Energy regulation and control method for low-carbon building

Abstract

The invention provides an energy regulation and control method for a low-carbon building. The energy regulation and control method comprises the steps that missing value filling and normalization preprocessing are conducted on building energy data collected in real time; preliminarily classifying the preprocessed data into a first energy consumption mode group through K-means clustering; matching and combining the first energy consumption mode group based on the load demand to obtain a second energy consumption mode group; continuously recombining the second energy consumption mode group into a third energy consumption mode group according to the energy consumption time period; judging and associating the third energy consumption mode group to form a fourth energy consumption mode group according to the inter-group energy consumption intensity association degree; and finally, generating an energy optimization scheme corresponding to the fourth energy consumption mode group by using the gradient boosting decision tree model. The energy consumption and carbon emission can be reduced, and the building energy efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building energy management, and more particularly, to an energy regulation method for low-carbon buildings. BACKGROUND

[0002] With the increasing emphasis on environmental protection and sustainable development worldwide, low-carbon buildings have gradually become an important development direction in the field of architecture. The core of low-carbon buildings lies in optimizing energy use, reducing carbon emissions, and achieving sustainable operation of buildings. In the prior art, building energy management mainly relies on traditional energy monitoring and simple scheduling strategies. Although these methods can achieve preliminary management of energy to some extent, they often lack deep analysis and precise regulation capabilities for energy consumption characteristics. For example, traditional energy management systems can only perform simple statistics and monitoring on energy consumption, and cannot classify and optimize complex energy use patterns. In addition, existing technologies often ignore the integrity and accuracy of energy data, resulting in a lack of scientific basis for generating energy regulation strategies.

[0003] In the implementation of the embodiments of the present application, the inventors found that the prior art has at least the following problems or defects: the prior art cannot effectively preprocess and deeply analyze building energy data, making it difficult to accurately identify different energy use patterns; it lacks analysis of the correlation between energy use patterns, and cannot achieve fine energy regulation; the generation of regulation strategies is not intelligent, and cannot be dynamically adjusted according to real-time data, making it difficult to meet the high requirements of low-carbon buildings for energy optimization. SUMMARY

[0004] The present application provides an energy regulation method for low-carbon buildings, comprising: Step S1, performing missing value filling and normalization preprocessing on the energy consumption characteristics in the real-time collected building energy data sequence; Step S2, performing preliminary classification on the preprocessed building energy data sequence based on K-means energy consumption clustering to obtain a first energy use pattern group; Step S3, performing preliminary merging of the first energy use pattern group based on load demand matching to obtain a second energy use pattern group; Step S4, performing re-merging of the second energy use pattern group based on energy use period continuity judgment to obtain a third energy use pattern group; Step S5, performing group association of the third energy use pattern group based on inter-group energy consumption intensity correlation degree judgment to obtain a fourth energy use pattern group; Step S6, generating regulation strategies for the fourth energy use pattern group based on a gradient boosting decision tree model to obtain energy optimization schemes corresponding to each group in the fourth energy use pattern group.

[0005] Further, in the K-means-based energy consumption clustering process in step S2, it includes: Step S2-1, for the pre-processed data, the normalized temperature, energy consumption intensity and one-hot encoding are converted into points in a multi-dimensional device operation state mapping in a multi-dimensional space; Step S2-2, principal component analysis dimension reduction processing is performed on the multi-dimensional data, and the clustering centers satisfying the preset group number and the maximum iteration number are found in the reduced space by an iterative method; Step S2-3, according to the principle of minimizing the Euclidean distance, each point is assigned to the nearest group, and the energy consumption information corresponding to the point is stored in the energy use mode library to generate a first energy use mode group.

[0006] Further, in the preliminary merging process of the first energy use mode group in step S3, it includes: Step S3-1, for each data point in each group in the first energy use mode group, the time stamps are arranged in ascending order, and the typical load demand of each group is calculated; Step S3-2, based on the load demand distribution histogram of each group, the representative load demand of each group is calculated; Step S3-3, the representative load demands of each group in the first energy use mode group are compared and calculated for matching degree, and the groups satisfying the matching degree merging threshold are merged to obtain a second energy use mode group.

[0007] Further, the calculation process of the representative load demand of each group in step S3-2 includes: 1) a load demand distribution histogram is established, in which the horizontal axis is the load demand value and the vertical axis is the frequency; 2) a sliding window with a fixed window width and a sliding step is set in the histogram, and the frequency sum in each window is recorded while the window is sliding from the left side of the histogram; 3) the load demand value in the window with the maximum frequency sum is selected and weighted averaged as the representative load demand.

[0008] Further, the calculation method of the matching degree in step S3-3 is: comparing the absolute value of the difference between the representative load demands of two groups with the ratio of the preset load tolerance, and if the ratio is less than the set threshold, it is determined that the two groups can be merged.

[0009] Further, in the re-merging process of the second energy use mode group in step S4, it includes: Step S4-1, arrange the data in each group of the second energy use mode group in ascending order of time stamp, identify the energy use time period breakpoints and split the non-continuous groups; Step S4-2, calculate the start time and end time of each group after splitting, and extract the time period features; Step S4-3, merging the groups with time period features overlapping or adjacent according to time continuity to obtain a third energy consumption mode group.

[0010] Further, the merging condition in step S4-3 is that the time interval of the two groups is less than a preset time threshold and the load demand fluctuation amplitude is less than a set tolerance.

[0011] Further, in the group association process of the third energy consumption mode group in step S5, the following is included: Step S5-1, calculating the energy consumption intensity correlation coefficient of the two groups in the third energy consumption mode group based on the Pearson correlation coefficient formula; Step S5-2, if the correlation coefficient is greater than an association threshold, the two groups are marked as the same energy consumption behavior mode.

