Heating load forecasting method

Abstract

The present invention discloses a heating load prediction method, comprising obtaining a historical data set that affects the heating load and processing it to obtain a training data set; constructing a heating load prediction model based on a generalized additive model; using the training data set to train the constructed heating load prediction model and obtain a temperature coefficient; and calculating a final heating load prediction value according to the temperature coefficient. The heating load prediction method provided by the present invention not only realizes the prediction of the heating load by innovatively establishing a prediction model and prediction means, but also has high reliability, high accuracy and a wide range of applications.

Description

Technical Field

[0001] The invention belongs to the field of electrical automation, and in particular relates to a heating load prediction method. Background Art

[0002] With the development of economy and technology and the improvement of people's living standards, electric energy has been widely used in people's production and life, bringing endless convenience to people's production and life. Therefore, ensuring the stable and reliable supply of electric energy has become one of the most important tasks of the power system.

[0003] Power load forecasting is one of the important tasks of the power system. Through accurate power load forecasting, the power system can accurately formulate dispatching plans and operation plans, thereby ensuring the stable, safe and efficient operation of the power system.

[0004] As people pursue more comfort, the proportion of heating load in the power system load is increasing. Therefore, the prediction of heating load is particularly important. At present, the existing heating load prediction method mainly adopts the regression method. When using the regression method to calculate the heating load, the generalized linear model or the Lagrange nonlinear regression model is mainly used to establish a strict mathematical model between load and temperature, and the heating load is predicted through the established mathematical model. However, the existing regression method has problems in dealing with the influencing factors of the heating load, which makes the existing heating load prediction method based on the regression method not accurate and has poor reliability. Summary of the invention

[0005] The object of the present invention is to provide a heating load prediction method with high reliability, high accuracy and wide application range.

[0006] The heating load prediction method provided by the present invention comprises the following steps:

[0007] S1. Obtain historical data sets that affect heating load;

[0008] S2. Process the historical data set obtained in step S1 to obtain a training data set;

[0009] S3. Based on the generalized additive model, a heating load prediction model is constructed;

[0010] S4. Using the training data set obtained in step S2, the heating load prediction model constructed in step S3 is trained to obtain the temperature coefficient;

[0011] S5. Calculate the final heating load forecast value based on the temperature coefficient obtained in step S4.

[0012] The historical data set affecting the heating load described in step S1 specifically obtains a historical meteorological data set affecting the heating load.

[0013] Step S2 processes the historical data set obtained in step S1 to obtain a training data set, which specifically includes the following steps:

[0014] A. Determine outliers on the acquired historical data set and replace outliers:

[0015] If the abnormal value is a null value, the vertical and horizontal cross method is used to fill it: the average value of the values ​​at the same time of the two days before and after is used to replace the abnormal value; if the value at the same time of the two days before and after is still a null value, the average value of the values ​​at the two times before and after on the same day is used to replace the abnormal value;

[0016] If the outlier value is not empty and is a single outlier point, the outlier is replaced by the average of the two values ​​adjacent to the outlier value; the rule for determining a single outlier point is: calculate the relative change rate between the decision point and the values ​​of the two adjacent moments, if the relative change rate between the decision point and the values ​​of the two adjacent moments is greater than the set value, then the decision point is identified as a single outlier point;

[0017] If the outlier value is not empty and is a continuous outlier point, the vertical and horizontal cross method is used to fill it: the average value of the values ​​at the same time of the two days before and after is used to replace the outlier value; the rule for determining the continuous outlier point is: if the number of continuous and unchanged determination points is greater than the set value, the continuous and unchanged determination points are considered to be continuous and unchanged outliers;

[0018] B. Construct a training data set according to the following data format:

[0019] The data format is [DATE,W,HD,POS,L i ,L i-24 ,T i ,T i-24 ,Trend], where DATE is the Gregorian calendar date; W is the day of the week, with 1 for Monday and 7 for Sunday; HD is the indicator of whether it is a heating load day, and HD = 1 indicates a heating load day, and HD = 0 indicates a non-heating load day; POS is the annual position information of the day, with a value range of [0,1], and the value corresponding to January 1 is 0, and the value corresponding to December 31 is 1. The dates between January 1 and December 31 correspond linearly to the values ​​in the range of (0,1); L i is the unified load value at time i on the day; L i-24 is the coordinated load value at time i on the previous day; T i is the temperature at time i on the day; T i-24 is the temperature value of the day before the current moment; Trend is the linear growth trend value.

