Prefabricated panel thermal resistance and cost prediction method based on QPSO random forest

Abstract

The invention discloses a prefabricated panel thermal resistance and cost prediction method based on a QPSO random forest, and the method comprises the following steps: collecting data related to the thermal resistance and cost of a prefabricated panel, and carrying out the preprocessing of the collected data; performing normalization operation on the preprocessed data to obtain a prefabricated wallboard data set; dividing the prefabricated wallboard data set into a training set and a test set; initializing the particle swarm scale, the particle dimension, the quantum coefficient L and the contraction and expansion coefficient beta; meanwhile, determining a basic structure and an initial hyper-parameter of the random forest model; training a random forest model by using the training set and the random forest parameter represented by the current particle, and calculating a fitness function value; using a quantum particle swarm algorithm to optimize hyper-parameters of the random forest model; and selecting a parameter corresponding to the particle with the optimal fitness as an optimal hyper-parameter of the random forest model. According to the method, the uncertainty of RF model prediction can be reduced, and the model prediction precision is improved.

Description

Technical Field

[0001] The invention relates to the technical field of building materials, and in particular to a method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO random forest. Background Art

[0002] In the field of building energy conservation, the thermal performance of the wall directly affects the energy consumption and indoor comfort of the building. As a common component of prefabricated buildings, the thermal performance and cost-effectiveness of prefabricated wall panels are key factors in design and material selection. However, existing thermal performance prediction methods mainly rely on field tests, which is not only time-consuming and labor-intensive, but also affected by the environment and instrument accuracy, resulting in low accuracy. It is difficult to fully consider the specific effects of material type, thickness, connector design, etc. on composite sandwich wall panels. In addition, it is difficult to customize thermal performance and cost predictions for specific building design schemes. Summary of the invention

[0003] Purpose of the invention: The present invention aims to provide a method for predicting the thermal resistance and cost of prefabricated wall panels based on QPSO optimized random forest, so as to accurately and quickly obtain the thermal resistance cost according to different influencing factors such as thickness, material, connectors, etc., and to reversely recommend suitable wall panels based on the thermal resistance cost.

[0004] Technical solution: A method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO random forest, including the following steps:

[0005] S1, collect data related to thermal resistance and cost of prefabricated wall panels, and pre-process the collected data;

[0006] S2, normalizing the preprocessed data to obtain a prefabricated wall panel data set;

[0007] S3, divide the prefabricated wall panel dataset into training set and test set;

[0008] S4, initialize the particle swarm size, particle dimension, quantum coefficient L and expansion coefficient β; at the same time, determine the basic structure and initial hyperparameters of the random forest model;

[0009] S5, using the training set data and the random forest parameters represented by the current particle to train the random forest model and calculate the fitness function value;

[0010] S6, using quantum particle swarm algorithm to optimize the hyperparameters of random forest model;

[0011] S7, select the parameters corresponding to the particles with the best fitness as the optimal hyperparameters of the random forest model, use the training set to train the random forest model; use the test set to predict the random forest model, calculate the error index between the predicted result and the true value, evaluate the model performance, and save the random forest model with the best performance.

[0012] Furthermore, among the collected data, the inner / outer board thickness, insulation layer thickness, insulation layer material, connector type and connector diameter are used as input feature parameters of the random forest model, and the wall panel thermal resistance and cost are used as output feature parameters of the random forest model, where the cost includes construction cost and maintenance cost.

[0013] Further, in step S4, the particle dimension value is equivalent to the number of hyperparameters to be optimized in the random forest model, and the hyperparameters include the number of trees in the random forest, the maximum number of features considered when splitting the decision tree, the maximum depth of the tree, the minimum number of samples for splitting internal nodes, and the minimum number of samples for leaf nodes.

[0014] Furthermore, in step S5, the coordinate values ​​of the particles in each dimension correspond to the hyperparameter values ​​of a specific random forest model, and the change of the particle position will cause the represented hyperparameter to change, thereby constructing different random forest models and calculating the fitness function value.

