CN116341372B - Heat exchanger performance prediction and optimization method based on artificial neural network - Google Patents

Abstract

The invention provides a heat exchanger performance prediction and optimization method based on artificial neural networks, which includes: constructing performance databases of multiple heat exchangers; based on the performance databases of multiple heat exchangers, using particle swarm algorithm to predict BP The neural network is optimized to obtain the corresponding PSO‑BP‑ANN prediction model; the PSO‑BP‑ANN model is combined with the multi-objective genetic algorithm to optimize the design of the structural parameters of the target heat exchanger, and the optimization results are obtained. The results are verified. The invention solves the problem in the prior art that the optimization design results of the heat exchanger have large errors.

Description

A heat exchanger performance prediction and optimization method based on artificial neural network

Technical field

The invention relates to the technical field of heat exchanger optimization design, and in particular to a heat exchanger performance prediction and optimization method based on artificial neural networks.

Background technique

Compact heat exchangers have broad development potential in the nuclear industry, electric power, refrigeration and other industrial fields because of their compact structure and efficient heat exchange. Achieving both enhanced heat transfer and flow resistance reduction is a goal that researchers at home and abroad are constantly pursuing. However, in the process of heat exchanger optimization design, selecting different performance calculation correlations will produce large errors in the obtained optimization design results.

Contents of the invention

In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a heat exchanger performance prediction and optimization method based on artificial neural networks. The present invention solves the problem of large errors in heat exchanger optimization design results in the prior art.

In order to achieve the above objects, the present invention provides the following solutions:

A heat exchanger performance prediction and optimization method based on artificial neural networks, including:

Construct performance database of various heat exchangers;

Based on the performance database of various heat exchangers, the particle swarm algorithm is used to optimize the BP neural network and the corresponding PSO-BP-ANN prediction model is obtained;

The PSO-BP-ANN model was combined with the multi-objective genetic algorithm to optimize the design of the structural parameters of the target heat exchanger, obtain the optimization results, and verify the optimization results.

Preferably, the performance database of the various heat exchangers includes:

Plate fin heat exchangers and printed circuit board heat exchangers. Plate fin heat exchangers include: straight fin heat exchangers, corrugated fin heat exchangers, louvered fin heat exchangers and sawtooth fin heat exchangers. Heater.

Preferably, the method of constructing a performance database of multiple heat exchangers includes:

Use literature research and data simulation to construct performance databases for straight fin heat exchangers, corrugated fin heat exchangers, louvered fin heat exchangers, sawtooth fin heat exchangers, and printed circuit board heat exchangers. Performance database.

Preferably, the use of the particle swarm algorithm to optimize the BP neural network and obtain the PSO-BP-ANN prediction model includes:

Select different iteration times and population sizes to train and test the BP neural network, and determine the target iteration times and population size;

Based on the target number of iterations and the population size, the acceleration factor is determined by the j factor and the f factor;

The PSO-BP-ANN prediction model is determined based on the acceleration factor, target iteration number and population size.

Preferably, the optimization design of the structural parameters of the target heat exchanger by combining the PSO-BP-ANN model with the multi-objective genetic algorithm includes:

Initialize population parameters;

Set current population parameters;

Determine the objective function of the heat exchanger, and calculate and optimize the objective function according to the PSO-BP-ANN model and the multi-objective genetic algorithm to obtain the current optimization results;

Determine whether the current optimization result has converged. If so, output the Pareto front. If not, add 1 to the current population parameter and continue the calculation.

Preferably, the objective function includes:

The maximum heat transfer performance of the heat exchanger and the minimum resistance performance of the heat exchanger.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

The present invention provides a heat exchanger performance prediction and optimization method based on artificial neural networks. The present invention establishes a variety of heat exchanger performance prediction models, uses particle swarm algorithm to optimize the models, and uses the optimized model combination Multi-objective genetic algorithm is used to optimize the heat exchanger, which improves the accuracy of the optimization results.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the drawings of the present invention. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 is a flow chart of a heat exchanger performance prediction and optimization method provided by an embodiment of the present invention;

Figure 2 is a schematic diagram of a sawtooth fin grid provided by an embodiment of the present invention;

Figure 3 is a schematic structural diagram of a Z-channel PCHE single-cycle model provided by an embodiment of the present invention;

Figure 4 is a flow chart of the optimization process provided by the embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Reference herein to "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those skilled in the art understand, both explicitly and implicitly, that the embodiments described herein may be combined with other embodiments.

