KR100648296B1 - Energy Demand Forecast Method and System - Google Patents

Abstract

The present invention relates to a method and system for predicting energy demand, which includes: a) reviewing trends and seasonality of local maximum power using previous non-synchronous maximum power data, and using simultaneous maximum power data to determine simultaneous maximum power. Reviewing the growth rate and the occupancy rate in the region, and examining whether there is a factor that can cause structural changes in the time series characteristics of the maximum power; b) modeling the volatility of energy demand as the seasons change; c) adapting energy demand based on performance to a cointegrating model with time variance coefficients through appropriate explanatory variables; d) an out-of-sample prediction step to confirm the predictive power in advance; And e) performing prediction using the cointegration model and the error correction model.

Energy demand forecast, cointegration, time error

Description

Energy Demand Forecast Method and System             

1A is a flowchart illustrating a method for predicting energy demand according to an embodiment of the present invention.

1B is a table dividing the whole country into 27 regions according to an embodiment of the present invention and giving a code thereof.

2 is a distribution diagram showing the occupancy ratio of the transformer maximum power in the Gangbuk region of the past Seoul according to an embodiment of the present invention.

FIG. 3 is a distribution diagram showing the quarterly non-simultaneous direct transaction maximum power and the annual non-simultaneous direct transaction maximum power in the Gangbuk area of Seoul according to an embodiment of the present invention.

Figure 4 shows the temperature distribution function of the August 2002 Seoul Gangbuk area according to an embodiment of the present invention.

5 is a result of estimating the temperature response function for the maximum power transformer in the Seoul Gangbuk area according to an embodiment of the present invention.

Figure 6 shows the monthly average temperature effect on the estimated monthly maximum power of the Gangbuk region of Seoul according to an embodiment of the present invention.

7 is a diagram illustrating a process of estimating the actual temperature effect by the estimated temperature distribution function and the temperature response function according to an embodiment of the present invention.

8 is a diagram showing the elasticity trend of the maximum power of the Seoul Gangnam transformer according to an embodiment of the present invention.

9 is a diagram showing the elasticity trend of the maximum power direct trading in Seoul Gangbuk according to an embodiment of the present invention.

10 is a view showing the simultaneous / asynchronous ratio of the Seoul Gangbuk transformer maximum power according to an embodiment of the present invention.

11 is a cointegration model of the Gangbuk area of Seoul according to an embodiment of the present invention.

12 is a multiple regression model of the Seoul Gangbuk region according to an embodiment of the present invention.

FIG. 13 is a cointegration model of a direct trade Seoul Gangbuk region according to an embodiment of the present invention.

14 is a multiple regression model of the direct trading Seoul Gangbuk region according to an embodiment of the present invention.

FIG. 15 is a comparison model of monthly transformer maximum power predictive power in Gangbuk, Seoul, according to an embodiment of the present invention. FIG.

16 is a long-term power sales forecast result according to an embodiment of the present invention.

17 is a long-term maximum power prediction result according to an embodiment of the present invention.

18 is a result of predicting the maximum power of the Seoul Gangbuk asynchronous transformer according to an embodiment of the present invention.

19 is a monthly asynchronous transformer maximum power prediction result according to an embodiment of the present invention.

20 is a result of predicting the maximum power of the weekly transformer in the Gangbuk region of Seoul according to an embodiment of the present invention.

21 is an overall configuration diagram of a demand forecasting system according to an embodiment of the present invention.

FIG. 22 is an internal block diagram of a server for storing various data required for demand forecast and performing a preparation step required for demand forecast according to an embodiment of the present invention.

The present invention relates to a method for estimating energy demand, and more particularly, to a method for estimating a more accurate and reliable estimate and to compensate for the disadvantages of the existing demand system for local demand.

Domestic power demand has been steadily increasing with economic growth, and has provided stable and reliable electric power service even in various harsh environments. The increase in electric power demand has continued until recently, and therefore, the necessity of increasing the capacity of power generation facilities as well as the transmission and transmission facilities for transporting electric power is urgently raised. However, investment in electric power business, such as reinforcement or new system, requires long investment and optimal cost in terms of national economy. In this case, the most essential factor is regional maximum electricity demand. Is a prediction.

