KR20150041906A - Power load forecasting method using a hybrid dynamic and fuzzy time series model - Google Patents

Abstract

The present invention relates to a method for predicting electric power demand, comprising the steps of: calculating predicted electric power demand through a dynamic model applied with the autoregressive effect and temperature effect; and calculating the residual of the predicted electric power demand through the fuzzy time series model. Accordingly, electric power demand can be predicted using a hybrid model mixed with the dynamic model and fuzzy time series model.

Description

{Power load forecasting method using a hybrid model and a fuzzy time series model}

The present invention relates to a power demand forecasting method, and more particularly, to a power demand forecasting method for predicting power demand using a hybrid model in which a dynamics model and a fuzzy time series model are mixed.

Stable power supply is a must in modern society, which must be fundamentally guaranteed for the nation's industrial, commercial, transportation and daily life. Korea's electricity consumption per GDP is 1.7 times that of the Organization for Economic Cooperation and Development (OECD), and electricity consumption per manufacturing value added is more than twice that of Japan and Germany. In addition, due to the shortage of power reserve due to the shortage of power plant facilities, it is very vulnerable to instability of power supply. This unstable supply-and-demand situation caused power blackouts that occurred on September 15, 2011, and left many homeworks on power management and demand forecasting in Korea. Therefore, more accurate forecasting of power demand is crucial to a more stable and effective balance of power supply and demand. In Korea, the Korea Power Exchange and KEPCO have short - term and medium - term forecasting models that take into account the meteorological factors on power demand, but the correlation and characteristics analysis between meteorological factors are still insufficient.

Prior art related to the prediction of electric power demand is disclosed in Korean Patent Laid-Open Publication No. 10-2009-0073937 (published on July 3, 2009), "Short-term electric power load prediction method using genetic algorithm and fuzzy system".

The first problem to be solved by the present invention is to provide a power demand forecasting method for predicting electric power demand using a hybrid model in which a dynamic model and a fuzzy time series model are mixed.

A second problem to be solved by the present invention is to provide a power demand forecasting apparatus for predicting power demand using a hybrid model in which a dynamics model and a fuzzy time series model are mixed.

According to another aspect of the present invention, there is provided a method for estimating a power demand using a dynamic model using an autocorrelation effect and a temperature effect, And computing a predicted power demand residual through a fuzzy time series model.

According to an embodiment of the present invention, the step of calculating the power prediction demand includes using a dynamic model implemented in the form of a quadratic function as an independent variable indicating a temperature at a previous time point, And a power demand prediction method.

According to an embodiment of the present invention, the step of calculating the residual includes: fuzzing residual data using existing residual data; Deriving a fuzzy logic relationship of the fuzzy data; And calculating the residual in accordance with the fuzzy logic relationship. The fuzzing of the residual data may comprise: determining a total set; Dividing the entire set by determining a size of the interval; And a degree of belonging, and fuzzing the data for each section.

According to an embodiment of the present invention, the step of calculating the residual according to the fuzzy logic relation may be a method of predicting the power demand using the IF-THEN rule.

In order to achieve the second object of the present invention, there is provided a power supply system comprising: a storage unit for storing power demand data and temperature data; And a processor for calculating a power prediction demand through a dynamics model applying an autocorrelation effect and a temperature effect using the stored power demand data and temperature data and calculating a residual of the predicted power demand through a fuzzy time series model Provides a power demand forecasting device.

According to an embodiment of the present invention, the power demand prediction apparatus may further include a communication unit that transmits the calculated power prediction demand and the residual to the outside.

According to the present invention, the accuracy of power demand prediction can be improved. Further, according to the present invention, the overall standard error can be greatly improved.

1 is a block diagram of a power demand predicting apparatus according to an embodiment of the present invention.
2 is a flowchart of a power demand forecasting method according to an embodiment of the present invention.
3 to 4 are flowcharts of a power demand forecasting method according to another embodiment of the present invention.
5 is a graph showing power demand in Seoul from 2000 to 2011. FIG.
Figure 6 shows the coincidence index of the power demand forecasting method and other methods according to an embodiment of the present invention.
7 is a graph illustrating an average prediction error rate of the power demand prediction method and other methods according to an embodiment of the present invention.
FIG. 8 shows prediction errors of the power demand prediction method and other methods according to the embodiment of the present invention.