The terms “first”, “second”, “third” and “fourth” in the description, claims and drawings of this application are used to distinguish different objects, rather than to describe a specific sequence. . Furthermore, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a series of steps, processes, methods, etc. are not limited to the listed steps, but optionally also include steps that are not listed, or optionally also include steps inherent to these processes, methods, products or equipment. Other steps.

The purpose of the present invention is to provide a heat exchanger performance prediction and optimization method based on artificial neural networks. The present invention solves the problem of large errors in heat exchanger optimization design results in the prior art.

In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

As shown in Figure 1, the present invention provides a heat exchanger performance prediction and optimization method based on artificial neural networks, including:

Step 100: Construct a performance database of various heat exchangers;

Step 200: Based on the performance database of the various heat exchangers, use the particle swarm algorithm to optimize the BP neural network to obtain the corresponding PSO-BP-ANN prediction model;

Step 300: Combine the PSO-BP-ANN model with the multi-objective genetic algorithm to optimize the design of the structural parameters of the target heat exchanger, obtain the optimization results, and verify the optimization results.

Further, the performance database of various heat exchangers includes:

Plate fin heat exchangers and printed circuit board heat exchangers. Plate fin heat exchangers include: straight fin heat exchangers, corrugated fin heat exchangers, louvered fin heat exchangers and sawtooth fin heat exchangers. Heater.

Further, the method of constructing a performance database of multiple heat exchangers includes:

Use literature research and data simulation to construct performance databases for straight fin heat exchangers, corrugated fin heat exchangers, louvered fin heat exchangers, sawtooth fin heat exchangers, and printed circuit board heat exchangers. Performance database.

Establish a correlation analysis of heat transfer and resistance performance on different fin structural parameters in the database:

Straight fin heat exchanger

The main geometric parameters are fin spacing Fp, height Fh, thickness t and channel length Ld, among which Fp and Fh are positively correlated with the j factor, t and Ld are negatively correlated with the j factor; Fp, Fh and t are positively correlated with the f factor, Ld has a negative correlation with the f factor; t and Ld have a positive correlation with FTEF, and Fp and Fh have a negative correlation with FTEF.

Wave fin heat exchanger

The main geometric parameters are fin spacing Fp, height Fh, channel length Ld, thickness t, twice the amplitude 2A and wavelength L. Among them, Fh and 2A are positively related to the j factor, and Fp and L are negatively related to the j factor; Fp, Fh and 2A are positively correlated with f factor, L is negatively correlated with f factor; Fh and 2A are positively correlated with FTEF, and Fp and L are negatively correlated with FTEF.

Louvered fin heat exchanger

The main geometric parameters are fin spacing Fp, height Fh, length Ld, thickness t, inclination angle La and spacing Lp. Among them, Lp and La are positively related to the j factor, Fp and t are negatively related to the j factor; Lp and La are related to f The factors are positively correlated, Fp and t are negatively correlated to the f factor; Lp and La are positively correlated to FTEF, and Fp and t are negatively correlated to FTEF.

Specifically, numerical simulation is used to supplement domestic sawtooth fin channel and Z-type channel PCHE performance data:

ANSYS Fluent 17.0 was used to analyze the flow and heat transfer characteristics of the sawtooth fin channel and Z-shaped channel PCHE. The schematic diagram of the domestic sawtooth fin channel structure and grid is shown in Figure 2. The Z-shaped channel PCHE study uses the minimum period model, and the structure is shown in Figure 3. The above calculation model uses ICEM software to generate a structured grid, divide the boundary layer grid on the wall surface and perform local densification. The continuity equation, momentum equation and energy equation are as follows:

The continuity equation is:

The momentum equation is:

The energy equation is:

Energy equation (solid domain):

Materials and working fluids in the solid domain are treated with variable physical properties. Among them, u is the flow velocity, u _i and u _k are the components of the flow velocity in the xz direction, P is the pressure, the kinematic viscosity μ, ρ is the density, T is the temperature, cp is the constant pressure specific heat capacity, λ is the thermal conductivity coefficient, x _i / x _k is the corresponding direction vector. The model is solved using pressure-velocity coupling using the SIMPLE algorithm, and the energy and momentum equations using the second-order upwind formula. When the calculation residual is less than 10-6, the solution is considered to have converged. Set the inlet velocity, temperature, outlet pressure, temperature or heat flux density of the upper and lower partitions, and the left and right sides of the channel are periodic boundaries.

In order to verify the correctness of the numerical simulation of the sawtooth fin channel, the calculated values of the numerical simulation are compared with the experimental values in the literature. The calculated values of the numerical simulation of j factor and f factor are in good agreement with the experimental values. The average absolute percentage errors are respectively are 8.07% and 9.28%. Therefore, the results of numerical simulation are considered reliable.

