A method for predicting building hourly electricity consumption based on dynamic and static hybrid data analysis

By combining dynamic and static data analysis in the building BIM operation and maintenance platform, the time convolution network and time-by-time power consumption coefficient are used to solve the problem of low prediction accuracy and credibility of time-by-time power consumption, and high efficiency is achieved. Accurate energy consumption prediction, supporting more effective energy management.

CN115860171BActive Publication Date: 2024-11-29ARMY ENG UNIV OF PLA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211281696.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-19
Publication Date
2024-11-29
Estimated Expiration
2042-10-19

AI Technical Summary

Technical Problem

The prior art has low accuracy in the prediction results of time-by-time power consumption in the building BIM operation and maintenance platform and is low in credibility, making it difficult to meet the demand for energy regulation of smart buildings.

Method used

Using a method based on dynamic and static mixed data analysis, the weighted fusion of dynamic data and static data is predicted through the weighted fusion of time-by-time power consumption coefficient, and the concept of time-by-time power consumption coefficient is proposed, combining static data such as building electricity consumption history records and spatial location information. Achieve high-precision time-by-time power consumption prediction.

Benefits of technology

It improves the credibility of the energy consumption prediction results of the building BIM operation and maintenance platform, reduces manpower and material costs, provides more accurate energy regulation strategies, and improves operation and maintenance efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115860171B_ABST
    Figure CN115860171B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for predicting hourly electricity consumption of a building based on dynamic and static hybrid data analysis, which aims to assist building operation and maintenance personnel in optimizing building energy configuration and realizing green and efficient building energy management. By combining dynamic data analysis with static data analysis, the dynamic data and static data of the building are analyzed to obtain a predicted value of hourly electricity consumption of the building and an estimated value of hourly electricity consumption of the building, respectively, and then the two are weightedly fused to obtain a corrected value of hourly electricity consumption prediction of the building. The dynamic data of the method is the historical time series of building electricity consumption obtained from the building Internet of Things, and the static data is the building space location, type of power-consuming equipment and rated power obtained from BIM. The present invention provides a new implementation method for predicting hourly electricity consumption of a building, which can assist in optimizing building energy configuration with high-precision and high-reliability prediction results, thereby promoting green and energy-saving building operation and maintenance management.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for predicting hourly electricity consumption based on a BIM operation and maintenance platform for large public buildings, and in particular to a method for predicting hourly electricity consumption based on dynamic and static hybrid data analysis. Background Art

[0002] The composition structure of the BIM operation and maintenance platform is the BIM model, the building Internet of Things system and the building operation and maintenance management platform. The main body of the entire architecture is the BIM model, and the core is the various types of data generated throughout the entire cycle from the engineering design stage to the construction stage to the building operation and maintenance stage. Its goal is to use BIM, Internet of Things, big data analysis, artificial intelligence and other technologies to comprehensively realize the integration, calculation and mutual control of building information, thereby realizing the intelligent operation and maintenance management of the building. Its basic architecture is shown in the figure. Since the BIM model has very superior three-dimensional visualization capabilities, in order to realize the building operation and maintenance management function in the BIM operation and maintenance platform, the BIM model will be used as a core carrier, and all data information will be integrated into the operation and maintenance management, and the security management, facility and equipment management, energy consumption management, space management, comprehensive management, property management and other subsystems will communicate and communicate with each other, and various data analysis technologies will be used to calculate and analyze them, and obtain conclusions that can be used to assist building operation decisions and reduce operating costs.