[0020] The step S3 described in which a heating load prediction model is constructed based on a generalized additive model specifically includes the following steps:

[0021] The following formula is used as the heating load prediction model:

[0022] L i =f0(T i )+f1(L i-24 ,T i-24 )+f3(Trend)+f4(W)+f5(T i-24 )

[0023] Where L i is the predicted value of heating load at time i on the day; f0(T i ) is the independent variable T i and heating load forecast value L i Function between; f1(L i-24 ,T i-24 ) is the independent variable L i-24 , T i-24 and heating load forecast value L i f3(Trend) is the function between the independent variable Trend and the heating load forecast value L i f4(W) is the function between the independent variable W and the heating load forecast value L i Function between; f5(T i-24 ) is the independent variable T i-24 and heating load forecast value L i The function between .

[0024] Step S4 uses the training data set obtained in step S2 to train the heating load prediction model constructed in step S3 using a generalized additive model, thereby obtaining the temperature coefficient, which specifically includes the following steps:

[0025] Using the training data set obtained in step S2, the generalized additive model is applied to solve the heating load prediction model constructed in step S3 to obtain the first function f0(T i ) and the second function f5(T i-24 ), and based on the first function and the second function, obtain the temperature sensitivity coefficient k1 and the temperature conduction coefficient k2;

[0026] The temperature sensitivity coefficient k1 is used to describe the increase in heating load at the current moment when the temperature index drops by 1°C. When calculating, the relationship between the current temperature and the heating load is decoupled to obtain the first function f0(T i ) function curve, and then f0(T i) The function curve is divided into n segments, and the slope k of each segment is calculated 1i , i=1,2,...,n, and finally the temperature sensitivity coefficient k1 is

[0027] The temperature conduction coefficient k2 is used to describe the increase in heating load at the current moment today when the temperature index at the current moment yesterday drops by 1°C. When calculating, the relationship between yesterday's temperature index and today's heating load is decoupled to obtain the second function f5(T i-24 ) function curve, and then f5(T i-24 ) The function curve is divided into n segments, and the slope k of each segment is calculated 2i , i=1,2,...,n, and finally the temperature sensitivity coefficient k2 is

[0028] The step S5 of calculating the final heating load prediction value according to the temperature coefficient obtained in step S4 specifically includes the following steps:

[0029] The predicted heating load HL at the current time is calculated using the following formula: d,i :

[0030] HL d,i =I(T d,i ≤15)[k1(T d,i -15)+k2(T d-1,i -15)]

[0031] Where I(T d,i ≤15) is the indicator function, when T d,i ≤15 when I(T d,i ≤15)=1, when T d,i ≤15 is not true when I(T d,i ≤15)=0;T d,i is the temperature data at the current moment; T d-1,i The temperature data for the current time yesterday.

[0032] The heating load prediction method provided by the present invention not only realizes the prediction of heating load by innovatively establishing a prediction model and prediction means, but also has high reliability, high accuracy and wide application range. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The figure is a schematic diagram of the method flow of the present invention.

[0034] Figure 2 This is a graph of hourly temperature coefficients for a province in an embodiment of the method of the present invention.

[0035] Figure 3This is the calculation result of the heating load in a certain province in 2021 according to an embodiment of the method of the present invention. DETAILED DESCRIPTION

[0036] like Figure 1 The method flow chart of the method of the present invention is shown as follows: The heating load prediction method provided by the present invention comprises the following steps:

[0037] S1. Obtain a historical data set that affects heating load; specifically, obtain a historical meteorological data set that affects heating load;

[0038] S2. Process the historical data set obtained in step S1 to obtain a training data set; specifically, the steps include:

[0039] A. Determine outliers on the acquired historical data set and replace outliers:

[0040] If the abnormal value is a null value, the vertical and horizontal cross method is used to fill it: the average value of the values ​​at the same time of the two days before and after is used to replace the abnormal value; if the value at the same time of the two days before and after is still a null value, the average value of the values ​​at the two times before and after on the same day is used to replace the abnormal value;

[0041] If the outlier value is not empty and is a single outlier point, the outlier value is replaced by the average value of the two adjacent values ​​of the outlier value; the rule for determining a single outlier point is: calculate the relative change rate between the determination point and the values ​​of the two adjacent moments, if the relative change rate between the determination point and the values ​​of the two adjacent moments is greater than the set value (for example, 50%), then the determination point is identified as a single outlier point;

[0042] If the outlier value is not empty and is a continuous outlier point, the vertical and horizontal cross method is used to fill it: the average value of the values ​​at the same time of the two days before and after is used to replace the outlier value; the rule for determining the continuous outlier point is: if the number of continuous and unchanged determination points is greater than the set value (for example, 6 consecutive points), the continuous and unchanged determination points are considered to be continuous and unchanged outliers;

