A method for predicting heating demand based on artificial intelligence

Abstract

The application discloses a heating demand prediction method based on artificial intelligence and relates to the technical field of building heating. The method comprises the following steps: collecting building characteristic environment data and heating heat data, performing chi-square test and Pearson correlation analysis to obtain significant correlation values and correlation coefficients of the heating heat, calculating a heating heat importance index, and screening the building characteristic environment data. The screened data and the heating heat data are input into a support vector machine regression model for training; the building characteristic environment data are analyzed based on Fourier's law and Newton's cooling law, wall heat conduction loss values and window heat convection loss values are calculated, and a building heat balance dynamic equation is established; the building characteristic environment screening data are collected in real time and input into the support vector machine regression model to obtain a first heating heat prediction value; the real-time building characteristic environment data are input into the dynamic equation to obtain a second heating heat prediction value. Comprehensive prediction values are generated by fusing the two types of prediction values, and the heating demand prediction is realized.

Description

Technical Field

[0001] This invention relates to the field of building heating technology, specifically to a heating demand prediction method based on artificial intelligence. Background Technology

[0002] Heating demand forecasting plays a crucial role in building energy management, particularly in energy conservation, emission reduction, and improving energy efficiency. With the impact of global climate change and fluctuating energy prices, accurate heating demand forecasting has become increasingly critical for controlling building energy consumption. In recent years, with the development of artificial intelligence and big data technologies, heating demand forecasting methods have gradually evolved towards intelligence and precision. Machine learning algorithms, especially support vector machines and neural networks, are being widely applied in heating forecasting. These methods can handle complex nonlinear relationships and, by combining large amounts of real-time data such as meteorological data and building energy consumption data, improve forecast accuracy. Furthermore, intelligent forecasting methods can dynamically adjust models based on real-time data, further optimizing the forecast results.

[0003] Existing machine learning methods for heating demand forecasting often use raw features directly, lacking the ability to quantify the degree of influence of each raw feature on heating volume. The inclusion of redundant features in the prediction model leads to a decrease in the accuracy of heating demand forecasting. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based heating demand prediction method to solve the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a heating demand prediction method based on artificial intelligence, comprising the following steps:

[0006] Step S1: Collect building heating data and classify it by time to obtain a building heating dataset, which includes building characteristic environmental data and heating heat data;

[0007] Step S2: Calculate the correlation coefficient of heating heat by performing Pearson correlation analysis on the building characteristic environmental data and heating heat data; calculate the significant correlation value of heating heat by performing a chi-square test on the building characteristic environmental data and heating heat data.

[0008] Step S3: Combine the heating heat correlation coefficient and the heating heat significant correlation value to calculate the heating heat importance index; use the heating heat importance index to perform feature filtering on the building characteristic environment data to obtain building characteristic environment filtered data;

[0009] Step S4: Input the building characteristic environment screening data and heating data into the support vector machine regression model for training to obtain a trained support vector machine regression model; analyze the building characteristic environment data using Fourier's law to obtain the heat conduction loss value of the building walls; analyze the building characteristic environment data using Newton's law of cooling to obtain the heat convection loss value of the building windows; combine the heat conduction loss value of the building walls and the heat convection loss value of the building windows to establish a dynamic equation for building heat balance.

[0010] Step S5: Collect the building characteristic environment screening data in real time and input it into the trained support vector machine regression model to obtain the first heating heat prediction value; at the same time, input the real-time collected building characteristic environment data into the building heat balance dynamic equation and output the second heating heat prediction value.

[0011] Step S6: Combine the first and second heating heat forecast values ​​to calculate the comprehensive heating heat forecast value, thereby achieving the prediction of heating demand.

[0012] Preferably, the step of calculating the heating heat correlation coefficient by performing Pearson correlation analysis on building characteristic environmental data and heating heat data includes the following steps:

[0013] By performing Pearson correlation analysis on building characteristic environmental data and heating heat data, the correlation coefficient of heating heat was calculated:

[0014]

[0015] in, The first feature in the building characteristic environmental data The correlation coefficient of heating heat for each characteristic, The first feature in the building characteristic environmental data The first feature One data point, The first feature in the building characteristic environmental data The average of the features, The first in the heating heat data One data point, This represents the average value of the heating data. This indicates the total number of samples.

[0016] Preferably, the step of calculating the significant correlation value of heating heat by performing a chi-square test on building characteristic environmental data and heating heat data includes the following specific steps:

[0017] By statistically analyzing the building characteristic environmental data, the first The number of samples for each feature under the heating heat value label is used to construct a contingency table:

[0018]

[0019] in, The first in the building characteristic environmental data A contingency table of features and heating load labels. The first in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating load labels.

[0020] The first in the calculation of building characteristic environmental data DF The first feature The categories in the Expected frequency of heating heat label:

[0021]

[0022] in, The first characteristic in the building characteristic environmental data The first feature The categories in the The expected frequency of each heating heat value label. The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat labels, The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating load labels.