[0012] Further, in the gradient boosting decision tree model training in step S6, historical energy consumption data and corresponding control strategies are used as a training set to generate a decision tree set through multiple rounds of iteration; and in prediction, the feature parameters of the fourth energy consumption mode group are input to output the control strategies of each group.

[0013] Further, the control strategies include device start-stop timing, energy storage system charging and discharging plan, and renewable energy priority scheduling rules, and the control strategies are executed through a building energy management system to realize low-carbon optimization of building energy consumption.

[0014] The low-carbon building energy regulation method of the present application can perform fine preprocessing and multi-dimensional analysis on building energy data, accurately identify and classify different energy consumption modes through a series of steps such as K-means clustering, load demand matching, energy consumption time period continuity judgment, and energy consumption intensity correlation analysis between groups, and provide a solid data foundation for subsequent energy optimization. In addition, the control strategies generated based on the gradient boosting decision tree model can realize dynamic optimization of building energy consumption, and can adjust the device start-stop timing, energy storage system charging and discharging plan, and renewable energy priority scheduling rules according to real-time data, thereby reducing the energy consumption and carbon emissions of the building.

[0015] The present application can also generate more intelligent and accurate control strategies through the gradient boosting decision tree model trained through multiple rounds of iteration, improve energy utilization efficiency, and realize automation and intelligentization of building energy management. At the same time, the method can execute the optimized energy control strategies through the building energy management system to ensure the effective implementation of the control strategies, further improve the low-carbon level of building energy management, and provide strong support for realizing sustainable development in the building field. BRIEF DESCRIPTION OF DRAWINGS

[0016] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description read in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of an energy regulation method for a low-carbon building according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] The principles and spirits of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only so that those skilled in the art can better understand and implement the present application, and do not limit the scope of the present application in any way. On the contrary, these embodiments are provided so that the present application is more thorough and complete, and the scope of the present application is fully conveyed to those skilled in the art.

[0018] Those skilled in the art will appreciate that the embodiments of the present application can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present application can be embodied in the form of entire hardware, entire software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0019] It should be noted that any number of elements in the drawings are used only for example, and not limitation, and any naming is only for distinction, and does not have any limiting meaning.

[0020] The principles and spirits of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only so that those skilled in the art can better understand and implement the present application, and do not limit the scope of the present application in any way. On the contrary, these embodiments are provided so that the present application is more thorough and complete, and the scope of the present application is fully conveyed to those skilled in the art. Figure 1 Figure 1 A flowchart of an energy regulation method for a low-carbon building according to an embodiment of the present application. As shown in Figure 1 The energy regulation method for a low-carbon building includes: Step S1, missing value filling and normalization preprocessing are performed on energy consumption features in a real-time collected building energy data sequence; Step S2, based on K-means energy consumption clustering, the preprocessed building energy data sequence is preliminarily classified to obtain a first energy use mode group; Step S3, based on load demand matching, the first energy use mode group is preliminarily merged to obtain a second energy use mode group; Step S4, based on energy use period continuity judgment, the second energy use mode group is re-merged to obtain a third energy use mode group; Step S5, based on inter-group energy consumption intensity correlation degree judgment, the third energy use mode group is associated to obtain a fourth energy use mode group; Step S6, based on a gradient boosting decision tree model, a regulation strategy is generated for the fourth energy use mode group to obtain an energy optimization scheme corresponding to each group in the fourth energy use mode group.​

[0021] Building energy data sequence refers to various energy-related data generated by buildings during operation, including but not limited to consumption data of electric energy, thermal energy, water energy, etc. Energy consumption characteristics refer to specific patterns or attributes embodied by these data, such as energy intensity, equipment operating status, etc. Missing value filling is to handle missing cases that may occur during data collection to ensure data integrity; normalization preprocessing is to convert data into a unified range for subsequent processing. These steps are the basis of data preprocessing, which can effectively improve data quality and provide reliable data support for subsequent energy consumption clustering and control strategy generation.

[0022] In the K-means-based energy consumption clustering, the first energy use mode group refers to the data set with similar energy consumption characteristics obtained by preliminary classification through clustering algorithm. K-means clustering is a clustering algorithm that divides data into different clusters by calculating the distance between data points. In this embodiment, the normalized temperature, energy intensity, and device operating status converted by one-hot encoding are mapped into points in a multi-dimensional space, with these parameters representing the temperature conditions of the building environment, the energy consumption intensity per unit time, and the operating status of the equipment. Temperature is an important factor affecting building energy consumption, energy intensity reflects the energy use efficiency per unit area or per unit time, and equipment operating status is converted into multi-dimensional data through one-hot encoding to better represent the operating status of different equipment. Principal component analysis dimension reduction processing is to convert high-dimensional data into low-dimensional data to more efficiently find the clustering centers in the reduced space. The clustering center is the representative point of each cluster, and these centers are found through an iterative method under the conditions of meeting the preset group number and the maximum number of iterations, thereby completing the preliminary classification.

[0023] Preferably, in the step of generating the regulation strategy based on the gradient boosting decision tree model, the gradient boosting decision tree model is an ensemble learning method that builds multiple decision trees to gradually optimize the prediction performance of the model. In training the model, historical energy consumption data and corresponding regulation strategies are used as the training set, which contains past energy consumption and the regulation measures taken at the time. Through multiple iterations, a set of decision trees is generated. In the prediction stage, the feature parameters of the fourth energy consumption mode group are input, including but not limited to energy intensity, device operating status, time period characteristics, etc. The model will output the regulation strategy for each group according to these input parameters. The specific content of the regulation strategy includes device start-stop timing, energy storage system charging-discharging plan, and renewable energy priority scheduling rules. These strategies are executed through the building energy management system to achieve low-carbon optimization of building energy consumption. Device start-stop timing refers to arranging the start and stop times of devices according to energy demand and device characteristics. The energy storage system charging-discharging plan refers to arranging the charging and discharging times of energy storage devices according to energy supply and demand. The renewable energy priority scheduling rule refers to prioritizing the use of renewable energy under the premise of meeting energy demand. The implementation of these regulation strategies can effectively reduce building energy consumption and carbon emissions, achieving the goal of low-carbon buildings.