Comparing the Z-channel PCHE numerical simulation results with the experimental values in the literature, the maximum errors between the Nusselt number and resistance factor numerical results and the experimental values are 6.7% and 9.3% respectively. Therefore, the results of the numerical simulation can be considered is reliable.

Further, the particle swarm algorithm is used to optimize the BP neural network, and the PSO-BP-ANN prediction model is obtained including:

Select different iteration times and population sizes to train and test the BP neural network, and determine the target iteration times and population size;

Based on the target number of iterations and the population size, the acceleration factor is determined by the j factor and the f factor;

The PSO-BP-ANN prediction model is determined based on the acceleration factor, target iteration number and population size.

Specifically, determine the optimal parameter configuration of the particle swarm algorithm:

Straight fin heat exchanger

The input layer of the straight fin channel PSO-BP-ANN model is 5 dimensionless parameters. The input parameters are input1(F _p /D _e ), input2(F _h /D _e ), and input3(t/D _e ). , input4(L _d /D _e ) and input5(Re), the output parameters are j factor or f factor, and their BP neural network structures are 5-10-1 and 5-4-1 respectively. Therefore, the dimensions D of the particle swarm are 71 (5×10+10+10×1+1) and 29 (5×4+4+4×1+1) respectively. The parameters of the particle swarm algorithm are initially set as follows: the maximum number of evolution iterations is 200, the population size is 100, the acceleration factors are both 1.5, and the inertia weight is 1.

First, determine the maximum number of iterations, and use different iteration numbers of 50, 100, 150, 200, and 250 to train and test the model. Then the population size is determined, and different population sizes of 30, 50, 70, 100 and 150 are used for model training and testing. Finally, the optimal acceleration factor is selected through j factor and f factor according to the selected number of iterations and population size.

For straight fin heat exchangers, when the maximum number of iterations is 150, the population sizes are 150 and 100, and the acceleration factors are both 1.5, the PSO-BP-ANN model of j factor and f factor has the best performance in predicting the test set .

Wave fin heat exchanger

The input layer of the corrugated fin channel PSO-BP-ANN model is 5 dimensionless parameters. The input parameters are input1(F _p /F _h ), input2(F _p /t), input3(L _d /L), input4( F _d /2A) and input5(Re), the output parameter is j factor or f factor. The BP neural network structures of j factor and f factor are 5-7-1 and 5-8-1 respectively. Therefore, the dimension D of the particle swarm is 50 (5×7+7+7×1+1) and 57 respectively. (5×8+8+8×1+1). The parameters of the particle swarm algorithm are initially set as follows: the maximum number of evolution iterations is 200, the population size is 100, the inertia weight is 1, and the acceleration factors are all 1.5.

First, determine the maximum number of iterations, and use different iteration numbers of 50, 100, 150, 200, and 250 to train and test the model. Then the population size is determined, and different population sizes of 30, 50, 70, 100 and 150 are used for model training and testing. Finally, the optimal acceleration factor is selected through j factor and f factor according to the selected number of iterations and population size.

For corrugated fin heat exchangers, when the maximum number of iterations is 150 and 250 respectively, the population size is 100, and the acceleration factors are both 2.5 and 0.5, the PSO-BP-ANN model with j factor and f factor has a better performance on the test set. Best prediction performance.

Louvered fin heat exchanger

The input layer of the louver fin channel PSO-BP-ANN model has 6 dimensionless parameters. The input parameters are input1(F _p /L _p ), input2(F _h /L _p ), and input3(L _d /L _p ). , input4(t/L _p ), input5(L _a /90) and input6(Re), the output parameter is j factor or f factor. The BP neural network structures of j factor and f factor are 6-9-1 and 6-13-1 respectively. Therefore, the dimension D of the particle swarm is 73 (6×9+9+9×1+1) and 105 respectively. (6×13+13+13×1+1). The parameters of the particle swarm algorithm are initially set as follows: the maximum number of evolution iterations is 200, the population size is 100, the inertia weight is 1, and the acceleration factors are all 1.5.

First, determine the maximum number of iterations, and use different iteration numbers of 50, 100, 150, 200, and 250 to train and test the model. Then the population size is determined, and different population sizes of 30, 50, 70, 100 and 150 are used for model training and testing. Finally, the optimal acceleration factor is selected through j factor and f factor according to the selected number of iterations and population size.

For louvered fin heat exchangers, when the maximum number of iterations is 100 and 150, the population size is 50 and 100, and the acceleration factors are 2, 1, 0.5, and 2.5, respectively, the prediction performance of the PSO-BP-ANN model is optimal.