However, to date, the forecasting model of regional maximum power demand has a disadvantage that it is difficult to use it in actual operation because the number of forecasting predicates required for forecasting is too large and complex. In addition, using data that is difficult to predict or that takes a long time to be published resulted in less predictive power.

Accordingly, the present invention has focused on establishing a predictive model that can be used more easily by a user under such a background, and which provides more accurate predictions.

Due to the continuing trend of increasing power demand, the expansion of power transmission and transmission facilities is necessary. However, as mentioned above, the expansion of transmission facilities requires a long time and a large investment, so an optimal transmission capacity expansion plan is required.

The most important factor for this is the prediction of peak power demand. By deriving the optimal transmission and substation facilities in the future based on the rigorously calculated maximum power demand forecasts for each region, it is possible to use resources efficiently at the national level as well as the relevant institutions.

However, regional maximum demand forecasts have shorter data accumulation periods than forecast target periods and are difficult to obtain reliable data. Therefore, in the present invention, it is set as an important goal to have a DB system that can accumulate and efficiently manage existing data.

Scientific models are urgently needed to accurately predict long-term power demand over the next 14 years using data accumulated for up to 13 years. The present invention attempts to compensate for the shortcomings resulting from a lack of data using state-of-the-art econometric techniques. The use of these models, which have already been validated in forecasting energy demand for power and gas, has led to much more accurate and reasonable estimates than existing models.

In addition, in the present invention, by designing with a fully automated prediction system, by providing a convenient user environment, it is possible to reduce the errors that can be caused by conventional manual prediction, and to significantly reduce the time and human resources required to make predictions.

An object of the present invention is to provide an energy demand prediction method and system that can compensate for the shortcomings of the local demand prediction system in use as described above and calculate a more accurate and reliable forecast.

In order to achieve the above object, the energy demand prediction method of the present invention a) examines the trend and seasonality of the local maximum power using previous non-synchronous maximum power data, and simultaneously uses the previous simultaneous maximum power data. Examining the rate of increase of maximum power and the occupancy ratio in the region, and examining whether there is a factor that can cause structural changes in the time series of the maximum power; b) modeling the volatility of energy demand as the seasons change; c) adapting energy demand to a cointegrating model with time variance coefficients through appropriate explanatory variables, based on performance values; d) an out-of-sample prediction step to confirm the predictive power in advance; And e) performing prediction using the cointegration model and the error correction model.

In the present invention, it is preferable that the asynchronous and simultaneous maximum power data of step a) include transformer maximum power by region and direct power maximum by region.

In the present invention, the step of modeling the variability of the energy demand according to the seasonal change of step b) comprises the steps of: b1) determining the relative frequency of how often the temperature during a given period of time appeared at each temperature; b2) determining the sensitivity of power demand to each temperature; And b3) modeling the total nonlinear temperature effect.

In the present invention, the step c) preferably includes c1) long-term prediction and c2) short-term prediction model.

In the present invention, c1) the long-term prediction model is c11) predicting asynchronous maximum power per region; c12) incorporating short-term asynchronous peak power prediction and long-term asynchronous peak power prediction; And c13) predicting the simultaneous maximum power for each region.

C2) the short-term prediction model in the present invention c21) predicting the monthly non-synchronous maximum power; And c22) predicting weekly asynchronous maximum power.

In the present invention, the power demand predicted in step e) is preferably a long-term maximum power demand and a short-term maximum power demand amount.

In the present invention, the long-term maximum power demand amount preferably includes a long-term non-simultaneous transformer maximum power amount, a long-term non-simultaneous direct trade maximum power amount, a long-term simultaneous transformer maximum power amount, and a long-term simultaneous direct-selling maximum power amount.

In the present invention, the short-term maximum power demand amount may include a short-term monthly transformer maximum power, a short-term monthly direct sales maximum power, a short-term weekly transformer maximum power amount, and a short-term weekly direct sales maximum power amount.

Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings.