[0003] With the widespread development of Internet of Things technology and information technology, the building Internet of Things system is a new type of building control management system that has emerged in recent years. Before it, the operation and management of traditional buildings relied on a single building automation system (BMS) and an integrated building management system (IBMS). In the past, most buildings were traditional buildings with simple equipment and simple functions. The interior of the building mainly contained an operating system based on the HVAC system, supplemented by lighting and power monitoring. A single building automation system could meet the needs of building operation and management, but it still needed to rely on the experience of the facility manager when using it. Later, as the equipment systems in the building expanded to functional requirements such as fire protection, power, and environment, an integrated building management system integrated in a traditional integrated manner came into being. However, with the development of intelligent buildings, traditional building operation and maintenance management methods can no longer meet the needs. What intelligent buildings need is an intelligent operation and maintenance model and a management system that can support intelligent operation and maintenance. The building Internet of Things system connects all the equipment and systems of the building in an Internet of Things manner, uses a sensor network to monitor the status of each device, each system mode, and environmental changes during the operation of the building, and stores the operation data of each system. It can be seen that most of the data contained in the building Internet of Things system is dynamic data that changes over time.

[0004] To sum up, the data in the building BIM operation and maintenance platform can be divided into two categories. One is static data without time attributes, such as building component attributes and equipment system parameters, which are mainly stored in the BIM model; the other is dynamic data with time attributes, which can be obtained from the building Internet of Things system, such as the operating status of building equipment, environmental change parameters, etc.

[0005] With the development of the world's population and economy, the growing demand for energy has attracted more attention from humans. At the same time, in order to solve problems such as pollution, carbon emissions, and greenhouse gases, energy consumption has always been a matter of great concern to humans. Through the energy review of the U.S. Energy Information Administration, it was found that 40% of energy consumption comes from buildings. In recent decades, my country's building energy consumption has grown rapidly, and the goal of reducing building energy consumption has been proposed, requiring precise control of building energy consumption to avoid unnecessary energy waste. With the development of new technologies and new ideas such as artificial intelligence, big data, and the Internet of Things, the development and implementation of smart buildings and smart cities has become a trend and trend in modern urban construction. However, the increasingly complex power energy consumption system in smart buildings has brought huge challenges to the balance of load supply and demand. Summary of the invention

[0006] The purpose of the present invention is to provide a method for predicting hourly electricity consumption of buildings based on dynamic and static mixed data analysis. From the perspective of improving the credibility of energy consumption prediction results, a new method for predicting hourly electricity consumption of buildings is established on the basis of analyzing the data characteristics of the building BIM operation and maintenance platform. This method makes full use of the large amount of dynamic and static data contained in the building BIM operation and maintenance platform, and solves the problem of low accuracy and low credibility of single time series prediction results. The efficiency of building BIM operation and maintenance is improved, and the cost of manpower and material resources is reduced. The building BIM operation and maintenance platform using this hourly electricity consumption prediction method can not only realize the rapid and efficient prediction of the hourly electricity consumption of buildings, but also the prediction results with high credibility are sufficient to assist building operation and maintenance personnel in making more accurate building energy regulation strategies and plans.

[0007] A method for predicting hourly electricity consumption of a building based on dynamic and static hybrid data analysis comprises the following steps:

[0008] Step 1: Obtain data from the building BIM operation and maintenance platform;

[0009] Step 2: Data preprocessing;

[0010] Step 3: Dynamic data analysis;

[0011] Step 4: Static data analysis;

[0012] Step 5: Finally, the predicted value of the building's hourly electricity energy consumption is weighted and fused with the estimated value of the building's hourly electricity energy consumption to obtain the energy consumption forecast correction value.

[0013] Aiming at the complex and changeable problem of energy consumption prediction of intelligent buildings for BIM operation and maintenance, this paper proposes an office building energy consumption prediction method based on mixed analysis of dynamic and static data. A temporal convolutional network prediction model is established using dynamic data to predict and output the hourly energy consumption value of office buildings. The concept of hourly power consumption coefficient is proposed, and the hourly energy consumption value of office buildings is estimated by combining static data. The hourly energy consumption prediction value is weighted and fused with the hourly energy consumption budget value to output the energy consumption correction prediction value.

