An office building indoor natural lighting glare influence element analysis method based on feature selection

Abstract

The application provides an office building indoor natural lighting glare influence element analysis method based on feature selection. The method comprises the following steps: step 1, collecting potential glare elements and subjective evaluation data to construct a glare perception evaluation data set; step 2, combining an ANOVA variance analysis method to develop a single variable linear feature analysis; step 3, combining a tree model algorithm under different machine learning to complete nonlinear combination feature analysis under different glare influence elements; and step 4, combining a recursive feature elimination algorithm to complete multivariate combination feature analysis under different glare influence elements. The method breaks through the technical bottleneck of glare research and analysis, fills the gap in the research on the interactive combination of glare elements, and improves the prediction accuracy and scientific level of indoor glare discomfort.

Description

Technical Field

[0001] This invention belongs to the field of indoor natural lighting technology for office buildings, and aims to provide an innovative method for constructing subjective evaluation indicators of glare in office buildings, particularly a method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection. Background Technology

[0002] In studies on the comfort of natural lighting, glare perception caused by natural lighting has always been a key issue in office space design and lighting comfort research. Prolonged exposure to glare can lead to visual fatigue, decreased attention span, vision loss, and chronic eye diseases in office workers, and even physiological problems such as headaches, nausea, or loss of vision. Glare has negative psychological and physiological impacts on visual comfort, mental and physical health, and work efficiency. my country's "Standard for Daylighting Design of Buildings GB 50033-2013" defines uncomfortable glare from windows as an important indicator for evaluating lighting quality. Therefore, research on glare perception is essential.

[0003] Existing research has created various indicators for predicting subjective evaluations of glare comfort. These indicators are derived based on empirical assumptions and statistical derivations, and their types and compositions are diverse. Furthermore, in comparative literature on different indicators, the predictive results for the same glare comfort indicator are inconsistent, and the differences in predictive performance between different indicators often vary across different studies. The essence of this lies in the differences in the correlation between the influencing factors of the glare indicator and glare, and these differences have not been effectively analyzed, resulting in insufficient understanding of the indicators' use. Therefore, a scientific quantitative analysis of the influencing factors of glare, and an understanding of the degree of influence of these factors on glare discomfort, is essential.

[0004] Existing studies often employ traditional statistical methods such as linear regression and logistic regression to analyze the correlation between independent glare influencing factors and glare perception evaluation results. However, more effective research methods are lacking to explore the nonlinear relationship between the two, and the impact of multiple glare influencing factors on the perceived comfort of office space users. Therefore, there is an urgent need for more effective methods to analyze the glare influencing factors of natural lighting in office buildings, to determine effective glare evaluation and prediction indicators, and to provide evaluation support for optimizing natural lighting in office buildings. Summary of the Invention

[0005] The purpose of this invention is to address the problems of incomplete and in-depth research on the factors affecting glare perception and the lack of effective analysis methods. It proposes a method for analyzing the factors affecting glare from indoor natural lighting in office buildings based on feature selection.

[0006] This invention is achieved through the following technical solution: This invention proposes a method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection. The method includes the following steps:

[0007] Step 1: In the selected typical office building space, collect data on potential influencing factors of indoor glare and glare evaluation data to construct a glare perception evaluation dataset.

[0008] Step 2: For the linear influence of a single variable on the glare environment, use analysis of variance to perform feature analysis on the glare perception evaluation dataset and obtain the correlation results of the linear influence of different features.

[0009] Step 3: To analyze the nonlinear effects of variables on the glare environment and the combined correlations between variables, a tree model based on machine learning is used to perform feature analysis on the glare perception evaluation dataset to obtain the correlation results of nonlinear effects of different features.

[0010] Step 4: For the combined effect of multiple variables on the glare environment, recursive feature elimination under machine learning is used to perform feature analysis on the glare perception evaluation dataset to obtain the optimal combination of multiple features on the glare effect.

[0011] Further, step 1 specifically includes:

[0012] Step 1.1: To determine the experimental site, conduct on-site surveys of multiple office buildings and collect information on the characteristics and types of typical office buildings;

[0013] Step 1.2: To obtain spatial influencing factors, select several representative office building spaces based on statistical data and obtain building space information and usage information;

[0014] Step 1.3: In order to construct and analyze the dataset, glare perception experiments were conducted under a certain experimental period in a typical season. Subjective glare perception evaluation data were collected through questionnaires, and data on potential influencing factors of glare in the natural light environment were also collected.

[0015] Step 1.4: To ensure data validity, conduct reliability and validity analysis on the subjective perception evaluation of glare to determine whether the experiment needs to be repeated.

