Parametric representation method for quantifying prediction accuracy and difficulty of building energy consumption curve

Abstract

The application provides a parameterized representation method for quantifying building energy consumption curve prediction accuracy and difficulty. The method comprises: dividing building groups according to types; constructing a building type consistency index, which is used to measure the similarity of the same type of buildings in electricity consumption behavior; obtaining the building type consistency index of the same type of building group; constructing an electricity load prediction model of the same type of building group; inputting the historical electricity consumption data set of the same type of building group into the electricity load prediction model of the same type of building group to obtain the prediction accuracy of the electricity load prediction model of the same type of building group; and establishing a first mapping relationship between the building type consistency index of the same type of building group and the prediction accuracy of the electricity load prediction model of the same type of building group. The application can assist in completing the data screening of large-scale building group energy consumption prediction tasks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building energy consumption prediction, and in particular to a parameterized characterization method for quantifying building energy consumption curve prediction accuracy and difficulty. BACKGROUND

[0002] The deep coupling of buildings and power systems has become an important direction of energy transformation. The new power system is facing the challenges of high load in winter and summer and the fluctuation and uncertainty of renewable energy penetration. This challenge is particularly prominent on the building side. With the continuous increase in distributed photovoltaic installation capacity and the gradual application of energy storage technology, building energy consumption load presents stronger dynamics and uncertainty, further exacerbating the regulation pressure of the power system. Under this background, building load prediction and energy optimization are important means to realize source-load collaboration, alleviate power regulation pressure, and improve system stability. In particular, accurate prediction and feature characterization of power load are the key foundation for building energy efficiency management, load active response, and flexible regulation of the power system. With the continuous improvement of smart meters and the digitalization of the power grid, building energy consumption data has exploded, providing rich data support for building high-precision load prediction models. Current mainstream building load prediction methods are mostly based on the combination of historical data and machine learning algorithms, from early traditional neural network models to long short-term memory networks (LS1TM, Long S1hort-Term Memory), attention mechanism networks, and the rise of active learning and deep reinforcement learning methods in recent years. These methods have gradually shown a trend of reducing data dependence and improving data utilization efficiency while improving prediction accuracy.

[0003] However, existing technologies mostly focus on improving prediction performance from the perspective of data structure or data screening. For example, pre-processing is performed with variational mode decomposition (VMD, Variational Mode DecompoS1ition) or clustering analysis to eliminate noise and interference from unpredictable factors; or dimensionality reduction and active learning mechanisms are used to screen data with effective information for model training. These strategies have certain effects in prediction tasks for small and medium-sized buildings or specific regions, but when faced with massive energy consumption data generated by regional or even city-level building groups, the computational complexity, model adaptability, and data processing cost of related algorithms significantly increase, severely restricting their practicality and generalizability. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a parameterized characterization method for quantifying building energy consumption curve prediction accuracy and difficulty, which can assist in completing data screening in large-scale building group energy consumption prediction tasks.

[0005] One aspect of the present application provides a parameterized representation method for quantifying the prediction accuracy and difficulty of building energy consumption curves. The method comprises: dividing a building group according to types; constructing a building type consistency index, which is used to measure the similarity of electricity consumption behavior of buildings of the same type; obtaining the building type consistency index of the building group of the same type; constructing an electricity load prediction model of the building group of the same type; inputting the historical electricity consumption data set of the building group of the same type into the electricity load prediction model of the building group of the same type to obtain the prediction accuracy of the electricity load prediction model of the building group of the same type; and establishing a first mapping relationship between the building type consistency index of the building group of the same type and the prediction accuracy of the electricity load prediction model of the building group of the same type.

[0006] Further, the construction of the building type consistency index comprises: performing correlation calculation on the electricity load curves between each two buildings in the building group of the same type to obtain a correlation coefficient sample set of the building group of the same type; performing probability distribution fitting on the correlation coefficient sample set to select an optimal probability distribution; extracting parameters of the optimal probability distribution and taking the parameters of the optimal probability distribution as the building type consistency index.