[0024] In some embodiments, in the K-means-based energy consumption clustering process in step S2, it includes: Step S2-1, for the pre-processed data, the normalized temperature, energy intensity, and one-hot encoded device operating status are mapped into a multi-dimensional space as points in the multi-dimensional space; Step S2-2, principal component analysis dimensionality reduction is performed on the multi-dimensional data, and the clustering centers that meet the preset number of groups and the maximum number of iterations are found in the reduced space through an iterative method; Step S2-3, according to the principle of minimizing Euclidean distance, each point is assigned to the nearest group, and the corresponding energy consumption information of the points is stored in the energy consumption mode library to generate the first energy consumption mode group.

[0025] It should be noted that in the energy consumption clustering process based on K-means, the pre-processed building energy data is mainly classified through a specific algorithm to identify different energy use patterns. The pre-processed data here refers to the building energy data sequence after missing value filling and normalization processing, which is more accurate and uniform, and can provide a reliable basis for clustering analysis. Normalized temperature refers to converting temperature data into a unified range for comparison and analysis with other data. Energy intensity refers to the amount of energy consumed per unit time, which is an important indicator of building energy efficiency. One-hot encoding to multi-dimensional device operating state is a data processing method that converts different operating states of devices into numerical form for use in clustering analysis. By mapping these data into points in a multi-dimensional space, clustering analysis can be more intuitive, providing a basis for subsequent energy regulation.

[0026] Specifically, the point in the multi-dimensional space in step S2-1 is a high-dimensional data point composed of temperature, energy intensity, and device operating state, with each dimension representing a feature. For example, temperature can be one dimension, energy intensity is another dimension, and device operating state can be converted to multiple dimensions through one-hot encoding. The principal component analysis dimension reduction processing in step S2-2 is a data dimension reduction technique that aims to reduce the dimension of data while retaining the main features of the data. Through dimension reduction, the computational complexity can be reduced, and it is easier to find patterns in the data. The iterative method to find clustering centers that meet the preset group number and maximum iteration number means adjusting the position of the clustering center until the set clustering group number and iteration number are met. The clustering center is the center point of each cluster, representing the average characteristics of the data points in the cluster. The Euclidean distance minimization principle in step S2-3 means that each data point is assigned to the group with the nearest clustering center, and the distance is calculated using Euclidean distance, which is a common distance measurement method. Finally, the energy consumption information corresponding to each point is stored in the energy use pattern library, forming the first energy use pattern group, providing basic data for subsequent energy regulation.

[0027] Preferably, when performing principal component analysis dimension reduction processing, the number of principal components to be retained can be selected, usually determined according to the cumulative variance contribution rate reaching a certain percentage, such as 95%, to ensure that the data after dimension reduction can retain most of the information of the original data. In the process of iteratively finding the cluster center, a convergence condition can be set, for example, when the change of the cluster center is less than a certain threshold, it is considered that the clustering has converged, and the iteration can be stopped. In addition, when calculating the Euclidean distance, weight adjustment can be performed on the data of each dimension to reflect the importance of different features in the clustering process. For example, if the energy intensity is more critical in energy regulation, a higher weight can be given to the energy intensity dimension. Through these refined operations, the accuracy and efficiency of clustering can be improved, and different energy use patterns can be better identified, providing more accurate data support for subsequent energy optimization strategies.

[0028] In some embodiments, in the preliminary merging process of the first energy use pattern groups in step S3, the following steps are included: Step S3-1, arrange the data points in each group of the first energy use pattern groups in ascending order according to the timestamp, and calculate the typical load demand of each group; Step S3-2, based on the load demand distribution histogram of each group, calculate the representative load demand of each group; Step S3-3, compare the representative load demands of each group in the first energy use pattern groups pairwise to calculate the matching degree, and merge the groups that meet the matching degree threshold to obtain the second energy use pattern groups.

[0029] It should be noted that in the process of preliminary merging of the first energy use pattern groups, the main purpose is to analyze the load demand characteristics of each group and merge energy use patterns with similar load demands, thereby simplifying the classification of energy use patterns and providing a clearer pattern basis for further energy regulation. The first energy use pattern groups refer to the preliminary classification results obtained by K-means clustering, each group containing a series of data points with similar energy consumption characteristics. Load demand refers to the amount of energy required by a building in a specific time period, which is an important indicator for measuring energy use patterns. The typical load demand refers to the average or representative load demand value of each energy use pattern group under normal operation conditions, which is used to represent the energy use characteristics of the group. The representative load demand is a value obtained by analyzing the load demand distribution, which can represent the load demand level of the group and is used for subsequent merging judgment.

[0030] Specifically, the time stamp ascending order arrangement in step S3-1 refers to sorting the data points in each energy consumption mode group in chronological order to better analyze the time variation law of its load demand. The calculation of typical load demand can be obtained by statistical analysis of the sorted data points, for example, calculating its average value or median, etc., to reflect the load level of the group under normal operation. The load demand distribution histogram in step S3-2 is a graphical representation method for showing the distribution of load demand values, with the horizontal axis representing the load demand value and the vertical axis representing the frequency. Through this distribution, the concentration trend of load demand can be intuitively seen. The calculation process of the representative load demand involves setting a sliding window in the histogram and finding the frequency sum and the maximum interval through the sliding window, and then performing a weighted average on the load demand values in the interval to obtain a value that can represent the load demand of the group. The matching degree in step S3-3 refers to the similarity between the representative load demands of the two energy consumption mode groups, which is determined by calculating the ratio of the absolute value of the difference between the two representative load demands to the preset load tolerance. If the ratio is less than the set threshold, it is considered that the load demands of the two groups are similar enough to be merged.