Prior to the description of the concrete contents of the present invention, for the sake of understanding, the outline of the solution of the problem to be solved by the present invention or the core of the technical idea is first given.

The power demand forecasting method according to an embodiment of the present invention includes the steps of calculating a power prediction demand through a dynamics model applying an autocorrelation effect and a temperature effect, and calculating the predicted power demand residual through a fuzzy time series model .

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. It will be apparent to those skilled in the art, however, that these examples are provided to further illustrate the present invention, and the scope of the present invention is not limited thereto.

BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings, in which: It is to be noted that components are denoted by the same reference numerals even though they are shown in different drawings, and components of different drawings can be cited when necessary in describing the drawings. In the following detailed description of the principles of operation of the preferred embodiments of the present invention, it is to be understood that the present invention is not limited to the details of the known functions and configurations, and other matters may be unnecessarily obscured, A detailed description thereof will be omitted.

1 is a block diagram of a power demand predicting apparatus according to an embodiment of the present invention.

The power demand predicting apparatus 100 according to an embodiment of the present invention includes a storage unit 110 and a processing unit 120. Further, it may further include a communication unit. The power demand predicting apparatus 100 according to an embodiment of the present invention can be used for short-term power demand prediction, such as day, and may be used for mid-term or device power demand prediction.

The storage unit 110 stores power demand data and temperature data.

More specifically, the power demand data and the temperature data are stored in order to predict the power demand using the power demand data and the temperature data. The power demand data and the temperature data may be stored separately by region or time, and may further store the power demand prediction data. And may store the residual, which is a difference between the power demand data and the power demand forecast data.

The processing unit 120 calculates a power prediction demand through a dynamics model to which an autocorrelation effect and a temperature effect are applied using the stored power demand data and temperature data, and calculates a residual of the predicted power demand through a fuzzy time series model .

More specifically, a hybrid model that combines a dynamic model and a fuzzy time series model is used for accurate power demand prediction. The dynamic model may be an autoregressive integrated moving average (ARIMA). The autocorrelation moving average model is a method to predict the future change pattern based on the univariate historical data. The biggest advantage is that it is possible to find a process that can describe the change of the data well. The autocorrelation moving average model includes an autocorrelation calculation, a moving average calculation, and a difference calculation, and can be expressed by the following equation (1).

Where Y _t is time series data, B and ε _t are backshift operators and white noise, respectively. The autocorrelation moving average model can use the method presented by Box and Jenkins (1970).

Unlike other time series models, the fuzzy time series model does not require conditions for checking and linearity of mathematical assumptions and can be applied to a small number of data. The fuzzy time series model can use the method presented by Song and Chissom (1993).

The hybrid model that combines the dynamics model and the fuzzy time series model can model the dynamics model to take into account the seasonal and biennial changes of the time series for the power demand forecast and simultaneously explain the residual using the fuzzy time series model. The hybrid model can be expressed by the following equation (2).

Here, Y _{1 , t} and Y _{2 , t} are time series data predicted by the dynamics model and the fuzzy time series model, respectively, and Δ and f (T _a ) are the autocorrelation effect and the temperature effect. To fuzzy time series data, we can use the interval average method.

Factors affecting power demand are affected by factors related to the electricity market, social and economic factors, and meteorological factors. If these factors are included as independent variables, it is expected that more efficient and accurate power demand forecasting will be possible. However, when performing prediction including all the variables, noise is added to the prediction model due to overlapping information or correlation between the respective variables And there is a disadvantage that a complicated model is formed. By using the hybrid model, more efficient prediction is possible by using a minimum number of variables. The hybrid model will be described in detail as follows.

First, the power demand forecast is calculated through the dynamics model applying the autocorrelation effect and the temperature effect. In other words, forecasts are made using a model that considers power demand and meteorological factors over time. Generally, the determinants of electricity demand are economic factors such as GDP, and non-economic variables such as weather factors. It is necessary to introduce the concept of time into the typical demand model and to incorporate the weather element into the model. For this purpose, the processor 120 may calculate the power prediction demand using a dynamic model implemented in the form of a quadratic function having an independent variable representing the temperature of the previous point of time. The dynamic model can be expressed by the following equation (3).