[0023] Calculate the chi-square statistic:

[0024]

[0025] in, The first feature in the building characteristic environmental data Chi-square statistic for each feature The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat labels, The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating load labels.

[0026] Perform a significance test: ,in, The first feature in the building characteristic environmental data The degrees of freedom of each feature are determined using the chi-square distribution critical value table, and the corresponding degrees of freedom are found to be... and significance level Chi-square distribution critical value ,when At that time, the first characteristic in the building characteristic environmental data is... These characteristics are significantly correlated with heating capacity;

[0027] The first characteristic of building property environmental data The Cramer correlation coefficients for each feature were used to calculate the significant correlation value for heating heat:

[0028]

[0029] in, The first feature in the building characteristic environmental data The heating heat value of each feature is significantly correlated. The first feature in the building characteristic environmental data Chi-square statistic for each feature For degrees of freedom and significance level Chi-square distribution critical value , The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat labels, The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating energy labels.

[0030] Preferably, the step of calculating the importance index of heating heat by combining the correlation coefficient and the significant correlation value of heating heat includes the following steps:

[0031] By combining the correlation coefficient and significant correlation value of heating heat, the importance index of heating heat is calculated:

[0032]

[0033] in, The first feature in the building characteristic environmental data The heating heat importance index of each characteristic, The first feature in the building characteristic environmental data The correlation coefficient of heating heat for each characteristic, The first feature in the building characteristic environmental data The heating heat value of each feature is significantly correlated.

[0034] Preferably, the step of using the heating heat importance index to filter building characteristic environmental data to obtain building characteristic environmental screening data includes the following specific steps:

[0035] The building characteristic environmental data is filtered for features using the heating heat importance index. The building characteristic environmental data is then further filtered for features. If the heating heat importance index of the first feature is greater than or equal to a preset threshold, then the second feature is retained. If the first feature is less than a preset threshold, then delete the first feature. These features ultimately yield building characteristic environment screening data.

[0036] Preferably, the step of analyzing building characteristic environmental data using Fourier's law to obtain the building wall heat conduction loss value includes the following specific steps:

[0037] By analyzing the environmental data of building characteristics using Fourier's law, the heat conduction loss value of the building walls was obtained:

[0038]

[0039] in, This represents the heat conduction loss value of the wall surface. The thermal conductivity of the wall material is determined by the material itself. The heat transfer area of ​​the wall. For the temperature inside the building, Building exterior temperature, This refers to the wall thickness.

[0040] Preferably, the step of analyzing building characteristic environmental data using Newton's law of cooling to obtain the heat convection loss value of building windows includes the following specific steps:

[0041] By analyzing building characteristic environmental data using Newton's law of cooling, the heat convection loss values ​​of building windows were obtained:

[0042]

[0043] in, This indicates the heat convection loss value of building windows. This represents the heat transfer coefficient of the form, which is determined by the material of the form itself. Represents the surface area of ​​the form. For the temperature inside the building, Building exterior temperature.

[0044] Preferably, the step of establishing a dynamic equation for building heat balance by combining the heat conduction loss value of the building walls and the heat convection loss value of the building windows includes the following specific steps:

[0045] By combining the heat conduction loss values ​​of the building walls and the heat convection loss values ​​of the building windows, a dynamic equation for the building's heat balance is established:

[0046]

[0047] in, for The building's heating capacity at all times For building heat capacity, ,in For building material density, For building volume, Specific heat capacity of building materials For the building's internal temperature at Rate of change at time t = , For time intervals, for The current indoor temperature for The current indoor temperature for The wall heat conduction loss value at that time. express The value of heat convection loss of building windows at any time. for Solar radiation gain at a given time = ,in, The solar radiation absorption rate of a window is determined by the material of the window itself. Represents the surface area of ​​the form. for The intensity of solar radiation at that moment.

[0048] Preferably, the real-time collection of building characteristic environment screening data and input into the trained support vector machine regression model to obtain the first predicted heating heat value; simultaneously, the real-time collected building characteristic environment data is input into the building heat balance dynamic equation to output the second predicted heating heat value, including the following specific steps:

[0049] Real-time data on building characteristics and environmental screening is collected and input into a trained support vector machine regression model to obtain the first predicted heating heat value.

[0050]

[0051] in, for The first heating heat forecast value at any given time. This indicates the number of support vectors. and All are Lagrange multipliers for the support vector machine regression model. Indicates the first The index of each support vector. The kernel function calculates the input building characteristics and environmental filtering data. With the support vectors The similarity is given by , where b is the bias term of the support vector machine regression model;

[0052] Simultaneously, the real-time collected building characteristic environmental data is input into the building heat balance dynamic equation, and the second predicted heating heat value is output:

[0053]

[0054] in, for The predicted second heating load at that time. For building heat capacity, ,in For building material density, For building volume, Specific heat capacity of building materials Temperature inside the building over time rate of change for The wall heat conduction loss value at that time. For solar radiation gain, express The heat convection loss value of building windows at a given time. = ,in, The solar radiation absorption rate of a window is determined by the material of the window itself. Represents the surface area of ​​the form. for The intensity of solar radiation at that moment.