[0015] Furthermore, the fitness function value adopts 1-R 2 As the calculation result, R 2 is the determination coefficient.

[0016] Compared with the prior art, the present invention has the following significant effects:

[0017] 1. The present invention uses the QPSO algorithm to optimize the hyperparameters of the RF model, because the optimization mechanism of QPSO can not only effectively reduce the uncertainty of RF model prediction, but also prevent the algorithm from falling into local optimality, thereby improving the prediction accuracy of the model;

[0018] 2. The QPSO optimization parameters and data preprocessing in the present invention work together to reduce the time and resource consumption required for model training; through two error criteria (MSE, R 2 ) clearly confirmed that QPSO-RF is a good prediction model that can improve the prediction accuracy to a certain extent; the use of this prediction method greatly reduces the amount of calculation of the thermal resistance and cost of prefabricated wall panels, can provide timely decision support for building design and cost control, reduce the resources and time costs consumed by experimental testing, achieve more accurate cost control, and meet the construction industry's demand for efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the present invention;

[0020] Figure 2 This is the structural diagram of RF random forest;

[0021] Figure 3 It is the fitness change diagram of QPSO-RF model;

[0022] Figure 4 This is the prediction effect diagram of thermal resistance by QPSO-RF model;

[0023] Figure 5 This is the prediction effect diagram of the QPSO-RF model on the cost. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0025] Reference Figure 1 The present invention proposes a method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO optimization (Quantum Particle Swarm Optimization) random forest, which specifically includes the following steps:

[0026] Step 1: Use solidworks and other modeling software to model wall panels with different wall panel thicknesses, different connector types, and connector diameters, and then use fluent series software to mesh and perform numerical calculations to obtain the thermal resistance of the wall panels. Then use the full life cycle method to calculate the wall panel cost, where the construction cost of the wall panel includes production costs, transportation costs, and installation costs. The maintenance cost of the wall panel takes into account winter heating costs and summer cooling costs.

[0027] Step 2: Collect data related to thermal resistance and cost of prefabricated wall panels, including but not limited to material type, thickness, connector design, etc. Clean the collected data to ensure the integrity and accuracy of the data, remove errors, duplications, missing values ​​and other abnormalities in the data, so as to improve data quality and provide a reliable data basis for subsequent model training.

[0028] The data includes six input features, namely, outer panel thickness, inner panel thickness, insulation layer thickness, insulation layer material, connector type, and connector diameter; the output features are wall panel thermal resistance and cost (including construction cost and maintenance cost). The size of the input features directly affects the size of the output features. For example, thicker wall panels, insulation layer materials with higher thermal insulation effects, and smaller connector diameters will have smaller wall panel thermal resistance and lower maintenance costs, as well as higher construction costs. Therefore, the wall panel thermal resistance, construction cost, and maintenance cost have opposite requirements for the size of the input features. The higher the maintenance cost, the lower the thermal resistance and construction cost, and vice versa.

[0029] Step 3: Use the Min-Max normalization method to process the preprocessed data to obtain the prefabricated wall panel data set; convert all data features to the same dimension to improve the stability and accuracy of model training.

[0030] The data preprocessed in step 2 is normalized to eliminate the differences in attribute values ​​due to large differences in order of magnitude, ensure the stability of data calculation, and improve the credibility of the experiment. The normalization formula is as follows:

[0031]

[0032] Among them, x is the attribute value corresponding to each data feature, x max and x min is the maximum and minimum value of each attribute value, x new It is the attribute value of the original indicator data after normalization. After normalization, the attribute values ​​corresponding to all features of the data sample are converted to between 0 and 1, ensuring the consistency of the data dimension level.

[0033] Step 4: Divide 80% of the prefabricated wall panel dataset into a training set and 20% into a test set. The training set is used for model training, and the test set is used for model verification.

[0034] Step 5: Initialize the parameters of the quantum particle swarm algorithm, including the particle swarm size, particle dimension, quantum coefficient L, and expansion coefficient β.