[0007] Further, the correlation calculation on the electricity load curves between each two buildings in the building group of the same type comprises: performing correlation calculation on the electricity load curves between each two buildings in the building group of the same type using Pearson correlation coefficient.

[0008] Further, the probability distribution fitting of the correlation coefficient sample set and the selection of the optimal probability distribution comprises: performing probability distribution fitting of the correlation coefficient sample set using multiple candidate probability distributions respectively; evaluating the fitting degrees of the multiple candidate probability distributions using Akaike information criterion respectively to select the optimal probability distribution from the multiple candidate probability distributions.

[0009] Further, the probability distribution fitting of the correlation coefficient sample set using multiple candidate probability distributions respectively comprises: performing probability distribution fitting of the correlation coefficient sample set using normal distribution, gamma distribution and Beta distribution respectively; and the selection of the optimal probability distribution from the multiple candidate probability distributions comprises: selecting a probability distribution with minimum AIC from the normal distribution, the gamma distribution and the Beta distribution as the optimal probability distribution, wherein the optimal probability distribution comprises Beta distribution.

[0010] Further, the method further comprises: extracting an order index of the building individual, the order index being used to reflect periodicity and regularity of the electricity load curve of the building individual in the time domain; constructing an electricity load prediction model of the building individual; inputting a historical electricity data set of the building individual into the electricity load prediction model of the building individual to obtain a prediction accuracy of the electricity load prediction model of the building individual; and establishing a second mapping relationship between the order index of the building individual and the prediction accuracy of the electricity load prediction model of the building individual.

[0011] Further, the extraction of the order index of the building individual comprises: performing time-domain to frequency-domain conversion on the electricity load curve of the building individual to obtain an electricity load spectrum graph; extracting amplitudes of main frequency components on a predetermined period of the electricity load spectrum graph; and calculating a ratio of the sum of the amplitudes of the main frequency components on the predetermined period to a total spectral amplitude of the electricity load spectrum graph, and taking the ratio as the order index.

[0012] Further, the method further comprises: classifying the building group and the building individual according to prediction difficulty based on the first mapping relationship and the second mapping relationship; and adopting a corresponding electricity data screening strategy based on the prediction difficulty classification.

[0013] Further, the classification of the building group and the building individual according to prediction difficulty comprises: dividing the building group and the building individual into three categories of easy prediction, medium-difficulty prediction and difficult prediction according to prediction difficulty; and the adoption of the corresponding data screening strategy based on the prediction difficulty classification comprises: adopting a unified electricity load prediction model for batch electricity load prediction for the easy prediction category of buildings; further optimizing the structure of the electricity load prediction model for the medium-difficulty prediction category of buildings; and supplementing the sampling frequency, improving the data quality or constructing an individualized electricity load prediction model for the difficult prediction category of buildings.

[0014] Further, the method further comprises: assigning a prediction difficulty classification label to all buildings to facilitate subsequent calling.

[0015] The parameterization representation method for quantifying electricity load prediction accuracy and difficulty of the building according to the present application can form a whole prediction feasibility evaluation system for a class of buildings by constructing a building type consistency index and establishing a first mapping relationship between the building type consistency index and the prediction accuracy of the electricity load prediction model thereof, and can be used to quickly evaluate the load prediction potential of an arbitrary building type group.

[0016] The parameterized representation method of quantifying the prediction accuracy and difficulty of building energy consumption curves of the application can quickly judge the data quality and modeling difficulty of individual buildings by constructing the order index of individual buildings and establishing a second mapping relationship between the order index of individual buildings and the prediction accuracy of the electricity load prediction model, thereby assisting in model design and dispatching strategy formulation.

[0017] The parameterized representation method of quantifying the prediction accuracy and difficulty of building energy consumption curves of the application can quickly evaluate the predictability of building energy consumption data in the load prediction task without complex modeling or deep learning algorithms, and clearly determine the utilization value and modeling difficulty of the data, thereby realizing the quick screening and classification of high-quality load data, i.e., quantifying the prediction accuracy and prediction difficulty, to assist in quickly screening key data sets with high information quality and strong model adaptability from massive data. This not only helps to improve the overall efficiency, applicability and scalability of subsequent building energy consumption prediction models, but also provides support for higher-level building energy consumption management and power system dispatching optimization. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 Flowchart of the parameterized representation method of quantifying the prediction accuracy and difficulty of building energy consumption curves of an embodiment of the application.