[0055] Preferably, the step of calculating a comprehensive heating demand forecast by combining the first and second heating demand forecasts includes the following specific steps:

[0056] By combining the first and second predicted heating heat values, a comprehensive predicted heating heat value is calculated:

[0057]

[0058] in, for The comprehensive predicted value of heating heat at any given time. for The first heating heat forecast value at any given time. for The predicted second heating load at that time. This is a sensitivity coefficient used to adjust the rate at which the accuracy of the support vector machine regression model is affected. For the precision threshold, The accuracy of the support vector machine regression model is determined by the mean squared error of the support vector machine regression model.

[0059] Beneficial Effects: This invention provides a heating demand prediction method based on artificial intelligence, involving machine learning and deep learning technologies, which has the following beneficial effects:

[0060] (1) By combining the correlation coefficient and significant correlation value of heating heat to calculate the importance index of heating heat, the association analysis of categorical and continuous variables can be covered by multivariate statistical methods, avoiding the limitations of a single method on data types. Among them, the significant correlation value quantifies the nonlinear correlation between categorical features and heating demand, and the correlation coefficient measures the linear correlation strength of continuous features. The two together constitute a unified quantitative standard, enhancing the comprehensiveness and reliability of feature evaluation, providing multidimensional basis for feature selection, ensuring the retention of key environmental parameters that significantly affect heating heat, and improving the predictive efficiency of subsequent models.

[0061] (2) A dynamic equation for building heat balance is established by combining the heat conduction loss value of building walls and the heat convection loss value of building windows. Based on physical principles such as Fourier's law and Newton's law of cooling, the conduction, convection effects and solar radiation gain in the building heat transfer process are incorporated into a unified framework to dynamically describe the relationship between the internal temperature change of the building and the heating demand. This equation fully considers the building material properties such as thermal conductivity and specific heat capacity, structural parameters such as wall thickness and window area, and the external environment such as indoor and outdoor temperatures and solar radiation intensity, so that the prediction process has clear physical meaning. It can provide mechanistic heat demand analysis for different building types and climatic conditions, and enhance the interpretability and scenario adaptability of the prediction results.

[0062] (3) Combine the first and second heating heat prediction values ​​to calculate the comprehensive prediction value. Integrate the statistical learning ability of the data-driven model with the prior knowledge of the physical mechanism model to break through the limitations of the single modeling logic and achieve high accuracy, strong robustness and scenario adaptability of heating demand prediction. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a flowchart illustrating the steps of an artificial intelligence-based heating demand prediction method proposed in this invention.

[0065] Figure 2 This is a step-by-step diagram of an artificial intelligence-based heating demand prediction method proposed in this invention. Detailed Implementation

[0066] 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.

[0067] Please see Figures 1-2 The present invention provides a technical solution: a heating demand prediction method based on artificial intelligence.

[0068] Step S1: Collect building heating data and classify it by time to obtain a building heating dataset, which includes building characteristic environmental data and heating heat data.

[0069] The specific steps for collecting building heating data and classifying it by time to obtain a building heating dataset are as follows: First, multi-source data collection: This includes building characteristic environmental data and heating data. Building characteristic environmental data includes meteorological data and building feature data. Meteorological data includes real-time external temperature, humidity, wind speed, precipitation, sunshine duration, etc., covering the climate characteristics of different seasons and day and night. Building feature data covers physical properties such as building type (e.g., residential, office, commercial), building volume, building envelope parameters (wall material, thickness, thermal conductivity, window type, area, and insulation performance), material density, specific heat capacity, and insulation level. Heating data is historical heating operation data, including heating data for each time period. The collected data undergoes cleaning and preprocessing: missing value imputation (e.g., by interpolation between adjacent time periods or statistical mean) and outlier detection (based on statistical methods or box plots for identification and correction) to ensure data integrity and accuracy. Simultaneously, numerical data is standardized (e.g., Z-score standardization) to eliminate the impact of dimensional differences on subsequent analysis. Time-based classification and feature engineering: For meteorological and heating data, features are categorized into hourly, daily, and monthly time scales. Building characteristic data, being static attributes, are directly incorporated into the dataset as fixed features. Finally, dataset integration: Building characteristic environment screening data and heating data are aligned and merged by timestamp to form a structured building heating dataset.

[0070] Step S2: Calculate the correlation coefficient of heating heat by performing Pearson correlation analysis on the building characteristic environmental data and heating heat data; calculate the significant correlation value of heating heat by performing a chi-square test on the building characteristic environmental data and heating heat data.

[0071] By performing Pearson correlation analysis on building characteristic environmental data and heating heat data, the correlation coefficient of heating heat was calculated:

[0072]

[0073] in, The first feature in the building characteristic environmental data The correlation coefficient of heating heat for each characteristic, The first feature in the building characteristic environmental data The first feature One data point, The first feature in the building characteristic environmental data The average of the features, The first in the heating heat data One data point, This represents the average value of the heating data. This indicates the total number of samples.