[0016] Furthermore, the spatial influencing factors include office type, density of office work areas, orientation of the lighting facade, window-to-wall ratio, and office space height; usage information includes whether it is a single-person or multi-person office, distance between office seats and windows, presence or absence of direct sunlight perceived by the human eye at the time of measurement, and frequency of direct sunlight occurrence over a cumulative period.

[0017] Furthermore, step 2 specifically includes:

[0018] Step 2.1: To meet the conditions for analysis of variance, the data is standardized by one-hot coding, the data is translated into typographic data, and the homogeneity of variance and the data are guaranteed to follow a normal distribution.

[0019] Step 2.2: To obtain the linear correlation between architectural space information, usage information, natural light environment information and user glare evaluation data, a one-way ANOVA is conducted.

[0020] Step 2.3: To improve the accuracy of the model in tree-based feature analysis, linearly uncorrelated variables are eliminated, thereby determining the feature data used in Step 3.

[0021] Furthermore, step 3 specifically includes:

[0022] Step 3.1: To meet the learning conditions of the tree model algorithm, the data is one-hot encoded and standardized, the data type is translated and the data amplitude is kept consistent, thereby removing the influence of feature amplitude on the correlation.

[0023] Step 3.2: To improve the accuracy of the training and testing tree models, the encoded data is divided into training and testing sets.

[0024] Step 3.3: To avoid the influence of algorithm type on feature selection results, four different tree model algorithms are used to train the hyperparameter-optimized model on the same dataset, and then the prediction performance is verified using the validation set. The optimal model is selected based on the model prediction performance index.

[0025] Step 3.4: To obtain the information's nonlinear impact on glare, the contribution values ​​of the information to the output prediction are ranked based on the feature coefficients in the optimal tree model.

[0026] Step 3.5: To further analyze the interaction and specific feature influence between information, the Model Interpretable Analysis (SHAP) method is used to interpret the features of different information in the data.

[0027] Furthermore, the four tree model algorithms are random forest algorithm, extreme random tree algorithm, gradient boosting tree algorithm, and extreme gradient boosting tree algorithm.

[0028] Furthermore, the model interpretability analysis method SHAP uses TreeExplainer under SHAP to perform different types of model interpretability analysis, including SHAP value analysis of prediction results of different features for typical samples, SHAP value distribution of different features for all test samples, average SHAP value analysis of different features, feature hierarchical clustering analysis, and feature interaction analysis.

[0029] Furthermore, step 4 specifically includes:

[0030] Step 4.1: To meet the learning requirements of the recursive feature elimination kernel evaluator, the data is standardized by one-hot encoding, the type data is translated and the data amplitude is kept consistent, thereby removing the influence of feature amplitude on correlation.

[0031] Step 4.2: In order to obtain the impact of the optimal combination of elements on glare perception, a recursive feature elimination method is used to successively delete different element features, and at the same time, the data model after removing the feature is iteratively learned.

[0032] Step 4.3: To verify the data model with different feature combinations, the five-fold cross-validation method is used to carry out data training and validation, and the average result of five training iterations is calculated.

[0033] Step 4.4: To determine the optimal features, the elbow method is used to observe the changes in the number and performance of features, obtain the optimal number of features at the performance inflection point, and at the same time, the SHAP decision method is used to further observe the changes in the cumulative effect of features, thereby clarifying the optimal information parameters in the natural lighting glare prediction model.

[0034] This invention proposes an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection.

[0035] This invention proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection.

[0036] The present invention has the following beneficial effects:

[0037] This invention proposes a method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection. This method aims to address a series of problems, including incomplete and superficial research on glare influencing factors and the limitations of traditional research methods that can only study linear correlations of single variables. By employing feature selection technology and the interpretable feature analysis method of the SHAP model in machine learning, it utilizes univariate analysis, tree-based model analysis, and recursive feature elimination analysis algorithms to construct interpretable models that cater to multi-perspective feature analysis. Through SHAP analysis, it obtains the importance distribution of office space parameters, building lighting environment parameters, and their interaction combinations on the perceived discomfort caused by glare. This provides a scientific basis for clarifying the impact of common natural lighting environment elements on the perceived discomfort, breaking through the technical bottlenecks in glare research and analysis, filling the gap in research on the interaction and combination of glare elements, and improving the accuracy and scientific level of indoor glare discomfort prediction. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0039] Figure 1 This is a flowchart of a method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection, according to the present invention.

[0040] Figure 2 This is a schematic diagram showing the characteristic ranking results (F-statistics) of different glare influencing factors under the test of one-way variance F-value.

[0041] Figure 3 This is a schematic diagram of the feature ranking results (-Log(p_{value}) under different glare influencing factors tested using the one-way variance F-value test.

[0042] Figure 4 This is a schematic diagram showing the magnitude of the weight coefficients of different features in the XGboost model after training in the case study.