Where s is a self-dependent parallax and T _{t - 1} is an independent variable representing the temperature of the exhibition point. Basically, considering the dependence of the second - order function on the temperature of the power demand, which is well - known, the temperature variable is transformed into the parabolic form to include the squared variable. The model is an autoregressive distributed lag model ARDL (p, q), depending on the number of parallax variables affecting y _t .

After calculating the power prediction demand, the predicted power demand residual is calculated through a fuzzy time series model. The processing unit 120 fuses the residual data using the existing residual data, derives the fuzzy logic relation of the fuzzy data, and calculates the residual according to the fuzzy logic relation. Before describing the fuzzy time series, we define a few.

First, U is a premise set defined by U = {u1, u2, ..., u _n }, where u _i is defined as possible values of U. In addition, the fuzzy set of the fuzzy variable Ai of U is defined by the following equation (4).

Where f _Ai is the membership function of the fuzzy set A _i , such as f _Ai : U → [0,1], and f _Ai (u _j ) is the degree of membership of u _j in A _i .

Next, a fuzzy time series {F (t): t = 0, 1, ...} is defined as f _Ai (t).

을 만족하면, F(t-1) → F(t)로 정의한다. 여기서,

는 최대 및 최소 합성 연산자(composition operator)이고, R은 퍼지 시계열간의 퍼지 관계를 나타낸다. 나아가,

인 경우, λ차 퍼지 관계가 있다고 정의하고, A_{i λ}는 F(t-λ)로 정의한다.Finally, the fuzzy relation R (t, t-1)

Is defined as F (t-1)? F (t). here,

Is a maximum and minimum composition operator, and R represents a fuzzy relation between fuzzy time series. Furthermore,

, It is defined that there is a? -Order fuzzy relation, and A _{i ?} Is defined as F (t-?).

Using the above definitions, the process of calculating the residual using the fuzzy time series will be described in detail.

The residual can be expressed by the following equation (5).

h _t is the residual obtained from the dynamics model, and the following procedure can be performed to calculate the residual using the fuzzy time series model.

로 결정할 수 있다. 여기서, D_min과 D_max는 h_t의 최솟값과 최댓값이고, c₁, c₂는 전체집합의 양 끝점을 단순화시키는 역할을 하는 양수인 실수이다.Step 1: Determine the overall set.

. Where D _min and D _max are the minimum and maximum values of h _t , and c ₁ and c ₂ are real positive integers that simplify both endpoints of the whole set.

퍼지시계열의 예측은 폐구간의 길이와 수에 따라 달라지는바, 구간의 크기를 결정한다. 또한, 동일한 구간 길이로 전체집합을 분할한다.Step 2: Determine the size of the interval and divide the whole set.

The prediction of the fuzzy time series depends on the length and number of the closed interval, and determines the size of the interval. Also, the entire set is divided by the same section length.

Step 3: Define membership level and fuzzy data. In order to establish the fuzzy time series A _i , the degree of affiliation at the i th time interval (u i) is defined as the following Equation (6).

The fuzzy time series Ai is as shown in Equation (4), and has a maximum menbership value in the interval ui.

Step 4: Determine the λ-order of the fuzzy time series considering AFEP (the average forecasting error percentage) and d (the index of agreement). This can be expressed by Equation (7).

D and AFEP can be expressed by the following equation (8).

는 관측된 수요량의 평균이다. d는 시스템상 오차의 랜덤 오차에 대한 상대적인 기여도를 나타내며, 이상적인 모델에서 1의 값을 갖는다.Where N is the total number of data and Oi and Pi are the observed and predicted demand.

Is the average of the observed demand. d represents the relative contribution to the random error of the systematic error and has a value of 1 in the ideal model.

Step 5: Derive fuzzy logic relations. The fuzzy logic relationship is defined as the change in state at time t-1 relative to the state at time t. This can be expressed by the following equation (9).

Here, i means a state at a time point t-1, and j means a state at a time point t.