[0074] It's important to note that Pearson correlation is a statistical method used to measure the strength and direction of the linear relationship between two continuous variables. In eigenvalue correlation analysis, the Pearson correlation coefficient helps determine whether a linear correlation exists between two characteristics. The Pearson correlation coefficient ranges from -1 to 1, where: 1 indicates a perfect positive correlation, with both characteristics increasing or decreasing accordingly; -1 indicates a perfect negative correlation, where an increase in one characteristic leads to a decrease in the other; and 0 indicates no linear correlation. Pearson correlation is commonly used to analyze relationships between numerical data. By calculating the Pearson correlation coefficient, we can help identify strong or weak associations between characteristics, thus providing a basis for further data modeling and prediction.

[0075] The heating data is binned into discrete categories, resulting in heating labels: high, medium, and low.

[0076] It should be noted that the heating heat data is binned into discrete categories to obtain heating heat labels. For example, if the heating heat... If the heating capacity label is low, then the heating energy label will be low. The heating capacity label is medium. The heating capacity label is high.

[0077] A chi-square test was performed on the building characteristic environmental data and the heating heat data after discretization.

[0078] It's important to note that the chi-square test is a statistical test primarily used to examine relationships or differences between categorical data. In feature association analysis, the chi-square test is mainly used to test whether a significant association exists between two categorical variables. By comparing observed frequencies with expected frequencies, the chi-square test helps determine whether two variables are independent or dependent. If there is no association between the two variables, the chi-square value is usually small, close to zero; conversely, if a significant association exists, the chi-square value is large, indicating a significant difference between the observed data and the hypothesized expected value. In short, the chi-square test reveals whether there is a statistically significant association between features and is widely used in market analysis, social surveys, medical research, and other fields to help analyze the relationships between different factors.

[0079] By statistically analyzing the building characteristic environmental data, the first The number of samples for each feature under the heating heat value label is used to construct a contingency table:

[0080]

[0081] in, The first in the building characteristic environmental data A contingency table of features and heating load labels. The first in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating energy labels.

[0082] It should be noted that the first part of the building characteristic environmental data... The first feature The first category, in the building characteristics environmental data If the first feature is a raw categorical variable, such as "Building Type" = Residential / Commercial / Office, or "Insulation Rating" = Grade A / Grade B / Grade C, it can be used directly; the second feature in the building characteristic environmental data... If a feature is a continuous variable (such as external temperature or wall thickness), it needs to be converted into a categorical variable (such as "external temperature" = low temperature / medium temperature / high temperature) by discretization methods, such as dividing it into fixed intervals.

[0083] The first in the calculation of building characteristic environmental data DF The first feature The categories in the Expected frequency of heating heat label:

[0084]

[0085] in, The first characteristic in the building characteristic environmental data The first feature The categories in the The expected frequency of each heating heat value label. The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat labels, The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating energy labels.

[0086] Calculate the chi-square statistic:

[0087]

[0088] in, The first feature in the building characteristic environmental data Chi-square statistic for each feature The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat labels, The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating energy labels.

[0089] Perform a significance test: ,in, The first feature in the building characteristic environmental data The degrees of freedom of each feature are determined using the chi-square distribution critical value table, and the corresponding degrees of freedom are found to be... and significance level Chi-square distribution critical value ,when At that time, the first characteristic in the building characteristic environmental data is... These characteristics are significantly correlated with heating heat, when Then the first characteristic in the building characteristic environmental data These characteristics are not significantly correlated with heating capacity.

[0090] It should be noted that the significance level Hypothesis testing is a core concept in statistics, used to quantify the maximum risk probability that a researcher will accept a "false rejection of the null hypothesis." It represents the likelihood of mistakenly believing a significant association exists when the null hypothesis (e.g., "the variables are not related") is actually true. For example, setting α=0.05 means allowing a maximum of 5 "false positive" conclusions (i.e., incorrectly classifying a relationship as related when there is none) in 100 independent tests. The choice of this threshold is usually a disciplinary convention (e.g., α=0.05 is commonly used in social sciences, while high-precision fields such as particle physics may require α=0.0000003), essentially controlling the risk of Type I errors (false positives).

[0091] The first characteristic of building property environmental data The Cramer correlation coefficients for each feature were used to calculate the significant correlation value for heating heat:

[0092]

[0093] in, The first feature in the building characteristic environmental data The heating heat value of each feature is significantly correlated. The first feature in the building characteristic environmental data Chi-square statistic for each feature For degrees of freedom and significance level Chi-square distribution critical value , The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat labels, The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating energy labels.

[0094] Step S3: Combine the heating heat correlation coefficient and the heating heat significant correlation value to calculate the heating heat importance index; use the heating heat importance index to perform feature filtering on the building characteristic environment data to obtain building characteristic environment filtered data.

[0095] By combining the correlation coefficient and significant correlation value of heating heat, the importance index of heating heat is calculated:

[0096]

[0097] in, The first feature in the building characteristic environmental data The heating heat importance index of each characteristic, The first feature in the building characteristic environmental data The correlation coefficient of heating heat for each characteristic, The first feature in the building characteristic environmental data The heating heat value of each feature is significantly correlated.