[0043] B. Construct a training data set according to the following data format:

[0044] The data format is [DATE,W,HD,POS,L i ,L i-24 ,T i ,T i-24,Trend], where DATE is the Gregorian calendar date; W is the day of the week, with 1 for Monday and 7 for Sunday; HD is the indicator of whether it is a heating load day, and HD = 1 indicates a heating load day, and HD = 0 indicates a non-heating load day; POS is the annual position information of the day, with a value range of [0,1], and the value corresponding to January 1 is 0, and the value corresponding to December 31 is 1. The dates between January 1 and December 31 correspond linearly to the values ​​in the range of (0,1); L i is the unified load value at time i on the day; L i-24 is the coordinated load value at time i on the previous day; T i is the temperature at time i on the day; T i-24 is the temperature value of the day before the current moment; Trend is the linear growth trend value;

[0045] S3. Based on the generalized additive models (GAMs), a load forecasting model is constructed; specifically, the steps include:

[0046] The following formula is used as the load forecasting model:

[0047] L i =f0(T i )+f1(L i-24 ,T i-24 )+f3(Trend)+f4(W)+f5(T i-24 )

[0048] Where L i is the predicted value of heating load at time i on the day; f0(T i ) is the independent variable T i and heating load forecast value L i Function between; f1(L i-24 ,T i-24 ) is the independent variable L i-24 , T i-24 and heating load forecast value L i f3(Trend) is the function between the independent variable Trend and the heating load forecast value L i f4(W) is the function between the independent variable W and the heating load forecast value L i Function between; f5(T i-24 ) is the independent variable T i-24 and heating load forecast value L i Functions between

[0049] S4. Using the training data set obtained in step S2, the heating load prediction model constructed in step S3 is trained to obtain the temperature coefficient; specifically comprising the following steps:

[0050] Using the training data set obtained in step S2, the generalized additive model is applied to solve the heating load prediction model constructed in step S3 to obtain the first function f0(T i ) and the second function f5(T i-24 ), and based on the first function and the second function, obtain the temperature sensitivity coefficient k1 and the temperature conduction coefficient k2;

[0051] The temperature sensitivity coefficient k1 is used to describe the increase in heating load at the current moment when the temperature index drops by 1°C. When calculating, the relationship between the current temperature and the heating load is decoupled to obtain the first function f0(T i ) function curve, and then f0(T i ) The function curve is divided into n segments, and the slope k of each segment is calculated 1i , i=1,2,...,n, and finally the temperature sensitivity coefficient k1 is

[0052] The temperature conduction coefficient k2 is used to describe the increase in heating load at the current moment today when the temperature index at the current moment yesterday drops by 1°C. When calculating, the relationship between yesterday's temperature index and today's heating load is decoupled to obtain the second function f5(T i-24 ) function curve, and then f5(T i-24 ) The function curve is divided into n segments, and the slope k of each segment is calculated 2i , i=1,2,...,n, and finally the temperature sensitivity coefficient k2 is

[0053] The hourly temperature sensitivity coefficient k1 and temperature conduction coefficient k2 of a province in this embodiment obtained by fitting the generalized additive model are as follows: Figure 2 As shown;

[0054] S5. Calculate the final heating load prediction value according to the temperature coefficient obtained in step S4; specifically comprising the following steps:

[0055] The predicted heating load HL at the current time is calculated using the following formula: d,i :

[0056] HL d,i =I(T d,i ≤15)[k1(T d,i -15)+k2(T d-1,i -15)]

[0057] Where I(T d,i ≤15) is the indicator function, when T d,i ≤15 when I(T d,i≤15)=1, when T d,i ≤15 is not true when I(T d,i ≤15)=0;T d,i is the temperature data at the current moment; T d-1,i The temperature data for the current moment yesterday;

[0058] Based on the above heating load calculation method, the heating load is calculated; the monthly heating load forecast results of a certain province from January to March and November to December 2021 obtained in this embodiment are as follows: Figure 3 shown.