[0014] The present invention adopts the above technical solution and has the following advantages compared with the prior art:

[0015] 1. The present invention uses a dynamic and static hybrid data analysis method to predict the hourly power consumption of future office buildings. By analyzing the current status of research on building power consumption prediction and the data characteristics of the building BIM operation and maintenance platform, it is proposed to use a combination of dynamic data analysis and static data analysis to achieve high-precision and high-efficiency hourly power consumption prediction. This solves the problem that the prediction results are often less reliable due to the high content of outliers in dynamic data.

[0016] 2. The present invention mines a large amount of valuable data in building BIM operation and maintenance, obtains operation and maintenance decisions based on data analysis results, and makes building BIM operation and maintenance more efficient and less costly. In particular, the static data that expresses the characteristics of the building itself contained in the BIM operation and maintenance platform is simultaneously integrated into the hourly electricity consumption forecast of the building, which greatly improves the problem of low credibility of the forecast results caused by the high content of abnormal values ​​in dynamic data represented by the historical record sequence value of building electricity consumption.

[0017] 3. The present invention defines the electricity consumption values ​​with time attributes monitored and recorded in the building Internet of Things as dynamic data, and divides the entire building into functional areas according to the differences in electricity consumption patterns, and monitors and records the electricity consumption values ​​of different functional areas respectively, so as to perform dynamic data analysis. In the dynamic data analysis module, the present invention uses the efficient time series data analysis capability of the time convolution network, and only uses the historical sequence values ​​of the building's hourly electricity consumption to establish a prediction model. In order to achieve the goal of predicting and outputting the energy consumption values ​​at 24 moments in the future on a daily basis, the present invention designs a supervised learning feature construction method that meets the input and output structure requirements of the time convolution network.

[0018] 4. The present invention defines numerical quantities such as spatial location information, types of power-consuming equipment, and rated power related to building electricity energy consumption as static data. These static data that do not change with time can be obtained from the BIM model. In order to make full use of the time-invariant stability of static data, the problem of low credibility of prediction results caused by high abnormal value content of dynamic data is solved. The present invention proposes the concept of hourly power consumption coefficient to quantitatively characterize the hourly power consumption value of a certain space in the building. On the basis of fully analyzing the characteristics of the building space and the regularity and periodicity of the historical sequence of building electricity energy consumption, a calculation formula for the hourly power consumption coefficient is designed, and a calculation method for the estimated value of the hourly power consumption of the building is given.

[0019] 5. The present invention realizes high-precision and high-reliability hourly electricity consumption forecasting in office buildings facing the BIM operation and maintenance platform, improves the efficiency of building BIM energy operation and maintenance, and reduces manpower and material costs. The method provides a favorable basis for the management personnel of the building BIM operation and maintenance platform to formulate more efficient, reasonable, green and environmentally friendly building energy control and adjustment strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0021] Figure 1 It is an architectural element in the TCN described in the present invention.

[0022] Top left: dilated causal convolution with dilation factors d=1,2,4 and filter size k=3.

[0023] Top right: TCN residual block. When the residual input and output have different dimensions, a 1x1 convolution is added.

[0024] Bottom: An example of a residual connection in a TCN. The blue line is the filter in the residual function, and the red line is the identity mapping. ;

[0025] Figure 2 This is a research idea diagram of the method for predicting hourly electricity consumption of office buildings according to the present invention;

[0026] Figure 3 It is the overall framework diagram of the energy consumption prediction method described in the present invention;

[0027] Figure 4 A flowchart of the construction of the supervised learning feature of the present invention;

[0028] Figure 5 It is a performance variation diagram of the prediction method under different weights described in the present invention;

[0029] Figure 6This is a comparison chart of the results of different prediction methods described in the present invention. DETAILED DESCRIPTION

[0030] The technical solution of the present invention is further explained below in conjunction with specific embodiments.

[0031] A method for predicting building hourly electricity consumption based on dynamic and static hybrid data analysis, which simultaneously utilizes dynamic data such as historical hourly electricity consumption series and static data such as rated power of building power-consuming equipment, functional zoning of building space, etc. It includes the following steps:

[0032] Step 1: Obtain data from the building BIM operation and maintenance platform; first obtain the historical time series of building electricity consumption through the building Internet of Things system, and then use the BIM model to obtain the spatial location information of each functional partition of the entire building and the types and rated powers of all power-consuming equipment.

