Heating heat load real-time prediction method based on building use behavior identification

Through a multivariate linear regression method based on building usage behavior and combined with physical laws, the problem of insufficient data mining of factors affecting meteorological and human behavior in heating heat load prediction is solved, and high-precision and rapid heating heat load is achieved Predictions solve the problem of excessive or insufficient heating.

CN119090044APending Publication Date: 2024-12-06CHINA IPPR INT ENG CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410989123.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing data-driven method in the prediction of heating heat loads in insufficient data mining of outdoor meteorological changes and human behavior factors, resulting in poor prediction accuracy, especially when the sample data is insufficient in the early stage, the phenomenon of excessive or insufficient heating is frequent, and the sun cannot be accurately described. Radiation and the impact of humans on building use behavior.

Method used

Using a real-time prediction method of heating heat load based on building usage behavior identification, the heating system data is continuously collected, the time series data vector is constructed, and the daily time-by-day data packets are cleaned and packaged, and load prediction is generated by using the use scenario. Coefficients are used to establish models for real-time predictions in combination with physical laws.

Benefits of technology

It realizes high-precision heating load prediction, fast computing speed and strong robustness, eliminates excessive or insufficient heating, and meets heating requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090044A_ABST
    Figure CN119090044A_ABST
Patent Text Reader

Abstract

The invention discloses a heating thermal load real-time prediction method based on building use behavior identification, and the method comprises the steps: S1, continuously collecting a building heating system thermal load, indoor temperature and humidity parameters and outdoor meteorological parameters hour by hour within k days, and constructing hour-by-hour data vectors according to a time sequence; s2, cleaning the hourly data vector, and generating a daily hourly data packet; s3, classifying the daily hourly data packets into a workday data set, a weekend data set and a holiday and festival data set according to different building use scenes; s4, after the kth day is ended, respectively generating a workday training set, a weekend training set and a holiday training set at the ith moment based on the data sets in the S; and S5, on the (k + 1) th day, selecting the corresponding training set at the ith moment generated on the kth day according to a building use scene, solving a load prediction coefficient vector at the ith moment on the (k + 1) th day by adopting a multiple linear regression method, and then calculating to obtain a thermal load prediction value at the ith moment on the (k + 1) th day.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of smart heating technology, and in particular to a real-time prediction method for heating heat load based on building usage behavior identification. Background Art

[0002] There are two main methods for heating load forecasting: physical law method and data-driven method. The physical law method starts from the physical concepts and laws of the heating system and uses the heat load formula for related calculations. Since the heating model required by the physical law method requires many parameters and is usually difficult to obtain, it is only used for rough load forecasting and is rarely used in smart heating systems with high prediction accuracy requirements. At present, with the advent of the big data era, data-driven methods have been widely used in heating load forecasting. This method mathematically describes the system based on the relationship between given input and output data, and further obtains prediction results. There are mainly statistical regression methods, support vector machine methods, neural network methods, decision tree methods, etc.

[0003] The amount of sample data is the key factor that determines the prediction effect of the data-driven method. Only after sufficient sample data training can satisfactory prediction accuracy be obtained. In actual engineering, due to insufficient training sample data, the heating system of newly-built buildings in the initial stage of operation often needs to accumulate sample data for a long period of time before satisfactory prediction accuracy can be gradually achieved. Due to insufficient initial sample data, excessive or insufficient heating caused by poor prediction accuracy frequently occurs, which does not meet the heating requirements. In addition, the heating heat load is greatly affected by outdoor meteorological parameters, especially solar radiation heat gain. Due to the intermittent, variable and uncertain nature of solar energy, there is no definite model that can accurately describe and predict the hourly solar heat gain. At the same time, people's random use of buildings also has a great impact on the heating load, and there is no definite model that can describe this impact. Due to the above technical problems, the data-driven method has shortcomings in exploring the actual characteristics of the heating load and improving the prediction accuracy.

[0004] Therefore, it is expected to develop new technical methods that can at least partially overcome the deficiencies in the prior art. Summary of the invention

[0005] In view of the shortcomings of existing data-driven heating heat load prediction technology in data mining of factors affecting outdoor meteorological changes and human behavior, the present invention proposes a real-time heating heat load prediction method based on building usage behavior identification.

