space and divide space, DB sample is divided into corresponding vector file according to the nearest distance, and the initial RBF global agent model is constructed.The first stage selects the crowding density minimum non-dominated solution from each vector file as the initial population, evaluates the individual using the global agent model, fills the solution according to the rules, updates the local agent model, and enters the second stage when the hyper volume increment value is greater than the threshold value, otherwise directly enters S5, and the second stage dynamically updates the reference vector according to the solution quality.The application uses the above method, by combining agent model and phased evolution strategy, significantly improves the efficiency and accuracy of building energy-saving optimization.

Description

Technical Field

[0001] This invention relates to the field of building energy conservation technology, and in particular to a method for intelligently generating building energy conservation solutions based on two-stage agent-assisted evolution. Background Technology

[0002] In practice, building energy-saving design is essentially a typical multi-objective optimization problem, which has many contradictory performance indicators such as building energy consumption and user discomfort.

[0003] Evolutionary algorithms (EAs) are a class of global search methods based on the principles of biological evolution. Currently, EEAs have demonstrated powerful performance in multi-objective optimization, constrained optimization, and dynamic optimization. Since the 1990s, the combination of building energy consumption simulation and multi-objective evolutionary optimization algorithms has gradually gained widespread attention from scholars, significantly reducing complex manual trial-and-error processes, improving the quality of optimal solutions, and achieving high-performance building design. Some methods, such as the one proposed by Asadi et al., use Chebyshev decomposition to optimize energy consumption, investment costs, and thermal comfort, and verify the feasibility of the proposed multi-objective model in real-world scenarios using an existing residential building requiring renovation as an example. Malatji et al. established a multi-objective optimization model, which is formulated as a multi-objective optimization problem constrained by net present value, initial investment, energy targets, and investment payback period, and solves it using a genetic algorithm to help decision-makers make optimal decisions when investing in energy-efficient building renovations. Han et al. designed a fast and convenient multi-objective optimization scheme for early planning problems in eco-cities and communities. This scheme uses building energy consumption, carbon emissions, and initial costs as three sub-objectives, assigning different weights to these sub-objectives according to the decision-maker's plan, and employing the BIN method to calculate building energy consumption. Harkouss et al. used the NSGA-II algorithm to minimize heat and electricity demand and life-cycle costs to improve the design of net-zero energy buildings, addressing the energy performance problems of specific buildings. Fesanghary et al. proposed a multi-objective optimization method using the Harmony Search (HS) algorithm, selecting several structural parameters as design variables. Bui et al., considering adaptive facade system design, studied a building energy-saving design algorithm based on multi-objective firefly intelligent optimization, which can effectively respond to changes in temperature environment. Ferdinando et al., considering the renovation problem of existing buildings, designed a building energy-saving design method integrating non-dominated sorting genetic algorithm and EnergyPlus, and applied it to the renovation of typical local residential buildings.

[0004] The evolutionary optimization methods described above have achieved good results in building energy-saving design. However, these algorithms still suffer from high operating costs because they require computationally expensive building energy management software to continuously evaluate the fitness values ​​of individual components. To address the problem of time-consuming evaluation costs, scholars began to study Surrogate-Assisted Evolutionary Algorithms (SAEAs) at the end of the 20th century. The idea behind SAEAs is to construct a surrogate model of the original problem based on historical solution data and use it to replace the computationally expensive original optimization problem to predict the objective or constraint values ​​of candidate solutions. Because they can significantly reduce the algorithm's running time, SAEAs have become an important method for solving building energy-saving optimization problems. For example, Magnier and Haghighat used TRNSYS simulations to train neural networks and coupled the trained neural networks with genetic algorithms to optimize thermal comfort and energy consumption. This method significantly shortened the optimization time for building energy-saving design problems. Chen et al. used a feedforward neural network with one hidden layer to identify temperature in smart buildings, presenting a method for designing building structural parameters based on particle swarm optimization. Khan et al. used the "moving least squares method" to build a surrogate model from an extremely expensive CFD simulation and used this surrogate model to optimize the design of a hospital ward ventilation system. Gengembre et al. combined the Kriging surrogate model and particle swarm optimization algorithm to optimize the life cycle cost of a single-area building model.

[0005] The methods mentioned above are mainly aimed at unconstrained building energy-saving design problems. However, in actual building energy-saving design, there are many constraints, such as restrictions on the ratio of window area to wall area, building orientation angle restrictions, and the coordination of window area in thermal zones. Therefore, the building energy-saving design problem is essentially an Expensive Constrained Multi-objective Optimization problem (ECMOPs). To solve ECMOPs, Li et al. proposed a data-driven CMOEA, which effectively balances the convergence, diversity, and feasibility of the algorithm by using a feasible region location strategy and a performance improvement exploration mechanism. Gu et al. proposed a random forest-assisted adaptive multi-objective particle swarm optimization algorithm. This algorithm uses a random forest model to accelerate the population search speed and uses an adaptive random ranking strategy and particle state update rules to balance the population search performance. Wang et al. proposed a surrogate-assisted global optimization algorithm, which uses a constraint improvement strategy and multiple indicators to build a two-layer selection framework. The adaptive selection framework dynamically adjusts the search direction of the population, making the population closer to the feasible region. Long et al. proposed a data-driven co-evolutionary algorithm, constructing a co-evolutionary exploration framework and improving the accuracy of the surrogate model through an adaptive selection strategy. Sun et al. proposed a constraint-hybrid problem optimization algorithm based on a multi-objective adaptive surrogate model, using multiple evolutionary operators to handle different types of decision variables and constructing a constraint model with a classification surrogate model, optimizing the solution generation process through a Kriging model. While these methods demonstrate good performance in solving ECMOPs, they have not been applied to building energy efficiency optimization problems. Furthermore, many studies indicate that this type of problem has irregular power factor (PF), a factor rarely considered in existing methods. Therefore, directly applying existing ECMOPs solutions to building energy efficiency optimization design problems may result in suboptimal design solutions. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent generation method for building energy-saving solutions based on two-stage agent-assisted evolution, so as to solve the problems existing in the background technology.