[0019] Figure 2 Another flowchart of the parameterized representation method of quantifying the prediction accuracy and difficulty of building energy consumption curves of an embodiment of the application.

[0020] Figure 3 Comparison diagram of AICs for different building types.

[0021] Figure 4 Distribution fitting effect diagram of correlation coefficients for five types of buildings.

[0022] Figure 5 Mapping relationship between building type consistency and prediction accuracy.

[0023] Figure 6 Mapping relationship between average correlation coefficient and prediction accuracy.

[0024] Figure 7 Effect diagram of order index extraction and parameterized representation of individual buildings.

[0025] Figure 8 Mapping relationship between order index and prediction accuracy for office buildings.

[0026] Figure 9 Prediction difficulty level division and data screening strategy diagram of the application. DETAILED DESCRIPTION

[0027] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses consistent with some aspects of this application as detailed in the appended claims.

[0028] This application provides a parametric characterization method for quantifying the accuracy and difficulty of building energy consumption curve prediction. It can quickly assess the predictability of building energy consumption data in load forecasting tasks without the need for complex modeling or deep learning algorithms, clarifying the data's utilization value and modeling difficulty. This enables rapid screening and classification of high-quality load data, i.e., quantifying its prediction accuracy and difficulty, to assist in quickly identifying key datasets with high information quality and strong model adaptability from massive datasets. This not only helps improve the overall efficiency, applicability, and scalability of subsequent building energy consumption prediction models but also provides support for higher-level building energy consumption management and power system dispatch optimization.

[0029] The parametric characterization method for predicting the accuracy and difficulty of quantitative building energy consumption curves according to this application will be described in detail below with reference to the accompanying drawings. Unless otherwise specified, the features in the following embodiments and implementation methods can be combined with each other.

[0030] Figure 1 A flowchart illustrating a parametric characterization method for predicting the accuracy and difficulty of building energy consumption curves according to an embodiment of this application is provided. Figure 1 As shown, a parametric characterization method for predicting the accuracy and difficulty of building energy consumption curves according to an embodiment of this application may include steps S11 to S16.

[0031] In step S11, the building groups are divided according to type.

[0032] For example, based on type, building complexes can be divided into offices, hospitals, shopping malls, schools, hotels, etc.

[0033] In step S12, a building type consistency index is constructed, which is used to measure the similarity of electricity consumption behavior among buildings of the same type.

[0034] Buildings with the same purpose or function often exhibit similar daily electricity consumption patterns, which is crucial for constructing a unified prediction model. Therefore, this application uses a building type consistency index to measure the degree of consistency in electricity consumption behavior among buildings of the same type.

[0035] In some embodiments, the building type consistency index in step S12 may further include steps S121 to S123.

[0036] In step S121, the correlation between the power consumption load curves of each two buildings in the same type of building group is calculated to obtain a correlation coefficient sample set of the same type of building group.

[0037] Alternatively, the Pearson correlation coefficient can be used to calculate the correlation between the power consumption load curves of each two buildings in the same type of building group. The closer the value of the correlation coefficient is to 1, the more consistent the power consumption load mode between the buildings is, and the more conducive to building a shared model. For buildings of the same type, it is generally expected that the correlation coefficient is close to 1. On the contrary, if the correlation coefficient is close to 0 or negative, it means that the power consumption behavior between the buildings is significantly different, and the general model has poor adaptability. By constructing a building group correlation coefficient matrix, the overall consistency of the same type of buildings in the load behavior can be preliminarily described.

[0038] The calculation formula of the Pearson correlation coefficient is as follows:

[0039] (1)

[0040] In the formula, denotes the Pearson correlation coefficient, and denotes the power consumption load of the two buildings of the same type at the i-th sampling time, and denotes the average power consumption load of the two buildings of the same type.