[0033] Step 2: Data preprocessing: fill the missing values ​​of the historical time series of building energy consumption through linear interpolation and normalize them to obtain dynamic data values ​​between 0 and 1. The normalization formula is: Among them, X i Indicates the current input data, X min is the minimum value in the entire set of data, X max is the maximum value in the entire set of data, and X' is the normalized value. The building space is divided into functional zones, and the types of power-consuming equipment and rated power of each functional zone are counted.

[0034] Step 3: Dynamic data analysis; Considering the spatiotemporal nature of dynamic data and the use background of large public buildings, a temporal convolutional network is used for dynamic data analysis; the historical series of building energy consumption that has undergone data preprocessing is converted into an hourly time series, and according to the input sample requirements and prediction result requirements of the temporal convolutional network, hourly electricity consumption prediction input and output samples are constructed, and then a temporal convolutional network is constructed based on the data input and output samples. After training, verification and testing, a building hourly electricity consumption prediction model is established to predict and output the hourly electricity consumption value of the building.

[0035] The input and output sample construction of the hourly electricity energy consumption prediction of the building that meets the actual needs includes the following specific implementation process: first, the energy consumption values ​​at 24 hours a day are combined into one data, so the amount of energy consumption data collected from a certain building in one year is 365; secondly, the N, N+1, N+2, ..., N+a data are arranged in sequence and combined into an array as the input data of the prediction model; finally, the N+a+1 data is used as the output data of the prediction model, then the N, N+1, N+2, ..., N+a and N+a+1 data are constructed into an input and output data sample that meets the requirements of the prediction model; wherein the a value is an adjustable parameter, which is determined by analyzing the periodicity and regularity of the electricity energy consumption value of the building.

[0036] The construction of the temporal convolutional network first establishes a 1D fully convolutional network structure in order to achieve the principle that the output of the network has the same length as the input and the network can only use the information of the past time steps. That is, all convolutional layers have the same length and are filled with zeros to ensure that higher layers have the same length as previous layers. In addition, in the constructed temporal convolutional network, causal convolution is used so that the output of each time step t of the network is only calculated no later than the area of ​​the previous time step t. Dilated convolution is used to realize that the receptive field of a finite layer network has an exponential size by skipping the size of a given step and then applying a filter on an area larger than the size of the dilated convolution.

[0037] Step 4: Static data analysis; firstly, statistical analysis is performed on the energy consumption data set that has undergone data preprocessing to analyze the temporal changes in building electricity consumption in different seasons, weekends, and working days; secondly, by investigating the spatial area settings of office buildings, the design and use standards of electrical equipment, and the general operating mode experience of office buildings, the energy consumption law statistical analysis is assisted to determine the hourly power consumption coefficient; finally, the hourly power consumption coefficient is combined with static data parameters such as the area ratio of different functional partitions and the rated power of power-consuming equipment to estimate the hourly energy consumption values ​​of office buildings in different seasons, working days, and weekends; the estimation formula is:

[0038]

[0039] Where n represents the functional areas of office buildings, k represents different types of models, and p n The rated power of power-consuming equipment in each functional area, It is the hourly power consumption coefficient of each functional partition.

[0040] Step 5: Finally, the predicted value of the building's hourly electricity consumption and the estimated value of the building's hourly electricity consumption are weighted and fused. According to the error between the fusion result and the true value, the weight group of the predicted value and the estimated value of the hourly electricity consumption is dynamically adjusted, and finally the weight group that can minimize the error is determined. The energy consumption forecast correction value is obtained.

[0041] The method for predicting hourly electricity consumption of buildings based on dynamic and static hybrid data analysis proposed in the present invention is mainly divided into three modules, namely a dynamic data analysis module, a static data analysis module and a building hourly electricity consumption prediction correction result output module.