1A is a flowchart illustrating a method for predicting energy demand according to an embodiment of the present invention.

The above embodiment schematically illustrates a method for predicting long term maximum power demand and short term maximum power demand.

In order to be able to link with the actual system plan, the regional classification divides the predicted area into 27 as shown in FIG. 1B, and predicts the non-synchronous transformer and direct power, simultaneous transformer, and direct power for each of the 27 areas. . However, in Jeju, there is no demand for direct sales at present, so the maximum power of direct sales is forecasted for 26 regions excluding Jeju. For the above 27 regions to be forecasted, the present invention provides long-term and short-term forecasts. Long-term and short-term forecasting models differ not only in the duration of the forecast, but also in their targets.

Referring to Figure 1a, the process of predicting the maximum power demand is as follows.

Step 101 is a process of analyzing time series characteristics of power usage using previously consumed maximum power data.

The time series analysis of the maximum power is performed for the transformer maximum power and the direct selling maximum power. The maximum power of the transformer refers to the maximum power of the home, and usually goes through a power transmission transformer before power is transmitted to the home. On the contrary, the direct power maximum power means a maximum power supplied to a large factory or a large customer through a transmission line in which a transmission line exists separately from a transformer.

The maximum power includes simultaneous maximum power and asynchronous maximum power. The simultaneous maximum power refers to the regional electricity demand at the point in time when the nation's electric power demand is the maximum. The asynchronous maximum power refers to the power demand of the region at the time when the power demand for each region is the maximum.

The quarterly asynchronous maximum power data is used to examine the trend and seasonality of the regional maximum power. And, using the previous annual simultaneous maximum power data, the increase rate of the maximum simultaneous power and the occupancy rate in the region are reviewed. In addition, the factors that can cause structural changes in the time series characteristics of the maximum power, such as the construction of a new town, the construction of a substation, etc., are also considered.

The transformer maximum power generally has the following time series characteristics. First, regions that supply large-scale electricity demand, such as metropolitan cities and metropolitan cities, show distinct seasonality, while other regions have recently experienced seasonal changes or quarterly deviations. This is because, in case of metropolitan city, the maximum power is generated by cooling demand in summer, while in other regions, the maximum power is generated in winter due to heating demand. Second, there is an increasing trend that changes with time. Third is the structural change in time series due to the construction of substations. The structural change in the maximum power time series is due to changes in the characteristics of the maximum power due to the influx of new demands such as the construction of new cities or housing complexes. Directly traded maximum power is greatly affected by production activities and economic conditions of local customers, so it is more variable and irregular than transformer maximum power.

2 is a distribution diagram showing the occupancy ratio of the transformer maximum power in the Gangbuk region of the past Seoul according to an embodiment of the present invention.

Referring to FIG. 2, when looking at the quarterly asynchronous maximum power time series of the Gangbuk area of Seoul, first, a distinct seasonality can be found. Asynchronous peak power occurs in the third quarter of the year, and this pattern has been maintained since 1991. In addition, the demand in the second and third quarters is greater than in the first and fourth quarters, indicating that the demand for cooling is greater than the demand for heating. The next trend is an increasing trend. Looking at annual asynchronous maximum power, annual asynchronous maximum power decreased by -0.5% and -0.4% in 1993 and 1995 respectively, and since 2000, the rate of increase is gradually decreasing. Finally, the time series characteristic of Seoul's Gangbuk area is the increase in maximum power due to structural changes. In 1994, the maximum rate of asynchronous power consumption was 24.7%, the highest since 1991. This is due to the fact that the Majang substation was newly established in 1994, and the demand for substations in Mia, Sanggye, Sejong-ro, Sinchon, Wonnam, and Hwikyung was greatly increased.

On the other hand, the annual maximum asynchronous power consumption in Gangbuk, Seoul was 24.8% in 1991, and the share of asynchronous maximum power dropped to 17.9% in 2003. Simultaneous maximum power is also increasing less than the simultaneous maximum power of Gyeongin area.

Collect the information on the maximum power of the past transformer by region for the whole country as described above.