[0041] Using the Pearson correlation coefficient to calculate the correlation of each two buildings in the same type of building group can obtain the correlation coefficient of each two buildings, and finally the correlation coefficient sample set of the same type of building group can be obtained.

[0042] In step S122, the probability distribution fitting of the correlation coefficient sample set is performed, and the optimal probability distribution is selected.

[0043] In some embodiments, the probability distribution fitting of the correlation coefficient sample set and the selection of the optimal probability distribution in step S122 can further include step S1221 and step S1222.

[0044] In step S1221, the correlation coefficient sample set is fitted with multiple candidate probability distributions, respectively.

[0045] For example, the correlation coefficient sample set can be fitted with multiple candidate probability distributions such as normal distribution, gamma distribution and Beta distribution, respectively.

[0046] In step S1222, Akaike Information Criterion (AIC) is used to evaluate the goodness of fit of the plurality of candidate probability distributions respectively, and the optimal probability distribution is selected from the plurality of candidate probability distributions.

[0047] For example, Akaike Information Criterion can be used to select the probability distribution with the minimum AIC from the plurality of candidate probability distributions such as normal distribution, gamma distribution and Beta distribution as the optimal probability distribution. In the present application, Beta distribution is finally selected as the optimal probability distribution.

[0048] The calculation formula of Akaike Information Criterion is as follows:

[0049] (2)

[0050] Wherein, AIC represents Akaike Information Criterion; k represents the number of free parameters (to be estimated parameters) in the model; L represents the likelihood function.

[0051] In step S123, the parameters of the optimal probability distribution are extracted, and the parameters of the optimal probability distribution are used as the building type consistency index.

[0052] The present application finally selects Beta distribution as the optimal probability distribution, uses it as the core representation method of the building type consistency index of the present application, and extracts its shape parameters a and β as the building type consistency index. The shape parameters of Beta distribution can flexibly describe the concentration degree and skewness of data, which makes it have significant advantages in describing building electricity consistency.

[0053] The probability density function of Beta distribution is as follows:

[0054] (3)

[0055] Wherein, represents the probability density function (PDF) of Beta distribution; represents the Beta function, which is used for normalization to ensure that the integral of the probability density on [0, 1] is 1; represents the normalized sample data (i.e. the Pearson correlation coefficient between buildings in the present application ); and are the two shape parameters of Beta distribution.

[0056] (4)

[0057] Wherein, denotes the integral variable (without practical meaning); denotes the Gamma function.

[0058] In step S13, an architectural type consistency index of the same type of building group is obtained.

[0059] In step S14, a unified electricity consumption load prediction model of the same type of building group is constructed, such as based on XGBoost (eXtreme Gradient Boosting), LSTM (Long Short-Term Memory), etc.

[0060] In step S15, the historical electricity consumption data set of the same type of building group is input into the electricity consumption load prediction model of the same type of building group, and the prediction accuracy of the electricity consumption load prediction model of the same type of building group is obtained, such as RMSE (Root Mean Squared Error), MAPE (Mean Absolute Percentage Error), etc.

[0061] In step S16, a first mapping relationship between the architectural type consistency index of the same type of building group and the prediction accuracy of the electricity consumption load prediction model of the same type of building group is established.

[0062] The parameterized characterization method of the present application can form a whole building prediction feasibility evaluation system by constructing an architectural type consistency index and establishing a first mapping relationship between the architectural type consistency index of the same type of building group and the prediction accuracy of the electricity consumption load prediction model of the same type of building group. The first mapping relationship can be used to quickly evaluate the load prediction potential of any building type group.

[0063] The above is for the building group level. At the building individual level, in order to evaluate the prediction difficulty of the single building data, Figure 2 Another flowchart of a parameterized characterization method for quantifying the prediction accuracy and difficulty of building energy consumption curves according to an embodiment of the present application is disclosed. As shown in Figure 2 In some embodiments, the parameterized characterization method of the present application can further include steps S21 to S24.

[0064] In step S21, an order index of the building individual is extracted, and the order index is used to reflect the periodicity and regularity of the electricity consumption load curve of the building individual in the time domain.