[0042] Dynamic data analysis module: In order to achieve a relatively accurate prediction of the building's electricity consumption value in the future by only using the historical time electricity consumption series, the dynamic data analysis part is a time series prediction model established using the time convolutional network (TCN). The time convolutional network (TCN) has two important characteristics. The first is that the network can achieve the same output and input length. The second is that the past information will not be "leaked" in the future. In order to achieve the first point, TCN establishes a network architecture based on the 1D fully convolutional network (FCN). The 1D fully convolutional network keeps the length of subsequent layers the same as that of previous layers by padding with zeros. In order to achieve the second point, TCN uses causal convolution. The output at time t is only convolved with the elements at time t and earlier in the previous layer. The architectural elements of TCN are as follows Figure 1 As shown, it mainly includes examples of dilated convolution, residual blocks and residual connections.

[0043] Based on a full understanding of the construction principles and functional characteristics of temporal convolutional networks, a temporal convolutional network hourly energy consumption prediction model based on dynamic data analysis was established.

[0044] The specific implementation steps are as follows.

[0045] Step 11: First, the original energy consumption series obtained from the building IoT system is processed for missing values ​​and the data set is normalized;

[0046] (1) Processing of missing values: fill missing values ​​through linear interpolation.

[0047] (2) Data normalization: Normalization can reduce model prediction errors and improve the convergence speed and training efficiency of the model. In the current study, the minimum-maximum normalization method is used to normalize the data, and the processed data is distributed between 0 and 1. The normalization formula is as follows:

[0048]

[0049] Among them, X i Indicates the current input data, X min is the minimum value in the entire set of data, Xmax is the maximum value in the entire set of data, and X' is the normalized value;

[0050] Step 12: Then, the preprocessed energy consumption time series is constructed into supervised learning features to make them conform to the input and output features of the temporal convolutional network. The construction process of supervised learning features is as follows: Figure 4 As shown;

[0051] Step 13: Finally, a temporal convolutional network prediction model is established based on the historical time series of energy consumption to predict and output the hourly energy consumption value of the office building.

[0052] Static data analysis module: Based on the very stable operation rules of office buildings, the strong correlation between the energy consumption of each system and the time, and the high credibility of static data related to building electricity consumption, this module proposes a concept of hourly power consumption coefficient, which is used to characterize the quantitative value of the hourly power consumption equipment utilization rate that can integrate the power consumption characteristics of all systems in office buildings and the law of building energy consumption changes. In view of the distinction between different seasons and working days and holidays, the hourly power consumption coefficient and the rated power of all power-consuming equipment contained in different functional partitions can be used to estimate the hourly energy consumption value of a certain working day or holiday in the corresponding season of the office building.

[0053] The specific implementation steps are as follows:

[0054] Step 21: Statistical analysis is performed on the energy consumption data set that has undergone data preprocessing to analyze the temporal variation regularity of building electricity consumption in different seasons, rest days, working days, etc.;

[0055] Take Nanjing, a city without heating and with four distinct seasons, as a research case. On the one hand, the whole year can be divided into cooling season (July, August, September, October), heating season (January, February, November, December), and transition season (March, April, May, June); on the other hand, it can be divided into working days (Monday to Friday) and rest days (Saturday and Sunday). The operating energy consumption data of an office building for one year is divided into six categories, namely ① working days in the cooling season, ② rest days in the cooling season, ③ working days in the heating season, ④ rest days in the heating season, ⑤ working days in the transition season, and ⑥ rest days in the transition season. The regularity of the six types of data is statistically analyzed by taking the average and integration, and finally the quantitative value P of the hourly energy consumption for 24 hours a day is extracted, which can represent the comprehensive data change characteristics of each type of data. k (t), where k = 1, 2, 3, 4, 5, 6, t = 0 to 23.