[0007] To achieve the above objectives, this invention provides an intelligent generation method for building energy-saving solutions based on two-stage agent-assisted evolution, comprising the following steps:

[0008] S1. Based on the decision variables and their value range of the residential building model, use the rejection-based sampling method to generate several uniformly distributed feasible solutions, and input the solutions into the constructed EnergyPlus simulation model to obtain the true target value of the solution, and put it into the sample set DB.

[0009] S2. Construct N uniformly in the target space. W The target space is divided by a reference vector; the initial sample in DB is assigned to the reference vector in the target space that is closest to the sample, forming the vector file V of that vector. i And use the samples in DB to construct an initial RBF global proxy model with M objective functions;

[0010] S3. Enter the first stage. Select the non-dominated solution with the smallest crowding density from each vector file as the initial population. Then, execute the evolution operator based on the fixed vector. Use the global surrogate model to evaluate individuals and obtain the target prediction value of the offspring. Select the filling solution according to the model management method of the first stage. Update the local surrogate model according to the current sample. Calculate the hypervolume increment ΔHV value of the current sample every h generations. When ΔHV is greater than the threshold α, the algorithm enters S4. Otherwise, it directly enters S5.

[0011] S4. Enter the second stage. Dynamically update the reference vector based on the quality of the solution obtained in each generation. Construct an RBF local proxy model for each reference vector using samples in each vector file. After constructing the local proxy model, execute the vector classification strategy and execute the adaptive selection mechanism of the search strategy based on the classification results. After obtaining the offspring according to the selected search strategy, select the filling solution according to the second stage model management strategy, dynamically update the reference vector based on the quality of the solution obtained in each generation, and update the local proxy model based on the current sample.

[0012] S5. Output several non-dominated feasible solutions.

[0013] Preferably, the steps for partitioning the target space using reference vectors and constructing the surrogate model in S2 are as follows:

[0014] S21. Using uniformly distributed fixed reference vectors to guide population evolution aims to overcome infeasible regions caused by constraints, ensuring the population covers the feasible solution space as early as possible and gradually approaches the Pareto front. Specifically, uniformly distributed reference vectors are generated in the M-dimensional target space, dividing the target space into several sub-regions. Each reference vector corresponds to a sub-space, and the population evolves based on these fixed reference vectors.

[0015] S22. Based on the sample information in the database, radial basis functions are used to obtain fitting functions for the building's annual energy consumption and the number of hours of user discomfort, which are then defined as surrogate models. In subsequent evolutionary processes, when a new solution is generated, this surrogate model can be used to predict the predicted values ​​of the building's annual energy consumption and the number of hours of user discomfort for the new solution.

[0016] Preferably, step S3 is as follows:

[0017] S31. Based on the individual assigned to each vector, calculate the APD value for each individual, and obtain the non-dominated solution based on the obtained APD value;

[0018] S32. Using the non-dominated solutions as the initial population and the surrogate model obtained in S22 as the objective function, perform the RVEA evolutionary operation to obtain multiple new individual solutions.

[0019] S33. Calculate the constraint violation values ​​of all newly generated solutions, select the NM individuals with the smallest constraint violation values, and use the surrogate model to predict the objective function values ​​of these individuals. Select the individual that contributes the most to the improvement of HV as the filling solution, evaluate the objective function value of the filling solution, and update the surrogate model.

[0020] Preferably, S31 contains the following:

[0021] RVEA uses the Angle Penalized Distance (APD) to select suitable individuals for each reference vector; for the current reference vector, the smaller the APD value of an individual, the higher its probability of being selected; the specific expression of APD is as follows:

[0022] APd j =(1+Pe(θ) j ))·||f j || (1)

[0023] Among them, ||f j || is the distance from the j-th individual in the target space to the origin; ||f j The smaller the value of θ, the better the convergence of the individual. j Pe(θ) is the angle between the j-th individual and its assigned reference vector. j ) is the penalty function.

[0024] Preferably, S33 contains the following:

[0025] For a candidate solution set HS, suppose its corresponding reference point is... Individual x i The objective function value of ∈HS is (f1(x) i ),f2(x i ),...,f m (x i If the individual's HV contribution relative to the reference point Z is:

[0026]

[0027] in, Let Z represent the maximum point of the objective function in the j-th dimension of set HS; set the reference point Z as... 1.1 times.

[0028] Preferably, the content of S4 is as follows:

[0029] S41. Dynamically update the reference vector based on the quality of the solutions obtained in each generation. The goal of the second stage is to accurately locate the true power field (PF) of the problem and find more solutions located on the PF. Based on the feasible solution information obtained in the first stage, the population is further guided to evolve towards the most promising feasible region by dynamically adjusting the reference vector. In this process, in order to effectively utilize computing resources, the algorithm will perform a fine search on the explored feasible region, add new reference vectors in the region to enhance the search density of the algorithm in that region; update the vectors of those explored as infeasible regions and delete the reference vectors corresponding to infeasible regions.