[0031] Preferably, in calculating the typical load demand, a weighted average method can be used to give different weights according to the time distribution or importance of the data points to more accurately reflect the load characteristics of the energy consumption mode. When constructing the load demand distribution histogram, the width and step of the sliding window can be adjusted according to the distribution of the actual data, for example, for data with relatively stable load demand changes, a larger window width can be used; while for data with frequent load demand changes, a smaller window width is needed to better capture the details. In calculating the representative load demand, the weights of the weighted average can be set according to the position or frequency of the load demand values within the window, for example, load demand values with higher frequency can be given higher weights. In judging the matching degree, the preset load tolerance can be set according to the actual energy consumption situation of the building and the control accuracy requirement, for example, for buildings with high energy control accuracy requirements, a smaller load tolerance can be set. Through these detailed operations, energy consumption modes with similar load demands can be more accurately identified and merged, providing more accurate input data for the subsequent energy control strategy generation.

[0032] In some embodiments, the calculation process of the representative load demand of each group in step S3-2 includes: 1) Establish a load demand distribution histogram, with the horizontal axis representing the load demand value and the vertical axis representing the frequency; 2) Set a sliding window with a fixed window width and a sliding step in the histogram, starting from the left side of the histogram and recording the frequency sum in each window; 3) Select the frequency and the maximum window position, and take the weighted average of the load demand values in the window as the representative load demand.

[0033] It should be noted that the calculation process of the representative load demand of each group in step S3-2 is mainly to determine the typical load demand value of each energy consumption mode group by analyzing the load demand distribution histogram. The load demand distribution histogram here is a chart used to show the distribution of load demand values, through which the concentrated area of load demand can be intuitively seen. The fixed window width and sliding step sliding window is a data analysis method, which finds the high frequency area of load demand by moving the window on the histogram. The frequency and the maximum window position refer to the area in the window where the load demand value appears most frequently during the sliding window movement. Weighted average is a calculation method, which calculates the average by assigning different weights to the load demand values in the window to obtain a more representative load demand value. The purpose of this process is to extract the load demand characteristic value that can represent the energy consumption mode group from the complex load demand data, providing a basis for subsequent energy consumption mode merging.

[0034] Specifically, the load demand distribution histogram in step S3-2 is a chart with load demand values as the horizontal axis and frequency as the vertical axis, which reflects the frequency of different load demand values. Each column in the histogram represents the frequency of load demand values in a certain range. The fixed window width means that the size of the window moving on the histogram is fixed, and this width determines the range of load demand values considered each time. The sliding step refers to the distance the window moves each time, which affects the search precision. During the sliding window process, the frequency sum of the window is recorded each time, which is the sum of the heights of all columnar bars in the window. The window position with the maximum frequency sum is the position of the window with the maximum frequency sum among all sliding windows, and the load demand value at this position is considered to be the most representative. Weighted average is calculated by multiplying the load demand values in the window by their corresponding frequency or weight, summing them up, and then dividing by the total frequency. The value obtained in this way can better reflect the average level of load demand in the window. Through this method, the characteristic value that can represent the load demand of the energy consumption mode group can be effectively extracted from a large amount of data.

[0035] Preferably, when constructing the load demand distribution histogram, the window width and sliding step can be dynamically adjusted according to the distribution of actual data. For example, if the load demand data distribution is relatively concentrated, the window width can be appropriately reduced to improve accuracy; if the data distribution is relatively dispersed, the window width can be appropriately increased to reduce the amount of calculation. When calculating the weighted average, the weights can be set according to the stability and importance of the load demand values. For example, for those load demand values that appear frequently within the window, higher weights can be assigned because these values are more likely to be typical load demands of the energy use mode group. In addition, a normalization step can be introduced to normalize the calculated representative load demand values to a standard range, facilitating subsequent comparison and merging operations. Through these refined operations, the representative load demand values of each energy use mode group can be more accurately calculated, thereby improving the accuracy and reliability of energy use mode merging.

[0036] In some embodiments, the matching degree in step S3-3 is calculated by comparing the ratio of the absolute value of the difference between the two groups of representative load demands to the preset load tolerance, and if the ratio is less than a set threshold, it is determined that the two groups can be merged.

[0037] It should be noted that the calculation method of the matching degree in step S3-3 is used to determine whether the two energy use mode groups have similar load demand characteristics, thereby deciding whether to merge them into one group. The matching degree here refers to the degree of similarity between the representative load demands of the two energy use mode groups, which is determined by calculating the ratio of the absolute value of the difference between the two groups of representative load demands to the preset load tolerance. If this ratio is less than a set threshold, it is considered that the load demands of the two groups are similar enough to be merged. This calculation method can quantify the degree of difference between the two energy use mode groups, providing a clear basis for subsequent merging operations. The preset load tolerance is a key parameter that is set according to the actual energy use demand and the requirement for regulation accuracy, and is used to measure the acceptable difference range of the load demands of the two energy use mode groups. The set threshold is a critical value for determining whether the matching degree meets the merging condition, and only when the ratio is less than this threshold, it is considered that the two groups can be merged.

[0038] Specifically, the preset load tolerance is a parameter set according to the actual demand of building energy management, which reflects the tolerance of load demand difference in the process of energy regulation. For example, if the building energy management system has a high requirement for the accuracy of energy consumption, the preset load tolerance can be set smaller; on the contrary, if the requirement for the accuracy of energy consumption is relatively low, the preset load tolerance can be set larger. The setting threshold is a critical value for judging whether the matching degree meets the merging condition, which is usually determined according to actual experience or simulation experiment. For example, the setting threshold can be set to 0.1 or 0.2, etc., and the specific value depends on the diversity of building energy use mode and the flexibility of regulation strategy. When calculating the matching degree, the absolute value of the difference between the two groups of representative load demand needs to be calculated first, and then compared with the preset load tolerance to get the ratio. If the ratio is less than the setting threshold, it means that the load demand of the two energy use mode groups is similar enough to be merged; otherwise, the two groups will remain independent. This calculation method can effectively screen out energy use mode groups with similar load demand characteristics, providing a clearer classification basis for subsequent energy optimization.