[0043] Figure 5 The figure shows the evaluation results of AUROC on random forest, extreme random tree, gradient boosting tree and extreme boosting tree models in the case study.

[0044] Figure 6 To explain the trained XGBoost model using the SHAP method in the case study, we obtain the SHAP values ​​of glare influencing factors with respect to the prediction results for all test samples.

[0045] Figure 7 This is a schematic diagram illustrating how the SHAP method can be used in the case study to explain the predictive influence of different features at each level of perceived glare discomfort.

[0046] Figure 8 The figure shows the results of the analysis of the model accuracy under the change of the number of features in the case study, combined with recursive feature elimination.

[0047] Figure 9 This is a schematic diagram illustrating the cumulative predictive impact of candidate features determined using the SHAP decision method in the case study. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] This invention proposes a method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection, comprising the following steps:

[0050] Step 1: In the selected typical office building space, collect data on potential influencing factors of indoor glare and glare evaluation data to construct a glare perception evaluation dataset.

[0051] Step 2: For the linear influence of a single variable on the glare environment, use analysis of variance to perform feature analysis on the glare perception evaluation dataset and obtain the correlation results of the linear influence of different features.

[0052] Step 3: To analyze the nonlinear effects of variables on the glare environment and the combined correlations between variables, a tree model based on machine learning is used to perform feature analysis on the glare perception evaluation dataset to obtain the correlation results of nonlinear effects of different features.

[0053] Step 4: For the combined effect of multiple variables on the glare environment, recursive feature elimination under machine learning is used to perform feature analysis on the glare perception evaluation dataset to obtain the optimal combination of multiple features on the glare effect.

[0054] Step 1 specifically involves:

[0055] Step 1.1: To determine the experimental site, conduct on-site surveys of multiple office buildings and collect information on the characteristics and types of typical office buildings;

[0056] Step 1.2: To obtain spatial influencing factors, select several representative office building spaces based on statistical data and obtain building space information and usage information;

[0057] Step 1.3: In order to construct and analyze the dataset, glare perception experiments were conducted under a certain experimental period in a typical season. Subjective glare perception evaluation data were collected through questionnaires, and data on potential influencing factors of glare in the natural light environment were also collected.

[0058] Step 1.4: To ensure data validity, conduct reliability and validity analysis on the subjective perception evaluation of glare to determine whether the experiment needs to be repeated.

[0059] The spatial influencing factors specifically include office type, density of office work areas, orientation of the lighting facade, window-to-wall ratio, and office space height; usage information includes whether it is a single-person or multi-person office, distance between office seats and windows, presence or absence of direct sunlight as perceived by the human eye at the time of measurement, and frequency of direct sunlight occurrence over a cumulative period.

[0060] The questionnaire uses a seven-level semantic scale to describe the comfort level of indoor lighting, screen lighting, and window lighting, and a four-level semantic scale to describe the overall glare perception level (unbearable, disturbing, perceptible, unperceptible). The questionnaire design not only provides overall glare evaluation data, but also verifies whether the source of glare is natural lighting.

[0061] The collected potential glare elements in natural light environments include luminance, luminance ratio, and illuminance. Luminance includes the average luminance of the window, the average luminance of the desktop, and the maximum luminance within the viewing angle (glare luminance); luminance ratio includes the luminance ratio of the glare source to the background and the luminance ratio of the desktop; illuminance includes the vertical illuminance to the human eye and the average illuminance of the desktop. These elements are all derived from commonly used components in existing glare evaluation and prediction indices, and have been studied extensively.

[0062] The questionnaire results underwent reliability and validity analysis. For reliability analysis, Cronbach's coefficient alpha method was used, which is the most commonly used reliability analysis method. A reliability score of 0.7 or higher was considered excellent, with 0.8 being the best. For validity analysis, factor analysis was performed using the Kaiser Meryer Olkin method (KMO) and Bartlett's test of sphericity to obtain KMO values, chi-square values, and p-values. The KMO value needed to be greater than 0.8, and the p-value needed to be less than 0.05. If reliability and validity were satisfactory, the data could be used for further correlation analysis.

[0063] Step 2 specifically involves:

[0064] Step 2.1: To meet the conditions for analysis of variance, perform one-hot coding and standardized coding on the data, translate the typographic data and ensure homogeneity of variance and that the data follows a normal distribution;

[0065] Step 2.2 involves conducting a one-way ANOVA to determine the linear strength of the correlation between architectural space information, usage information, natural light environment information, and user glare evaluation data. An F-test is used to rank the importance of different feature selections, based on the size of the F-statistic and the p-value. To emphasize the differences in p-values, -Log(p_{value}) is used to logarithmize the p-values.