[0098] It should be noted that when performing the chi-square test, if the characteristic in the building characteristic environmental data is... This characteristic is not significantly correlated with heating heat, that is When, the default is the first The importance index of heating heat for each characteristic It equals 0.

[0099] It should be noted that the heating heat importance index is calculated by combining the heating heat correlation coefficient and the heating heat significant correlation value. This utilizes multivariate statistical methods to cover the association analysis of categorical and continuous variables, avoiding the limitations of single methods on data types. Double validation enhances the comprehensiveness and reliability of feature assessment, while a unified quantitative standard is constructed to support feature importance ranking. The heating heat significant correlation value is a quantitative representation of the correlation strength of categorical features in the importance index, directly reflecting the non-linear correlation between categorical variables (such as building type and insulation level) and heating demand, thus contributing to the categorical dimension of the importance index. The heating heat correlation coefficient, on the other hand, is the core indicator of the linear correlation strength of continuous features in the importance index. By quantifying the linear correlation between continuous variables (such as external temperature and wall thermal conductivity) and heating heat, it lays the foundation for continuous dimension assessment of the importance index. Together, these two factors constitute the core of the importance index calculation, enabling multi-dimensional value mining of building characteristic environmental data.

[0100] The building characteristic environmental data is filtered for features using the heating heat importance index. The first feature in the building characteristic environmental data... If the heating heat importance index of the first feature is greater than or equal to a preset threshold, then the second feature is retained. If the first feature is less than a preset threshold, then delete the first feature. These features ultimately yield building characteristic environment screening data.

[0101] Step S4: Input the building characteristic environment screening data and heating data into the support vector machine regression model for training to obtain a trained support vector machine regression model; analyze the building characteristic environment data using Fourier's law to obtain the heat conduction loss value of the building walls; analyze the building characteristic environment data using Newton's law of cooling to obtain the heat convection loss value of the building windows; combine the heat conduction loss value of the building walls and the heat convection loss value of the building windows to establish a dynamic equation for building heat balance.

[0102] After standardizing the building characteristic environment screening data and heating data, the building characteristic environment screening data was used as the independent variable and the heating data as the dependent variable. Both sets were divided into training and test sets in a 7:3 ratio and input into a Support Vector Machine (SVM) regression model for training. The Radial Basis Function (RBF) of the SVM regression model was prioritized, and the regularization parameter C and kernel function parameter γ of the SVM regression model were optimized using a grid search method. Cross-validation was used to evaluate the model's generalization ability and avoid overfitting. During training, the mean squared error (MSE) trend of the model was monitored. Training was stopped when the error no longer decreased significantly after several consecutive iterations. Finally, the model's predictive performance was quantified using indicators such as the mean absolute error (MAE) and root mean square error (RMSE) of the test set to ensure that the trained SVM regression model has reliable generalization ability.

[0103] By analyzing the environmental data of building characteristics using Fourier's law, the heat conduction loss value of the building walls was obtained:

[0104]

[0105] in, This represents the heat conduction loss value of the wall surface. The thermal conductivity of the wall material is determined by the material itself. The heat transfer area of ​​the wall. For the temperature inside the building, Building exterior temperature, This refers to the wall thickness.

[0106] It's important to note that Fourier's law is a fundamental physical law in the field of heat conduction, used to quantify the rate at which heat is transferred through solid materials. Its core principle reveals that, under steady-state conditions, the heat flow rate through a homogeneous material per unit time is directly proportional to the temperature gradient across the material and the cross-sectional area perpendicular to the heat flow direction, and inversely proportional to the material's thickness in the heat flow direction. This proportionality constant is the material's thermal conductivity, an inherent physical property characterizing its ability to conduct heat. Higher thermal conductivity, such as in metals, indicates stronger thermal conductivity; lower thermal conductivity, such as in insulating foam, indicates better thermal insulation.

[0107] By analyzing building characteristic environmental data using Newton's law of cooling, the heat convection loss values ​​of building windows were obtained:

[0108]

[0109] in, This indicates the heat convection loss value of building windows. This represents the heat transfer coefficient of the form, which is determined by the material of the form itself. Represents the surface area of ​​the form. For the temperature inside the building, Building exterior temperature.

[0110] It's important to note that Newton's law of cooling is a fundamental physical law describing convective heat transfer between a fluid and a solid surface, used to quantify the rate of heat transfer caused by airflow. Its core principle states that the heat lost through surface convection per unit time is directly proportional to the temperature difference between the solid surface and the surrounding fluid, and the surface area involved in the heat transfer. This proportionality constant is called the convective heat transfer coefficient, which comprehensively reflects the influence of fluid properties (such as air density, viscosity, and specific heat capacity), flow state (natural or forced convection), and surface characteristics (such as roughness and orientation) on the intensity of heat transfer. A higher heat transfer coefficient (e.g., a window in strong wind) results in faster convective heat dissipation; conversely, a lower coefficient (e.g., a window in still air) leads to slower heat dissipation.