Unlike the maximum power of non-simultaneous direct sales, the maximum power of non-simultaneous direct sales is greatly influenced by the contract power of the customers included in the region, and thus moves according to the local characteristics of the local industry rather than the overall power demand. Thus, long-term time trends cannot be seen to increase as asynchronous transformer maximum power, and seasonality is not apparent. In addition, the peak demand will be discontinuous due to the new or removed consumer.

FIG. 3 is a distribution diagram showing the quarterly non-simultaneous direct transaction maximum power and the annual non-simultaneous direct transaction maximum power in the Gangbuk area of Seoul according to an embodiment of the present invention.

3, it can be seen that the value is very small and the variation is large. This is because there are only two substations introducing direct trade power in the region. In Gangbuk, Seoul, direct sales power is supplied from the Sanggye Substation and the Hwayang Substation. However, due to the self-development of the Nowon cogeneration plant, which is a direct sales customer supplied by the offsetting substation, it often appears that power is not supplied directly, so the maximum power of direct sales in this region is severely fluctuating.

In addition, the region's peak direct trade power in 2002 has dropped significantly compared to other years. This is due to the fact that both the substation and Hwayang substations have undervalued values. As shown in FIG. 3, the local direct sales maximum power time series has no particular trend or seasonality.

Collect and analyze information on the past direct trading power by region in the country as described above.

Step 102 is a process of estimating the regional temperature distribution function.

The temperature distribution function is a function of smoothing the relative frequency of how often the temperature during a given period of time appeared at each temperature. The temperature distribution function is preferably estimated for each region.

As a step for obtaining a regional temperature distribution function using the regional maximum power data obtained above, the present invention estimates using a kernel non-parametric estimation method. Specifically, in the present invention, using the temperature distribution function as the temperature data

는 커널이라고 부르는 함수이며,

는 띠너비(bandwidth)이다. 일반적으로

은

을 만족하는 확률밀도함수이며, 본 발명에서는 정규 분포(normal distribution)의 확률밀도(probability density) 함수를 이용하고 있다.

는 평활화 모수(smoothing parameter)라고 부르며, 본 발명에서는 실버만(Silverman)이 제안한 값을 사용하여 계산한다.Estimate as here

Is a function called the kernel,

Is the bandwidth. Generally

silver

Is a probability density function satisfying. In the present invention, a probability density function of a normal distribution is used.

Is called a smoothing parameter, and is calculated using the value proposed by Silerman in the present invention.

Figure 4 shows the temperature distribution function of the August 2002 Seoul Gangbuk area according to an embodiment of the present invention.

Using the same method as above, the temperature distribution function of each country is derived.

Step 103 is a process of estimating the national temperature distribution function by weighting the regional demand using the temperature distribution function of each region, wherein the regional temperature distribution function is estimated to estimate the temperature distribution function of Korea as a whole in response to the energy demand in Korea. The weighted average of each region's demand is estimated to estimate the temperature distribution function of Korea.

Step 104 is a process of estimating the temperature response function.

, 를 고려하여 다음과 같은 모형을 고려한다.The temperature response function is a concept newly introduced in the present invention in order to estimate the nonlinear temperature effect, and may be viewed as a function representing the sensitivity of power demand to each temperature. Other variables that go into the predictive model to estimate the temperature response function

Consider the following model, taking into account.

의

추정치를 얻고 이를 이용하여 에너지 수요에 대한 기온반응함수(g)를 After estimating the above equation

of

Get an estimate and use it to calculate the temperature response function (g) for energy demand.

Estimate as

5 is a result of estimating the temperature response function for the maximum power transformer in the Seoul Gangbuk area according to an embodiment of the present invention.

The temperature response function is represented by a U-shape having a large value at a low temperature at which a heating load is generated and at a high temperature at which a cooling load is generated. This means that the power demand's response to temperature is nonlinear. The estimated temperature response function appears to have a U-shape with a relatively high value at high or low temperatures. This U-shape means that the demand for power is relatively higher at higher and lower temperatures, and this temperature response function indicates that the demand for power is increased due to cooling and heating loads in summer and winter, respectively. Prove well.