[0030] S42. Based on the newly generated vectors, allocate the existing building schemes in DB to obtain new vector files, and use the building schemes in each vector file to construct an RBF local proxy model for each reference vector.

[0031] S43. Introduce a search strategy adaptive selection mechanism driven by reference vector type, classify each reference vector, and adaptively select the next generation population update mechanism according to the obtained type.

[0032] S44. Randomly obtain an initial population, and predict newly generated individuals based on the search strategy selected in S43 and the surrogate model obtained in S42.

[0033] Preferably, the reference vector update strategy in S41 is as follows:

[0034] For any reference vector V i and the corresponding vector file VS i If VS within h iterations i If no feasible solution is found in the reference vector set, then V is removed from the reference vector set. i If VS is within h iterations i If the improvement of the non-dominated solutions is always less than the threshold ε, then the reference vector is adjusted according to the position of the superior solutions in the population, and equation (3) is used in V. i With reference vector V j Generate a new reference vector V between them k :

[0035]

[0036] Among them, V j To be with adjacent V i It also contains a reference vector with a relatively large number of feasible solutions; λ is an interpolation coefficient;

[0037] At the same time, VS is redistributed based on the distance between the individual and the reference vector. i and VS j If VS i The number of feasible nondominated solutions exceeds a preset threshold N. th Calculate V i Two adjacent reference vectors (let's say V) i-1 and V i+1 The number of nondominated feasible solutions is calculated, and the one with the fewest feasible solutions is selected, let's say it's V. i+1 Subsequently, a new reference vector is generated according to equation (3), and non-dominated feasible solutions are redistributed.

[0038] Preferably, S43 contains the following:

[0039] For any reference vector V i and its corresponding files VS i The method for calculating the correlation between the constraint violation value minimization direction and the objective optimization direction is as follows:

[0040] First, use equation (4) to calculate the file VS i The constraint violation value of each individual is calculated, and all individuals are sorted from smallest to largest according to the constraint violation value to obtain the ranking rank CVN(x) of each individual. j Simultaneously, calculate VS. i Average rank of all individuals

[0041] Subsequently, based on the objective function value, the VS... i The individuals in the dataset are non-dominated and ordered, meaning that the lower the order, the better the objective optimization performance of the solution. The non-dominated order rank MN(x) of each individual is obtained. i and average rank

[0042] Finally, the correlation coefficient ρ between the constraint violation direction and the target convergence direction is calculated according to equation (5). i .

[0043]

[0044] Where n is the vector file VS i The number of candidate solutions; g i (x) represents the constraint value of the solution x;

[0045] Based on formula (5), all N can be obtained. W The correlation coefficient ρ corresponding to the reference vector i i = 1, 2, ..., N w For any reference vector, if its correlation coefficient ρ iIf the value is greater than μ, it is considered that the target optimization direction of the reference vector is consistent with the direction of minimizing the constraint violation value, and the population has not yet entered the optimal feasible region. Individuals can continue to search along the optimization direction of the reference vector. Therefore, individuals are selected from the vector file in non-dominated sorting as the initial population, and then the DE operator is executed for exploration.

[0046] If ρ i If the value is less than -μ, then the objective optimization direction of the reference vector is considered to be opposite to the direction of minimizing the constraint violation value. In this case, the individuals contained in the vector may have crossed the feasible region and reached the infeasible region. Therefore, the exploration position is readjusted using a local search method. The feasible solution with the optimal objective value in the vector file is taken as the current best point, and subsequent local search strategies are executed.

[0047] If -μ < ρ i If μ < μ, then the objective optimization direction of the reference vector is considered weakly correlated with the direction of minimizing the constraint violation value, and the population on this vector has not yet reached the constraint feasible region. For such vectors, flexible search strategies can be adopted, so the DE operator or subsequent local search operators can be executed randomly.

[0048] Preferably, S44 contains the following:

[0049] According to the S43 vector classification and search adaptive mechanism, each vector file obtains corresponding candidate solutions, which are stored in HS. For positively correlated vectors, the DE operator is executed to obtain candidate solutions. All candidate solutions are sorted according to the surrogate model's predicted values, and the N solutions with the best predicted values ​​are selected. Their constraint violation values ​​are calculated, and the solution with the smallest constraint violation value is selected as the filling solution. For negatively correlated vectors, the local search operator TR-LS is executed to obtain a new individual x. c Then, calculate the constraint violation value for that individual. If the individual is a feasible solution, use a surrogate model to calculate x. c The predicted value, if If the solution is better than the one in the existing archives, then choose x. c As filler samples, a true evaluation is performed. For weakly correlated vectors, the constraint violation value of each newly generated candidate solution is calculated, and the solutions with smaller constraint violation values ​​are saved. The target value of the saved candidate solutions is evaluated using a surrogate model, and the solution with the best predicted target value is selected as the filler sample to update the vector archive and its corresponding surrogate model.

[0050] Therefore, the intelligent generation method for building energy-saving solutions based on two-stage agent-assisted evolution, as described above, has the following beneficial effects:

[0051] (1) A two-stage evolution strategy is adopted. The population is guided by a uniformly distributed fixed reference vector, which can quickly cover the feasible solution space and approach the Pareto front (PF), cross the infeasible region caused by constraints, ensure the diversity of the initial population, encrypt the search of feasible regions, delete infeasible region vectors, accurately locate the real PF, improve the quality and density of solutions, and the phased strategy balances global exploration and local development, avoiding traditional algorithms from getting trapped in local optima too early.

