CN-116680779-B - Model feature analysis method based on building indoor sunlight perception evaluation prediction model

CN116680779BCN 116680779 BCN116680779 BCN 116680779BCN-116680779-B

Abstract

The invention provides a model feature analysis method based on a building indoor sunlight perception evaluation prediction model. The method comprises the steps of 1, collecting building indoor light environment characteristic data and subjective sunlight perception evaluation results, establishing a basic database of sunlight perception evaluation, 2, expanding exploratory factor analysis and verification factor analysis to obtain sunlight perception commonality factors, 3, screening an optimal performance machine learning algorithm to construct a sunlight perception evaluation prediction model, and 4, performing characteristic interpretation analysis on the sunlight perception evaluation prediction model by using a SHAP method to obtain influence weights of light environment characteristic parameters and interaction combinations thereof on the sunlight perception evaluation results. The sunlight perception evaluation prediction model provided by the invention not only has higher prediction precision, but also has good model interpretability, can analyze the influence weights of various environmental factors in sunlight perception evaluation, and improves the accuracy and scientificity of natural lighting design of a building.

Inventors

SUN CHENG
LUO ZHAOYANG
QI XUANNING
DONG QI
LIU LEI
QU DAGANG

Assignees

哈尔滨工业大学

Dates

Publication Date: 20260512
Application Date: 20230517

Claims (8)

1. A model feature analysis method based on a building indoor sunlight perception evaluation prediction model is characterized by comprising the following steps: Step 1, developing a sunlight perception evaluation experiment in a typical building space, collecting indoor light environment characteristic parameters and subjective sunlight perception evaluation results of a building, and establishing a basic database of sunlight perception evaluation; Step 2, invoking data of a sunlight perception evaluation basic database, and expanding exploratory factor analysis and verification factor analysis to obtain a sunlight perception commonality factor; step 3, screening an optimal performance machine learning algorithm based on a sunlight perception commonality factor classification result, and constructing a sunlight perception evaluation prediction model; step 4, performing feature interpretation analysis on the sunlight perception evaluation prediction model by using a SHAP method, and respectively calculating an average SHAP value of each light environment feature parameter, a SHAP value of a light environment feature parameter interaction combination of a typical sample and a SHAP value of a light environment feature parameter interaction combination to obtain an influence weight of the light environment feature parameter and the interaction combination thereof on a sunlight perception evaluation result; the step 3 is specifically as follows: Step 3.1, carrying out feature selection on a data set in a basic database according to a classification result of the sunlight perception commonality factors, and preprocessing light environment feature parameters; Step 3.2, respectively constructing a sunlight perception evaluation prediction model by adopting a decision tree algorithm, a random forest algorithm and a XGBoost algorithm, comparing the precision of the prediction model by using model evaluation indexes, and screening an optimal prediction model; Step 3.3, a grid search method and a cross validation method are used for expanding the super-parameter optimization of the prediction model, and finally an effective sunlight perception evaluation prediction model is obtained; The method comprises the steps of carrying out characteristic interpretation on a sunlight perception evaluation prediction model by utilizing a SHAP method, analyzing the influence degree of light environment characteristic parameters on the sunlight perception evaluation result, the sunlight perception evaluation classification result of the light environment characteristic parameters, the sunlight perception evaluation result of the light environment characteristic parameters based on typical samples and the sunlight perception evaluation result of the light environment characteristic parameter interaction combination so as to discuss the quantitative relation between light environment characteristic elements and the sunlight perception evaluation result, and further obtaining the influence weight of the light environment characteristic parameters and the interaction combination thereof on the sunlight perception evaluation result.
2. The method according to claim 1, wherein the step 1 is specifically: step 1.1, acquiring typical space environment information of different building types by using a questionnaire investigation and field investigation method; step 1.2, developing a sunlight perception evaluation experiment in a typical building space, collecting indoor light environment characteristic parameters of a building, and collecting subjective sunlight perception evaluation results; and 1.3, screening and classifying the light environment characteristic parameters and the sunlight perception evaluation results according to the building space type and the lighting form, and establishing a basic database of the sunlight perception evaluation model.
3. The method according to claim 2, characterized in that: the content of the sunlight perception evaluation is 15 groups of bipolar adjectives for evaluating indoor light environment characteristics, and the bipolar adjectives are respectively as follows: dim- - -bright, low contrast- - -high contrast, obvious- - -blurred, uniform- - -uneven, stable- - -varying, simple- - -complex, continuous- - -discontinuous, neat- - -random, interesting- - -uninteresting, pleasant- - -uncomfortable, satisfying- - -unsatisfactory, pleasant- - -unpleasant, strenuous- - -relaxed, inhibitory- - -irritating and attractive- - -unattractive.
4. The method of claim 1, wherein step 2 comprises the steps of performing reliability and effectiveness analysis on sunlight perception evaluation results, extracting potential commonality factors among sunlight perception evaluation problem items by using a principal component analysis method in exploratory factor analysis, expanding verification factor analysis on the commonality factors, performing trial-and-error deletion on the problem items with low standard load coefficients by taking standard values of model fitting indexes as references according to polymerization effectiveness, distinguishing effectiveness and structure effectiveness analysis results, and obtaining the sunlight perception commonality factors until all indexes reach a threshold standard.
5. The method of claim 1, wherein the SHAP value is used for deep mining of importance of the input feature to the predicted result, and the specific calculation formula is: Wherein, the The representation of the predictive model is given by, Representing feature vectors, M representing the number of simplified input features, In the feature vector, "1" indicates that the corresponding feature value "exists" and "0" indicates "does not exist"; representing the model output where all simplified inputs are turned off, i.e., missing.
6. The method of claim 5, wherein the SHAP method is used to analyze the dataset characteristics of the learned prediction model and wherein the SHAP method is used to analyze the final XGBoost model when constructing the solar perception evaluation prediction model using XGBoost algorithm The SHAP method determines SHAP values of all the features by constructing a subset of the features S, and the specific calculation formula is as follows: Wherein, the Representing the number of non-zero entries in S, Representing all S vectors where the non-zero elements are a subset of the non-zero elements in M.
7. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1-6 when the computer program is executed.
8. A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the method of any one of claims 1-6.