[0006] More specifically, the present invention provides a real-time prediction method for heating heat load based on building usage behavior identification, comprising:

[0007] S1: Within k days, the heat load of the building heating system, indoor temperature and humidity parameters, and outdoor meteorological parameters are continuously collected hourly, and hourly data vectors are constructed according to the time series;

[0008] S2: Clean the hourly data vectors to remove the data vectors at the time when unqualified data exists, and pack the cleaned data daily to generate daily hourly data packets;

[0009] S3: Classify the daily hourly data packets into weekday data sets, weekend data sets, and holiday data sets according to different building usage scenarios;

[0010] S4: After the kth day ends, the most recent n (n>4) data vectors at the i-th moment (1≤i≤24) are extracted from the workday data set, weekend data set, and holiday data set to generate the workday training set, weekend training set, and holiday training set at the i-th moment respectively; and

[0011] S5: On the k+1th day, according to the building usage scenario on the k+1th day, the corresponding training set at the i-th moment generated on the k+1th day is selected, and the load prediction coefficient vector at the i-th moment on the k+1th day is obtained by using the multivariate linear regression method, and then the heat load prediction value at the i-th moment on the k+1th day is calculated.

[0012] According to an embodiment of the present invention, in step S1, the heat load of the heating system, indoor temperature and humidity parameters and outdoor meteorological parameters are obtained through actual measurement or an existing database.

[0013] According to the embodiment of the present invention, in step S1, the heat load of the building heating system, the indoor temperature and humidity parameters and the outdoor meteorological parameters, specifically including the indoor air temperature test value T of the building heat user n , Building heat user indoor air enthalpy test value H n , Outdoor air temperature test value T w , Outdoor air enthalpy test value H w , Outdoor horizontal surface total solar radiation test value S z , Outdoor wind speed test value V f And the building heating heat load test value Q t ; The hourly data vector is a row vector containing 7 elements in the format of:

[0014] According to an embodiment of the present invention, the heating system operation condition correlation coefficient J is set A , J B , J C , J D , where J A =T n -T w , J B =(Hn -H w )·V f 1.34 , J C =1, J D =S z ;

[0015] In step S2, the generation of the daily hourly data packet includes: reorganizing the data at each moment according to the above correlation coefficient to generate a row vector containing 5 elements, the format of which is: [J A J B J C J D Q t ]; Then, the row vectors containing 5 elements generated by the above reorganization at each moment are arranged in time sequence and packaged to generate daily hourly data packets. The data at the vacant moments due to cleaning remain vacant. The daily hourly data packets are the data matrix U i×5 , where i is the number of sampling moments per day.

[0016] According to the embodiment of the present invention, in step S3, the weekday data set, the weekend data set and the holiday data set are all three-dimensional matrices, respectively denoted as W i×5×r , L i×5×j and M i×5×p , where r is the number of working days stored in the data set, j is the number of weekend days stored in the data set, and p is the number of holidays stored in the data set.

[0017] According to an embodiment of the present invention, in step S4, the matrix WX of the working day training set at the i-th moment is generated by:

[0018]

[0019] The method for generating the weekend training set matrix LX at the i-th moment is:

[0020]

[0021] The method for generating the "holiday training set" matrix MX at the i-th moment is:

[0022]

[0023] According to an embodiment of the present invention, in step S5, obtaining the load forecast coefficient vector at the i-th moment on the k+1-th day includes:

[0024] On the K+1th day, according to the building usage scenario, the multivariate linear regression method is used to obtain the working day training set matrix WX generated in the k+1th selection step S4 at the i-th moment n×5 , weekend training set matrix LX n×5Or holiday training set matrix MX n×5 , and then construct the matrix equations according to the following formula (14):

[0025]

[0026] Then, solve the equation, and since n>4, obtain the building usage behavior correlation coefficient matrix [ABCD]′ as the least squares solution of the equation;

[0027] Q t =A.J A +B.J. B +C.J. C +D.J. D (14)

[0028] Among them, A, B, C and D are the correlation coefficients of building use behavior at a certain moment during the heating period, A = k·F / 1000, B = m, C = -Q b , D = -α / 1000, k is the average heat transfer coefficient of the building envelope; F is the surface area of ​​the building envelope; m is the amount of cold air infiltrated into the building under the combined action of wind pressure and thermal pressure at unit outdoor wind speed; Q b is the heat load of indoor equipment in the building; α is the heat load of solar radiation of the building at a certain moment Q4 and the total solar radiation of the outdoor horizontal surface S z ratio.