Sawtooth fin heat exchanger

The input layer of the sawtooth fin channel PSO-BP-ANN model is 4 dimensionless parameters, and the input parameters are input1(F _p /F _h ), input2(t/L _f ), input3(t/F _p ) and input4 respectively. (Re), the output parameter is j factor or f factor. The BP neural network structures of j factor and f factor are 4-11-1 and 4-10-1 respectively. Therefore, the dimension D of the particle swarm is 67 (4×11+11+11×1+1) and 61 respectively. (4×10+10+10×1+1). The parameters of the particle swarm algorithm are initially set as follows: the maximum number of evolution iterations is 200, the population size is 100, the inertia weight is 1, and the acceleration factors are all 1.5.

First, determine the maximum number of iterations, and use different iteration numbers of 50, 100, 150, 200, and 250 to train and test the model. Then the population size is determined, and different population sizes of 30, 50, 70, 100 and 150 are used for model training and testing. Finally, the optimal acceleration factor is selected through j factor and f factor according to the selected number of iterations and population size.

For sawtooth fin heat exchangers, when the maximum number of iterations are 150 and 50 respectively, the population size is 30 and 100 respectively, and the acceleration factors are both 1.5 and 1.5 combinations, the PSO-BP-ANN model prediction test of j factor and f factor The performance of the set is optimal.

Z-channel printed circuit board heat exchanger

The input layer of the Z-channel printed circuit board heat exchanger PSO-BP-ANN model has 4 parameters. The input parameters are input1(D), input2(Ф), input3(P) and input4(Re), and the output parameters are Nu or f factor. Therefore, the dimensions D of the particle swarm are 49 (4×8+8+8×1+1) and 31 (4×5+5+5×1+1) respectively. The parameters of the particle swarm algorithm are initially set as follows: the maximum number of evolution iterations is 200, the population size is 100, the inertia weight is 1, and the acceleration factors are all 1.5.

First, determine the maximum number of iterations, and use different iteration numbers of 50, 100, 150, 200, and 250 to train and test the model. Then the population size is determined, and different population sizes of 30, 50, 70, 100 and 150 are used for model training and testing. Finally, the optimal acceleration factor is selected through j factor and f factor according to the selected number of iterations and population size.

Z-type channel printed circuit board heat exchanger When the maximum iteration times are 100 and 200 respectively, the population size is 100, and the acceleration factors are 1.5 and 1.5, the PSO-BP-ANN model of Nu and f factors predicts the performance of the test set Optimal.

Further, the optimization design of the structural parameters of the target heat exchanger by combining the PSO-BP-ANN model with the multi-objective genetic algorithm includes:

Initialize population parameters;

Set current population parameters;

Determine the objective function of the heat exchanger, and calculate and optimize the objective function according to the PSO-BP-ANN model and the multi-objective genetic algorithm to obtain the current optimization results;

Determine whether the current optimization result has converged. If so, output the Pareto front. If not, add 1 to the current population parameter and continue the calculation.

Specifically, taking the sawtooth fin channel of the plate-fin heat exchanger and the Z-channel printed circuit board heat exchanger as the research objects, the optimal PSO-BP-ANN model was combined with the multi-objective genetic algorithm to optimize the structural parameters. design:

Optimization study of sawtooth fin channel

The constraints of the multi-objective optimization problem of sawtooth fin channels can be expressed by the following formula.

Minimize goals:-j＝-net _j (α,γ,δ,Re)＝-net _j (F _p /F _h ,t/F _p ,t/L _f ,Re)

f＝net _f (α,γ,δ,Re)＝net _f (F _p /F _h ,t/F _p ,t/L _f ,Re)

Subjected to:2.7≤F _h ≤9.3; 1.2≤F _p ≤3.2; 0.1≤F _p ≤0.5; 3≤L _f ≤9

Among them, j is the heat transfer factor, f is the resistance factor, L _f is the fin length, F _p is the spacing, F _h is the height, t is the thickness, and net is the neural network model. Taking the maximum heat transfer performance and minimum resistance performance of the heat exchanger as two objective functions, the optimal PSO-BP-ANN prediction model of the sawtooth fin channel of the plate-fin heat exchanger is combined with NSGA-II to predict The structural parameters are optimized and designed and the Pareto optimization solution set is obtained. The following table lists the operating parameters of NSGA-II. The optimization process is shown in Figure 4. The operating parameters of NSGA-II are shown in Table 1.

Table 1 Operating parameters of NSGA-II

Optimization study of Z-channel printed circuit board heat exchanger

The constraints of the multi-objective optimization problem of Z-channel printed circuit board heat exchanger can be expressed by the following formula.