[0035] Specifically, the quantum particle swarm algorithm (QPSO) is a random search algorithm that simulates biological activities and swarm intelligence in nature. It has made innovative improvements to the traditional particle swarm optimization (PSO) and introduced the quantum behavior of particles to optimize the search range of the particle swarm. In addition, the QPSO algorithm introduces a new term mbest, which represents the average value of pbesti, that is, the average historical optimal position of the particle. In this way, the randomness of the particles in the solution space is greatly enhanced, the probability of the solution falling into the local extreme value is greatly reduced, and the global search capability is greatly enhanced. It mainly has the following calculation steps:

[0036]

[0037] In the formula, X i,d (t) is the position of the ith particle at the tth iteration in the dth dimension; P i,d (t) is the historical optimal position of particle i in the dth dimension; u is a random number that obeys the normal distribution, u∈[0,1]; L is the quantum coefficient used to control the intensity of the search.

[0038] Update of quantum coefficient L:

[0039] L(t+1)=2β|mbest-X(t)| (3)

[0040] Where mbest is the average value of the optimal position of all particles in the particle swarm in the current iteration; X(t) is the position vector of the particle swarm at time t; β is the expansion coefficient, which is used to adjust the search range.

[0041] Calculation of the average particle historical optimal position mbest:

[0042]

[0043] Where M is the size of the particle swarm, and pbest is the current optimal position of the particle. The current optimal position update formula of particle i is:

[0044] Update of individual optimal position pbest:

[0045] pbest=φ·pbest±(1-φ)·p g,d (t) (5)

[0046] Where φ is a random number that follows a normal distribution, φ∈[0,1]; p g,d (t) is the global optimal position.

[0047] Update of the expansion coefficient β:

[0048] β=m-(mn)t / t max (6)

[0049] Where m and n are the expansion coefficient parameters; t is the current iteration number; t max is the maximum number of iterations.

[0050] By simulating quantum behavior, the QPSO algorithm is able to make probabilistic jumps in the search space, increasing the possibility of finding the global optimal solution. This probability-based position update strategy helps the algorithm avoid falling into the local optimal solution while improving its ability in the global search space. Moreover, the parameters in the algorithm (such as quantum factors and learning rates) can be fine-tuned according to specific problems to better adapt to diverse optimization challenges.

[0051] In the prediction method based on QPSO optimization of random forest, the particle swarm size, particle dimension, quantum coefficient L and expansion coefficient β are closely related to the random forest parameters and jointly affect the model performance.

[0052] The particle dimension value is equivalent to the number of hyperparameters to be optimized in the random forest model, so as to establish a mapping relationship between the two. For example, this algorithm optimizes the five parameters of the number of trees in the random forest, the maximum number of features considered when splitting a decision tree, the maximum depth of the tree, the minimum number of samples for splitting internal nodes, and the minimum number of samples for leaf nodes, and the particle dimension is 5. In this way, the particle position can characterize the random forest hyperparameter combination and provide an intuitive representation for hyperparameter optimization.

[0053] The size of the particle swarm determines the search scope. A larger size means more particles participating, providing a richer initial combination of random forest hyperparameters, increasing the possibility of finding the optimal solution, but also increasing the computational complexity. Although a smaller size is computationally simple, it may not be able to fully cover the hyperparameter space due to the limited number of particles, reducing the possibility of finding the global optimum.

[0054] The quantum coefficient L and the expansion coefficient β control the search behavior. L affects the search intensity. The larger L is, the easier it is for particles to deviate from the current position, expanding the exploration range of random forest hyperparameters; the smaller L is, the more conservative the search is, focusing on local areas. β dynamically adjusts the search range. At the beginning of the iteration, β is large, the L update amplitude is large, and the search range is wide, which is conducive to the global search of random forest hyperparameters; with the iteration, β decreases, the search range is narrowed, and it is convenient to search for the local optimum in detail. The two work together to guide particles to update their positions according to the fitness function value during the iteration, that is, to adjust the random forest hyperparameters.