[0056] Step 22: Through research on office building space area settings, electrical equipment design and use standards, and general office building operation experience, assist in energy consumption law statistical analysis to determine the hourly power consumption coefficient;

[0057] By investigating the functional distribution characteristics of multiple office buildings and combining with general office building design specifications and standards, the functional divisions of the building are determined to include offices, corridors and stairwells, toilets, equipment rooms and basements, halls, etc. Since there are huge differences in the hourly changes in the density of people in different functional divisions, and the density of people has a direct impact on the power consumption of the building, the hourly power consumption coefficients of different functional divisions will be different. After comprehensively considering the change of the density of people in different functional divisions and the experience of experts, the hourly energy consumption ratio of each functional division is determined. n (t) (n represents different functional areas) and the total energy consumption design capacity P of the office building. Combined with the hourly energy consumption quantification value P in the previous step k (t), determine the hourly power consumption coefficient of different functional partitions as

[0058]

[0059] Step 23: Combine the hourly power consumption coefficient with the area ratio of different functional zones and the rated power of power-consuming equipment and other static data parameters to estimate the hourly energy consumption of office buildings in different seasons, working days and rest days. The estimation formula is:

[0060]

[0061] Where n represents the functional areas of office buildings, k represents different types of models, and p n The rated power of power-consuming equipment in each functional area, It is the hourly power consumption coefficient of each functional partition.

[0062] Forecast correction result output module: Based on the comparative analysis of the predicted value of building hourly electricity energy consumption based on dynamic data analysis and the estimated value of building hourly electricity energy consumption based on static data analysis, try to merge the two in different proportions to obtain the final predicted correction value of building hourly electricity energy consumption.

[0063] like Figure 1 Shown are the architectural elements in a Temporal Convolutional Network (TCN).

[0064] (1) Dilated convolution

[0065] For a 1-D sequence input x∈R n and a filter f:{0,...,k-1}→R, then the dilated convolution operation F on the sequence element s is defined as

[0066]

[0067] Where d is the dilation factor, k is the filter size, and sd·i represents the past direction. When d=1, the dilated convolution is reduced to a regular convolution. Using a larger dilation function, the output of the top layer can be made to represent a wider range of inputs, effectively expanding the receptive field of the ConvNet. This gives us two ways to increase the receptive field of a TCN: choose a larger filter size k and increase the dilation factor d, where the effective history of a layer is (k-1)d. This ensures that there are filters that can hit every input in the effective history, while also allowing deep networks to generate very large effective histories. Examples of dilated convolutions are Figure 1 shown.

[0068] (2) Residual Block

[0069] The residual block of the baseline TCN is as follows Figure 1 As shown in the upper right corner, in the residual block, TCN has two layers of dilated causal convolutions and nonlinear units ReLU. In it, we apply weight normalization to the convolution filters. In addition, spatial attenuation is added after each dilated convolution for regularization: at each training step, the entire channel is zeroed. However, in TCN, the input and output may have different widths. To account for different input and output widths, we rely on another 1x1 convolution to ensure that we receive tensors of the same shape.

[0070] (3) Residual Connection

[0071] Increasing the network depth can improve the network expression ability of the proposed model. However, if the network layer is very deep, the value of the gradient information may "disappear". However, the residual network with increased network depth can solve the problem of "gradient disappearance" during network training. TCN is stacked by residual connection model. And each residual block contains a branch for learning the transformation function f(x). The output of the transformation function is connected to the input of the next residual block, which can be described as:

[0072] o = activation(x + F(x))

[0073] like Figure 2 The figure shows the research idea of ​​the hourly electricity consumption prediction method for office buildings. The research idea follows three steps: What is the problem? How to solve the problem? How to achieve the solution? The research idea of ​​this section is as follows: Figure 2 shown.

[0074] what is the problem?

[0075] By analyzing the characteristics of the measured data (dynamic data) of the building Internet of Things system, we know that due to the influence of factors such as aging or failure of sensor equipment and changes in the surrounding environment, the dynamic data collected from the building Internet of Things system often contains a large number of abnormal values. However, the hourly energy consumption prediction of buildings based solely on dynamic data analysis has the problem of low accuracy and low credibility. Therefore, how to achieve high-precision and high-reliability hourly energy consumption prediction of office buildings is the problem to be solved in this paper.