[0066] Step 2.3: To improve the accuracy of the tree-based feature analysis in Step 3, linearly uncorrelated variables are eliminated based on the magnitude of differences between features, thereby determining the feature data to be used in Step 3.

[0067] Step 3 specifically involves:

[0068] Step 3.1: To meet the learning conditions of the tree model algorithm, the data is one-hot encoded and standardized, the data type is translated and the data amplitude is kept consistent, thereby removing the influence of feature amplitude on the correlation.

[0069] Step 3.2: To improve the accuracy of the training and testing tree models, the encoded data is divided into training and testing sets.

[0070] Step 3.3: To avoid the influence of algorithm type on feature selection results, four different tree model algorithms are used to train the hyperparameter-optimized model on the same dataset. The prediction performance is then validated using a validation set, and the optimal model is selected based on the model's prediction performance metrics. The four tree model algorithms are Random Forest, Extremely Random Tree, Gradient Boosting Tree, and Extreme Gradient Boosting Tree.

[0071] Step 3.4: To obtain the information on the nonlinear effect of glare, compare the four tree models and select the optimal tree model, and rank them according to the contribution of their characteristic coefficients to the output prediction.

[0072] Step 3.5, to further analyze the interactions and influences of specific features among the information, the model interpretability analysis method SHAP (SHapley Additive exPlanations) is used to interpret the features of different information in the data. SHAP is a model-agnostic method for interpreting its predictions and can be used with any machine learning estimator. It calculates the importance of each feature's influence by converting each feature into a SHAP value and incorporating it into the conditional expectation function of the model output. The calculation formula is as follows:

[0073] f x (S)=E|f(X)|do(X S =x S )

[0074] Where S is the conditional feature set, and X is a random variable representing the input features of the model. S It is the current predicted model input vector, and the changes produced at each step are attributed to the introduced features, and then the process is averaged over all possible feature rankings.

[0075] Specifically, TreeExplainer under SHAP is used to perform different types of model interpretation and analysis, including SHAP value analysis of prediction results for typical samples by different features (flow graph analysis and force graph analysis), SHAP value distribution of different features for all test samples (bee colony graph analysis), average SHAP value analysis of different features (summary graph analysis), feature hierarchical clustering analysis (clustering tree graph), and feature interaction analysis (interaction graph analysis). These analyses examine the nonlinear and interactive effects of glare impact information on subjective glare perception from different perspectives.

[0076] Step 4 specifically involves:

[0077] Step 4.1: To meet the learning requirements of the recursive feature elimination kernel evaluator, the data is standardized by one-hot encoding, the type data is translated and the data amplitude is kept consistent, thereby removing the influence of feature amplitude on correlation.

[0078] Step 4.2: To obtain the impact of the optimal combination of elements on glare perception, a recursive feature elimination method is used to successively delete different element features, while simultaneously iteratively learning the data model after removing the feature. The kernel evaluator selects the optimal tree model algorithm from step 3;

[0079] Step 4.3: To verify the data model with different feature combinations, the five-fold cross-validation method is used to carry out data training and validation, and the average result of five training iterations is calculated.

[0080] Step 4.4: To determine the optimal features, the elbow method is used to observe the changes in the number and performance of features, obtain the optimal number of features at the performance inflection point, and at the same time, the SHAP decision method is used to further observe the changes in the cumulative effect of features, thereby clarifying the optimal information parameters in the natural lighting glare prediction model.

[0081] Example

[0082] Combination Figure 1 This embodiment illustrates a method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection, comprising the following steps:

[0083] Step 1: In the selected typical office building space, collect data on potential influencing factors of indoor glare and glare evaluation data to construct a glare perception evaluation dataset.

[0084] Step 2: For the linear influence of a single variable on the glare environment, use analysis of variance to perform feature analysis on the glare perception evaluation dataset and obtain the correlation results of the linear influence of different features.

[0085] Step 3: To analyze the nonlinear effects of variables on the glare environment and the combined correlations between variables, a tree model based on machine learning is used to perform feature analysis on the glare perception evaluation dataset to obtain the correlation results of nonlinear effects of different features.

[0086] Step 4: For the combined effect of multiple variables on the glare environment, recursive feature elimination under machine learning is used to perform feature analysis on the glare perception evaluation dataset to obtain the optimal combination of multiple features on the glare effect.

[0087] Step 1 specifically involves:

[0088] Step 1.1: To determine the experimental site, conduct on-site surveys of multiple office buildings and collect information on the characteristics and types of typical office buildings;

[0089] Step 1.2: To obtain spatial influencing factors, select several representative office building spaces based on statistical data and obtain building space information and usage information;

[0090] Step 1.3: In order to construct and analyze the dataset, glare perception experiments were conducted under a certain experimental period in a typical season. Subjective glare perception evaluation data were collected through questionnaires, and data on potential influencing factors of glare in the natural light environment were also collected.