[0052] (2) The adaptive mechanism enables the algorithm to dynamically switch strategies according to the population position, adaptively switching DE / local search between positive / negative / weakly correlated vectors to solve the problem of population deviation from the feasible region and effectively handle complex PF structures.

[0053] (3) For engineering constraints in building design (such as window-to-wall ratio, orientation angle, heat transfer coefficient combination, etc.), individuals are screened by constraint violation value, and feasible solutions are retained first to improve the engineering practicality of the solution; at the same time, the two contradictory objectives of building annual energy consumption (BEC) and user discomfort hours (UDH) are optimized, and diverse choices are provided through Pareto front solution set to support decision-makers in balancing energy consumption and comfort.

[0054] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the intelligent generation method for building energy-saving solutions based on two-stage agent-assisted evolution according to the present invention.

[0056] Figure 2 This is a flowchart illustrating the overall process of the intelligent generation method for building energy-saving solutions based on two-stage agent-assisted evolution, as described in this invention.

[0057] Figure 3 This is a second-stage vector update in an embodiment of the present invention, wherein (a) represents a schematic diagram of the vector set after the previous iteration; and (b) represents a schematic diagram after re-sorting and renaming the vectors according to their target values.

[0058] Figure 4 This illustrates the relationship between population location and feasible region in an embodiment of the present invention.

[0059] Figure 5 This is a schematic diagram of the exterior of a residential building in an embodiment of the present invention;

[0060] Figure 6 This is a schematic diagram of the solution framework for building energy conservation problems according to an embodiment of the present invention;

[0061] Figure 7 This is a schematic diagram of the PF curves obtained by the eight algorithms in the embodiments of the present invention on the problem of building energy-saving design. Detailed Implementation

[0062] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0063] This embodiment uses EnergyPlus simulation software to describe the mathematical model of the problem. Nineteen parameters are used as decision variables in this model, including building orientation, window length and width in each thermal zone, window heat transfer coefficient and solar heat gain coefficient, wall insulation layer thickness, external wall solar absorptivity, lighting power density, and heating and cooling system set temperatures. The building energy efficiency optimization model can be described as follows:

[0064] minF = (BEC(X), UDE(X))

[0065] stX=(x or ,x tolw ,x srar ,x ghtc ,x shgc ,x lwl ,x lww ,x bwl ,x bww ,x kwl (6)x kww ,x wwl ,x www ,x lepd ,x bepd ,x kepd ,x wepd ,x hst ,x cst )

[0066] Where BEC is the building’s annual energy consumption and UDH is the number of hours of user discomfort.

[0067] Based on the above description, the five cost constraints to be considered in this embodiment are expressed as follows:

[0068] Constraint 1: Limitation on the ratio of window area to wall area:

[0069] Engineering Basis: According to the "Energy Conservation Design Standard for Civil Buildings" (JGJ26-2018), the upper limit for the window-to-wall ratio of residential buildings in each orientation is: North-facing ≤25%, South-facing ≤35%, East-West-facing ≤30%. Mathematical Expression: Window parameter x for each thermal zone (living room, bedroom, etc.) lwl ,x lww (Length, width) and corresponding wall area A wall The constraints are:

[0070]

[0071] Where γ is the orientation-related threshold (e.g., γ = 0.35 for southward direction).

[0072] Constraint 2: Building orientation angle restrictions:

[0073] Engineering Basis: Based on the "Beijing Municipal Standard for Energy-Saving Design of Residential Buildings"

[0074] (DB11 / 891-2020) To improve the efficiency of solar energy utilization in winter, the orientation of the main facade should be controlled within 15° east / west of south.

[0075] The constraints are specifically stated as follows:

[0076] g2(x)=∣x or -180° | -15° ≤ 0 (due south is 180°) (8)

[0077] Constraint 3: Combined constraint of window heat transfer coefficient and solar heat gain coefficient:

[0078] Engineering basis: Referencing the "Energy-saving performance standard for building doors and windows" (GB / T 8484-2020), windows with high heat transfer coefficients need to be matched with low solar heat gain coefficients to balance energy consumption.

[0079] The constraint is specifically expressed as (x ghtc x is the heat transfer coefficient. shgc (Solar heat gain coefficient):

[0080]

[0081] Constraint 4: Thermal zone window area compatibility constraint:

[0082] Engineering Basis: According to the "Standard for Daylighting Design of Buildings" (GB 50033-2013), the different daylighting needs of different functional areas of residential buildings need to be coordinated and balanced through window area to avoid excessive lighting in the living room leading to insufficient lighting in the bedroom.

[0083] Mathematical expression:

[0084]

[0085] Function: To prevent energy consumption compensation caused by unbalanced lighting (such as turning on extra lights in the bedroom), and to reduce invalid layouts that require simulation verification by approximately 22%.

[0086] Constraint 5: Heat transfer-heat gain product constraint:

[0087] Engineering basis: The principle of thermal performance balance requires that windows with high heat transfer coefficients be matched with low solar heat gain coefficients to reduce the fluctuation of total energy consumption throughout the year.

[0088] Mathematical expression:

[0089] g4 = x ghtc ·x shgc -0.9≤0 (11)

[0090] Function: Eliminates the combined degradation of heat transfer and heat gain (such as ordinary single-layer glass), improving screening efficiency by 31%.