[0111] By combining the heat conduction loss values ​​of the building walls and the heat convection loss values ​​of the building windows, a dynamic equation for the building's heat balance is established:

[0112]

[0113] in, for The building's heating capacity at all times For building heat capacity, ,in For building material density, For building volume, Specific heat capacity of building materials For the building's internal temperature at Rate of change at time t = , For time intervals, for The current indoor temperature for Indoor temperature at time -1 for The wall heat conduction loss value at that time. express The value of heat convection loss of building windows at any time. for Solar radiation gain at a given time = ,in, The solar radiation absorption rate of a window is determined by the material of the window itself. Represents the surface area of ​​the form. for The intensity of solar radiation at that moment.

[0114] Step S5: Collect the building characteristic environment screening data in real time and input it into the trained support vector machine regression model to obtain the first heating heat prediction value; at the same time, input the real-time collected building characteristic environment data into the building heat balance dynamic equation and output the second heating heat prediction value.

[0115] Real-time data on building characteristics and environmental screening is collected and input into a trained support vector machine regression model to obtain the first predicted heating heat value.

[0116]

[0117] in, for The first heating heat forecast value at any given time. This indicates the number of support vectors. and All are Lagrange multipliers for the support vector machine regression model. Indicates the first The index of each support vector. The kernel function calculates the input building characteristics and environmental filtering data. With the support vectors The similarity is given by , where b is the bias term of the support vector machine regression model.

[0118] Simultaneously, the real-time collected building characteristic environmental data is input into the building heat balance dynamic equation, and the second predicted heating heat value is output:

[0119]

[0120] in, for The predicted second heating load at that time. For building heat capacity, ,in For building material density, For building volume, Specific heat capacity of building materials For the building's internal temperature at Rate of change at time t for The wall heat conduction loss value at that time. For solar radiation gain, express The heat convection loss value of building windows at a given time. = ,in, The solar radiation absorption rate of a window is determined by the material of the window itself. Represents the surface area of ​​the form. for The intensity of solar radiation at that moment.

[0121] It should be noted that real-time collected building characteristic environmental data is input into the building heat balance dynamic equation, and the output yields a second predicted heating heat value. For example, if the heat capacity C is 1000 kJ / ℃, and the indoor temperature needs to be raised from 19℃ to 20℃, then... =1℃ / h, then ==1000kw, if If the total value during the change process is 50kW, then the predicted value of the second heating load is 1050kW.

[0122] Step S6: By combining the first and second predicted heating heat values, a comprehensive predicted heating heat value is calculated to achieve the prediction of heating demand.

[0123] By combining the first and second predicted heating heat values, a comprehensive predicted heating heat value is calculated:

[0124]

[0125] in, for The comprehensive predicted value of heating heat at any given time. for The first heating heat forecast value at any given time. for The predicted second heating load at that time. This is a sensitivity coefficient used to adjust the rate at which the accuracy of the support vector machine regression model is affected. For the precision threshold, The accuracy of the support vector machine regression model is determined by the mean squared error of the support vector machine regression model.

[0126] It should be noted that the comprehensive prediction value is calculated by combining the first predicted heating heat value (output of the support vector machine regression model) and the second predicted heating heat value (output of the building heat balance dynamic equation). The weighting function for the first predicted heating load is the sigmoid function, which is the mean squared error of the support vector machine regression model on the test set. As input, the weights are dynamically assigned: when the support vector machine regression model has high accuracy, A value close to 1 indicates a preference for data-driven support vector machine (SVM) regression models. This is especially true when the accuracy of the SVM regression model is low. Approaching zero, the prediction relies on the physical model of the building's thermal balance dynamic equation. By integrating the statistical learning capabilities of data-driven models with the prior knowledge of physical mechanism models, the limitations of single modeling logic are overcome, achieving high accuracy, strong robustness, and scenario adaptability in heating demand prediction. This addresses the inherent limitations of single models: Data-driven models, such as support vector machines, can mine nonlinear correlations from historical data, but they depend on the coverage and quality of the data. In extreme climates (such as cold waves and abnormally high temperatures) or new building types (such as passive energy-saving buildings), the lack of sufficient historical samples can lead to data blind spots and prediction bias. Physical mechanism models, while based on physical principles such as Fourier's law and Newton's law of cooling, can characterize the essential laws of heat conduction and convection, but they are insufficient in capturing real-time environmental variables (such as indoor temperature fluctuations caused by user behavior and unsteady meteorological conditions) and dynamic changes in building parameters (such as thermal conductivity drift caused by wall aging). By combining the two types of models, machine learning fills the gaps in the physical model's characterization of complex nonlinear relationships, while the physical model provides mechanistic constraints for machine learning, preventing it from falling into the data fitting trap, and improving prediction reliability, especially when data is sparse or the scene changes abruptly.