The estimated temperature response function has an asymmetric U-shape with a difference between high temperature and low temperature values, and the sensitivity is high when the temperature is low. This means that the maximum power in the Seoul Gangbuk area receives greater cooling load than heating effect.

In addition, the estimated temperature response function can explain the phenomenon in which the effect on temperature change varies depending on the level of temperature. In the Gangbuk area of Seoul, even if the temperature rises by 1 ° C, the demand for electricity increases more than when the temperature rises near the heating load. Since the difference in sensitivity between 1 ° C near the temperature at which the cooling load appears in the estimated temperature response function is greater than the difference between 1 ° C of the smaller temperature, the measured temperature effect is different from the temperature change depending on the temperature level. It can be explained.

Step 105 is to consider the overall nonlinear temperature effect.

The seasonality of the power time series comes from the cooling load and the heating load, and thus can be determined by the temperature distribution. In order to model the relationship between the seasonality of the power time series and the temperature distribution, the present invention introduces the concept of the temperature response function.

를 t시점에서의 기온 분포 함수라 하고

를 기온 반응 함수라고 하면 t시점에서의 기온효과는Specifically

Is the temperature distribution function at time t

Is the temperature response function, the temperature effect at time t is

Is defined as The actual temperature effect is obtained by estimating the temperature distribution function at time t and integrating it into the temperature response function.

 The temperature effect is obtained by integrating the estimated temperature response function as the temperature distribution function for a given period as defined in the above equation. FIG. 6 shows the monthly average temperature effect on the estimated monthly maximum power. In the case of the Seoul Gangbuk area shown in Figure 6, the monthly maximum power temperature effect has a high value in summer and winter, it can be seen that the summer value is relatively larger than the winter value. 7 is a diagram illustrating a process of estimating the actual temperature effect by the estimated temperature distribution function and the temperature response function according to an embodiment of the present invention.

In the model developed in the present invention, the temperature effect is measured using the temperature distribution function at the corresponding time point, whereas in the conventional power time series prediction model, only the average, the highest and the lowest temperature are used. The model that predicts the power demand using only average temperature cannot measure the temperature effect due to the temperature distribution properly. Even though the average temperature is the same, the day with the large daily cross is relatively higher than the day with the smallest day, so the average temperature alone cannot accurately predict the power demand. In addition, if you use the highest and lowest temperatures in addition to the average temperature, you can better measure the effect of the temperature distribution over the period than if you are using the average temperature alone, but this model inevitably applies when the temperature is high or low for a moment. The effect is overestimated. On the contrary, the model developed in the present invention can measure the temperature effect more accurately because it uses the temperature distribution.

Step 106 is a process of fitting energy demand as a cointegrating model with time variance coefficients through appropriate explanatory variables such as temperature effects and economic variables, based on performance values.

Many economic time series are known to have a unit root, and power demand time series are very similar. Because time series with unit roots have different properties from normal time series, analysis with a normal regression model gives incorrect results. Therefore, in case of power time series with unit root, it is reasonable to set and analyze time series model using cointegration model and error correction model.

The present invention sets and estimates the relationship between time series having unit roots as a cointegrating regression model having the following time variation coefficients.

는 에너지 수요를 나타내며,

는 에너지 수요를 설명하기 위한 설명변수로서의 경제변수이며 이 때

는 정상시계열이 된다. 이와 같이 추 정된 공적분 모형은 에너지 수요와 다른 변수들 사이의 장기균형(longrun equilibrium) 관계를 나타낸다. here

Represents energy demand,

Is an economic variable as an explanatory variable for explaining energy demand.

Becomes the normal time series. The estimated cointegration model represents a longrun equilibrium relationship between energy demand and other variables.

In the present invention, an error correction model that takes into account the short-term dynamic relationship of the energy demand time series having the cointegration relationship is used.