[0029] According to the embodiment of the present invention, in step S5, the calculation of the heat load forecast value at the i-th moment on the k+1th day includes: obtaining the building use behavior correlation coefficients A, B, C, D at the i-th moment, and the heating system operation condition correlation coefficient J at the i-th moment on the k+1th day A·T , J B·T , J C , J D Substituting the real-time test data into the following formula ⒂, we can get the predicted value Q of the building heating load at the i-th moment on the K+1 day: y :

[0030] Q y =A.J A·T +B.J. B·T +C.J. C +D.J. D (15)

[0031] Among them, J A·T =T n·s -T w , J B·T =(H n·s -H w )·V f 1.34 , Tn·s is the design value of indoor air temperature of building heat users (unit: °C); H n·s Design value of indoor air enthalpy for building heat users.

[0032] According to an embodiment of the present invention, the real-time prediction method for heating heat load based on building usage behavior identification also includes restarting steps S1 to S5 after step S5 is completed to obtain the hourly prediction value of the building heating heat load on and after the K+2th day.

[0033] The real-time prediction method for heating heat load based on building use behavior identification of the present invention combines the respective advantages of the physical law method and the data-driven method, adopts the physical law method to establish an identification model of human behavior influencing factors and outdoor meteorological change factors, and uses a multiple linear regression method based on scenario day analysis to identify the hourly use behavior of the building through learning the historical operation data of heating; then the hourly measured data of outdoor meteorological parameters and the hourly use behavior of the building identified above are substituted into the mathematical model established by the physical law method to achieve real-time high-precision prediction of heating heat load. The method of the present invention has the advantages of fast computing speed and strong robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic flow chart of a method for real-time prediction of heating load based on building usage behavior identification according to an embodiment of the present invention; and

[0035] Figure 2 A comparison diagram of a daily heating load prediction value and a theoretical value of a real-time heating load prediction method based on building usage behavior identification according to an embodiment of the present invention;

[0036] Figure 3 A daily heating supply prediction error diagram of a real-time heating load prediction method based on building usage behavior identification according to an embodiment of the present invention;

[0037] Figure 4 The hourly load diagram of the day with the maximum positive value of the daily heating quantity prediction error of the real-time heating heat load prediction method based on building usage behavior identification according to the implementation scheme of the present invention;

[0038] Figure 5 It is a diagram of the identification coefficient (usage behavior correlation coefficient) of the day with the maximum positive value of the daily heating quantity prediction error of the real-time heating heat load prediction method based on building usage behavior identification according to the embodiment of the present invention;

[0039] Figure 6 The hourly load diagram of the day with the minimum negative value of the daily heating quantity prediction error of the real-time heating heat load prediction method based on building usage behavior identification according to the implementation scheme of the present invention;

[0040] Figure 7 It is an identification coefficient diagram of the day with the minimum negative value of the daily heating quantity prediction error of the real-time heating heat load prediction method based on building usage behavior identification according to the implementation scheme of the present invention;

[0041] Figure 8 The hourly load diagram of the day with the minimum value of the daily heating quantity prediction error of the real-time heating heat load prediction method based on the identification of building usage behavior according to the implementation scheme of the present invention;

[0042] Fig. 9 It is an identification coefficient diagram of the minimum value day of daily heat supply prediction error of the real-time heating heat load prediction method based on building usage behavior identification according to the implementation scheme of the present invention;

[0043] Fig.10 An hourly load diagram of the day with the maximum heat load according to the real-time prediction method for heating heat load based on building usage behavior identification according to an embodiment of the present invention; and

[0044] Fig.11 It is an identification coefficient diagram of the maximum heat load day of the real-time heating load prediction method based on building usage behavior identification according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention is further described in detail below through specific embodiments in conjunction with the accompanying drawings. The shown contents are used to fully illustrate the contents of the present invention, but are not used to limit the present invention.

[0046] In order to explain the present invention more clearly, the design principle of the present invention is first described in detail below.

[0047] According to the physical laws of building heating systems, the following heat balance relationship exists during the heating period:

[0048] Q t =Q1+Q2+Q3+Q4 (1)

[0049] in:

[0050] Q1=k·F·(T n -T w ) / 1000 ⑵

[0051] Q2=m·V f 1.34 ·(H n -H w )⑶

[0052] Q3=-Q b ⑷

[0053] Q4=-α·S z / 1000 ⑸

[0054] Where:

[0055] Q t — Test value of building heating heat load (unit: kW); Q1—Heat load of building envelope (unit: kW); Q2—Heat load of cold air infiltration (unit: kW); Q3—Heat load of indoor equipment (unit: kW); Q4—Heat load of solar radiation (unit: kW); k—Average heat transfer coefficient of building envelope (unit: W / m 2 ·℃); F—surface area of ​​building envelope (unit: m 2 );T n —Test value of average indoor air temperature of building heat users (unit: °C); T w —Outdoor air temperature test value (unit: ℃); m—Building infiltration cold air volume under the combined action of wind pressure and thermal pressure at unit outdoor wind speed (unit: kg / h); V f —Outdoor wind speed test value (unit: m / s); H n — Test value of indoor air enthalpy of building heat users (unit: kJ / kg.dry air); H w —Outdoor air enthalpy test value (unit: kJ / kg.dry air); Q b —Heat load of indoor equipment in the building (unit: kW); α—Heat load of solar radiation of the building at a certain moment Q4 and total solar radiation of outdoor horizontal surface S z Ratio of (unit: dimensionless); S z —Outdoor horizontal surface total solar radiation test value (unit: W / m 2 ).

[0056] Set the heating system operating condition correlation coefficient J A , J B , J C , J D , the calculation formula is as follows:

[0057] J A =T n -T w ⑹

[0058] J B =(H n -H w )·V f 1.34 ⑺

[0059] J C =1 ⑻

[0060] J D =S z ⑼

[0061] From formulas ⑹~⑼, it can be seen that the operating condition correlation coefficient J at a certain moment in the heating period is A , J B , J C , J D The influencing factors are the indoor air parameters and outdoor meteorological parameters at that moment, which can be obtained through testing.

[0062] Set the building usage behavior correlation coefficients A, B, C and D, and the calculation formula is as follows:

[0063] A=k·F / 1000⑽

[0064] B=m⑾

[0065] C=-Q b ⑿

[0066] D=-α / 1000 ⒀

[0067] It can be seen from formulas ⑽~⒀ that for the correlation coefficients A, B, C, and D of building use behavior at a certain moment in the heating period, the influencing factors are the building use behavior at that moment, such as the room use pattern, window opening and ventilation pattern, indoor electrical equipment use pattern, and the change in α value caused by changes in the solar hour angle and altitude angle. These cannot be obtained through direct testing but can be obtained through system digital identification.

[0068] Substitute coefficient formulas (6) to (7) into formulas (2) to (5), and then substitute into formula (1), and we get the following heat balance equation:

[0069] Q t =A.J A +B.J. B +C.J. C +D.J. D ⒁

[0070] For a certain usage scenario of a building (such as weekdays, weekends, holidays, etc.), based on the assumption that the building usage behavior at the same time under the usage scenario for n consecutive days remains basically unchanged (such as getting up at 7 o'clock every day on weekdays to open windows for ventilation), on the Kth day, the heating system operating condition correlation coefficient J at the i-th moment (1≤i≤24) of the same usage scenario in the last n days (n>4) is obtained. A , J B , J C , J D and heat load Q tAccording to the test data, according to formula ⒁, a heat balance equation can be established at the time i of each day. According to the data of the last n days, a total of n independent equations can be established. The number of unknowns of the building usage behavior correlation coefficients A, B, C, and D at the time i that need to be solved is 4. Since n>4, the least squares solution of the building usage behavior correlation coefficients A, B, C, and D at the time i can be obtained by identifying the multivariate linear regression method.

[0071] Based on the assumption that the building usage behavior at the i-th moment of the K+1 day and the building usage behavior at the i-th moment of the K day are basically unchanged under the same building usage scenario, the building usage behavior correlation coefficients A, B, C, and D at the i-th moment obtained by identification on the K-th day and the heating system operation condition correlation coefficient J at the i-th moment of the K+1 day are calculated. A·T , J B·T , J C , J D Substituting the real-time test data into formula ⒂, we can get the predicted value Q of the building heating load at the i-th moment on the K+1 day: y , the calculation formula is as follows:

[0072] Q y =A.J A·T +B.J. B·T +C.J. C +D.J. D ⒂

[0073] in:

[0074] J A·T =T n·s -T w ⒃

[0075] J B·T =(H n·s -H w )·V f 1.34 ⒄

[0076] Where: Q y —Building heating load forecast value (unit: kW); T n·s —Design value of indoor air temperature of building heat users (unit: °C); H n·s —Design value of indoor air enthalpy of building heat users (unit: kJ / kg.dry air).

[0077] The above is the heat load prediction principle of the present invention. Based on this, the present invention proposes a real-time prediction method for heating heat load based on building usage behavior identification.