Claims

1. A heating load prediction method, comprising the following steps: S1. Obtain historical data sets that affect heating load; S2. Process the historical data set obtained in step S1 to obtain a training data set; specifically, the steps include: A. Determine outliers on the acquired historical data set and replace outliers: If the abnormal value is a null value, the vertical and horizontal cross method is used to fill it: the average value of the values ​​at the same time of the two days before and after is used to replace the abnormal value; if the value at the same time of the two days before and after is still a null value, the average value of the values ​​at the two times before and after on the same day is used to replace the abnormal value; If the outlier value is not empty and is a single outlier point, the outlier is replaced by the average of the two values ​​adjacent to the outlier value; the rule for determining a single outlier point is: calculate the relative change rate between the decision point and the values ​​of the two adjacent moments, if the relative change rate between the decision point and the values ​​of the two adjacent moments is greater than the set value, then the decision point is identified as a single outlier point; If the outlier value is not empty and is a continuous outlier point, the vertical and horizontal cross method is used to fill it: the average value of the values ​​at the same time of the two days before and after is used to replace the outlier value; the rule for determining the continuous outlier point is: if the number of continuous and unchanged determination points is greater than the set value, the continuous and unchanged determination points are considered to be continuous and unchanged outliers; B. Construct a training data set according to the following data format: The data format is [DATE,W,HD,POS,L i ,L i-24 ,T i ,T i-24 ,Trend], where DATE is the Gregorian calendar date; W is the day of the week, with 1 for Monday and 7 for Sunday; HD is the indicator of whether it is a heating load day, and HD = 1 indicates a heating load day, and HD = 0 indicates a non-heating load day; POS is the annual position information of the day, with a value range of [0,1], and the value corresponding to January 1 is 0, and the value corresponding to December 31 is 1. The dates between January 1 and December 31 correspond linearly to the values ​​in the range of (0,1); L i is the unified load value at time i on the day; L i-24 is the coordinated load value at time i on the previous day; T i is the temperature at time i on the day; T i-24 is the temperature value of the day before the current moment; Trend is the linear growth trend value; S3. Based on the generalized additive model, a heating load prediction model is constructed; specifically, the steps include: The following formula is used as the heating load prediction model: L i =f 0 (T i )+f 1 (L i-24 ,T i-24 )+f 3 (Trend)+f 4 (W)+f 5 (T i-24 ) Where L i is the predicted value of heating load at time i on the day; f 0 (T i ) is the independent variable T i and heating load forecast value L i The function between 1 (L i-24 ,T i-24 ) is the independent variable L i-24 , T i-24 and heating load forecast value L i The function between 3 (Trend) is the independent variable Trend and the heating load forecast value L i The function between 4 (W) is the independent variable W and the heating load forecast value L i The function between 5 (T i-24 ) is the independent variable T i-24 and heating load forecast value L i Functions between S4. Using the training data set obtained in step S2, the heating load prediction model constructed in step S3 is trained to obtain the temperature coefficient; S5. Calculate the final heating load forecast value based on the temperature coefficient obtained in step S4.

2. The heating load prediction method according to claim 1, Features The historical data set affecting the heating load described in step S1 specifically obtains a historical meteorological data set affecting the heating load.

3. The heating load prediction method according to claim 2, Features Step S4 uses the training data set obtained in step S2 to train the heating load prediction model constructed in step S3, thereby obtaining the temperature coefficient, which specifically includes the following steps: Using the training data set obtained in step S2, the generalized additive model is applied to solve the heating load prediction model constructed in step S3 to obtain the first function f that reflects the relationship between temperature and heating load. 0 (T i ) and the second function f 5 (T i-24 ), and based on the first function and the second function, the temperature sensitivity coefficient k is obtained. 1 and air temperature conductivity k 2 ; Temperature sensitivity coefficient k 1 , which is used to describe the increase in heating load at the current moment when the temperature index drops by 1°C; when calculating, the relationship between the current temperature and the heating load is decoupled to obtain the first function f 0 (T i ) function curve, and then f 0 (T i ) The function curve is divided into n segments, and the slope k of each segment is calculated 1i , i=1,2,...,n, and finally we get the temperature sensitivity coefficient k 1 for Air temperature conductivity k 2 , which is used to describe the increase in heating load at today's current moment when the temperature index at yesterday's current moment drops by 1°C; when calculating, the relationship between yesterday's temperature index and today's heating load is decoupled to obtain the second function f 5 (T i-24 ) function curve, and then f 5 (T i-24 ) The function curve is divided into n segments, and the slope k of each segment is calculated 2i , i=1,2,...,n, and finally we get the temperature sensitivity coefficient k 2 for 4. The heating load prediction method according to claim 3, Features The step S5 of calculating the final heating load prediction value according to the temperature coefficient obtained in step S4 specifically includes the following steps: The predicted heating load HL at the current time is calculated using the following formula: d,i : HL d,i =I(T d,i ≤15)[k 1 (T d,i -15)+k 2 (T d-1,i -15)] Where I(T d,i ≤15) is the indicator function, when T d,i ≤15 when I(T d,i ≤15)=1, when T d,i ≤15 is not true when I(T d,i ≤15)=0; T d,i is the temperature data at the current moment; T d-1,i The temperature data for the current time yesterday.

Images