[0055] The fitness function is calculated by the fitness method, and the fitness function value is 1-R 2 As a result of the calculation, the goal is to make the smaller the fitness function value, the better the model performance (because R 2 The closer it is to 1, the better the model fit is, so 1-R 2The smaller the better). In the code, QPSO optimizes the RF model parameters in many ways. The particle swarm dimension corresponds to the number of hyperparameters that need to be optimized in RF. In the particle swarm initialization method, the particle dimension value is set according to the hyperparameter range, and a mapping between particle position and hyperparameter is established so that particle movement can traverse different combinations. The number of particle swarms determines the size of the particle swarm. Although a large size is computationally complex, it can increase the possibility of finding the optimal solution. It expands the hyperparameter search range by generating multiple particle initial positions. In the update method, the particle position update is affected by many factors. The random coefficient and the value range limit control the search. Special treatment is given to hyperparameters such as the maximum depth of the tree to balance global and local searches. Particles are updated according to the fitness function value (with 1-R 2 After multiple iterations, the optimal particle corresponding hyperparameters are finally selected as the optimal hyperparameters of the RF model to achieve the optimization of various parameters.

[0056] Finally, the parameters corresponding to the particle with the best fitness were selected to construct the QPSO-RF model, and the random forest parameters were optimized to improve the accuracy and stability of the model in predicting the thermal resistance and cost of prefabricated wall panels.

[0057] Step 6: Use the training set and the random forest (RF) parameters represented by the current particle to train the RF model. The particle dimension is consistent with the number of RF hyperparameters to be optimized. The coordinate values ​​of the particle in each dimension correspond to the values ​​of specific RF hyperparameters. The change of its position will change the represented hyperparameters, so as to construct different RF models to explore the optimal parameter configuration. And calculate the fitness function value. The model of random forest is as follows Figure 2 , the specific optimized random forest parameters are as follows:

[0058] n_estimators: The number of trees in the forest. More trees means a more stable and accurate model, but the computational cost increases.

[0059] max_features: The maximum number of features considered when splitting the decision tree. It can be an integer, percentage, or "log2". It can prevent overfitting and improve training efficiency.

[0060] max_depth: The maximum depth of the decision tree. Limiting it can prevent overfitting.

[0061] min_samples_split: The minimum number of samples for splitting internal nodes, which can be an integer or a proportion of the total number of samples. Increasing it can reduce the complexity of the tree, but may lead to underfitting.

[0062] min_samples_leaf: The minimum number of samples for a leaf node. Increasing it can reduce model complexity but may also lead to underfitting.

[0063] Step 7: Update the particle position according to the core update formula of the quantum particle swarm algorithm. Comprehensively consider the key factors such as the current position of the particle, the individual historical optimal position, and the group historical optimal position, and combine the quantum coefficient L and the expansion coefficient β for calculation. In the iterative process, the β value is dynamically adjusted according to the number of iterations, thereby changing the amplitude of the quantum coefficient L, so that the particles can perform wide-area exploration and local deep digging at different iteration stages, and gradually optimize the random forest hyperparameter combination mapped by the particle position to approach the optimal solution.

[0064] Step 8: Check whether the stopping condition is met, such as whether the maximum number of iterations is reached. If so, stop the iteration.

[0065] Specifically, the termination condition is that the maximum number of optimization iterations is reached and the fitness has no significant change (for example, the change in fitness after several consecutive iterations is less than the set threshold). If the above conditions are met, the global optimal particle is mapped to the hyperparameter of the random forest (RF) model, and this is used as the optimization result to obtain a well-established QPSO-RF composite model; otherwise, return to parameter optimization steps 6 and 7 to continue iteration.

[0066] Step 9: Select the parameters corresponding to the particle with the best fitness as the optimal parameters, and use the training set to train the RF model. Use the test set to make predictions, calculate the error index between the prediction results and the true value, and evaluate the model performance.