[0091] The variable value range for question (6) is shown in Table 1. Some initial parameters for this question are set as follows: the initial length and width of the windows are 1.8m and 1.2m, respectively. Referring to the recommendations in the EnergyPlus software manual, their return air volume coefficient is set to 0, the radiation coefficient is set to 0.37, the visible light coefficient is set to 0.18, and the coefficient for heat transfer from light to the regional air is set to 0.40.

[0092] Table 1. Decision variables and their value ranges for the residential building model.

[0093]

[0094]

[0095] Please see Figure 1 The intelligent generation method for building energy-saving solutions based on two-stage agent-assisted evolution includes the following steps:

[0096] S1. Based on the decision variables and their value range of the residential building model, use the rejection-based sampling method to generate several uniformly distributed feasible solutions, and input the solutions into the constructed EnergyPlus simulation model to obtain the true target value of the solution, and put it into the sample set DB.

[0097] S2. Construct N uniformly in the target space. W The target space is divided by a reference vector; the initial sample in DB is assigned to the reference vector in the target space that is closest to the sample, forming the vector file V of that vector. i And use the samples in DB to construct an initial RBF global proxy model with M objective functions.

[0098] S21. Using uniformly distributed fixed reference vectors to guide population evolution aims to overcome infeasible regions caused by constraints, ensuring the population covers the feasible solution space as early as possible and gradually approaches the Pareto front. Specifically, uniformly distributed reference vectors are generated in the M-dimensional objective space, dividing the objective space into several sub-regions. Each reference vector corresponds to a sub-space, and the population evolves based on these fixed reference vectors. Taking a two-objective optimization problem as an example... Figure 2 A schematic diagram of the individual assignment of the reference vector is shown.

[0099] S3. Enter the first stage. Select the non-dominated solution with the smallest crowding density from each vector file as the initial population. Then, execute the evolutionary operator based on the fixed vector. Use the global surrogate model to evaluate individuals and obtain the target prediction value of the offspring. Select the filling solution according to the model management method of the first stage. Update the local surrogate model according to the current sample. Calculate the hypervolume increment ΔHV value of the current sample every h generations. When ΔHV is greater than the threshold α, the algorithm enters S4. Otherwise, it directly enters S5.

[0100] S31. Based on the individuals assigned to each vector, calculate the APD value for each individual, and obtain the non-dominated solution based on the obtained APD value.

[0101] RVEA uses the Angle Penalized Distance (APD) to select suitable individuals for each reference vector; for the current reference vector, the smaller the APD value of an individual, the higher its probability of being selected; the specific expression of APD is as follows:

[0102] APd j =(1+Pe(θ) j ))·||f j || (1)

[0103] Among them, ||f j || is the distance from the j-th individual in the target space to the origin; ||f j The smaller the value of θ, the better the convergence of the individual. j Pe(θ) is the angle between the j-th individual and its assigned reference vector. j ) is the penalty function.

[0104] S32. Using the non-dominated solutions as the initial population and the surrogate model obtained in S22 as the objective function, perform the RVEA evolution operation to obtain multiple new individual solutions.

[0105] S33. Calculate the constraint violation values ​​of all newly generated solutions, select the NM individuals with the smallest constraint violation values, and use the surrogate model to predict the objective function values ​​of these individuals. Select the individual that contributes the most to the improvement of HV as the filling solution, evaluate the objective function value of the filling solution, and update the surrogate model.

[0106] For a candidate solution set HS, suppose its corresponding reference point is... Individual x i The objective function value of ∈HS is (f1(x) i ),f2(x i ),...,f m (x i If the individual's HV contribution relative to the reference point Z is:

[0107]

[0108] in, Let Z represent the maximum point of the objective function in the j-th dimension of set HS; set the reference point Z as... 1.1 times.

[0109] During algorithm execution, the hypervolume increment ΔHV value of the current sample is calculated every h iterations. When ΔHV is less than or greater than the threshold α, it indicates that the algorithm needs to explore the region with added constraints. At this point, the algorithm executes S4.

[0110] S4. Enter the second stage. Dynamically update the reference vector based on the quality of the solution obtained in each generation. Construct an RBF local proxy model for each reference vector using samples in each vector file. After constructing the local proxy model, execute the vector classification strategy and perform an adaptive selection mechanism for the search strategy based on the classification results. After obtaining offspring based on the selected search strategy, select the filling solution according to the second-stage model management strategy. Dynamically update the reference vector based on the quality of the solution obtained in each generation and update the local proxy model based on the current sample.

[0111] S41. Dynamically update the reference vector based on the quality of the solutions obtained in each generation. The goal of the second stage is to accurately locate the true power field (PF) of the problem and find more solutions located on the PF. Based on the feasible solution information obtained in the first stage, the population is further guided to evolve towards the most promising feasible region by dynamically adjusting the reference vector. In this process, in order to effectively utilize computing resources, the algorithm will perform a fine search on the explored feasible region, adding new reference vectors in the region to enhance the search density of the algorithm in that region; the vectors of those explored as infeasible regions will be updated, and the reference vectors corresponding to infeasible regions will be deleted.