Description

Model feature analysis method based on building indoor sunlight perception evaluation prediction model Technical Field The invention belongs to the technical field of indoor natural lighting design of buildings, and particularly relates to a model feature analysis method based on an indoor sunlight perception evaluation prediction model of a building. Background The "WELL health building standard" by the international health building institute (International WELLBuilding Institute) states that natural lighting has a substantial impact on the user's mood, circadian rhythm and work efficiency. The sunlight perception evaluation research of the user is unfolded, so that the indoor light environment can be accurately predicted and evaluated according to the view angle of the user, and the evaluation result can remarkably improve the lighting design precision and decision efficiency of the building. The visual perception process is a complex reaction mechanism formed by integrally processing, analyzing and understanding the external information by the brain when the human eyes are stimulated by the external world. Since visual perception is an essential consideration in the architectural lighting design, in order to more scientifically and effectively study the visual perception rule of a user, a quantitative study method needs to be introduced in the architectural lighting design. Through the quantitative research of visual perception, the influence of the indoor environment of the building on the visual perception of a user can be deeply known, the building design decision is optimized, and the comfort and the use effect of the indoor environment of the building are improved. In building design research, subjective questionnaires and semantic difference scales are the most used quantitative analysis methods at present. The existing research usually adopts different subjective evaluation scales to carry out experiments, and because the questionnaires have different questionnaire modes and strong subjective bias and experience, whether the difference of sunlight perception rules is influenced by the questionnaire modes is difficult to judge. Meanwhile, researchers often adopt traditional statistical methods such as linear regression, correlation analysis, factor analysis and the like to conduct relation research between a single environmental element and sunlight perception evaluation, consideration of a potential complex nonlinear relation between the environmental element and an evaluation result is lacked, influence of the combined effect of the environmental element on the sunlight perception evaluation result is difficult to judge, and precise lighting design decision support cannot be provided for a designer. Disclosure of Invention The invention aims to solve the problem that the existing sunlight perception evaluation research cannot comprehensively reflect subjective feelings of users and influence weights of environmental parameters are unclear, and provides a model feature analysis method based on a building indoor sunlight perception evaluation prediction model. The invention is realized by the following technical scheme, and provides a model feature analysis method based on a building indoor sunlight perception evaluation prediction model, which comprises the following steps: Step 1, developing a sunlight perception evaluation experiment in a typical building space, collecting indoor light environment characteristic parameters and subjective sunlight perception evaluation results of a building, and establishing a basic database of sunlight perception evaluation; Step 2, invoking data of a sunlight perception evaluation basic database, and expanding exploratory factor analysis and verification factor analysis to obtain a sunlight perception commonality factor; step 3, screening an optimal performance machine learning algorithm based on a sunlight perception commonality factor classification result, and constructing a sunlight perception evaluation prediction model; And 4, performing feature interpretation analysis on the sunlight perception evaluation prediction model by using a SHAP method, and respectively calculating an average SHAP value of each light environment feature parameter, a SHAP value of a light environment feature parameter interaction combination of a typical sample and a SHAP value of a light environment feature parameter interaction combination to obtain the influence weight of the light environment feature parameter and the interaction combination thereof on the sunlight perception evaluation result. Further, the step 1 specifically comprises the following steps: step 1.1, acquiring typical space environment information of different building types by using a questionnaire investigation and field investigation method; step 1.2, developing a sunlight perception evaluation experiment in a typical building space, collecting indoor light environment characteristic parameters of a buildi