Method suitable for power consumption feature extraction and power consumption mode classification of power consumption system of airport terminal

Abstract

The invention discloses a power consumption feature extraction and power consumption mode classification method suitable for an airport terminal power consumption system, and relates to the field of data mining. According to the method, the power consumption data of the terminal power utilization system are collected for correlation analysis, the main influence factors influencing the power consumption are screened out, the influence factors are classified by using a clustering method according to the characteristics of the influence factors, and then different power utilization modes of the power utilization system are obtained. By accurately analyzing the power consumption characteristics of the terminal power consumption system, mining the key factors influencing the power consumption, deeply analyzing the power consumption curve and clearly identifying different power consumption modes, a manager is helped to judge whether the current energy consumption is within a reasonable range, a basis is provided for adjustment of a management strategy, meanwhile, the potential energy waste problem is helped to be found, and the management efficiency is improved. Therefore, the system operation is optimized, and the energy-saving potential is improved.

Description

Technical Field

[0001] The present invention relates to the field of data mining, and is a method suitable for extracting the power consumption characteristics and pattern classification of airport terminals. Background Art

[0002] Airport terminals face huge challenges in achieving energy-saving goals. First of all, airport terminals are special large public buildings with large building scales, complex spaces, many functional areas, large passenger flows, complex energy-using systems and high energy consumption, which are 1.7 times that of general public buildings and 3.7 times that of residential buildings. Secondly, compared with visible resources such as materials and human resources, energy such as electricity and fuel is "invisible and intangible", and energy waste is often ignored due to its invisibility and relatively low past energy prices. In addition, the operation and management of airport terminals are difficult. There are thousands of various energy-using equipment, the energy consumption of the equipment is huge, there are many management personnel and technical departments involved, and a large number of compound talents and a sound management system matching them are required; the building energy management platform, although it has collected a large amount of energy data, has not been well utilized, making it a data "island".

[0003] To address this issue, this method proposes a method suitable for extracting the power consumption characteristics and pattern classification of airport terminals. By accurately analyzing the power consumption characteristics of the power consumption system of the terminal building, mining the key factors affecting power consumption, deeply analyzing the power consumption curve, clearly identifying different power consumption patterns, helping managers judge whether the current energy consumption is within a reasonable range, providing a basis for the adjustment of management strategies, and at the same time helping to discover potential energy waste problems, so as to optimize the system operation and improve the energy-saving potential. Summary of the Invention

[0004] The present invention relates to a method suitable for extracting the power consumption characteristics and pattern classification of airport terminals. Collect the power consumption data of the power consumption system of the terminal building for correlation analysis, screen out the main influencing factors affecting power consumption, classify them using a clustering method according to the characteristics of the influencing factors, and then obtain different power consumption patterns of the power consumption system. The method flow chart is shown in Figure 1 .

[0005] 1. Extraction of Power Consumption Characteristics

[0006] Step 1: First, according to expert experience screening, find the influencing factors F i (F1, F2, F3, F4, F5,...)(i = 1 to R) that may affect the hourly power consumption E of the power consumption system of the airport terminal. For example, outdoor air temperature, solar radiation intensity, number of passengers, proportion of departing passengers (ratio of the number of departing passengers to the number of arriving passengers), and holidays.

[0007] Step 2: For any one influencing factor F i , taking Fi Taking the x-axis as and the E as the y-axis, establish a matrix (as Figure 2 ). Along the F i direction, the grid step size B F is determined according to Equation (1); along the E direction, the grid step size B E is determined according to Equation (2). To prevent overfitting, it is stipulated that and (the variables f and g can take any values within this range but must be positive integers).

[0008]

[0009] Step 3: Assume that the total sample size of the hourly power consumption E of the power consumption system is n, and all fall within Figure 2 the E×F i matrix. Among them, n pq is the sample size of the hourly power consumption E falling in the grid of the p-th row and q-th column (p = 1 to f and is a positive integer, q = 1 to g and is a positive integer); n p is the sample size of the hourly power consumption E falling in the p-th row grid; n q is the sample size of the hourly power consumption E falling in the p-th row grid. Calculate the probability P(p,q) of the sample size of the hourly power consumption E falling in the grid of the p-th row and q-th column according to Equation (3); calculate the marginal probability P E (p) of the sample size of the hourly power consumption E falling in the p-th interval according to Equation (4); calculate the marginal probability P F (q) of the sample size of the hourly power consumption E falling in the q-th interval according to Equation (5).