[0091] Step 1.4: To ensure data validity, conduct reliability and validity analysis on the subjective perception evaluation of glare to determine whether the experiment needs to be repeated.

[0092] The spatial influencing factors specifically include office type, density of office work areas, orientation of the lighting facade, window-to-wall ratio, and office space height; usage information includes whether it is a single-person or multi-person office, distance between office seats and windows, presence or absence of direct sunlight as perceived by the human eye at the time of measurement, frequency of direct sunlight occurrence over a cumulative period, usage time, and usage season.

[0093] The questionnaire uses a seven-level semantic scale to describe the comfort level of indoor lighting, screen lighting, and window lighting, and a four-level semantic scale to describe the overall glare perception level (unbearable, disturbing, perceptible, unperceptible). The questionnaire design not only provides overall glare evaluation data, but also verifies whether the source of glare is natural lighting.

[0094] To determine whether glare perception stems from changes in natural lighting, correlation analyses were conducted between indoor lighting comfort, screen lighting comfort, and overall glare perception. To determine whether glare perception originates from window orientation, a correlation analysis was conducted between window lighting comfort and overall glare perception. Both analyses were performed using one-way ANOVA. ANOVA can compare the significant effects of two or more independent samples to determine if the statistical correlation hypothesis between the samples is invalid. One-way ANOVA studies whether different levels of a control variable have a significant effect on the observed variable. Its principle is to decompose the total sum of squares of deviations of the observed variable into the between-group sum of squares (SSR) and the within-group sum of squares (SSE). Differences in repeated observations for the same treatment of the sample can be classified as within-group bias (SSR), often influenced by noisy data; differences in expectations due to different treatments of the sample are classified as between-group bias (SSE). If the sum of squares of the inter-group deviations accounts for a large proportion, it proves that the change is caused by the control variable and has a significant impact. The specific approach is as follows:

[0095] 1. Establish the hypotheses, setting H0 (no difference in means among different factor levels) and H1 (significant difference in means among different factor levels); 2. Calculate the F-statistic, which is the quotient of the mean square between groups (MSA) and the mean square within groups (MSE), where MSA is SSR / (k-1) and MSE is SSE / (nk), where k is the number of sample groups and n is the sample size. The F-test is also called the homogeneity of variance test. Given a significance level (confidence level) α, if F > F0... a This indicates unequal variances, rejecting the null hypothesis. The tested factor has a significant impact on the observed values ​​and is correlated. Generally, the larger the F-value, the more significant the significance. 3. Determine the p-value using the probability value test: The significance level threshold for the p-value is 0.05. If the p-value is less than 0.05 and closer to 0, the more it indicates a normal distribution and better significance. 4. Determine significant differences: If significant differences exist, use multiple comparisons to identify which levels have significant differences. A commonly used method is the Tukey Honestly Significant Difference (HSD) test. For example, an ANOVA significance analysis of the subjective evaluation results of glare revealed the significance levels as shown in Table 1. Both indoor lighting evaluation and window lighting comfort evaluation were correlated with glare ratings, proving that light perception originates from changes in natural lighting and glare perception comes from window orientation.

[0096] Table 1. Comparison of significance between different subjective evaluation contents and evaluation levels.

[0097]

[0098] The collected potential glare elements in natural light environments include luminance, luminance ratio, and illuminance. Luminance includes the average luminance of the window, the average luminance of the desktop, and the maximum luminance within the viewing angle (glare luminance); luminance ratio includes the luminance ratio of the glare source to the background and the luminance ratio of the desktop; illuminance includes the vertical illuminance to the human eye and the average illuminance of the desktop. These elements are all derived from commonly used components in existing glare evaluation and prediction indices, and have been studied extensively.

[0099] The questionnaire results underwent reliability and validity analysis. For reliability analysis, Cronbach's coefficient alpha method was used, which is the most commonly used reliability analysis method. A reliability score of 0.7 or higher was considered excellent, with 0.8 being the best. For validity analysis, factor analysis was performed using the Kaiser Meryer Olkin method (KMO) and Bartlett's test of sphericity to obtain KMO values, chi-square values, and p-values. The KMO value needed to be greater than 0.8, and the p-value needed to be less than 0.05. If reliability and validity were satisfactory, the data could be used for further correlation analysis.

[0100] Step 2 specifically involves:

[0101] Step 2.1: To meet the conditions for analysis of variance, one-hot encoding and standardized encoding are performed on the data to translate the categorical data and ensure homogeneity of variance and that the data follows a normal distribution. One-hot encoding is a commonly used data preprocessing method for handling discrete features. It uses registers of the total number of categories to encode the number of states for that category using either 0 or 1, with each state having its own independent register bit. The advantage of one-hot encoding is that it extends the discrete feature values ​​to Euclidean space, while feature selection based on similarity involves mapping features to Euclidean space for computation.