Step 6: Calculate the residual using the purge IF-THEN rule. The IF-THEN rule can be divided into the following three cases, and each calculated value is expressed by Equation (9).

일 때 현재 상태 Aj가 최대 소속함수 값을 갖기 때문에, 산출되는 값은 u_j의 중간 값인 m_j이다.here,

, The calculated value is m _{j, which} is the median value of u _j , since the current state Aj has the maximum membership function value.

인 경우, 는 각각

에서 최대 소속함수 값을 갖는바, 산출되는 값은

이다.

Quot; Respectively

Has a maximum value of the membership function, and the calculated value is

to be.

인 경우,

이고, 여기서, w_k는 가중인자로서 여기서는 1로 설정될 수 있다.

Quot;

, Where w _k can be set to 1 as a weighting factor.

And the calculated value is made non-fuzzy, whereby the residual can be finally calculated.

The power demand prediction apparatus according to another embodiment of the present invention may further include a communication unit for transmitting the calculated power prediction demand and the residual to the outside.

More specifically, the communication unit may transmit data from the outside and store it in the storage unit 110 or may be used to predict power demand in the processing unit 120, and may transmit the power prediction demand and the residual calculated by the processing unit 120 to the outside Can be transmitted.

2 is a flowchart of a power demand forecasting method according to an embodiment of the present invention.

Step 210 is a step of calculating a power prediction demand through a dynamic model to which an autocorrelation effect and a temperature effect are applied.

More specifically, the predicted power demand is calculated using a dynamic model implemented in the form of a quadratic function as an independent variable representing the temperature of the previous point of time. The dynamic model may be an autocorrelation moving average model. The detailed description of this step corresponds to the detailed description of the process of calculating the power prediction demand by the processing unit 120 of FIG. 1, and is replaced with a detailed description of the processing unit 120 of FIG.

Step 220 is a step of calculating a residual of the predicted power demand through a fuzzy time series model.

More specifically, the predicted power demand is calculated through the dynamics model in step 210, and the predicted power demand is calculated through the fuzzy time series model. Using the hybrid model of the dynamics model and the fuzzy time series model, the accuracy of power demand prediction can be improved. The detailed description of this step corresponds to the detailed description of the process of calculating the residual of the power prediction demand by the processing unit 120 of FIG. 1, and is replaced with a detailed description of the processing unit 120 of FIG.

3 to 4 are flowcharts of a power demand forecasting method according to another embodiment of the present invention.

Step 310 is a step of fuzzing residual data using existing residual data.

More specifically, in order to calculate the residual, the data must first be fuzzy. For this, steps 410 to 430 may be performed. That is, in order to fuzzy residual data, a step of determining an entire set, a step of dividing the entire set by determining the size of a section, a degree of belonging, and fuzzing the data for each section are performed . The detailed description of this step corresponds to the detailed description of the process of calculating the residual of the power prediction demand by the processing unit 120 of FIG. 1, and is replaced with a detailed description of the processing unit 120 of FIG.

Step 320 is a step of deriving a fuzzy logic relation of the fuzzy data.

More specifically, it derives a fuzzy logic relationship from the fuzzy data. The fuzzy logic relation is the basis for calculating the residual. The detailed description of this step corresponds to the detailed description of the process of calculating the residual of the power prediction demand by the processing unit 120 of FIG. 1, and is replaced with a detailed description of the processing unit 120 of FIG.

Step 330 is a step of calculating the residual according to the fuzzy logic relation.

More specifically, the residual is calculated by using the fuzzy logic relation derived in step 320. [ The residual can be calculated using the IF-THEN rule. The detailed description of this step corresponds to the detailed description of the process of calculating the residual of the power prediction demand by the processing unit 120 of FIG. 1, and is replaced with a detailed description of the processing unit 120 of FIG.

FIG. 5 is a graph showing power demand in Seoul from 2000 to 2011 for households, public sector, service sector, and industrial sector.

Electricity demand is steadily increasing, except in the industrial sector. Peak values occur in the same interval of each year, which shows strong autocorrelation of the data. The annual cycle has a bimodal distribution over summer and winter, which shows that there is a significant correlation between demand and temperature. Indeed, it can be seen that temperature changes can account for 84.3%, 83.2%, 86.1%, and 72.7% of demand changes in the home, public sector, service sector, and industry, respectively.