[0112] The reference vector update strategy is as follows:

[0113] For any reference vector V i and the corresponding vector file VS i If VS within h iterations i If no feasible solution is found in the reference vector set, then V is removed from the reference vector set. i If VS is within h iterations i If the improvement of the non-dominated solutions is always less than the threshold ε, then the reference vector is adjusted according to the position of the superior solutions in the population, and equation (3) is used in V. i With reference vector V j Generate a new reference vector V between them k :

[0114]

[0115] Among them, V j To be with adjacent V i It also contains a reference vector with a relatively large number of feasible solutions; λ is an interpolation coefficient;

[0116] At the same time, VS is redistributed based on the distance between the individual and the reference vector. i and VS j If VS i The number of feasible nondominated solutions exceeds a preset threshold N. th Calculate V i Two adjacent reference vectors (let's say V) i-1 and V i+1 The number of nondominated feasible solutions is calculated, and the one with the fewest feasible solutions is selected, let's say it's V. i+1 Subsequently, a new reference vector is generated according to equation (3), and non-dominated feasible solutions are redistributed.

[0117] Figure 3 This demonstrates a specific example of reference vector updating, where, Figure 3 (a) V i (i = 1, 2, 3, 4) is the vector set after the previous iteration. It can be seen that the density of feasible non-dominated solutions in V2 exceeds the set threshold N. th Meanwhile, among its neighboring vectors V1 and V3, V1 has fewer feasible solutions. Therefore, a new reference vector V is generated based on V1. new V new After adding them to the reference vector set, re-sort and name them according to the target value of the vectors, such as... Figure 3 As shown in (b).

[0118] S42. Based on the newly generated vectors, allocate the existing building schemes in DB to obtain new vector files, and use the building schemes in each vector file to construct an RBF local proxy model for each reference vector.

[0119] S43. Introduce a search strategy adaptive selection mechanism driven by reference vector type, classify each reference vector, and adaptively select the next generation population update mechanism based on the obtained type.

[0120] The relationship between the optimal Pareto front (UPF) of an unconstrained optimization problem and the optimal Pareto front (CPF) of a constrained optimization problem is complex. For example... Figure 4 As shown, the algorithm population may fall into the following situations: ① The population has crossed the optimal feasible region and entered the infeasible region; ② The population is currently in the optimal feasible region; ③ The population has not yet entered the optimal feasible region. For these three situations, evolutionary algorithms based on fixed search strategies struggle to accurately guide the population's search direction. Therefore, this embodiment proposes a reference vector type-driven adaptive selection mechanism for search strategies to adaptively determine the evolutionary search strategy adopted by each individual in the population. This mechanism classifies reference vectors by analyzing the correlation between the constraint violation minimization direction and the objective optimization direction, and applies different evolutionary search strategies to different types of reference vectors to effectively guide the population towards CPF convergence.

[0121] For any reference vector V i and its corresponding files VS i The method for calculating the correlation between the constraint violation value minimization direction and the objective optimization direction is as follows:

[0122] First, use equation (4) to calculate the file VS i The constraint violation value of each individual is calculated, and all individuals are sorted from smallest to largest according to the constraint violation value to obtain the ranking rank CVN(x) of each individual. j Simultaneously, calculate VS. i The average rank CVN(x) of all individuals in the dataset;

[0123] Subsequently, based on the objective function value, the VS... i The individuals in the dataset are non-dominated and ordered, meaning that the lower the order, the better the objective optimization performance of the solution. The non-dominated order rank MN(x) of each individual is obtained. i and the average rank MN(x);

[0124] Finally, the correlation coefficient ρ between the constraint violation direction and the target convergence direction is calculated according to equation (5). i .

[0125]

[0126] Where n is the vector file VS i The number of candidate solutions; g i (x) represents the constraint value of the solution x;

[0127] Based on formula (5), all N can be obtained. W The correlation coefficient ρ corresponding to the reference vector i i = 1, 2, ..., N w For any reference vector, if its correlation coefficient ρ i If the value is greater than μ, it is considered that the target optimization direction of the reference vector is consistent with the direction of minimizing the constraint violation value, and the population has not yet entered the optimal feasible region. Individuals can continue to search along the optimization direction of the reference vector. Therefore, individuals are selected from the vector file in non-dominated sorting as the initial population, and then the DE operator is executed for exploration.

[0128] If ρ i If the value is less than -μ, then the objective optimization direction of the reference vector is considered to be opposite to the direction of minimizing the constraint violation value. In this case, the individuals contained in the vector may have crossed the feasible region and reached the infeasible region. Therefore, the exploration position is readjusted using a local search method. The feasible solution with the optimal objective value in the vector file is taken as the current best point, and subsequent local search strategies are executed.

[0129] If -μ < ρ i If μ < μ, then the objective optimization direction of the reference vector is considered weakly correlated with the direction of minimizing the constraint violation value, and the population on this vector has not yet reached the constraint feasible region. For such vectors, flexible search strategies can be adopted, so the DE operator or subsequent local search operators can be executed randomly.

[0130] This embodiment uses surrogate-assisted Trust Region-Based Local Search (TR-LS) as a local search operator to construct an approximate surrogate model within the local trust region to improve the algorithm's ability to handle discontinuous power fields (PF).

[0131] S44. Randomly obtain an initial population. Based on the search strategy selected in S43, and using the surrogate model obtained in S42, predict newly generated individuals. Perform vector classification and search adaptation mechanisms according to S43. For each vector file, obtain corresponding candidate solutions and store them in HS. For positively correlated vectors, execute the DE operator to obtain candidate solutions. Sort all candidate solutions according to the surrogate model's prediction values, select the N solutions with the best prediction values, calculate their constraint violation values, and select the solution with the smallest constraint violation value as the filling solution. For negatively correlated vectors, execute the local search operator TR-LS to obtain the new individual x. c Then, calculate the constraint violation value for that individual. If the individual is a feasible solution, use a surrogate model to calculate x. c The predicted value, if If the solution is better than the one in the existing archives, then choose x. cAs filler samples, a true evaluation is performed. For weakly correlated vectors, the constraint violation value of each newly generated candidate solution is calculated, and the solutions with smaller constraint violation values ​​are saved. The target value of the saved candidate solutions is evaluated using a surrogate model, and the solution with the best predicted target value is selected as the filler sample to update the vector archive and its corresponding surrogate model.