[0102] Step 2.2 involves conducting an F-test based on one-way variance to determine the linear strength of the correlation between architectural space information, usage information, natural light environment information, and user glare evaluation data. The F-test is used to rank the importance of different feature selections, based on the size of the F-statistic and the p-value. To emphasize the difference in p-values, -Log(p_{value}) is used to logarithmize the p-values. For example... Figure 2 , Figure 3The table shows the feature ranking results for different glare influencing factors based on the one-way variance F-value test (F-statistics and -Log(p_{value}). It can be seen that the vertical illuminance of the human eye has the strongest linear correlation with glare evaluation and prediction, while the influence of season and direct light does not show a significant effect.

[0103] Step 2.3: To improve the accuracy of the tree-based feature analysis in Step 3, linearly uncorrelated variables are eliminated based on the magnitude of differences between features, thereby determining the feature data to be used in Step 3.

[0104] Step 3 specifically involves:

[0105] Step 3.1: To meet the learning conditions of the tree model algorithm, the data is one-hot encoded and standardized, the data type is translated and the data amplitude is kept consistent, thereby removing the influence of feature amplitude on the correlation.

[0106] Step 3.2: To improve the accuracy of the training and testing tree models, the encoded data is divided into training and testing sets.

[0107] Step 3.3: To avoid the influence of algorithm type on feature selection results, four different tree model algorithms are used to train the hyperparameter-optimized model on the same dataset. The prediction performance is then validated using a validation set, and the optimal model is selected based on the model's prediction performance metrics. Hyperparameters define the model's complexity and learning ability. Different hyperparameter adjustments will produce different model performances; however, the patterns of performance changes are often difficult to obtain directly, requiring cross-validation to search the parameter space. In machine learning, this type of problem is collectively referred to as the hyperparameter optimization problem. Common solutions include grid search cross-validation, randomized search cross-validation, and Bayesian optimization search. Grid search cross-validation optimizes by exhaustively traversing all parameter combinations, which is often very time-consuming. Randomized search cross-validation randomly samples parameters based on spatial distribution and combines them into feasible solutions, reducing time significantly compared to grid search, but it is prone to missing key points. Bayesian optimization search uses a Gaussian distribution to obtain a surrogate model to fit the relationship between different hyperparameters and model evaluation, and iterates through combinations to obtain the best-performing hyperparameter combination. For small data samples, grid search can be used, while for large batches of samples, Bayesian optimization can be used.

[0108] The four tree model algorithms are Random Forest, Extremely Random Tree, Gradient Boosting Tree, and Extreme Gradient Boosting Tree. Random Forest and Extremely Random Tree belong to the averaging method of reducing variance (Bagging), while Gradient Boosting Tree and Extreme Gradient Boosting Tree benefit from the boosting method of reducing bias. The principle of averaging is to independently construct multiple estimators and then average their predictions by combining the estimators. In contrast, in the Boosting method, a series of weak learners are built on the basis of repeatedly modifying the data, and then all predictions are merged through weighted majority voting to obtain the final prediction. Since different algorithms have different adaptability to different data types, using different algorithms for unknown datasets under prior conditions helps to obtain the optimal model. For example, using the four tree models to perform feature selection on a dataset combining environmental data and glare evaluation data, the XGboost model showed the best performance after training, and was used for feature analysis. Figure 4 The values ​​of the weight coefficients for different features in the trained XGboost model can be used to compare the magnitude of the glare effects of different elements.

[0109] Model validation employs a combination of classification metrics for performance evaluation, including accuracy, precision, recall, F1 score, AUROC (area under the receiver operating characteristic), and Matthews correlation coefficient. The combined use of multiple metrics ensures the model is unaffected by imbalanced classification samples. For example, Figure 5 The AUC of the four tree models was evaluated using AUROC. It can be seen that the AUC of all four models is very high. However, other indicators are also very high, as shown in Table 2. Therefore, a multi-faceted evaluation is needed.

[0110] Table 2 shows the values ​​of different predictive performance evaluation indicators for the four models.

[0111]

[0112] Step 3.4: To obtain the information on the nonlinear effect of glare, compare the four tree models and select the optimal tree model, and rank them according to the contribution of their characteristic coefficients to the output prediction.

[0113] Step 3.5, to further analyze the interactions and influences of specific features among the information, the model interpretability analysis method SHAP (SHapley Additive exPlanations) is used to interpret the features of different information in the data. SHAP is a model-agnostic method for interpreting its predictions and can be used with any machine learning estimator. It calculates the importance of each feature's influence by converting each feature into a SHAP value and incorporating it into the conditional expectation function of the model output. The calculation formula is as follows:

[0114] f x (S)=E|f(X)|do(X S =x S )

[0115] Where S is the conditional feature set, and X is a random variable representing the input features of the model. S It is the current predicted model input vector, and the changes produced at each step are attributed to the introduced features, and then the process is averaged over all possible feature rankings.