[0076] What can I do to solve a problem?

[0077] The outlier problem of dynamic data leads to low accuracy and credibility of general hourly energy consumption prediction methods. However, the existence of outliers is determined by the temporal characteristics of dynamic data, so how to reduce the impact of outliers is the key to solving the problem. By analyzing the characteristics of static data related to building energy consumption, it is determined that predicting the hourly energy consumption of buildings by static data analysis can avoid the problem of low credibility of prediction results. Therefore, an hourly energy consumption prediction method based on mixed analysis of dynamic and static data is proposed.

[0078] How to implement the method?

[0079] Firstly, an hourly electricity consumption prediction model is established based on the temporal convolutional network, and the hourly electricity consumption prediction value is obtained by using dynamic data. Then, the characteristics of office building electricity consumption are analyzed to determine that the office building electricity consumption has good regularity and strong estimability. Then, the concept of hourly electricity consumption coefficient is proposed to realize the hourly electricity consumption estimation of office buildings based on static data analysis. Finally, the corrected value of the hourly electricity consumption prediction of office buildings is obtained by weighted fusion.

[0080] Figure 3 The figure shows the overall framework diagram of the energy consumption prediction method, which can clearly illustrate the hourly electricity consumption prediction method of office buildings based on dynamic and static hybrid data analysis described in the present invention.

[0081] The present invention mainly includes two parts. One part is the dynamic data analysis process, which is also the energy consumption series prediction process based on the time convolution network. First, the historical energy consumption time series obtained from the building Internet of Things system is preprocessed, such as missing value processing data set normalization. Then, the preprocessed data set is constructed with supervised learning features as the input and output of the time convolution network. Finally, a TCN prediction model is established, and the data is used to train the model and predict the hourly energy consumption values ​​at future moments.

[0082] The other part is the static data analysis process, which is the total energy consumption estimation process of office buildings hour by hour. The estimation formula is:

[0083]

[0084] Where n represents the functional areas of office buildings, k represents different types of models, and p n The rated power of power-consuming equipment in each functional area, The hourly power consumption coefficient is the hourly power consumption coefficient of each functional zone. The hourly power consumption coefficient is determined by statistically analyzing the regularity and periodicity of historical energy consumption sequences and combining expert experience and knowledge as well as the distribution of power-consuming equipment in different functional zones of the building.

[0085] Finally, based on the analysis in the first two parts, the hourly energy consumption budget value and the hourly energy consumption estimate value are weighted and fused to obtain the hourly energy consumption revised forecast value.

[0086] Figure 4 The figure shows the construction flow chart of supervised learning features. The preprocessed data is divided and arranged according to the timeline, and the energy consumption values ​​at 24 moments of the day are arranged in order. The 24*N energy consumption data permutations and combinations of N consecutive days before the forecast day are selected as a sample input of the time convolution network, and the 24 energy consumption data of the forecast day are used as a sample output of the time convolution network. The next sample input and output is to push the timeline forward by 1 day and use the same method to obtain it. This process is repeated until all data are covered, and finally the input and output samples of supervised learning features are formed.

[0087] Figure 5 The performance change diagram of the prediction method under different weights is shown. Taking mode category ③ as an example, the hourly energy consumption of a certain office building is predicted according to the present invention, and a weight selection comparison experiment is performed in the range of 0 to 1. The experimental results show that under mode category ③, when the weight of the static data analysis result, i.e., the hourly power consumption estimation value, is 0.3, the prediction performance of the present invention is optimal.

[0088] Figure 6 The results of different prediction methods are shown in the figure. From the curve in the figure, it can be seen intuitively that the prediction performance of the prediction method based on the hybrid analysis of dynamic and static data (dynamic and static hybrid method) is better than that of the prediction method based on dynamic data (dynamic method).