[0039] Preferably, when calculating the matching degree, a weight coefficient can be introduced to adjust the influence degree of the absolute value of the difference between the representative load demand. For example, if the load demand of a certain energy use mode group fluctuates greatly in a certain time period, the load demand value in that time period can be given a higher weight, so as to more accurately reflect its load demand characteristics. In addition, the preset load tolerance and the setting threshold can be dynamically adjusted according to the diversity of actual energy use mode and the flexibility of regulation strategy. For example, for some seasonal building energy use mode, the preset load tolerance and the setting threshold can be adjusted according to the seasonal change to better adapt to the energy use demand in different seasons. In actual operation, the effectiveness of the matching degree calculation method can also be verified through simulation experiment or historical data comparison to ensure that the merged energy use mode group can accurately reflect the actual energy use of the building, providing a reliable basis for subsequent energy optimization.

[0040] In some embodiments, the re-merging process of the second energy use mode group in step S4 includes: Step S4-1, arranging the data in each group of the second energy use mode group in ascending order of timestamp, identifying the time period breakpoint and splitting the non-continuous group; Step S4-2, calculating the start time and end time of each group after splitting, and extracting the time period characteristics; Step S4-3, merging the groups with overlapping or adjacent time period characteristics according to time continuity, to obtain the third energy use mode group.

[0041] It should be noted that in the process of re-merging the second energy mode groups in step S4, the main purpose is to further optimize the classification of energy modes by analyzing the continuity of energy periods. The energy period breakpoint here refers to the point in time series where the energy mode changes significantly, such as the start and stop of the device or the switching of energy supply. The period feature refers to the characteristics of each energy mode group in time, including start time, end time and duration, etc. By identifying these breakpoints and splitting non-continuous groups, the energy mode can be more accurately divided, providing a more detailed classification basis for subsequent energy optimization. Finally, by merging groups with overlapping or adjacent period features, the classification of energy modes can be further simplified, improving the efficiency of energy management.

[0042] Specifically, the second energy mode group refers to the energy mode group after preliminary merging, which has certain similarity but may still have differences in time continuity. The energy period breakpoint refers to the point in time series where the energy mode changes significantly, such as the start and stop of the device or the switching of energy supply. In step S4-1, the data in each group of the second energy mode group is arranged in ascending order of timestamp, and the energy period breakpoint is identified and the non-continuous group is split. This means that the time series data of each energy mode group needs to be analyzed to find the discontinuous points, and these points are used as the dividing line to separate the non-continuous energy mode. The period feature refers to the characteristics of each energy mode group in time, including start time, end time and duration, etc. In step S4-2, the start time and end time of each group after splitting are calculated, and the period feature is extracted. The purpose of this step is to quantify the time characteristics of each energy mode group for subsequent merging operations. In step S4-3, groups with overlapping or adjacent period features are merged according to time continuity to obtain the third energy mode group. Here, overlapping or adjacent period features refer to the time characteristics of two energy mode groups having overlapping or adjacent parts on the time axis, and meeting certain merging conditions, such as the period interval of the two groups being less than a preset time threshold and the load demand fluctuation amplitude being less than a set tolerance.

[0043] Preferably, in identifying the breakpoints of the energy consumption period, time series analysis methods can be employed, such as determining the breakpoints by calculating the rate of change of energy consumption between adjacent time points. If the rate of change of energy consumption exceeds a certain preset threshold, the time point is considered a breakpoint. In calculating the period characteristics, the concept of time window can be introduced, such as setting a fixed time window, and performing statistical analysis on the data within the window to determine the start time and end time of each energy consumption pattern group. In merging groups with overlapping or adjacent period characteristics, the preset time threshold and the load demand fluctuation tolerance can be adjusted according to the characteristics of the actual energy consumption patterns and the requirements of energy management. For example, for some buildings with high requirements for energy continuity, a smaller preset time threshold can be set; while for some buildings with low requirements for energy continuity, the preset time threshold can be appropriately increased. In addition, machine learning algorithms such as decision trees or support vector machines can also be introduced to automatically identify and merge energy consumption pattern groups, improving the accuracy and efficiency of merging. Through these detailed operations, energy consumption pattern groups with similar period characteristics can be more accurately identified and merged, providing a more accurate classification basis for subsequent energy optimization.

[0044] In some embodiments, the merging condition in step S4-3 is that the period interval of the two groups is less than the preset time threshold and the load demand fluctuation amplitude is less than the set tolerance.

[0045] It should be noted that in step S4-3, the setting of the merging condition is to ensure that the merged energy consumption pattern group has reasonableness and consistency in terms of time continuity and load demand fluctuation. The period interval here refers to the interval length of the two energy consumption pattern groups in time, and the load demand fluctuation amplitude refers to the degree of change of the load demand of the two energy consumption pattern groups. By setting the preset time threshold and the load demand fluctuation amplitude tolerance, energy consumption pattern groups that are close in time and have small changes in load demand can be effectively screened for merging, thereby further optimizing the classification of energy consumption patterns and improving the efficiency and accuracy of energy management.