[0065] In some embodiments, the step S21 of extracting the order index of the building individual can further include steps S211 to S213.

[0066] In step S211, Fourier transform is applied to the electrical load curve of each building to convert it from the time domain to the frequency domain, so as to obtain the electrical load spectrum diagram.

[0067] In step S212, the amplitude of the main frequency component on a predetermined period of the electrical load spectrum is extracted.

[0068] For example, the amplitude of the main frequency components in a specific period (such as 12 hours, 24 hours, 168 hours) of the electricity load spectrum can be extracted.

[0069] In step S213, the ratio of the sum of the amplitudes of the main frequency components in the predetermined period to the total spectral amplitude of the power load spectrum is calculated, and the ratio is used as an orderliness index.

[0070] A higher orderliness index indicates a more regular and periodic electricity load curve, which is beneficial for model learning and prediction. The orderliness index has unit normalization properties and can be compared across buildings.

[0071] The orderliness index can be represented as follows:

[0072] (5)

[0073] Here, OC represents the orderliness index. , , These represent the amplitudes of the main frequency components over 12 hours, 24 hours, and 168 hours, respectively. This represents the amplitude at the corresponding frequency.

[0074] (6)

[0075] in, This represents the frequency domain signal after the discrete Fourier transform (the complex representation of the k-th frequency component). This represents the value of the t-th sample point of the original discrete signal (time-domain signal).

[0076] (7)

[0077] in, This represents the amplitude of a frequency component, corresponding to the k-th frequency component. Representing complex numbers The model; Representing complex numbers The real part; Representing complex numbers The imaginary part.

[0078] Continue to refer to Figure 2In step S22, a power consumption load prediction model of the building individual is constructed.

[0079] For example, a lightweight model such as SVR (Support Vector Regression), KNN (K-Nearest Neighbors) or the like can be selected.

[0080] In step S23, the historical power consumption data set of the building individual is input into the power consumption load prediction model of the building individual, and the prediction accuracy of the power consumption load prediction model of the building individual is obtained.

[0081] In step S24, a second mapping relationship between the orderliness index of the building individual and the prediction accuracy of the power consumption load prediction model of the building individual is established.

[0082] The parameterized characterization method of the present application can be used to quickly judge the data quality and modeling difficulty of the individual building at the individual building level by constructing the orderliness index of the building individual and establishing the second mapping relationship between the orderliness index of the building individual and the prediction accuracy of the power consumption load prediction model thereof, thereby assisting in model design and dispatching strategy formulation.

[0083] In some embodiments, the parameterized characterization method of the present application can further include step S31 and step S32.

[0084] In step S31, the building group and the building individual are classified according to the prediction difficulty based on the first mapping relationship and the second mapping relationship.

[0085] For example, the building group and the building individual can be divided into three categories according to the prediction difficulty, i.e., easy prediction (i.e., low difficulty prediction), medium difficulty prediction, and difficult prediction (i.e., high difficulty prediction).

[0086] In step S32, a corresponding power consumption data screening strategy is adopted based on the prediction difficulty classification.

[0087] For example, for the easy prediction type building, a unified power consumption load prediction model can be used for batch power consumption load prediction; for the medium difficulty prediction type building, the power consumption load prediction model structure can be further optimized; for the difficult prediction type building, the sampling frequency can be supplemented, the data quality can be improved, or a personalized power consumption load prediction model can be constructed. Finally, the priority is sorted in combination with the index, which is used as the data selection basis for subsequent feature engineering and model training.

[0088] In some embodiments, the parameterized characterization method of the present application can further include step S33.

[0089] In step S33, a prediction difficulty classification label is assigned to all buildings to facilitate subsequent calling.