[0116] Specifically, the TreeExplainer under SHAP is used to perform different types of model interpretation and analysis, including SHAP value analysis of different features for the prediction results of typical samples (flow graph analysis and force graph analysis), SHAP value distribution of different features for all test samples (bee colony graph analysis), average SHAP value analysis of different features (summary graph analysis), feature hierarchical clustering analysis (clustering tree graph), and feature interaction analysis (interaction graph analysis). These analyses examine the nonlinear and interactive effects of glare impact information on subjective glare perception from different perspectives. For example, the SHAP method is used to interpret the trained XGBoost model, obtaining the SHAP values ​​of glare impact factors for the prediction results in all test samples, such as... Figure 6 As shown, the horizontal axis represents the magnitude of the SHAP value of different features in each sample and the positive or negative direction of their influence. The color represents the magnitude of the sample value of each feature, which can reflect the strength of the feature's influence and also provide feedback on the correlation between the feature and the sample size.

[0117] The SHAP method can be used to explain the predictive influence of different features at each level of perceived glare discomfort, such as... Figure 7 As shown, the horizontal axis represents the mean absolute SHAP of different features across all samples, while in the horizontal bar chart for each feature, different colors represent the mean absolute SHAP predicted by that feature at different glare rating levels. This allows for a more detailed analysis of the impact of different potential glare influencing factors on glare perception.

[0118] Step 4 specifically involves:

[0119] Step 4.1: To meet the learning requirements of the recursive feature elimination kernel evaluator, the data is standardized by one-hot encoding, the type data is translated and the data amplitude is kept consistent, thereby removing the influence of feature amplitude on correlation.

[0120] Step 4.2: To obtain the impact of the optimal combination of elements on glare perception, a recursive feature elimination method is used to successively delete different element features, while simultaneously iteratively learning the data model after removing the feature. The kernel evaluator selects the optimal tree model algorithm from step 3;

[0121] Step 4.3: To verify the data model with different feature combinations, the five-fold cross-validation method is used to carry out data training and validation, and the average result of five training iterations is calculated.

[0122] Step 4.4, to determine the optimal features, uses the elbow method to observe the changes in the number and performance of features, obtaining the optimal number of features at the performance inflection point. Simultaneously, the SHAP decision method is used to further observe the cumulative effect of features, thereby clarifying the optimal information parameters in the natural lighting glare prediction model. For example, the model accuracy under changes in the number of features is analyzed in conjunction with recursive feature elimination. Figure 8 As shown, the horizontal axis represents the number of features, and the vertical axis represents the average prediction accuracy after five-fold cross-validation. It can be seen that the model has the best prediction effect when the number of variables is 3, indicating that an effective glare prediction model can be constructed by taking the top three most important influence parameters among the candidates of glare prediction elements. Figure 9 To determine the cumulative predictive influence of candidate features using the SHAP decision method, the vertical axis is arranged with feature names in ascending order of influence, and different lines represent the SHAP value of each sample after being superimposed with different features. Figure 8 It can be seen that vertical illuminance of the human eye, glare brightness, and average window brightness are the three candidate influencing parameters, and they are the most suitable for constructing glare evaluation and prediction indicators.

[0123] This invention proposes a method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection. This method aims to address a series of problems, including incomplete and superficial research on glare influencing factors and the limitations of traditional research methods that can only study linear correlations of single variables. By employing feature selection technology and the interpretable feature analysis method of the SHAP model in machine learning, it utilizes univariate analysis, tree-based model analysis, and recursive feature elimination analysis algorithms to construct interpretable models that cater to multi-perspective feature analysis. Through SHAP analysis, it obtains the importance distribution of office space parameters, building lighting environment parameters, and their interaction combinations on the perceived discomfort caused by glare. This provides a scientific basis for clarifying the impact of common natural lighting environment elements on the perceived discomfort, breaking through the technical bottlenecks in glare research and analysis, filling the gap in research on the interaction and combination of glare elements, and improving the accuracy and scientific level of indoor glare discomfort prediction.

[0124] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection.

[0125] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection.