[0078] Figure 1This is a flow chart of a method for real-time prediction of heating load based on building usage behavior identification according to an embodiment of the present invention. As shown in the figure, the method of the embodiment includes the following steps:

[0079] S1: Within k days, the heat load of the building heating system, indoor temperature and humidity parameters, and outdoor meteorological parameters are continuously collected hourly, and hourly data vectors are constructed according to the time series, that is, the relevant original data are obtained.

[0080] The heat load of the building heating system, indoor temperature and humidity parameters, and outdoor meteorological parameters collected hourly in step S1 include the indoor air temperature test value T of the building heat user. n (Unit: °C), building heat user indoor air enthalpy test value H n (Unit: kJ / kg. dry air), outdoor air temperature test value T w (Unit: °C), outdoor air enthalpy test value H w (Unit: kJ / kg. dry air), outdoor horizontal surface total solar radiation test value S z (Unit: W / m 2 ), outdoor wind speed test value V f (Unit: m / s) and building heating heat load test value Q t (Unit: kW) and other 7 types of parameters. All raw data can be collected at 1 hour intervals, so 24 sampling points are set every day to collect the above 7 types of parameters hourly. Of course, you can also set a suitable time interval according to your needs.

[0081] The hourly "data vector" (raw data) constructed from the time series is a row vector containing 7 elements in the format: For example, there are 24 hours in a day, with a time interval of 1 hour, and a total of 24 sets of "data vectors" are generated.

[0082] It should be understood that the heat load of the building heating system, indoor temperature and humidity parameters, and outdoor meteorological parameters can be obtained through actual measurements or existing databases.

[0083] S2: Clean the hourly data vectors to remove the data vectors at the time when unqualified data exists, and pack the cleaned data daily to generate daily hourly data packets.

[0084] There is no particular restriction on the method of cleaning the original data, for example, a reasonable interval screening method can be used. This method sets a reasonable value interval for each of the above 7 types of parameters, and the data outside the interval is regarded as an abnormal value, and the "data vector" at the time when it is located is deleted. After the deletion, the data gap at that time is no longer filled.

[0085] Then, the cleaned raw data is packaged, and the data at each moment is rearranged according to formulas (6) to (9) to generate a row vector containing 5 elements in the format of: [J A J B J C J D Q t ]; Then, the row vectors containing 5 elements generated by the above reorganization at each moment are arranged in time sequence and packaged to generate a "daily hourly data packet". The data at the vacant moments due to cleaning remain vacant. When the sampling time interval is 1 hour, the "daily hourly data packet" is a 24×5 data matrix, denoted as matrix U 24×5 .

[0086] S3: Classify the daily hourly data packets into weekday data sets, weekend data sets, and holiday data sets according to different building usage scenarios.

[0087] For example, when a day is a working day, the matrix U 24×5 Sort by date and include them in the "Workday Dataset". The Workday Dataset is a 3D matrix, denoted as W 24×5×r , where r is the number of working days stored in the data set, and the sorting principle is W 24×5×1 is the data of the most recent day, W 24×5×r is the data of the earliest day. When a day is a weekend, the matrix U 24×5 Sort by date and include them in the "weekend data set". The weekend data set is a 3D matrix, denoted as L 24×5×j , where j is the number of weekend days stored in the data set, and the sorting principle is L 24×5×1 is the data of the most recent day, L 24×5×j is the data of the earliest day. When a day is a holiday, the matrix U 24×5 Sort by date and include them in the "holiday data set". The holiday data set is a 3D matrix, denoted as M 24×5×p , where p is the number of holidays stored in the data set, and the sorting principle is M 24×5×1 is the data of the most recent day, M 24×5×p The data is for the earliest day.

[0088] S4: After the kth day, extract the most recent n (n>4) data vectors at the i-th moment (1≤i≤24) from the above-mentioned weekday data set, weekend data set, and holiday data set to generate the weekday training set, weekend training set, and holiday training set at the i-th moment, respectively.

[0089] More specifically, the method for generating the "weekday training set" matrix WX at the i-th moment is:

[0090]

[0091] The generation method of the "weekend training set" matrix LX at the i-th moment is:

[0092]

[0093] The generation method of the "holiday training set" matrix MX at the i-th moment is:

[0094]

[0095] S5: On the k+1th day, according to the building usage scenario on the k+1th day, the corresponding training set at the i-th moment generated on the k+1th day is selected. For example, if the k+1th day is a working day, the "working day training set" is selected; then the multiple linear regression method is used to obtain the load prediction coefficient vector at the i-th moment on the k+1th day, and then the heat load prediction value at the i-th moment on the k+1th day is calculated.