[0090] Compared with the prior art, the parameterized characterization method of quantifying building energy consumption curve prediction accuracy and difficulty has the advantages that:

[0091] Traditional building load prediction methods generally rely on deep learning or complex feature engineering, and the model training process has high dependence on data volume and computing resources, which is difficult to quickly deploy in urban building groups. The parameterized characterization method of quantifying building energy consumption curve prediction accuracy and difficulty of the present application can evaluate the data quality and difficulty before prediction without constructing a complex model, which significantly improves the overall system response speed and application range. The present application constructs the building type consistency index of the building group and the orderliness index of the building individual through correlation analysis between building types and frequency domain orderliness analysis of building individuals, and establishes a quantifiable and interpretable prediction difficulty evaluation system through probability model fitting and prediction accuracy mapping. Compared with the error output result of the black box model, the parameterized characterization method of the present application can clearly explain the source of the prediction difficulty, which is more conducive to model optimization and data management, and is suitable for comparative analysis of different regions and different building types, and can realize horizontal evaluation and priority ranking in large-scale building groups such as regional and urban levels. The parameterized characterization method of the present application helps the power load prediction platform to prioritize high-value data in the case of limited resources, and improves the overall modeling efficiency and accuracy. With the popularization of smart meters, energy consumption sensors and building energy management systems, building load data will grow explosively, and the parameterized characterization method of the present application can adapt to this trend and provide an expandable and efficient electricity data management and evaluation tool, laying a foundation for building a large-scale and intelligent load prediction platform in the future.

[0092] The parameterized characterization method of quantifying building energy consumption curve prediction accuracy and difficulty of the present application will be described in detail below in combination with the drawings and specific examples.

[0093] Figure 3 A comparative diagram of AICs of different building types is disclosed. As can be seen from Figure 3 , compared with normal distribution and Gamma distribution, the AIC index of Beta distribution is smaller, and can better reflect the distribution characteristics of the correlation coefficients between buildings of the same type. Therefore, the present application finally selects Beta distribution as the optimal probability distribution.

[0094] Figure 4 The distribution fitting effect diagram of the correlation coefficients of five types of buildings is disclosed, and Table 1 below gives an example of the building type consistency index of the five building groups.

[0095] Table 1

[0096]

[0097] As Figure 4Based on Table 1, a Beta distribution was fitted to the correlation coefficients of five typical building types. The shape parameters obtained from the fitting allow for a direct assessment of the consistency of electricity load data across different building types. A larger α value indicates a more rightward sloping distribution curve, suggesting a stronger correlation between buildings of the same type; a larger β value indicates a more leftward sloping distribution curve, suggesting a weaker correlation between buildings of the same type. In the example, the α values ​​for office, hospital, shopping mall, and school buildings are significantly larger than their β values, with the correlation coefficients concentrated on the right side of the distribution curve, i.e., in the high-value range.

[0098] according to Figure 4 The correlation coefficient distribution shown in the figure is in Figure 5 The mapping relationship between building type consistency and prediction accuracy was constructed in the middle. Figure 6 The paper establishes a mapping relationship between the average correlation coefficient and prediction accuracy. In the example, a unified XGBoost model was trained for each type of building to predict electricity load, thereby deriving the average R-value reflecting the prediction accuracy of each type of building. 2 . Figure 5 The x and y axes represent the shape parameters of the Beta distribution, and the point size reflects the prediction accuracy for the corresponding building type. Type consistency based on shape parameters reflects the positive correlation between the correlation coefficient distribution and prediction accuracy. When α is greater than β, the correlation coefficient distribution among the corresponding building types is closer to a strong correlation; when predicting the load of this type of building, machine learning models are more likely to capture consistent electricity load change trends and characteristics, thus achieving higher prediction accuracy. Figure 6 The diagram further illustrates that the higher the correlation coefficient of the electricity load of individual buildings of the same type, the higher the prediction accuracy of that type of building. Based on the mapping relationship between type consistency index and prediction accuracy, model building strategies can be guided in large-scale building cluster energy consumption prediction tasks. For building types with high consistency, a unified prediction model can be used, while for building types with high degree of individuality, a more detailed model can be used to learn the characteristic differences between buildings or to adapt to different buildings by adjusting model parameters.