[0010]

[0011] Step 4: According to Equation (6), correspondingly Figure 2 calculate the mutual dependence degree A i between the hourly power consumption E and F i (E,F i ).

[0012]

[0013] Step 5: Standardize A i (E,F i ) according to Equation (7) to obtain A i标准 (E,F i ), that is, process the value range of A i (E,F i ) to be between 0 and 1, which is beneficial for different A i (E,F i) are compared and evaluated. The denominator in the formula takes the minimum value of f and g to ensure that the standardized A i (E,F i ) has a value between 0 and 1.

[0014]

[0015] Step 6: According to the value changes and combinations of f and g, E×F i There can be multiple groups in matrix form, assumed to be h groups (0 < h < n, h is a positive integer). Therefore, according to formulas (1) to (7), the degree value A of the corresponding E×F i matrix can be calculated. i标准 (E,F i )

[0016] Step 7: Select the maximum value among h A i标准 (E,F i ) as the correlation coefficient C i between the hourly power consumption E of the power consumption system and the influencing factor F i , and its value is between 0 and 1.

[0017] C i = max(A i标准 (E,F i )) (8)

[0018] Step 8: According to Step 2 to 6, the R correlation coefficients C i (F1, F2, F3, F4, F5,...) (i = 1 to R) corresponding to the influencing factor F i (C1, C2, C3, C4, C5,...) (i = 1 to R) can be calculated. The influencing factor corresponding to the correlation coefficient closer to 1 is used as the important characteristic parameter affecting the power consumption of the power consumption system.

[0019] 2. Classification of power consumption patterns

[0020] Step 1: Select 2 parameters from several important characteristic parameters determined in Step 8 of Section 1, assumed to be F1 and F2. Establish a rectangular coordinate system with F1 as the x-axis and F2 as the y-axis (as Figure 3 (a)). Still assume that the total sample size of the hourly power consumption E of the power consumption system is n, and all fall within the coordinate axis system; and assume there are K i types of power consumption pattern classification methods (K1, K2, K3,...) (i = 1 to Q, K is a positive integer). For example, assume K i = 2, and the corresponding initial clustering center points are α1, α2 ( Figure 3represented by × in the figure, and the coordinates of these two points are (β1, γ1) and (β2, γ2) respectively; the remaining power consumption sample data E c (c = 1, 2,..., (n - 2))( Figure 3 represented by · in the figure) has the coordinate (β c , γ c )(c = 1, 2,..., (n - 2)); calculate the distance d(E c from α1 and α2 according to Equation (9), i.e., d(E c , α1) and d(E c , α2).

[0021] Step 2: According to Figure 3 (b), compare the magnitudes of d(E c , α1) and d(E c , α1), and assign the power consumption sample data E c to the cluster corresponding to the closer clustering center (the dashed box in Figure 3 (b)).

[0022] Step 3: Recalculate the means α 1.1 and α 2.1 of all power consumption sample data in the two clusters according to Equation (10), and use them as the new clustering centers. Here, E c1 and E c2 are the power consumption sample data in each cluster respectively, and their coordinates are (β c1 , γ c1 ), (β c2 , γ c2 ), and e1 and e2 are the sample data amounts in the two clusters respectively ( Figure 3 (c)).

[0023]

[0024] Step 4: Loop Steps 2 - 3; until the two clustering centers no longer change, and the coordinates of the corresponding clustering centers are (β1, γ1) and (β2, γ2).

[0025] Step 5: Calculate the average distance a from a certain power consumption sample data point to other power consumption sample data points in its cluster according to Equation (11), and this value reflects the closeness of the power consumption sample data points in this cluster; calculate the average distance b from this power consumption sample data point to the power consumption sample data points in the nearest neighbor cluster (not the cluster it belongs to) according to Equation (12); calculate the compactness coefficient TC according to Equation (13).