[0046] Specifically, the time interval refers to the length of time interval between two energy usage pattern groups, usually measured in time units such as minutes, hours. For example, if one energy usage pattern group ends at 9 am and another energy usage pattern group starts at 10 am, their time interval is 1 hour. The preset time threshold is a parameter set according to the characteristics of actual energy usage patterns and energy management requirements, used to determine whether the time interval of two energy usage pattern groups is small enough to be merged. For example, if the energy usage patterns of the building change frequently, the preset time threshold can be set smaller, such as 30 minutes; if the energy usage patterns change slowly, the preset time threshold can be set larger, such as 2 hours. The load demand fluctuation amplitude refers to the degree of change in load demand between two energy usage pattern groups, usually measured by the range of load demand values. The set tolerance is a parameter set according to the actual energy usage demand and the accuracy requirement of energy management, used to determine whether the load demand fluctuation of two energy usage pattern groups is within an acceptable range. For example, if the energy management system of the building has a high requirement for the stability of load demand, the set tolerance can be set smaller; otherwise, the set tolerance can be set larger. Through the setting of these parameters, the energy usage pattern groups after merging can have reasonable and consistent time continuity and load demand fluctuation.

[0047] Preferably, when setting the preset time threshold and the load demand fluctuation amplitude tolerance, dynamic adjustment can be made according to historical energy usage data and the characteristics of actual energy usage patterns. For example, by analyzing historical energy usage data, it can be found that some energy usage pattern groups frequently appear within a certain time period and have small load demand fluctuations, at which time the preset time threshold and the load demand fluctuation amplitude tolerance can be appropriately reduced to improve the accuracy of merging. In actual operation, machine learning algorithms such as cluster analysis or decision tree can also be introduced to automatically identify and adjust these parameters. For example, through cluster analysis, the distribution of time intervals and load demand fluctuations of energy usage pattern groups can be found, so that reasonable preset time thresholds and load demand fluctuation amplitude tolerances can be automatically set. In addition, when merging energy usage pattern groups, the results after merging can be verified and adjusted. For example, by simulating the energy consumption of the merged energy usage pattern groups, it can be evaluated whether it meets the goal of energy management. If it is found that the merged energy usage pattern groups have abnormal energy consumption, the preset time threshold and the load demand fluctuation amplitude tolerance can be adjusted to optimize the merging results. Through these detailed operations, energy usage pattern groups with similar energy usage time period characteristics can be more accurately merged, providing a more accurate classification basis for subsequent energy optimization.

[0048] In some embodiments, in the group association process of the third energy usage pattern group in step S5, it includes: Step S5-1, calculate the energy consumption intensity correlation coefficient of two groups in the third energy usage pattern group based on the Pearson correlation coefficient formula; Step S5-2, if the correlation coefficient is greater than the correlation threshold, mark the two groups as the same energy use behavior pattern.

[0049] It should be noted that in the process of grouping correlation of the third energy use pattern group in step S5, the correlation between different energy use pattern groups is mainly judged by calculating the energy intensity correlation coefficient. The energy intensity here refers to the amount of energy consumed per unit time, which is an important indicator for measuring energy use patterns. Pearson correlation coefficient is a statistical indicator used to measure the linear correlation between two variables, with a value range of -1 to 1. The closer the value is to 1 or -1, the stronger the linear correlation between the two variables. By calculating the energy intensity correlation coefficient of the two groups in the third energy use pattern group, it can be determined whether their energy consumption patterns have similarities or correlations, thereby providing a more comprehensive analysis basis for subsequent energy optimization.

[0050] Specifically, the third energy use pattern group refers to the energy use pattern group after the energy use period continuity judgment and re-merging. These groups have certain similarities in terms of time continuity and load demand fluctuation. The energy intensity correlation coefficient is calculated by the Pearson correlation coefficient formula, which reflects the linear correlation between the energy intensity of the two energy use pattern groups. In the calculation process, the energy intensity data of each energy use pattern group in the same time interval needs to be collected, and then substituted into the Pearson correlation coefficient formula for calculation. If the calculated correlation coefficient is greater than the preset correlation threshold, for example, 0.8, it can be considered that there is a strong correlation between the two energy use pattern groups, and they can be marked as the same energy use behavior pattern. The correlation threshold here is a parameter set according to the characteristics of the actual energy use pattern and energy management requirements, used to judge whether the correlation between the two energy use pattern groups is strong enough to be marked as correlated.

[0051] Preferably, when calculating the energy intensity correlation coefficient, data preprocessing can be performed, such as removing outliers or data smoothing, to improve the accuracy of the correlation coefficient calculation. In addition, according to the characteristics of the actual energy use pattern, different time intervals can be selected to collect energy intensity data, such as hours, days or weeks, to more accurately reflect the periodic characteristics of the energy use pattern. In actual operation, machine learning algorithms such as random forest or neural network can also be introduced to automatically identify and judge the correlation between energy use pattern groups. These algorithms can automatically adjust the correlation threshold by learning patterns in historical data, and provide more accurate correlation judgments. Through these detailed operations, energy use pattern groups with similar energy consumption patterns can be more accurately identified and marked, providing a more comprehensive analysis basis for subsequent energy optimization.

[0052] In some embodiments, during the training of the gradient boosting decision tree model in step S6, historical energy consumption data and corresponding control strategies are used as the training set to generate a set of decision trees through multiple rounds of iteration. During prediction, the feature parameters of the fourth energy mode group are input, and the control strategies of each group are output.

[0053] It should be noted that in step S6, the generation of control strategies based on the gradient boosting decision tree model is the key link of the entire low-carbon building energy regulation method. Gradient boosting decision tree model is a powerful machine learning algorithm that can improve the accuracy and stability of prediction by integrating multiple decision trees. This model learns from historical energy consumption data and its corresponding control strategies to generate optimal energy optimization schemes for different energy mode groups. Here, historical energy consumption data refers to the energy consumption records of the building over a period of time, including the running status of various devices, energy intensity, timestamp, and other information. Control strategies refer to optimization measures developed based on energy consumption, such as device start-stop timing, energy storage system charging and discharging plans, and renewable energy priority scheduling rules. By training the gradient boosting decision tree model, dynamic optimization of building energy consumption can be achieved, thereby achieving the goal of low-carbon operation.