가 이에 해당한다. 이는 긴 기간동안 분석하게 되는 경우 시간의 흐름에 따라 전력기술의 진보, 경제의 성장정도, 전력정책, 소비자 성향 변화 등으로 말미암아 탄력성이 변하는 것이 일반적이므로 이를 모형에 반영한 것이다. 탄력성의 변화를 반영하지 못하는 기존의 고정계수를 갖는 예측 모형In the present invention, the long-term stable relationship between energy demand and economic variables is modeled as a cointegration relationship, and their relationship is used to reflect the change over time. In the cointegration model above

Is equivalent to this. This is reflected in the model as the elasticity changes due to the progress of power technology, economic growth, power policy, and consumer tendency over time when analyzed for a long period of time. Predictive Models with Existing Fixed Coefficients That Do Not Reflect Changes in Elasticity

In this case, this cannot reflect the change in elasticity, so the prediction accuracy is inferior to the time variation coefficient.

In this step, we forecast long-term peak power demand and short-term peak power demand. The long-term maximum power demand includes the maximum long-term asynchronous transformer maximum power, the long-term non-simultaneous direct trade maximum power, the long-term simultaneous transformer maximum power and the long-term simultaneous direct trade maximum power, and the short-term maximum power demand is the short-term monthly transformer maximum power and short-term monthly direct trade. Includes maximum power, short-term weekly transformer maximum and short-term weekly direct power.

The long-term prediction model consists of three parts. The first is regional non-synchronous maximum power prediction model. This model analyzes the long-term equilibrium relationship between regional asynchronous maximum power and total national electricity sales, and provides regional asynchronous maximum power forecasts. The second is a short and long term link that integrates short-term asynchronous peak power forecasts with long-term asynchronous peak power forecasts. The asynchronous maximum power forecast, calculated through short and long term linkage, provides more accurate forecasting by reflecting both short-term fluctuations and long-term trends. Finally, the third is the regional simultaneous maximum power forecasting model. This model converts the nation's simultaneous peak power forecasts into regional peak peak power estimates using the simultaneous / non-simultaneous ratios.

The asynchronous transformer peak power prediction model among the asynchronous peak powers aims to predict the annual maximum transformer power by region. To predict this, we use the quarterly regional transformer maximum power, quarterly national total sales, and quarterly dummy variables.

In particular, since the maximum power of the asynchronous transformer is mainly home and commercial power, the maximum power of the asynchronous transformer is significantly affected by temperature and economic conditions compared to the asynchronous direct power. In 1998, for example, during the financial crisis, the maximum power of asynchronous transformers decreased or slowed in most regions. In addition, in 2003, when the summer temperatures were unusually low due to heavy precipitation, the maximum rate of increase in the maximum power of asynchronous transformers was found to have slowed significantly in most regions. As described above, the asynchronous transformer peak power prediction model reflects this effect in the model through the national total sales volume.

Regional asynchronous transformer maximum power prediction model

는 지역별 비동시 최대전력을 나타내고

는 전국 총 판매량을 나타낸다.

는 분기별 더미이며

로써 지역 구분을 나타내고,

는 최대전력 시계열 상의 구조변화를 반영하기 위한 더미변수이다. 도 8은 본 발명의 일 실시예에 따른 서울 강남 변압기 최대전력의 탄력성 추이를 나타낸 도면이다.A cointegration model like this establishes a long-term equilibrium relationship between regional asynchronous transformer maximum power and gross national sales. here

Represents the maximum asynchronous power by region.

Represents the national total sales.

Is a quarterly pile

To represent regional divisions,

Is a dummy variable to reflect the structural change on the maximum power time series. 8 is a diagram showing the elasticity trend of the maximum power of the Seoul Gangnam transformer according to an embodiment of the present invention.

The asynchronous direct power forecast model aims to predict annual direct peak power by region. To forecast this, we use the quarterly regional direct trading power, the quarterly national total sales volume, and the quarterly dummy variable.

Since the maximum amount of non-simultaneous direct sales is mostly industrial power, it is more influenced by industrial production and investment in the area than by temperature. While new investments and industry trends can be important variables in predicting direct deals, it's almost impossible to predict those variables in practice, so we focus on analyzing the long-term trends of regional direct peak power.

The prediction model for the asynchronous direct power is the same as the model of the asynchronous transformer maximum power. 9 is a diagram showing the elasticity trend of the maximum power direct trading in Seoul Gangbuk according to an embodiment of the present invention.