[0126] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0127] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0128] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0129] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0130] The above provides a detailed description of the method for analyzing the influencing factors of indoor natural lighting glare in office buildings based on feature selection proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An office building indoor natural lighting glare influence factor analysis method based on feature selection, characterized in that: The method comprises the following steps: Step 1, in a selected typical office building space, collecting building indoor glare potential influencing element variable data and glare evaluation data, and constructing a glare perception evaluation data set; Step 2, for the linear influence of single variable on the glare environment, using variance analysis to analyze the feature of the glare perception evaluation data set, and obtaining different feature linear influence correlation results; Step 3, for the nonlinear influence of variables on the glare environment and the combined correlation between variables, using tree model under machine learning to analyze the feature of the glare perception evaluation data set, and obtaining different feature nonlinear influence correlation results; Step 4, for the joint influence of multiple variables on the glare environment, using recursive feature elimination under machine learning to analyze the feature of the glare perception evaluation data set, and obtaining the optimal combination of multiple features for the influence of glare; The step 3 is specifically: Step 3.1, in order to meet the learning conditions of tree model algorithm, the data is subjected to one-hot encoding and standardization encoding, the type data is translated and the data amplitude is ensured to be consistent, so as to remove the influence of feature amplitude on correlation; Step 3.2, in order to train the tree model and test the model accuracy, the coded data is divided into training set and test set; Step 3.3, in order to avoid the influence of algorithm type on feature selection results, four different tree model algorithms are used to train the model after super parameter optimization on the same data set, and the validation set is used to verify the prediction performance, and the optimal model is selected based on the model prediction performance index; Step 3.4, in order to obtain the nonlinear influence of information on glare, the contribution value of information in the output prediction is arranged based on the feature coefficient in the optimal tree model; Step 3.5, in order to further analyze the interaction between information and the influence of specific features, SHAP, a model interpretable analysis method, is used to explain the features of different information in the data; The step 4 is specifically: Step 4.1, in order to meet the learning of recursive feature elimination kernel evaluator, the data is subjected to one-hot encoding and standardization encoding, the type data is translated and the data amplitude is ensured to be consistent, so as to remove the influence of feature amplitude on correlation; Step 4.2, in order to obtain the optimal element combination for the influence of glare perception, the recursive feature elimination method is used to delete different element features one by one, and the data model after removing the feature is iteratively learned; Step 4.3, in order to verify the data model of different feature combinations, five-fold cross-validation method is used to carry out data training and verification, and the average result of five times training is obtained; Step 4.4, in order to determine the optimal feature, the elbow method is used to observe the change of feature quantity and performance, the best feature quantity at the performance inflection point is obtained, and SHAP decision method is used to further observe the feature cumulative effect change graph, so as to determine the optimal information parameter in the natural lighting glare prediction model.

2. The method of claim 1, wherein: The step 1 is specifically: Step 1.1, in order to determine the experimental site, a plurality of office buildings are investigated in the field, and the characteristic and type information of typical office buildings is counted; Step 1.2, in order to obtain the space influencing elements, according to the statistical data, a plurality of representative office building spaces are selected, and the building space information and use information are obtained; Step 1.3, for constructing analysis dataset, the glare perception experiment under a typical season and a certain experimental period is carried out, the subjective evaluation data of glare perception is collected by questionnaire survey, and the potential influencing factor data of glare in natural light environment is collected; Step 1.4, in order to ensure the validity of the data, the reliability and validity analysis of the subjective evaluation of glare is carried out to judge whether the experiment needs to be repeated.

3. The method of claim 2, wherein: The spatial influencing factors include office type, office work area density, lighting facade orientation, window-wall ratio size and office space height; the use information includes single or multiple office, office seat distance from window, whether direct light is perceived by human eye at the time of measurement and the frequency of direct light occurrence at the cumulative time.

4. The method of claim 1, wherein: The step 2 is specifically: Step 2.1, in order to meet the condition of variance analysis, the data is standardized by one-hot encoding, the type data is translated and the variance homogeneity is ensured, and the data meets the normal distribution; Step 2.2, in order to obtain the linear strength and weakness correlation between the building space information, use information, natural light environment information and the user's glare evaluation data, single factor variance analysis is carried out; Step 2.3, in order to improve the precision of the model in the tree model feature analysis, the linear irrelevant variables are removed, so as to determine the feature data used in step 3.

5. The method of claim 1, wherein: The four tree model algorithms are random forest algorithm, extreme random tree algorithm, gradient boosting tree algorithm and extreme gradient boosting tree algorithm.

6. The method of claim 5, wherein: The model interpretable analysis method SHAP uses TreeExplainer under SHAP to do different types of model interpretation analysis, including SHAP value analysis of different features on typical sample prediction results, SHAP value distribution of different features on all test samples, average SHAP value analysis of different features, feature hierarchical clustering analysis and feature interaction analysis. 7.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The processor executes the computer program to realize the steps of the method of any one of claims 1-6.

8. A computer readable storage medium for storing computer instructions, characterized in that, The computer instructions are executed by the processor to realize the steps of the method of any one of claims 1-6.

Images