[0127] This paper proposes an artificial intelligence-based method for predicting heating demand, achieving accurate predictions through a multi-step fusion of statistical data analysis and physical models. First, building heating data is collected and categorized by time, followed by data cleaning and feature engineering to construct a dataset. Then, significant correlation values ​​and correlation coefficients of heating heat are calculated using chi-square tests and Pearson correlation analysis. These two methods are combined to obtain an importance index for feature selection, and the selected data is then input into a support vector machine regression model for training. Simultaneously, a dynamic equation for building heat balance is established based on Fourier's law and Newton's law of cooling. Finally, the prediction results from the machine learning model and the physical model are integrated to achieve a comprehensive prediction of heating demand.

[0128] By combining the correlation coefficient and significant correlation value of heating demand to calculate the importance index of heating demand, a multivariate statistical method can be used to cover the association analysis of categorical and continuous variables, avoiding the limitations of a single method on data types. Specifically, the significant correlation value quantifies the non-linear correlation between categorical features and heating demand, while the correlation coefficient measures the linear correlation strength of continuous features. Together, they constitute a unified quantitative standard, enhancing the comprehensiveness and reliability of feature evaluation, providing multidimensional basis for feature selection, ensuring the retention of key environmental parameters that significantly affect heating demand, and improving the predictive performance of subsequent models.

[0129] A dynamic equation for building heat balance is established by combining the heat conduction loss values ​​of building walls and the heat convection loss values ​​of building windows. Based on physical principles such as Fourier's law and Newton's law of cooling, the equation incorporates the conduction, convection effects, and solar radiation gain in the building heat transfer process into a unified framework, dynamically describing the relationship between changes in building interior temperature and heating demand. This equation fully considers building material properties (such as thermal conductivity and specific heat capacity), structural parameters (such as wall thickness and window area), and the external environment (such as indoor and outdoor temperatures and solar radiation intensity), giving the prediction process clear physical meaning. It can provide mechanistic-level heat demand analysis for different building types and climatic conditions, enhancing the interpretability and scenario adaptability of the prediction results.

[0130] By combining the first heating heat forecast (output of the support vector machine regression model) and the second heating heat forecast (output of the building heat balance dynamic equation), a comprehensive forecast value is calculated. This integrates the statistical learning ability of the data-driven model with the prior knowledge of the physical mechanism model, breaking through the limitations of a single modeling logic and achieving high accuracy, strong robustness, and scenario adaptability in heating demand forecasting.

[0131] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0132] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A heating demand forecasting method based on artificial intelligence, characterized in that: Includes the following steps: Step S1: Collect building heating data and classify it by time to obtain a building heating dataset, which includes building characteristic environmental data and heating heat data; Step S2: Calculate the correlation coefficient of heating heat by performing Pearson correlation analysis on the building characteristic environmental data and heating heat data; calculate the significant correlation value of heating heat by performing a chi-square test on the building characteristic environmental data and heating heat data. Step S3: Combine the heating heat correlation coefficient and the heating heat significant correlation value to calculate the heating heat importance index; use the heating heat importance index to perform feature filtering on the building characteristic environment data to obtain building characteristic environment filtered data; Step S4: Input the building characteristic environment screening data and heating heat data into the support vector machine regression model for training to obtain a trained support vector machine regression model; analyze the building characteristic environment data using Fourier's law to obtain the heat conduction loss value of the building walls; specifically: ； in, This represents the heat transfer loss value of the building walls. The thermal conductivity of the wall material is determined by the material itself. The heat transfer area of ​​the wall. For the temperature inside the building, Building exterior temperature, The thickness of the wall; By analyzing building characteristic environmental data using Newton's law of cooling, the heat convection loss values ​​of building windows were obtained; specifically: ； in, This indicates the heat convection loss value of building windows. This represents the heat transfer coefficient of the form, which is determined by the material of the form itself. Represents the surface area of ​​the form. For the temperature inside the building, Building exterior temperature; Based on the heat conduction loss values ​​of the building walls and the heat convection loss values ​​of the building windows, a dynamic equation for the building's heat balance is established: ； in, for The building's heating capacity at all times For building heat capacity, ,in For building material density, For building volume, Specific heat capacity of building materials For the building's internal temperature at Rate of change at time t = , For time intervals, for The current indoor temperature for The current indoor temperature for The wall heat conduction loss value at a given time. express The value of heat convection loss of building windows at any time. for Solar radiation gain at a given time = ,in, The solar radiation absorption rate of a window is determined by the material of the window itself. Represents the surface area of ​​the form. for The intensity of solar radiation at that moment; Step S5: Collect the building characteristic environment screening data in real time and input it into the trained support vector machine regression model to obtain the first heating heat prediction value; at the same time, input the real-time collected building characteristic environment data into the building heat balance dynamic equation to output the second heating heat prediction value. Step S6: Combining the first and second predicted heating heat values, a comprehensive predicted heating heat value is calculated to predict heating demand. The formula for calculating the comprehensive predicted heating heat value is as follows: ； in, for The comprehensive predicted value of heating heat at any given time. for The first heating heat forecast value at any given time. for The predicted second heating load at that time. This is a sensitivity coefficient used to adjust the rate at which the accuracy of the support vector machine regression model is affected. For the precision threshold, The accuracy of the support vector machine regression model is determined by the mean squared error of the support vector machine regression model.