In order to link the short and long term, the present invention uses a method of weighted averaging long-term and short-term predictions at a constant ratio. Here, the long-term forecast refers to the prediction results provided by the asynchronous transformer peak power prediction model among regional asynchronous peak power forecast models, and the short-term forecast is the annual calculation calculated from the monthly transformer asynchronous peak power prediction model among the short-term forecast models discussed in the future. Say the prediction result.

The ratio of weighted average of long-term and short-term forecasts is shown in the following table.

year Short-term Prediction Results Long-term Prediction Results 2004 100% 0% 2005's 50% 50% 2006's 0% 100%

On the other hand, we did not link short-term forecasts with long-term forecasts for non-simultaneous direct transactions. Unlike non-simultaneous transformers, non-simultaneous direct sales are very irregular in the short run because they are affected by industrial production and economic conditions in the region, in addition to sales volume and temperature. In addition, if there is a direct consumer price, there is a big demand in the short term. On the contrary, when the consumer disappears, the demand disappears, so the short-term fluctuation assumed in the model is very large. Therefore, short-term forecasts that reflect short-term changes do not have more accurate predictive power than long-term forecasts that represent long-term trends. As a result, it is more reasonable to provide long-term forecasts for asynchronous co-occurrence maximum power only with forecasts that are calculated from long-term forecasting models.

Short- and long-term linkages yield final regional asynchronous transformers and direct power peak estimates. This estimate is used to obtain regional maximum peak power estimates.

If the asynchronous maximum power model is the so-called bottom-up method that analyzes the relationship between the asynchronous maximum power and the total national sales volume for each region, the simultaneous maximum power model distributes the national simultaneous maximum power forecast to each region. That's the way. The simultaneous maximum power model takes this form because, first, it is reasonable to divide the national forecast by region since regional simultaneous maximum power means regional power demand when the nation's maximum demand is highest throughout the year. Second, the results of the simultaneous maximum power prediction by region should always be smaller than the non-simultaneous maximum power, but it is difficult to meet such constraints when modeled by region. Third, long-term data are needed to model the simultaneous maximum power generated once per year, whereas the only available data was the 10-year period from 1994 to 2003. Therefore, in order to present the results of the simultaneous maximum power prediction in a reasonable way, it is appropriate to use the simultaneous / asynchronous ratio so that the national maximum power prediction results are divided by region but smaller than the asynchronous maximum power prediction result.

In order to provide 27 regional maximum peak power forecasts for 14 years from 2004 to 2017, the simultaneous peak power prediction model calculates simultaneous peak power forecasts from asynchronous peak power forecasts using regional simultaneous / asynchronous rates. Simultaneous / non-simultaneous ratios are derived from past performance but use the average three-year average concurrent / asynchronous ratios for which the simultaneous / asynchronous ratios have stabilized. The simultaneous / non-simultaneous ratios thus obtained are used to distribute the national maximum power forecasts to regional maximum power projections. Specifically, the distribution ratio of the simultaneous maximum power forecasts is the value obtained by multiplying the occupancy ratio of the non-simultaneous maximum power forecasts by the region's simultaneous / non-simultaneous ratios and resuming the sum of these ratios to 1.

10 is a view showing the simultaneous / asynchronous ratio of the Seoul Gangbuk transformer maximum power according to an embodiment of the present invention.

Short-term peak power forecasts monthly asynchronous peak power and weekly asynchronous peak power.

The monthly asynchronous maximum power, the monthly prediction model provides a prediction of the monthly non-synchronous maximum power for the next 24 months. Based on the monthly maximum power data by substation, it is added to each of 27 regions to generate and use the monthly asynchronous maximum power data of the region. These monthly maximum power data exist from 1998 to 2003, and based on these estimates.

In the monthly prediction model of the present invention, the maximum monthly power is predicted using the standardized monthly sales volume and the temperature effect. Standardized monthly sales volume is estimated using monthly sales forecasts. At this time, it is preferable to use the sales volume data from the data of the Korea Electric Power Exchange Short-Term Electricity Demand Forecast Information Disclosure (2004.2). The temperature effect is measured according to the process described in Section 2 of Chapter 3 after estimating the temperature distribution function and the temperature response function using hourly temperature data from 27 regions.