[0096] More specifically, on the K+1th day, when the building usage scenario is a working day, the steps of using the multivariate linear regression method to obtain the load forecast coefficient vector at the i-th moment on the k+1th day include:

[0097] First, select the "weekday training set" matrix WX generated in step S4 at the i-th moment n×5 , construct a linear equation system according to formula ⒁:

[0098] WX(:,1:4)×[ABCD]′=WX(:,5) ⒅

[0099] It can be written in matrix equation form as follows:

[0100]

[0101] Then, solve the above equation ⒆. Since n>4, the building usage behavior correlation coefficient matrix [AB CD]′ obtained by solving is the least square solution of the equation.

[0102] When the building usage scenario is the weekend, the steps of using the multivariate linear regression method to obtain the load forecast coefficient vector at the i-th moment on the k+1th day include:

[0103] First, select the "weekend training set" matrix LX generated in step S4 at the i-th moment n×5 , construct a linear equation system according to formula ⒁:

[0104] LX(:,1:4)×[ABCD]′=LX(:,5) ⒇

[0105] It can be written in matrix equation form as follows:

[0106]

[0107] Then, solve the above equation (21). Since n>4, the building usage behavior correlation coefficient matrix [A BC D]′ obtained by solving is the least square solution of the equation.

[0108] When the building usage scenario is a holiday, the steps of using the multivariate linear regression method to obtain the load prediction coefficient vector at the i-th moment on the k+1th day include:

[0109] First, select the "holiday training set" matrix MX at the i-th moment generated in step S4 n×5 , construct a linear equation system according to formula ⒁:

[0110] MX(:,1:4)×[ABCD]′=MX(:,5) (22)

[0111] Written in matrix equation form:

[0112]

[0113] Then, solve the above equation (23). Since n>4, the building usage behavior correlation coefficient matrix [A BC D]′ obtained by solving is the least square solution of the equation.

[0114] After obtaining the least squares solutions under the above scenarios, the building usage behavior correlation coefficients A, B, C, D at time i and the heating system operation condition correlation coefficient J at time i on day K+1 are obtained. A·T , J B·T , J C , J D Substituting the real-time test data into formula (15), we can get the predicted value Q of the building heating load at the i-th moment on the K+1 day: y .

[0115] Similarly, restarting step S1, the hourly forecast value Q of the building heating load on the K+2th day and later can be obtained hour by hour. y .

[0116] The present invention is further described in detail below in conjunction with specific examples.

[0117] Taking the heating system of a residential building in Beijing as an example, the original data for 30 days from November 15 to December 14, 2023 is collected to predict the heating load Q on the 31st day, that is, December 15. y .

[0118] Firstly, EnergyPlus software was used to simulate the hourly heat load of the example building. The outdoor meteorological parameters in Beijing used in the simulation were selected according to the EnergyPlus meteorological database, and the source was CSWD (China Typical Meteorological Year Data).

[0119] The above hourly simulation value is used as the building heating heat load test value Q t , and read the data in the meteorological database at the same time, construct the hourly "data vector" (raw data), including the indoor air temperature test value T of the building heat user n (Unit: °C), building heat user indoor air enthalpy test value H n (Unit: kJ / kg. dry air), outdoor air temperature test value T w (Unit: °C), outdoor air enthalpy test value H w (Unit: kJ / kg. dry air), outdoor horizontal surface total solar radiation test value S z (Unit: W / m 2 ), outdoor wind speed test value V f (Unit: m / s) and building heating heat load test value Q t (Unit: kW) and other 7 types of parameters.

[0120] The data collection interval was 1 hour, and the duration was 30 days from November 15 to December 14. A total of 720 sets of "data vectors" (raw data) were collected.

[0121] After data cleaning, all the data of the sample project are qualified. From November 15 to December 14, a total of 30 days, a total of 30 "daily hourly data packets" were generated.

[0122] The "Daily Hourly Data Pack" is classified into "Workday Data Pack" and "Weekend Data Pack" according to the different building usage behaviors. This example is a residential building and does not involve holiday operation scenarios (generally used in school dormitories, etc.). From November 15th to December 14th, there are 30 days, including 22 working days and 8 weekend days. According to the rules, the number of days classified into the "Workday Data Pack" is 22 days, and the number of days classified into the "Weekend Data Pack" is 8 days.