[0099] Based on the parameterization of prediction accuracy at the building type level, the predictability of individual buildings can be assessed through individual orderliness indicators. Figure 7 This paper presents a schematic diagram illustrating the extraction and parameterized characterization of the orderliness index. For the original building electricity load time series, the time-domain data is transformed to the frequency domain using Fourier transform. Besides the low-frequency, long-period trends, the short-term variations in building electricity consumption within 12 hours, 24 hours (day), and 168 hours (week) better reflect its regularity. Therefore, the amplitudes of the corresponding three periods are found in the frequency domain, and the ratio between these amplitudes and the total spectral amplitude is calculated as the orderliness index. A higher orderliness index indicates that the electricity load curve has a high amplitude at the corresponding frequency, exhibiting obvious periodic characteristics.Figure 7 Typical weekly electricity consumption curves of buildings with order indicators of 0.101 and 0.051 respectively are shown in FIG. 1, from which it can be seen that the higher the order indicator, the more obvious the periodicity of the electricity consumption load curve; and when the order indicator is low, the electricity consumption load curve presents a relatively disordered fluctuation state.

[0100] Figure 7 The electricity consumption order indicator of the building shown in FIG. 1 can further measure and explain the difficulty of building energy consumption prediction and generation. In theory, the greater the order indicator of a building individual, the easier it is to predict and generate through data-driven or physical modeling.

[0101] Figure 8 The mapping relationship between the order indicator of the office building and the prediction accuracy is disclosed. The order indicator of the single building has a strong positive correlation with the prediction accuracy, and presents a linear increasing relationship. Through the mapping relationship and the order indicator, the predictability of the corresponding building electricity consumption load can be quickly judged.

[0102] Figure 9 The prediction difficulty level division and data screening strategy of the present application are disclosed. The prediction difficulty level division and data screening strategy are based on the above-mentioned building type consistency indicator and building individual order indicator, as well as the mapping relationship between them and the prediction accuracy, to further realize the effective classification and preferential screening of building load data. The specific implementation is as follows:

[0103] S81, input data processing: obtaining historical load data of a building group, and completing preliminary classification according to building use and functional attributes.

[0104] S82, building type consistency indicator and order indicator calculation: calculating the Pearson correlation coefficient for the load curve of the same type of building, and performing Beta distribution fitting to extract the shape parameter (i.e. the building type consistency indicator); performing Fourier transform on each single building to calculate the amplitude ratio in the 12h, 24h and 168h frequency bands, and extracting the order indicator.

[0105] S83, prediction accuracy estimation model establishment: using part of the data to construct an electricity consumption load prediction model, and establishing a mapping relationship between the building type consistency indicator and the prediction accuracy of the electricity consumption load prediction model, and between the order indicator and the prediction accuracy of the electricity consumption load prediction model.

[0106] S84, prediction difficulty level division: according to the mapping relationship, the building data is divided into three categories of low difficulty, medium difficulty and high difficulty; the division standard can be set according to the actual accuracy threshold, such as MAPE < 10%, 10% ~ 20%, > 20% corresponding to the three levels respectively.

[0107] S85, data screening and use strategy: S851: for low difficulty building data, into the unified prediction model as the main training data; S852: medium difficulty data for model optimization, verification and migration training and other auxiliary purposes; S853: high difficulty data into a separate processing flow, can be customized modeling, data enhancement or not used.

[0108] S86, form a difficulty level label library: all building samples are given a prediction difficulty label, which is used for subsequent model scheduling, energy management and data governance system calls.

[0109] The parameterization representation method of the quantitative building energy consumption curve prediction accuracy and difficulty provided by the embodiments of the present application is described in detail. In this paper, specific examples are used to illustrate the parameterization representation method of the quantitative building energy consumption curve prediction accuracy and difficulty provided by the embodiments of the present application. The above description of the embodiments is only used to help understand the core idea of the present application and does not limit the present application. It should be pointed out that for ordinary skilled persons in the art, without departing from the spirit and principles of the present application, some improvements and modifications can be made to the present application, and these improvements and modifications should also fall within the protection scope of the appended claims of the present application.