Minimize goals:-Nu＝-net _Nu (D,φ,P,Re)

f＝net _f (D,φ,P,Re)

Subjected to:5°≤φ≤45°; 1.25≤D≤2.25; 10≤P≤30

Among them, f is the resistance factor, Ф is the channel angle, P is the pitch, D is the inlet diameter, Re is the Reynolds number, and Nu is the Nusselt number. The optimal PSO-BP-ANN prediction model of the Z-channel printed circuit board heat exchanger was combined with NSGA-II to optimize the design of the structural parameters and obtain the Pareto optimization solution set. Table 2 lists NSGA-II. operating parameters.

Table 2 Operating parameters of NSGA-II

Compare the optimization results with the CFD calculation results to verify the accuracy.

Specifically, the objective function includes:

The maximum heat transfer performance of the heat exchanger and the minimum resistance performance of the heat exchanger.

The beneficial effects of the present invention are as follows:

This invention uses particle swarm algorithm to optimize the initial weights and thresholds of the BP neural network. In addition, the comparison of the prediction performance of the PSO-BP-ANN model and the correlation model on independent experimental data proves that the PSO-BP-ANN prediction model is better than the correlation model in terms of coverage and prediction accuracy. This invention combines the optimal PSO-BP-ANN model with a multi-objective genetic algorithm to optimize the design of the structural parameters of the sawtooth fin channel of the plate-fin heat exchanger and the Z-channel printed circuit board heat exchanger. The accuracy of the optimization results was verified through CFD calculations. The results proved that the method combining numerical simulation, artificial neural network and multi-objective genetic algorithm can accurately optimize the design of the heat exchanger.

Each embodiment in this specification is described in a progressive manner. Each embodiment focuses on its differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and the core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the present invention There will be changes in the specific implementation methods and application scope of the ideas. In summary, the contents of this description should not be construed as limitations of the present invention.

Claims (4)

1. A heat exchanger performance prediction and optimization method based on artificial neural network, which is characterized by including: Construct performance database of various heat exchangers; Based on the performance database of various heat exchangers, the particle swarm algorithm is used to optimize the BP neural network and the corresponding PSO-BP-ANN prediction model is obtained; The PSO-BP-ANN model is combined with the multi-objective genetic algorithm to optimize the design of the structural parameters of the target heat exchanger, obtain the optimization results, and verify the optimization results; The combination of the PSO-BP-ANN model and the multi-objective genetic algorithm to optimize the structural parameters of the target heat exchanger includes: Initialize population parameters; Set current population parameters; Determine the objective function of the heat exchanger, and calculate and optimize the objective function according to the PSO-BP-ANN model and the multi-objective genetic algorithm to obtain the current optimization results; Determine whether the current optimization result has converged. If so, output the Pareto front. If not, add 1 to the current population parameter and continue the calculation; Specifically, the constraint formula of the multi-objective optimization problem of sawtooth fin channel is: ; Among them, j is the heat transfer factor, f is the resistance factor, L _f is the fin length, F _p is the spacing, F _h is the height, t is the thickness, and net is the neural network model; The constraint formula of the multi-objective optimization problem of Z-channel printed circuit board heat exchanger is: ; Among them, f is the resistance factor, Ф is the channel angle, P is the pitch, D is the entrance diameter, Re is the Reynolds number, and Nu is the Nusselt number; The objective function includes: The maximum heat transfer performance of the heat exchanger and the minimum resistance performance of the heat exchanger. 2. A heat exchanger performance prediction and optimization method based on artificial neural network according to claim 1, characterized in that the performance database of the multiple heat exchangers includes: Plate fin heat exchangers and printed circuit board heat exchangers. Plate fin heat exchangers include: straight fin heat exchangers, corrugated fin heat exchangers, louvered fin heat exchangers and sawtooth fin heat exchangers. Heater. 3. A heat exchanger performance prediction and optimization method based on artificial neural networks according to claim 2, characterized in that the method of constructing a performance database of multiple heat exchangers includes: Use literature research and data simulation to construct performance databases for straight fin heat exchangers, corrugated fin heat exchangers, louvered fin heat exchangers, sawtooth fin heat exchangers, and printed circuit board heat exchangers. Performance database. 4. A heat exchanger performance prediction and optimization method based on artificial neural network according to claim 1, characterized in that the particle swarm algorithm is used to optimize the BP neural network to obtain PSO-BP-ANN prediction. Models include: Select different iteration times and population sizes to train and test the BP neural network, and determine the target iteration times and population size; Based on the target number of iterations and the population size, the acceleration factor is determined by the j factor and the f factor; The PSO-BP-ANN prediction model is determined based on the acceleration factor, target iteration number and population size.