[0132] S5. Output several non-dominated feasible solutions.

[0133] Example:

[0134] Taking rural residential building design in a certain region as an example, Figure 5 The basic shape of a single-room residential building is given. Based on different functions, the residential building is divided into four thermal zones: living room, bedroom, kitchen, and bathroom.

[0135] Figure 6 A framework diagram for this type of problem in this example is provided. Similar to compressor aerodynamic optimization design, the framework is divided into three modules: evolutionary optimization module, surrogate model construction module, and simulation module. In the simulation module, when it is necessary to evaluate a new sample point or candidate solution, the SketchUP building model parameters are first modified according to the sample input, and then, combined with the weather problem, the target value is obtained by substituting it into the EnergyPlus simulation software.

[0136] In this example, some initial parameters for this problem are set as follows: the initial length and width of the windows are 1.8m and 1.2m, respectively. Referring to the recommendations in the EnergyPlus software manual, their return air volume coefficient is set to 0, the radiation coefficient is 0.37, the visible light coefficient is 0.18, and the coefficient for heat transfer from light to the surrounding air is 0.40.

[0137] To verify the effectiveness, seven surrogate-assisted multi-objective evolutionary algorithms, namely ABSAEA, CSEA, EDN-ARMOEAD, HeEMOEAD, MCEAD, ParEGO, and REMO, were used as comparison algorithms.

[0138] Table 2. HV values ​​obtained by eight methods in building energy efficiency design.

[0139]

[0140] Table 2 shows the HV values ​​obtained by the eight algorithms. It can be seen that the HV value obtained in this embodiment is significantly higher than that of the seven selected comparison algorithms, and the variance of the results obtained is also smaller than that of the seven comparison algorithms. Furthermore, Figure 7The power factor (PF) curves obtained by eight algorithms for building energy efficiency design problems are presented. It can be seen that although the optimal solution of VD-SAMOEA is partially dominated by the solution obtained by ABSAEA in some regions, the result of VD-SAMOEA dominates the results of the other six comparative algorithms in most regions. In summary, the proposed VD-SAMOEA algorithm can obtain a better non-dominated solution set within a limited number of real evaluations. Compared with existing methods, VD-SAMOEA is more competitive in solving building energy efficiency design problems with complex PF. In conclusion, when solving building energy efficiency design problems with cost constraints, the algorithm in this embodiment can obtain a better non-dominated solution set within a limited number of real evaluations.

[0141] Therefore, the present invention adopts the above-mentioned intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution. By combining the agent model and the phased evolution strategy, the efficiency and accuracy of building energy-saving optimization are significantly improved.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligently generating building energy-saving solutions based on two-stage agent-assisted evolution, characterized in that, Includes the following steps: S1. Based on the decision variables and their value range of the residential building model, use the rejection-based sampling method to generate several uniformly distributed feasible solutions, and input the solutions into the constructed EnergyPlus simulation model to obtain the true target value of the solution, and put it into the sample set DB. S2. Construct uniformly in the target space The target space is divided by a reference vector; the initial sample in the DB is assigned to the reference vector in the target space that is closest to the sample, thus forming the vector file of that vector. And use the samples in DB to construct initial RBF global proxy models with M objective functions; the steps of partitioning the objective space and constructing the proxy models using reference vectors are as follows: S21. Using a uniformly distributed fixed reference vector to guide the population evolution, crossing the infeasible region caused by the constraints, so that the population covers the feasible solution space and gradually approaches the Pareto front. S22. Based on the sample information in DB, use radial basis functions to obtain fitting functions for building annual energy consumption and user discomfort hours, and define them as surrogate models. S3. Entering the first stage, select the non-dominated solution with the lowest crowding density from each vector file as the initial population. Then, execute the evolutionary operator based on the fixed vector, use the global surrogate model to evaluate individuals and obtain the target predicted value of the offspring, select filling solutions according to the model management method of the first stage, and update the local surrogate model according to the current sample. Calculate the hypervolume increment of the current sample every h generations. Value, when Greater than the threshold Then, the algorithm proceeds to S4; otherwise, it proceeds directly to S5. S4. Enter the second stage. Dynamically update the reference vector based on the quality of the solution obtained in each generation. Construct an RBF local proxy model for each reference vector using samples in each vector file. After constructing the local proxy model, execute the vector classification strategy and execute the adaptive selection mechanism of the search strategy based on the classification results. After obtaining the offspring according to the selected search strategy, select the filling solution according to the second stage model management strategy, dynamically update the reference vector based on the quality of the solution obtained in each generation, and update the local proxy model based on the current sample. S5. Output several non-dominated feasible solutions.

2. The intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution as described in claim 1, characterized in that, The steps for S3 are as follows: S31. Based on the individual assigned to each vector, calculate the APD value for each individual, and obtain the non-dominated solution based on the obtained APD value; S32. Using the non-dominated solutions as the initial population and the surrogate model obtained in S22 as the objective function, perform the RVEA evolution operation to obtain several new individual solutions; S33. Calculate the constraint violation values ​​of all newly generated solutions, select the NM individuals with the smallest constraint violation values, and use a surrogate model to predict the objective function values ​​of these individuals. The individual that contributes the most is selected as the filler solution, the objective function value of the filler solution is evaluated, and the surrogate model is updated.