Claims

1. A parametric characterization method for quantifying the accuracy and difficulty of predicting building energy consumption curves, characterized in that, include: The building complexes are divided according to type; A building type consistency index is constructed, which is used to measure the similarity of electricity consumption behavior among buildings of the same type; Obtain building type consistency indicators for similar building groups; Construct a power load prediction model for this type of building complex; Input the historical electricity consumption dataset of the same type of building group into the electricity load prediction model of the same type of building group to obtain the prediction accuracy of the electricity load prediction model of the same type of building group; Establish a first mapping relationship between the building type consistency index of the same type of building group and the prediction accuracy of the electricity load prediction model of the same type of building group. Extract the orderliness index of individual buildings, which is used to reflect the periodicity and regularity of the electricity load curve of the individual buildings in the time domain; Construct a power load prediction model for the individual buildings; The historical electricity consumption data of the building is input into the electricity load prediction model of the building to obtain the prediction accuracy of the electricity load prediction model of the building. Establish a second mapping relationship between the orderliness index of the building individual and the prediction accuracy of the electricity load prediction model of the building individual; Based on the first mapping relationship and the second mapping relationship, building groups and individual buildings are classified according to the prediction difficulty; Based on the difficulty level of prediction, corresponding electricity consumption data filtering strategies are adopted.

2. The parameterized characterization method as described in claim 1, characterized in that, The consistency indicators for the constructed building types include: Correlation calculations are performed on the electricity load curves of each pair of buildings in a building group of the same type to obtain a sample set of correlation coefficients for the building group of the same type. Fit the probability distribution of the correlation coefficient sample set and select the optimal probability distribution; Extract the parameters of the optimal probability distribution and use the parameters of the optimal probability distribution as the consistency index of the building type.

3. The parameterized characterization method as described in claim 2, characterized in that, The correlation calculation of the electricity load curves between any two buildings in a building group of the same type includes: The correlation between the electricity load curves of two buildings in the same building group was calculated using the Pearson correlation coefficient.

4. The parameterized characterization method as described in claim 2, characterized in that, The step of fitting a probability distribution to the correlation coefficient sample set and selecting the optimal probability distribution includes: The correlation coefficient sample set was fitted with probability distributions using multiple candidate probability distributions respectively; The goodness of fit of the multiple candidate probability distributions is evaluated using the Akaike information criterion, and the optimal probability distribution is selected from the multiple candidate probability distributions.

5. The parameterized characterization method as described in claim 4, characterized in that, The step of fitting the correlation coefficient sample set with multiple candidate probability distributions includes: The probability distribution of the correlation coefficient sample set was fitted using normal, gamma, and beta distributions, respectively. The step of selecting the optimal probability distribution from the plurality of candidate probability distributions includes: The probability distribution with the smallest AIC is selected from the normal distribution, the gamma distribution, and the Beta distribution as the optimal probability distribution, wherein the optimal probability distribution includes the Beta distribution.

6. The parameterized characterization method as described in claim 1, characterized in that, The extracted orderliness indicators for individual buildings include: The electricity load curves of the individual buildings are converted from the time domain to the frequency domain to obtain the electricity load spectrum diagram; Extract the amplitude of the main frequency components in the predetermined period of the power load spectrum diagram; Calculate the ratio of the sum of the amplitudes of the main frequency components in the predetermined period to the total spectral amplitude of the power load spectrum, and use the ratio as the orderliness index.

7. The parameterized characterization method as described in claim 1, characterized in that, The classification of building groups and individual buildings according to prediction difficulty includes: Building groups and individual buildings are classified into three categories according to the difficulty of prediction: easy to predict, medium difficulty to predict, and difficult to predict. The data filtering strategy based on the prediction difficulty classification includes: For buildings with easily predictable electricity loads, a unified electricity load prediction model is used to predict electricity loads in batches. For buildings requiring moderate-difficulty forecasting, the structure of the electricity load forecasting model will be further optimized. For buildings that are difficult to predict, increase the sampling frequency, improve data quality, or build personalized electricity load prediction models.

8. The parameterized characterization method as described in claim 1, characterized in that, The method further includes: Assign prediction difficulty level labels to all buildings for easy subsequent use.

Images