2. The heating demand forecasting method based on artificial intelligence according to claim 1, characterized in that: The process of calculating the correlation coefficient of heating heat by performing Pearson correlation analysis on building characteristic environmental data and heating heat data includes the following steps: By performing Pearson correlation analysis on building characteristic environmental data and heating heat data, the correlation coefficient of heating heat was calculated: ； in, The first feature in the building characteristic environmental data The correlation coefficient of heating heat for each characteristic, The first feature in the building characteristic environmental data The first feature One data point, The first feature in the building characteristic environmental data The average of the features, The first in the heating heat data One data point, This represents the average value of the heating data. This indicates the total number of samples.

3. The heating demand forecasting method based on artificial intelligence according to claim 2, characterized in that: The method of calculating a significant correlation value between building characteristic environmental data and heating heat data by performing a chi-square test includes the following specific steps: By statistically analyzing the building characteristic environmental data, the first The number of samples for each feature under the heating heat value label is used to construct a contingency table: ； in, The first in the building characteristic environmental data A contingency table of features and heating load labels. The first in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating load labels. The first in the calculation of building characteristic environmental data DF The first feature The categories in the Expected frequency of heating heat label: ； in, The first characteristic in the building characteristic environmental data The first feature The categories in the The expected frequency of each heating heat value label. The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat label The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating load labels. Calculate the chi-square statistic: ； in, The first feature in the building characteristic environmental data Chi-square statistic for each feature The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat label The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating load labels. Perform a significance test: ,in, The first feature in the building characteristic environmental data The degrees of freedom of each feature are determined using the chi-square distribution critical value table, and the corresponding degrees of freedom are found to be... and significance level Chi-square distribution critical value ,when At that time, the first characteristic in the building characteristic environmental data is... These characteristics are significantly correlated with heating capacity; The first characteristic of building property environmental data The Cramer correlation coefficients for each feature were used to calculate the significant correlation value for heating heat: ； in, The first feature in the building characteristic environmental data The heating heat value of each feature is significantly correlated. The first feature in the building characteristic environmental data Chi-square statistic for each feature For degrees of freedom and significance level Chi-square distribution critical value , The first characteristic in the building characteristic environmental data The first feature The categories in the Number of samples per heating heat label The first characteristic in the building characteristic environmental data The first feature Index of each category, Indicates the first An index of heating heat label The first element representing the environmental data of building characteristics The total number of categories for each feature. This indicates the number of heating load labels.

4. The heating demand forecasting method based on artificial intelligence according to claim 3, characterized in that: The calculation of the heating heat importance index by combining the heating heat correlation coefficient and the heating heat significant correlation value includes the following steps: By combining the correlation coefficient and significant correlation value of heating heat, the importance index of heating heat is calculated: ； in, The first feature in the building characteristic environmental data The heating heat importance index of each characteristic, The first feature in the building characteristic environmental data The correlation coefficient of heating heat for each characteristic, The first feature in the building characteristic environmental data The heating heat value of each feature is significantly correlated.

5. The heating demand forecasting method based on artificial intelligence according to claim 4, characterized in that: The process of filtering building characteristic environmental data using the heating heat importance index to obtain filtered building characteristic environmental data includes the following specific steps: The building characteristic environmental data is filtered for features using the heating heat importance index. The first feature in the building characteristic environmental data... If the heating heat importance index of the first feature is greater than or equal to a preset threshold, then the second feature is retained. If the first feature is less than a preset threshold, then delete the first feature. These features ultimately yield building characteristic environment screening data.

6. The heating demand forecasting method based on artificial intelligence according to claim 5, characterized in that: The process of collecting real-time building characteristic environment screening data and inputting it into a trained support vector machine regression model to obtain a first predicted heating heat value; simultaneously, inputting the real-time collected building characteristic environment data into the building heat balance dynamic equation to output a second predicted heating heat value, includes the following specific steps: Real-time data on building characteristics and environmental screening is collected and input into a trained support vector machine regression model to obtain the first predicted heating heat value. ； in, for The first heating heat forecast value at any given time. This indicates the number of support vectors. and All are Lagrange multipliers for the support vector machine regression model. Indicates the first The index of each support vector. The kernel function calculates the input building characteristics and environmental filtering data. With the support vectors similarity, This refers to the bias term in the support vector machine regression model; Simultaneously, the real-time collected building characteristic environmental data is input into the building heat balance dynamic equation, and the second predicted heating heat value is output: ； in, for The predicted second heating load at that time. For building heat capacity, ,in For building material density, For building volume, Specific heat capacity of building materials Temperature inside the building over time rate of change for The wall heat conduction loss value at a given time. For solar radiation gain, express The value of heat convection loss of building windows at a given time. = ,in, The solar radiation absorption rate of a window is determined by the material of the window itself. Represents the surface area of ​​the form. for The intensity of solar radiation at that moment.

Images