3. The intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution as described in claim 2, characterized in that, S31 contains the following: RVEA uses penalized angular distance (APD) to select the corresponding individual for each reference vector; for the current reference vector, the smaller the APD value of an individual, the higher its probability of being selected; the specific expression of APD is as follows: (1) in, To be in the target space j The distance of each individual from the origin; It is the first j The angle between each individual and its assigned reference vector; Let be the penalty function.

4. The intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution as described in claim 3, characterized in that, S33 contains the following: For a candidate solution set HS, suppose its corresponding reference point is... ,individual The objective function value is Then the individual is relative to the reference point. of Contribution value: (2) in, It means that set HS is in the th order. j The maximum point of the objective function; the reference point Set as 1.1 times.

5. The intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution as described in claim 4, characterized in that, The content of S4 is as follows: S41. In the second stage, the reference vector is dynamically updated based on the quality of the solution obtained in each generation. Based on the feasible solution information obtained in the first stage, the population is guided to evolve towards the most promising feasible region by dynamically adjusting the reference vector of the population. S42. Based on the newly generated vectors, allocate the existing building schemes in DB to obtain new vector files, and use the building schemes in each vector file to construct an RBF local proxy model for each reference vector. S43. Introduce a search strategy adaptive selection mechanism driven by reference vector type, classify each reference vector, and adaptively select the next generation population update mechanism according to the obtained type. S44. Randomly obtain an initial population, and predict newly generated individuals based on the search strategy selected in S43 and the surrogate model obtained in S42.

6. The intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution as described in claim 5, characterized in that, The reference vector update strategy in S41 is as follows: For any reference vector and the corresponding vector files If within h iterations If no feasible solution is found in the reference vector set, it is removed from the reference vector set. If within h iterations The improvement of the included non-dominated solutions has always been less than the threshold. Then, the reference vector is adjusted according to the position of the high-quality solution in the population, and equation (3) is used in... With reference vector Generate new reference vectors between them : (3) in, To be adjacent It also contains reference vectors with a large number of feasible solutions; It is an interpolation coefficient; Simultaneously, the data is redistributed based on the distance between the individual and the reference vector. and ;if The number of feasible nondominated solutions exceeds a preset threshold. ,calculate two adjacent reference vectors and The number of non-dominated feasible solutions is calculated, and the one with the smallest number of feasible solutions is selected, let's say it is... Subsequently, a new reference vector is generated according to equation (3), and non-dominated feasible solutions are redistributed.

7. The intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution as described in claim 5, characterized in that, S43 contains the following: For any reference vector and its corresponding archives The method for calculating the correlation between the constraint violation value minimization direction and the objective optimization direction is as follows: First, calculate the archive using equation (4). The constraint violation value of each individual is calculated, and all individuals are sorted from smallest to largest according to their constraint violation values ​​to obtain the ranking rank of each individual. Simultaneously, calculation Average rank of all individuals ; Subsequently, based on the objective function value, for Perform a non-dominated ranking on the individuals in the dataset to obtain the non-dominated ranking rank of each individual. and average rank ; Finally, the correlation coefficient between the constraint violation direction and the target convergence direction is calculated according to equation (5). : (4) (5) Where n is the vector file The number of candidate solutions; Representing the solution x The constraint value; Based on formula (5), all are obtained. Correlation coefficient corresponding to the reference vector For any reference vector, if its correlation coefficient If the target optimization direction of the reference vector is consistent with the direction of minimizing the constraint violation value, the population has not yet entered the optimal feasible region, and the individuals continue to search along the optimization direction of the reference vector; individuals are selected from the vector file in non-dominated sorting as the initial population, and then the DE operator is executed to explore; if If the direction of the objective optimization of the reference vector is opposite to the direction of minimizing the constraint violation value, then the individual contained in the vector has crossed the feasible region of the constraint and reached the infeasible region. The exploration position is readjusted through the local search method. The feasible solution with the best objective value in the vector file is taken as the current best point, and the subsequent local search strategy is executed. if If the objective optimization direction of the reference vector is weakly correlated with the direction of minimizing the constraint violation value, then the population on the vector has not yet reached the constraint feasible region; adopt a flexible search strategy and randomly execute the DE operator or subsequent local search operators.

8. The intelligent generation method for building energy-saving schemes based on two-stage agent-assisted evolution as described in claim 5, characterized in that, S44 contains the following: According to the S43 vector classification and search adaptive mechanism, each vector file obtains corresponding candidate solutions, which are stored in HS. For positively correlated vectors, the DE operator is executed to obtain candidate solutions. All candidate solutions are sorted according to the surrogate model prediction values, and the N solutions with the best prediction values ​​are selected. Their constraint violation values ​​are calculated, and the solution with the smallest constraint violation value is selected as the filling solution. For negatively correlated vectors, the local search operator TR-LS is executed to obtain new individuals. Then, calculate the constraint violation value for that individual; if the individual is a feasible solution, use the surrogate model to calculate... The predicted value, if Choose the solution that is better than the one in the existing archives. As filler samples, perform real evaluation; for weakly correlated vectors, calculate the constraint violation value of each new candidate solution and save the solutions with smaller constraint violation values; The target value of the saved candidate solutions is evaluated using a surrogate model. The solution with the best predicted target value is selected as the filling sample, and the vector file and its corresponding surrogate model are updated.

Images