[0054] Specifically, the gradient boosting decision tree model is an ensemble learning model based on boosting methods, which optimizes the predictive performance of the model by gradually adding decision trees. During training, the model learns from historical energy consumption data and its corresponding control strategies, generating a set of decision trees through multiple rounds of iteration. Each decision tree is trained based on the residual error of the previous tree, gradually improving the predictive accuracy of the model. Historical energy consumption data includes energy consumption records of the building under different time, different device running status, which is the basis for model training. Corresponding control strategies refer to energy optimization measures that match historical energy consumption data, which are developed based on past energy management experience. During model training, input parameters include various features in historical energy consumption data, such as energy intensity, device running status, timestamp, etc., and the output is the corresponding control strategy. In this way, the model can learn the mapping relationship between the characteristics of different energy mode groups and the optimal control strategy, providing decision support for subsequent energy optimization.

[0055] Preferably, when constructing the gradient boosting decision tree model, the parameters of the model can be optimized to improve the performance and generalization ability of the model. For example, the number of decision trees, the depth of each decision tree, the learning rate, and other parameters can be adjusted. The number of decision trees determines the complexity and prediction ability of the model, and usually needs to be selected according to the size and complexity of the training data. The depth of each decision tree affects the fitting ability of the model, and a too deep tree may cause overfitting, while a too shallow tree may not capture the complex patterns in the data. The learning rate controls the contribution of each decision tree to the model, and a lower learning rate can make the model more stable, but may require more decision trees to achieve the same prediction accuracy. During the model training process, techniques such as cross-validation can be used to evaluate the performance of the model, and the parameters can be adjusted according to the validation results. In addition, in the prediction stage, the feature parameters of the fourth energy mode group need to go through the same preprocessing steps as the training data to ensure that the model can correctly understand and process these data. Through these detailed operations, the prediction accuracy of the gradient boosting decision tree model can be improved, thereby providing more effective control strategies for low-carbon building energy optimization.

[0056] In some embodiments, the control strategy includes device start-stop timing, energy storage system charging and discharging plan, and renewable energy priority scheduling rules, which are implemented by the building energy management system to achieve low-carbon optimization of building energy consumption.

[0057] It should be noted that the implementation of the control strategy is a key link to achieve low-carbon building energy optimization. The control strategy includes device start-stop timing, energy storage system charging and discharging plan, and renewable energy priority scheduling rules, which are implemented by the building energy management system BEMS to achieve low-carbon optimization of building energy consumption. The device start-stop timing here refers to arranging the start and stop times of the devices according to the energy demand and device characteristics to optimize energy use efficiency. The energy storage system charging and discharging plan refers to arranging the charging and discharging times of the energy storage devices according to the energy supply and demand to balance energy supply and demand. The renewable energy priority scheduling rule refers to prioritizing the use of renewable energy under the premise of meeting energy demand to reduce carbon emissions. Through the synergistic effect of these three strategies, the energy consumption and carbon emissions of the building can be effectively reduced, and low-carbon operation can be achieved.

[0058] Specifically, the equipment start-stop timing refers to reasonably arranging the start and stop times of equipment according to the energy demand of the building and the operating characteristics of the equipment. For example, during periods of low energy demand, some non-essential equipment can be stopped to save energy. During periods of high energy demand, more equipment can be started to meet the demand. The energy storage system charging and discharging plan refers to reasonably arranging the charging and discharging times of the energy storage equipment according to the energy supply and demand situation. For example, during periods of energy surplus, the energy storage equipment can be charged; during periods of energy shortage, the energy storage equipment can be discharged to balance energy supply and demand. The renewable energy priority scheduling rule refers to prioritizing the use of renewable energy such as solar and wind energy under the premise of meeting energy demand. For example, when solar power generation is sufficient, solar power generation can be used first to reduce dependence on traditional energy sources. These three strategies are implemented through the building energy management system (BEMS), which is an integrated system that can monitor and control various energy equipment and systems in the building in real time to achieve optimal energy management.

[0059] Preferably, when implementing the regulation strategy, real-time data and prediction models can be combined to dynamically adjust the strategy. For example, by monitoring the energy consumption and equipment operation status of the building in real time, combined with weather forecasts and energy market information, the equipment start-stop timing and the energy storage system charging and discharging plan can be dynamically adjusted. In terms of equipment start-stop timing, the start and stop times of equipment can be dynamically adjusted according to real-time energy demand and equipment operating efficiency. In terms of the energy storage system charging and discharging plan, the charging and discharging times can be dynamically adjusted according to real-time energy supply and demand and the remaining capacity of the energy storage equipment. In terms of the renewable energy priority scheduling rule, the proportion of renewable energy use can be dynamically adjusted according to real-time renewable energy generation and energy demand. In addition, machine learning algorithms such as reinforcement learning can be introduced to automatically optimize the regulation strategy. Reinforcement learning algorithms can automatically adjust the strategy to achieve optimal energy management results by learning historical data and real-time feedback. Through these detailed operations, the flexibility and adaptability of the regulation strategy can be improved, further optimizing the energy consumption and carbon emissions of the building.

[0060] The above-mentioned various embodiments of the present application have the following beneficial effects: the method can improve the accuracy and efficiency of building energy consumption analysis through multi-stage data clustering and pattern optimization. First, K-means clustering and principal component analysis can accurately identify the initial energy use mode, and through load demand matching, time period continuity judgment and energy consumption intensity correlation analysis, the grouping results can be gradually optimized to finally form the most representative energy use mode group. The gradient boosting decision tree model can generate the optimal regulation strategy, including equipment start-stop timing, energy storage scheduling and renewable energy priority use scheme, thereby effectively reducing the overall energy consumption and carbon emissions of the building.