[0026]

[0027] The value of TC is between [-1, 1]. If TC is close to 1, it indicates that the clustering of power consumption sample data points is reasonable; if TC is close to -1, it means that the power consumption sample data points may be assigned to the wrong cluster; if TC is close to 0, it means that the power consumption sample data points are near the boundary of two clusters. Generally, for a sample set, the corresponding cohesion coefficient TC is the average of the cohesion coefficients of all power consumption sample data points. Therefore, when determining the optimal K value, try to select the K value corresponding to a larger TC value.

[0028] For example: TC is close to 1, such as 0.8; 0.9; 0.98; 0.99; for example, finally 0.99 and 0.98. Then we will also give priority to 0.99.

[0029] Step 6: Calculate the distance coefficient DC according to Equation (14). It is the sum of the distances from each power consumption sample data point to the center of its affiliated cluster. The smaller the DC value, the smaller the distance from the power consumption sample data point to the center of its affiliated cluster, indicating better clustering effect. Therefore, when selecting the optimal K value, try to select the K corresponding to a smaller DC value.

[0030]

[0031] Step 7: According to the calculation processes of Step 5 and Step 6, a set of K i will yield a set of TC i and DC i (i = 1 to Q). When selecting the optimal K value, the K corresponding to the combination of a larger TC i and a smaller DC i should be used as the optimal K value. It should be noted that the K value should follow the principle of moderation, neither too large nor too small. If the K value is too small, the role of clustering cannot be exerted; on the contrary, overfitting will occur, and the role of clustering cannot be exerted either. Step 8: Combining the calculation results of Step 8 in Section 1, the optimal clustering number K i corresponding to Step 7 in Section 2 i is the classification number of the power consumption patterns of the power consumption system. Description of the Drawings

[0032] Figure 1 is the method flow chart;

[0033] Figure 2 is the principle of constructing the E×F i matrix;

[0034] Figure 3 is the clustering principle diagram;

[0035] Figure 4 is the SC value and SSE value under different clustering numbers;

[0036] Figure 5 It is the classification of the electricity consumption of the baggage system;

[0037] Figure 6 It is the impact of the number of passengers per hour and the number of flights per hour on the hourly electricity consumption of the baggage. Specific implementation method

[0038] Taking the baggage system of a certain terminal as an example, the present method will be further described below:

[0039] The hourly electricity consumption data of the baggage system of this terminal from 2023 to 2024 were collected (sample size n = 17544), and the influencing factors F of the electricity consumption were preliminarily analyzed, including the outdoor air temperature (F1), solar radiation intensity (F2), number of passengers (F3), proportion of departing passengers (the ratio of the number of departing passengers to the number of arriving passengers, F4), and holidays (F5).

[0040] 1. Feature extraction

[0041] 1) Calculate the maximum value specified by the grid Therefore, the value range of E can be determined as 1 to 134 (positive integers).

[0042] 2) Substitute the probability P(p,q) and marginal probabilities P E (p), P F (q) in each grid into Equation (6) to calculate the degree of mutual dependence A i (E,F i ); among them, P(p,q) is calculated according to Equation (3), P E (p) is calculated according to Equation (4), and P F (q) is calculated according to Equation (5).

[0043] 3) Standardize A i (E,F i ) using Equation (7) to obtain A i标准 (E,F i ), and select the maximum value in A i标准 (E,F i ) as the correlation coefficient C i between the electricity consumption E of the terminal baggage system and the influencing factor F i (C1, C2, C3, C4, C5). The calculation results are shown in Table 1.

[0044] Table 1 Correlation analysis of influencing factors of the baggage system

[0045]

[0046] 4) According to Table 1, select F with C > 0.5 as the important influencing factor, that is, extract F3, F4, and F5 as the characteristic factors affecting the power consumption of the baggage system.