The following Table 1 shows coefficients applied to the dynamics model of the hybrid model (Equation 3) and can be calculated by the least squares method. The annual change in demand is the same as in FIG. 5, and the autocorrelation time difference is calculated as 12 months. The linear model used in the hybrid model is statistically appropriate.

Observed temperature and actual demand data from January 2000 to October 2010 were used to evaluate the sheck model, the ARIMA model and the hybrid model according to the embodiment of the present invention.

Table 2 above shows that it is difficult to explain the residuals only by the dynamic model. In order to establish a fuzzy time series model, the whole set is determined by the above residuals, and the data are divided into the categories of 10000, 5000, 40000, 7500 MWh, 24, 10, 13, And industrial sectors.

As shown in FIG. 6, the koyck model considers only temperature and time differences. The coincidence index is smaller than 0.7, which is worse than the ARIMA and hybrid models. In other words, it can be seen that considering only the temperature and time difference is insufficient for demand forecasting. Humidity, and wind, as well as government regulations. The ARIMA model and the hybrid model show good performance as shown in Table 3 below.

The hybrid model shows almost ideal results in the home, public sector, and service sectors, as shown in FIG. The hybrid model also shows an AFEP of 5% in the industrial sector and 1% in other sectors, as shown in FIG. Compared with the ARIMA model, it can be seen that it is suitable to predict the observed peak value of demand.

FIG. 8 shows prediction errors of the power demand prediction method and other methods according to the embodiment of the present invention.

As shown in FIG. 8, it can be seen that the hybrid model, which is a power demand prediction method according to the embodiment of the present invention, has an error rate close to zero as compared with the other models. The coincidence index d is close to 1, and the average prediction error rate is lower than 1%.

Embodiments of the present invention may be implemented in the form of program instructions that can be executed on various computer means and recorded on a computer readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and specific embodiments and drawings. However, it should be understood that the present invention is not limited to the above- And various modifications and changes may be made thereto by those skilled in the art to which the present invention pertains.

Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

100: Power demand forecasting device
110:
120:

Claims (11)

Calculating a power prediction demand through a dynamic model applying an autocorrelation effect and a temperature effect; And
And calculating the predicted power demand residual through a fuzzy time series model. The method according to claim 1,
The step of calculating the power prediction demand includes:
A power demand prediction method using a dynamics model implemented in the form of a quadratic function as an independent variable representing the temperature at a previous point of time. The method according to claim 1,
Wherein the dynamic model is an autocorrelated moving average model. The method according to claim 1,
Wherein the step of calculating the residual comprises:
Fuzzing residual data using existing residual data;
Deriving a fuzzy logic relationship of the fuzzy data; And
And calculating the residual according to the fuzzy logic relationship. 5. The method of claim 4,
Wherein the step of fuzzing the residual data comprises:
Determining an overall set;
Dividing the entire set by determining a size of the interval; And
And defining a degree of affiliation and fuzzing the data for each section. 5. The method of claim 4,
Wherein the step of calculating the residual according to the fuzzy logic relationship comprises:
And the residual is calculated using an IF-THEN rule. A computer-readable recording medium storing a program for causing a computer to execute the method according to any one of claims 1 to 6. A storage unit for storing power demand data and temperature data; And
And a processor for calculating a power prediction demand through a dynamics model applying an autocorrelation effect and a temperature effect using the stored power demand data and temperature data and calculating a residual of the predicted power demand through a fuzzy time series model, Demand forecasting device. 9. The method of claim 8,
Wherein,
Wherein the power demand forecasting unit calculates the power demand by using a dynamic model implemented in the form of a quadratic function having an independent variable representing the temperature of the previous time point. 9. The method of claim 8,
Wherein,
Wherein the residual data is fuzzy using the existing residual data, the fuzzy logic relation of the fuzzy data is derived, and the residual is calculated according to the fuzzy logic relation. 9. The method of claim 8,
And a communication unit for transmitting the calculated power prediction demand and the residual to the outside.

Images