[0123] At the end of the 30th day, the "data vectors" of the most recent n (n>4) moments (1≤i≤24) are extracted from the "weekday data set", "weekend data set", and "holiday data set" to generate the "weekday training set", "weekend training set", and "holiday training set" of the moment i, respectively. Specifically, as of December 14, the matrix of the "weekday training set" of the moment i (1≤i≤24) in this example is a 22×5 matrix, the matrix of the "weekend training set" is an 8×5 matrix, and the "holiday training set" is not involved.

[0124] On the 31st day (i.e. the k+1th day), according to the building usage scenario, the relevant "training set" generated on the 30th day at the i-th moment is selected, and the multiple linear regression method is used to obtain the load prediction coefficient vector at the i-th moment on the 31st day, and then the heat load prediction value at the i-th moment on the 31st day is calculated.

[0125] Specifically, December 15 is a working day. First, the "working day training set" generated up to December 14 is selected, and 22 equations are constructed for the i-th moment (1≤i≤24, the time interval is 1 hour). The multivariate linear regression method is used to solve the usage behavior correlation coefficients A, B, C, and D at that moment; then, the usage behavior correlation coefficients A, B, C, and D obtained above are combined with the heating system operation condition correlation coefficient J at the i-th moment. A·T , J B·T , J C , J D Substituting the real-time test data into formula (15), we can get the predicted value of the building heating load Q at the i-th moment on December 15 y .

[0126] This embodiment uses 30 days of simulation data from November 15 to December 14 as learning data, and uses all simulation data of the heating period as test data. Taking absolute percentage error as the evaluation index of the method of the present invention, experiments were conducted on the MATLAB program. The experimental results show that the total heat consumption prediction error of the heating period is 0.4%; the daily heat supply prediction error is between -3.76% and 11.08%; except for very few days, the daily heat supply prediction error is less than 5%. Analysis shows that the above prediction error level can eliminate the phenomenon of excessive heating that is common in existing heating systems, while ensuring the quality of heating, with significant results. The specific experimental results are as follows:

[0127] The predicted maximum value of heating heat load is 1244.17kW, the theoretical maximum value is 1244.17kW, and the prediction error is 0.0%;

[0128] The predicted value of total heat consumption during the heating period is 1.4263 million kWh, the theoretical value is 1.4211 million kWh, and the prediction error is 0.4%;

[0129] Among them, the daily heat supply forecast value and theoretical value are detailed in Figure 2 , daily heat supply prediction error is detailed in Figure 3 .

[0130] The maximum positive value of the daily heat supply forecast error is 11.08%, which occurs on November 23. The hourly load and identification coefficients (A, B, C, D) of this day are detailed in Figure 4 , Figure 5 .

[0131] The minimum negative value of the daily heat supply forecast error is -3.76%, which occurs on January 15. For details of the hourly load and identification coefficient of this day, see Figure 6 , Figure 7 .

[0132] The minimum error of daily heat supply forecast is 0.00%, which occurs on November 19. The hourly load and identification coefficient of this day are detailed in Figure 8 , Fig. 9 .

[0133] The forecast error of heat supply on the day with the maximum heat load (January 19) is 0.02%. For details of the hourly load and identification coefficient of this day, see Fig.10 , Fig.11 .

[0134] The above description of the embodiments is to facilitate the understanding and application of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the embodiments herein, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.

Claims

1. A real-time prediction method for heating load based on building usage behavior identification, It is characterized in that include: S1: Within k days, the heat load of the building heating system, indoor temperature and humidity parameters, and outdoor meteorological parameters are continuously collected hourly, and hourly data vectors are constructed according to the time series; S2: Clean the hourly data vectors to remove the data vectors at the time when unqualified data exists, and pack the cleaned data daily to generate daily hourly data packets; S3: Classify the daily hourly data packets into weekday data sets, weekend data sets, and holiday data sets according to different building usage scenarios; S4: After the kth day ends, extract the latest n (n>4) data vectors at the i-th moment (1≤i≤24) from the weekday data set, weekend data set, and holiday data set to generate the weekday training set, weekend training set, and holiday training set at the i-th moment respectively; as well as S5: On the k+1th day, according to the building usage scenario on the k+1th day, the corresponding training set at the i-th moment generated on the k+1th day is selected, and the load prediction coefficient vector at the i-th moment on the k+1th day is obtained by using the multivariate linear regression method, and then the heat load prediction value at the i-th moment on the k+1th day is calculated.

2. A real-time prediction method for heating heat load based on building usage behavior identification as claimed in claim 1, Features: In step S1, the heat load of the heating system, indoor temperature and humidity parameters, and outdoor meteorological parameters are obtained through actual measurement or an existing database.