[0047] 2. Classification of power consumption patterns

[0048] 1) First, randomly select 2 data points from the power consumption sample data of the baggage system in the airport terminal as the initial cluster centers. Use Equation (9) to calculate the distances between these two data points and other data points, compare the magnitudes of these two distances, and assign them to the cluster corresponding to the nearest cluster center.

[0049] 2) Recalculate the mean of all power consumption sample data in each cluster according to Equation (10) as the new cluster center.

[0050] 3) Continuously repeat the above steps until the cluster centers no longer change.

[0051] 4) Calculate the clustering situation of the hourly power consumption of the baggage system for different clustering numbers K values (K = 2 - 10) according to the above clustering method; calculate the compactness coefficient TC and the distance coefficient DC according to Equations (13) and (14). The calculation results are as Figure 4 shown.

[0052] 5) According to the principle of the optimal K value, the comprehensive optimal classification number of the power consumption pattern of the baggage system in this case is 5.

[0053] 6) Combining the extracted power consumption characteristic factors and the classification results of the power consumption patterns, the relationship between the hourly power consumption of the baggage system and the daily flight schedule can be established ( Figure 5 ), and combined with Figure 6 a relatively clear explanation of its power consumption law can be given:

[0054] (a) During the daytime period (7:00 - 20:00), although the number of passengers will change with the daily flight number, the hourly power consumption of the baggage system remains basically the same. In other periods, the power consumption will change significantly due to the fluctuations of the hourly flight schedule and the number of passengers, and classification is needed to distinguish their influence degrees on the power consumption.

[0055] (b) The characteristics of Mode ① are concentrated in the period from 2 am to 4 am in the second half of the night, during which both the electricity consumption and the number of people are at a very low level; Mode ② is concentrated in the periods from 0 am to 2 am and from 4 am to 5 am in the second half of the night, and the number of passengers during this period has not reached the lowest value within a day; the characteristics of Mode ③ are concentrated in the periods of rapid fluctuations in the number of passengers from 5 am to 7 am and from 20 pm to 21 pm, and the electricity consumption remains basically unchanged; Mode ④ mainly appears from 22 pm to 23 pm, during which its electricity consumption gradually decreases from a high level, and the electricity consumption will increase with the increase in the number of daily flight schedules; Mode ⑤ is in the period of high flight numbers during the day (from 7 am to 20 pm), and its characteristic is that the electricity consumption remains basically unchanged, but the electricity consumption will increase with the increase in the number of daily flight schedules.

Claims

1. A method for extracting power consumption features and classifying patterns in airport terminals, characterized in that The following steps are involved:

1. Power consumption feature extraction Step 1: First, find the factors F that may affect the hourly power consumption E of the airport terminal power system i , F1, F2, F3, F4, F5,...F R ; i = 1 ~ R, including outdoor air temperature, solar radiation intensity, number of passengers, ratio of departing passengers to arriving passengers, and holidays; Step 2: For any influencing factor F i , with F i is the x-axis, E is the y-axis to create a matrix; along F i Direction grid step B F Determine the calculation according to formula (1); the grid step size B along the E direction E Determine the calculation according to formula (2); in order to prevent overfitting, it is stipulated and The variables f and g take values ​​within this range but must be positive integers; Step 3: Set the total sample size of the hourly power consumption E of the power system to n, and all samples fall within E×F i In the matrix; where n pq is the sample size of the hourly electricity consumption E that falls in the grid of the pth row and the qth column, p = 1 to f and is a positive integer, q = 1 to g and is a positive integer; n p is the sample size of hourly electricity consumption E falling in the p-th row of the grid; n q is the sample size of hourly electricity consumption E falling in the p-th row of the grid; According to formula (3), the probability P(p,q) of the sample quantity of hourly electricity consumption E falling in the grid of the pth row and the qth column is calculated; according to formula (4), the marginal probability P of the sample quantity of hourly electricity consumption E falling in the pth interval is calculated E (p); According to formula (5), the marginal probability P of the sample volume of hourly electricity consumption E falling in the qth interval is calculated F (q); Step 4: Calculate the hourly power consumption E and F according to formula (6) i The degree of mutual dependence i (E,F i ); Step 5: According to formula (7), i (E,F i ) is standardized to obtain A i标准 (E,F i ), which is about to be A i (E,F i ) is processed to a value range between 0 and 1; the denominator in the formula takes the minimum value of f and g to ensure that the standardized A i (E,F i ) has a value between 0 and 1; Step 6: According to the value changes and combinations of f and g, E×F i There are multiple matrix forms, set as h groups, 0 < h < n, h is a positive integer; therefore, according to formulas (1) to (7), the degree value A of the corresponding E×F matrix is calculated i of the matrix i标准 (E, F i ) Step 7: Select h A according to formula (8) i标准 (E,F i ) as the maximum value of the hourly power consumption E of the power system and the influencing factor F i The correlation coefficient C i , its value is between 0 and 1; C i =max(A i标准 (E,F i )) (8) Step 8: According to Steps 2 to 6, calculate the corresponding influencing factor F i The R correlation coefficient C i That is, C1, C2, C3, C4, C5, ... C R , the influencing factors corresponding to the correlation coefficients greater than 0.5 are taken as important characteristic parameters affecting the power consumption of the power system; 2. Classification of power consumption patterns Step 1: Select two parameters from the important characteristic parameters determined in Step 8 of Section 1 and set them as F1 and F2; establish a rectangular coordinate system with F1 as the x-axis and F2 as the y-axis; set the total sample size of the hourly power consumption E of the power system to n, and all of them fall in the coordinate axis system; and set K i Classification of electricity consumption patterns; When K i =2, the corresponding initial cluster center points are α1 and α2, and the coordinates of these two points are (β1,γ1) and (β2,γ2) respectively; the remaining power consumption sample data E c The coordinates of (β c ,γ c );Calculate the power consumption sample data E according to formula (9) c The distance d(E c ,α1),d(E c ,α2); where c = 1, 2, ..., (n-2); Step 2: Compare d(E c ,α1),d(E c ,α1), the power consumption sample data E c Assign to the cluster corresponding to the cluster center that is close to you; Step 3: Recalculate the mean α of all power consumption sample data in the two clusters according to formula (10) 1.1 and α 2.1 , as the new cluster center; where E c1 and E c2 They are the power consumption sample data in each cluster, and their coordinates are (β c1 ,γ c1 ), (β c2 ,γ c2 ), e1 and e2 are the sample data amounts in the two clusters respectively; Step 4: Repeat Step 2 to 3 until the two cluster centers no longer change, and the coordinates of the corresponding cluster centers are (β1,γ1) and (β2,γ2). Step 5: According to formula (11), calculate the average distance a from a certain power consumption sample data point to other power consumption sample data points in its cluster. This value reflects the compactness of the power consumption sample data points in the cluster; according to formula (12), calculate the average distance b from the power consumption sample data point to the power consumption sample data points in the nearest neighboring cluster; according to formula (13), calculate the compactness coefficient TC; The value of TC is between [-1,1]. If TC is close to 1, it means that the clustering of the power consumption sample data points is reasonable; if TC is close to -1, it means that the power consumption sample data points may be assigned to the wrong cluster; if TC is close to 0, it means that the power consumption sample data points are near the boundary of two clusters; when determining the optimal K value, select the K value corresponding to the large TC value; Step 6: Calculate the distance coefficient DC according to formula (14), which is the sum of the distances from each power consumption sample data point to the center of the cluster to which it belongs. The smaller the DC value, the smaller the distance from the power consumption sample data point to the center of the cluster to which it belongs, which means the better the clustering effect. Therefore, when selecting the best K value, select the K corresponding to the small DC value. Step 7: Repeat the calculation process of Step 5 and Step 6. i A set of TC i and DC i , i = 1 ~ Q; when selecting the optimal K value, the corresponding larger TC i , Smaller DC i Combination K i As the optimal K value; K is within 2-10; Step 8: Combine the calculation results of Step 8 in Section 1 and the optimal number of clusters K in Step 7 in Section 2 i It is the classification number of power consumption modes of the power consumption system.

Images