[0061] The method can calculate typical load demand through a sliding window, and verify the pattern correlation in combination with a Pearson correlation coefficient, to ensure the scientificity and reliability of the energy use grouping. The period continuity merging can avoid fragmentation of the energy use patterns, and the load demand fluctuation tolerance judgment can improve the merging accuracy. The finally generated regulation strategy can be automatically executed through a building energy management system, to realize dynamic optimization, improve the energy utilization efficiency, and adapt to the low-carbon demands of different building types, and has wide applicability and operability.

[0062] Further, the storage medium of the embodiments of the present application stores program instructions capable of implementing all the methods described above, wherein the program instructions can be stored in the storage medium in the form of a software product, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor execute all or part of the steps of the methods described in various embodiments of the present application. The storage medium described above includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes, or a terminal device such as a computer, a server, a mobile phone, a tablet, etc.

[0063] The above description is only some of the preferred embodiments of the present application and the explanation of the technical principles applied. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combinations of the technical features described above, and should also cover other technical solutions formed by any combination of the above technical features or equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features and the technical features disclosed in the embodiments of the present application (but not limited to) having similar functions.

Claims

1. A method for energy regulation of a low-carbon building, characterized by, The method comprises the following steps: Step S1, missing value filling and normalization preprocessing are performed on the energy consumption features in the real-time collected building energy data sequence; Step S2, K-means-based energy consumption clustering is performed on the preprocessed building energy data sequence to obtain a first energy use mode group; Step S3, based on load demand matching, the first energy use mode group is preliminarily combined to obtain a second energy use mode group; Step S4, based on energy use period continuity judgment, the second energy use mode group is recombined to obtain a third energy use mode group; Step S5, based on the energy consumption intensity correlation degree between groups, the third energy use mode group is associated to obtain a fourth energy use mode group; Step S6, based on a gradient boosting decision tree model, an energy optimization scheme corresponding to each group in the fourth energy use mode group is generated.

2. The energy regulation method for low-carbon buildings according to claim 1, wherein in the K-means-based energy consumption clustering process in step S2, the following steps are included: Step S2-1, for the preprocessed data, the normalized temperature, energy consumption intensity and device running state converted by one-hot encoding are mapped into a point in a multi-dimensional space; Step S2-2, principal component analysis dimension reduction is performed on the multi-dimensional data, and the clustering centers satisfying the preset number of groups and the maximum number of iterations are found in the reduced space by an iterative method; Step S2-3, each point is distributed to the nearest group according to the principle of minimizing Euclidean distance, and the energy consumption information corresponding to the point is stored in the energy use mode library to generate the first energy use mode group.

3. The energy regulation method for low-carbon buildings according to claim 2, wherein the preliminary combination of the first energy use mode group in step S3 comprises: Step S3-1, for each data point in each group in the first energy use mode group, the time stamp is arranged in ascending order, and the typical load demand of each group is calculated; Step S3-2, based on the load demand distribution histogram of each group, the representative load demand of each group is calculated; Step S3-3, the representative load demands of each group in the first energy use mode group are compared with each other to calculate the matching degree, and the groups satisfying the matching degree threshold are combined to obtain the second energy use mode group.

4. The energy regulation method for low-carbon buildings according to claim 3, wherein the calculation of the representative load demand of each group in step S3-2 comprises: 1) a load demand distribution histogram is established, in which the horizontal axis is the load demand value and the vertical axis is the frequency; 2) a sliding window with a fixed window width and a sliding step is set in the histogram, and the frequency sum in each window is recorded by sliding the window from the left side of the histogram; 3) the load demand value in the window with the maximum frequency sum is selected and weighted averaged as the representative load demand.

5. The energy regulation method for low-carbon buildings according to claim 4, wherein the comparison of the representative load demands of each group in the first energy use mode group in step S3-3 to calculate the matching degree comprises: comparing the ratio of the absolute value of the difference between the representative load demands of two groups to the preset load tolerance, and if the ratio is less than the set threshold, it is determined that the two groups can be combined. ​ ​ ​ ​ 6. The low-carbon building energy regulation method according to claim 3, wherein, the re-merging of the second energy consumption mode groups in step S4 comprises: step S4-1, arranging each group of data in the second energy consumption mode groups in ascending order of time stamp, identifying the energy consumption period breakpoints and splitting the non-continuous groups; step S4-2, calculating the start time and end time of each group after splitting, and extracting the period characteristics; step S4-3, merging the groups with overlapping or adjacent period characteristics according to time continuity to obtain third energy consumption mode groups.

7. The low-carbon building energy regulation method according to claim 6, wherein, the merging condition for merging the groups with overlapping or adjacent period characteristics according to time continuity in step S4-3 is that the period interval of the two groups is less than a preset time threshold and the load demand fluctuation amplitude is less than a set tolerance.

8. The low-carbon building energy regulation method according to claim 6, wherein, the group association of the third energy consumption mode groups in step S5 comprises: step S5-1, calculating the energy consumption intensity correlation coefficient of two groups in the third energy consumption mode groups based on the Pearson correlation coefficient formula; step S5-2, if the correlation coefficient is greater than an association threshold, the two groups are marked as the same energy consumption behavior mode.

9. The low-carbon building energy regulation method according to claim 8, wherein, in step S6, the gradient boosting decision tree model is trained using historical energy consumption data and corresponding regulation strategies as the training set, and a set of decision trees is generated through multiple rounds of iteration; when predicting, the feature parameters of the fourth energy consumption mode groups are input, and the regulation strategies of each group are output.

10. The low-carbon building energy regulation method according to claim 9, wherein, the regulation strategies include device start-stop timing, energy storage system charging and discharging plan, and renewable energy priority scheduling rules, and the regulation strategies are executed through the building energy management system to realize low-carbon optimization of building energy consumption.

Images