3. A real-time prediction method for heating heat load based on building usage behavior identification as claimed in claim 1, Features: In step S1, the heat load of the building heating system, indoor temperature and humidity parameters and outdoor meteorological parameters, specifically including the indoor air temperature test value T of the building heat user n , Building heat user indoor air enthalpy test value H n , Outdoor air temperature test value T w , Outdoor air enthalpy test value H w , Outdoor horizontal surface total solar radiation test value S z , Outdoor wind speed test value V f And the building heating heat load test value Q t ; The hourly data vector is a row vector containing 7 elements in the format of:

4. A real-time prediction method for heating heat load based on building usage behavior identification as claimed in claim 3, Features: Set the heating system operating condition correlation coefficient J A , J B , J C , J D , where J A =T n -T w , J B =(H n -H w )V f 1.34 , J C =1, J D =S z ; In step S2, the generation of the daily hourly data packet includes: reorganizing the data at each moment according to the above correlation coefficient to generate a row vector containing 5 elements, the format of which is: [J A J B J C J D Q t ]; Then, the row vectors containing 5 elements generated by the above reorganization at each moment are arranged in time sequence and packaged to generate daily hourly data packets. The data at the vacant moments due to cleaning remain vacant. The daily hourly data packets are the data matrix U i×5 , where i is the number of sampling moments per day.

5. A real-time prediction method for heating heat load based on building usage behavior identification as claimed in claim 4, Features: In step S3, the weekday data set, weekend data set, and holiday data set are all three-dimensional matrices, denoted as W i×5×r , L i×5×j and M i×5×p , where r is the number of working days stored in the data set, j is the number of weekend days stored in the data set, and p is the number of holidays stored in the data set.

6. A method for real-time prediction of heating load based on building usage behavior identification as claimed in claim 5, Features: In step S4, the matrix WX of the working day training set at the i-th moment is generated as follows: The method for generating the weekend training set matrix LX at the i-th moment is: The generation method of the "holiday training set" matrix MX at the i-th moment is:

7. A method for real-time prediction of heating load based on building usage behavior identification as claimed in claim 6, Features: In step S5, obtaining the load forecast coefficient vector at the i-th moment on the k+1-th day includes: On the K+1 day, according to the building usage scenario, the multivariate linear regression method is used to obtain the working day training set matrix WX generated at the i-th moment in the k+1 selection step S4 n×5 , weekend training set matrix LX n×5 Or holiday training set matrix MX n×5 , and then construct the matrix equations according to the following formula (14): Then, solve the equation, and since n>4, obtain the building usage behavior correlation coefficient matrix [ABCD]′ as the least squares solution of the equation; Q t =A·J A +B·J B +C·J C +D·J D (14) Among them, A, B, C and D are the correlation coefficients of building use behavior at a certain moment during the heating period, A = k·F / 1000, B = m, C = -Q b , D = -α / 1000, k is the average heat transfer coefficient of the building envelope; F is the surface area of ​​the building envelope; m is the amount of cold air infiltrated into the building under the combined action of wind pressure and thermal pressure at unit outdoor wind speed; Q b is the heat load of indoor equipment in the building; α is the heat load of solar radiation in the building at a certain moment Q 4 Total solar radiation S z ratio.

8. A method for real-time prediction of heating load based on building usage behavior identification as claimed in claim 7, Features: In step S5, the calculation of the heat load forecast value at the i-th moment on the k+1th day includes: obtaining the building usage behavior correlation coefficients A, B, C, D at the i-th moment and the heating system operation condition correlation coefficient J at the i-th moment on the k+1th day A·T , J B·T , J C , J D Substituting the real-time test data into the following formula ⒂, we can get the predicted value Q of the building heating load at the i-th moment on the K+1 day: y : Q y =A·J A·T +B·J B·T +C·J C +D·J D (15) Among them, J A·T =T n·s -T w , J B·T =(H n·s -H w )·V f 1.34 , T n·s is the design value of indoor air temperature of building heat users (unit: °C); H n·s Design value of indoor air enthalpy for building heat users.

9. A method for real-time prediction of heating load based on building usage behavior identification as claimed in claim 1, Features: The method also includes restarting steps S1 to S5 after step S5 is completed to obtain hourly forecast values ​​of the building heating heat load on and after day K+2.

Citation Information

Cited By

  • Intelligent heating ventilation air conditioner prediction control system and method

    CN120740164A