[0089] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention without departing from the spirit and scope of the present invention, and all should be covered by the protection scope of the present invention.

Claims

1. A method for predicting building hourly electricity consumption based on dynamic and static hybrid data analysis, characterized in that: The following steps are involved: Step 1: Obtain data from the building BIM operation and maintenance platform; specifically, the following process is involved: first, obtain the historical time series of building electricity consumption through the building Internet of Things system, and then use the BIM model to obtain the spatial location information of each functional partition of the entire building and the types and rated powers of all power-consuming equipment; Step 2: Data preprocessing; specifically includes the following process: fill the missing values ​​of the historical time series of building energy consumption through linear interpolation, and normalize it to obtain dynamic data values ​​between 0 and 1. The normalization formula is: Among them, X i Indicates the current input data, X min is the minimum value in the entire set of data, X max is the maximum value in the whole set of data, and X' is the normalized value; the building space is divided into zones according to functional utility, and the types of power-consuming equipment and rated power of each functional zone are counted; Step 3: Dynamic data analysis; specifically includes the following processes: Considering the spatiotemporal nature of dynamic data and the use background of large public buildings, a time convolutional network is used for dynamic data analysis; the historical time series of building electricity consumption that has undergone data preprocessing is converted into an hourly time series, and according to the input sample requirements and prediction result requirements of the time convolutional network, hourly electricity consumption prediction input and output samples are constructed, and then a time convolutional network is constructed based on the data input and output samples. After training, verification and testing, a building hourly electricity consumption prediction model is established, and the predicted value of the building hourly electricity consumption is predicted and output; The construction of hourly electricity consumption prediction input and output samples includes the following specific implementation process: first, the energy consumption values ​​at 24 hours a day are combined into one data, so the amount of energy consumption data collected from a certain building for one year is 365; secondly, the Nth, N+1, N+2, ..., N+a data are arranged in order and combined into an array as the input data of the prediction model; finally, the N+a+1th data is used as the output data of the prediction model, so the Nth, N+1, N+2, ..., N+a and N+a+1th data are constructed into an input and output data sample that meets the requirements of the prediction model; wherein the a value is an adjustable parameter, which is determined by analyzing the periodicity and regularity of the building's electricity consumption value; The construction of the temporal convolutional network first establishes a 1D fully convolutional network structure in order to achieve the principle that the output of the network has the same length as the input and the network can only use the information of the past time steps. That is, all convolutional layers have the same length and are filled with zeros to ensure that the higher layers have the same length as the previous layers. In addition, in the constructed temporal convolutional network, causal convolution is used so that the output of each time step t of the network is only calculated no later than the area of ​​the previous time step t. Dilated convolution is used to realize the receptive field of a finite layer network with an exponential size by skipping the size of a given step and then applying a filter on an area larger than the size of the dilated convolution. Step 4: Static data analysis; specifically includes the following processes: first, statistical analysis is performed on the energy consumption data set that has undergone data preprocessing to analyze the temporal changes in building electricity consumption in different seasons, weekends, and working days; second, by investigating the spatial area settings of office buildings, the design and use standards of electrical equipment, and the general operating mode experience of office buildings, the energy consumption law statistical analysis is assisted to determine the hourly power consumption coefficient; finally, the hourly power consumption coefficient is combined with the static data parameters of the area ratio of different functional partitions and the rated power of power-consuming equipment to estimate the hourly power consumption of office buildings in different seasons, working days, and weekends; the estimation formula is: Where n represents the functional areas of office buildings, k represents different types of models, and p n The rated power of power-consuming equipment in each functional area, is the hourly power consumption coefficient of each functional area; Step 5: Finally, the predicted value of the building's hourly electricity energy consumption and the estimated value of the building's hourly electricity energy consumption are weighted and fused. According to the error between the fusion result and the true value, the weight combination of the predicted value and the estimated value of the hourly electricity energy consumption is dynamically adjusted. Finally, the weight combination that can minimize the error is determined to obtain the energy consumption forecast correction value.