[0067] The indicators for evaluating the model are mean square error (MSE), determination coefficient (R 2 ) indicator, the smaller the MSE, the higher the R 2 The closer it is to 1, the better the model performance is. If the model does not meet the conditions, first readjust the hyperparameters, such as optimizing the QPSO parameters and re-iteratively optimizing the RF hyperparameters; secondly, check and optimize the data processing to comprehensively improve the model performance to meet the expected requirements.

[0068] Specifically, the QPSO-RF composite model is constructed using the corresponding programming environment (such as Python, etc.), and the optimal hyperparameters obtained by quantum particle swarm algorithm optimization are loaded into the random forest model for prediction, and the thermal resistance and cost of the wall panel are output.

[0069] like Figure 3 The figure shows the relationship between the change of fitness and the number of iterations. In this embodiment, the fitness value is 1-R 2 Calculation shows that the fitness finally converges to around 0.0013, which has a high prediction accuracy. Figure 4 , Figure 5 The figures show the real data of thermal resistance and cost (the sum of maintenance cost and construction cost) and the fitting degree of the predicted values. The dots are the predicted data of QPSO-RF model. Figure 4 , Figure 5 It can be seen that the difference between the predicted value and the actual value is small, which can meet the actual needs; accurate wall panel thermal resistance and cost are crucial for enterprise optimization design and personal choice.

[0070] The present invention proposes a method for predicting the thermal resistance and cost of prefabricated wall panels based on QPSO optimization of RF, and uses the QPSO algorithm to optimize the hyperparameters of the RF model. The optimization mechanism of QPSO can effectively reduce the uncertainty of RF model prediction, avoid the algorithm from falling into local optimality, and improve the prediction accuracy of the model. The QPSO optimization parameters work together with data preprocessing to reduce the time and resource consumption of model training. 2 The two error criteria confirm that QPSO-RF is a good prediction model and can improve the prediction accuracy. The prediction method can reduce the calculation of thermal resistance and cost of prefabricated wall panels, provide decision support for building design and cost control in a timely manner, and meet the demand for efficiency in the construction industry.

Claims

1. A method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO random forest, characterized in that: The steps include: S1, collect data related to thermal resistance and cost of prefabricated wall panels, and pre-process the collected data; S2, normalizing the preprocessed data to obtain a prefabricated wall panel data set; S3, divide the prefabricated wall panel dataset into training set and test set; S4, initialize the particle swarm size, particle dimension, quantum coefficient L and expansion coefficient β; at the same time, determine the basic structure and initial hyperparameters of the random forest model; S5, using the training set data and the random forest parameters represented by the current particle to train the random forest model and calculate the fitness function value; S6, using quantum particle swarm algorithm to optimize the hyperparameters of random forest model; S7, select the parameters corresponding to the particles with the best fitness as the optimal hyperparameters of the random forest model, use the training set to train the random forest model; use the test set to predict the random forest model, calculate the error index between the predicted result and the true value, evaluate the model performance, and save the random forest model with the best performance.

2. The method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO random forest according to claim 1 is characterized in that: Among the collected data, the inner / outer board thickness, insulation layer thickness, insulation layer material, connector type and connector diameter are used as input feature parameters of the random forest model, and the wall panel thermal resistance and cost are used as output feature parameters of the random forest model, where the cost includes construction cost and maintenance cost.

3. The method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO random forest according to claim 1 is characterized in that: In step S4, the particle dimension value is equivalent to the number of hyperparameters to be optimized in the random forest model, including the number of trees in the random forest, the maximum number of features considered when splitting the decision tree, the maximum depth of the tree, the minimum number of samples for splitting internal nodes, and the minimum number of samples for leaf nodes.

4. The method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO random forest according to claim 1 is characterized in that: In step S5, the coordinate values ​​of the particles in each dimension correspond to the hyperparameter values ​​of a specific random forest model. The change of the particle position will change the represented hyperparameter, thereby constructing different random forest models and calculating the fitness function value.

5. The method for predicting thermal resistance and cost of prefabricated wall panels based on QPSO random forest according to claim 4 is characterized in that: The fitness function value adopts 1-R 2 As the calculation result, R 2 is the determination coefficient.

Images