The monthly maximum power forecasting model uses a standardized integral model that estimates the relationship between the standardized monthly sales volume, the temperature effect and the maximum integral power. In particular, we use the coefficient of variation in time as their relationship will change over time. After estimating the cointegration relationship that represents the long-term equilibrium relationship, the final prediction is obtained by using an error correction model that takes into account the short-term dynamic relationship.

The monthly maximum power includes a monthly transformer maximum power and monthly direct sales maximum power.

To predict the monthly transformer maximum power

는 각 지역의 월별 비동시 변압기 최대전력이고,

는 지역별 총체적 비선형적 기온효과를 나타내며,

는 표준화된 월별 판매량을 나타내는 변수이다.

은 예측의 대상이 되는 각 지역을 나타낸다. 특히 일부 지역의 경우, 이 모형의 균형관계가 시간에 따라서 변하는 것으로 보아

의 계수

를 시간변동계수로 설정한다.

는 최대전력의 판매량에 대한 탄력성을 의미하므로, 시간에 따라서 탄력성이 변하는 것을 모형화한 것으로 볼 수 있다.We set up a static equilibrium relationship with a cointegration model such as here

Is the maximum power of monthly asynchronous transformers in each region,

Represents the total nonlinear temperature effect by region,

Is a variable representing standardized monthly sales volume.

Represents each region to be predicted. Especially in some regions, the equilibrium of this model changes over time.

Coefficient of

Is set as the time variation coefficient.

Since the elasticity of the maximum power sales volume, it can be seen that the elasticity changes over time modeled.

The regional monthly maximum direct sales power varies greatly from region to region. In particular, unlike monthly transformer peak power, there are many areas where no special seasonality is found, and there are also areas where the peak power decreases with time. This phenomenon appears to be attributable to the characteristics of peak direct power, which is directly affected by the demand pattern of a small number of direct customers. In particular, the creation of new customers or changes in demand caused by internal development of customers are factors that significantly change the maximum power of direct trading.

For this reason, it is not reasonable to predict monthly direct peak power using temperature effects and standardized monthly sales across all regions, as in the transformer peak power model. Rather, it is possible to enhance the predictive power by selectively reflecting variables suitable to the characteristics of each region in the model.

The monthly direct trade maximum power prediction model of 26 regions of the present invention is divided into regions affected by the temperature effect and regions not affected by the temperature effect. Divide into rapidly changing areas and areas that do not. In addition, regions with small and fluctuating direct trading power, such as the Seoul Gangbuk or Gyeonggi Northwest regions, are modeled separately to estimate the forecasts.

The weekly forecast model provides a forecast of weekly asynchronous maximum power for the next 104 weeks. Performance of the weekly asynchronous maximum power is obtained from the daily maximum power data of each substation. If we obtain the weekly maximum from the daily maximum power data and add it to all substations in the area, we can generate the data for the weekly asynchronous maximum power for each region.

While monthly data existed from 1998 to 2003, currently available daily data are from 1999 to 2002. There are also problems such as severe data deficiency and many data with excessive values. Based on these daily data, the monthly maximum power data is generated and compared with the data used in the monthly maximum power prediction model. As a result, there is a big difference between the two data. The extremely low reliability of the data is the biggest difficulty of the weekly maximum power forecasting model.

The more reliable of the monthly maximum power data and the weekly maximum power data is the monthly maximum power data. Therefore, the monthly maximum power generated from the daily maximum power data is modified to match the value of the existing monthly maximum power data.

In the weekly prediction model of the present invention, the weekly maximum power is predicted using standardized weekly sales volume, temperature effects, and special days such as holidays and summer holidays. The standardized weekly sales volume is calculated using the monthly sales forecasts according to the process described in Section 3 of Chapter 3, and the temperature effect is calculated using the temperature distribution function and the temperature response function using hourly temperature data from 27 regions. It is measured after estimation. The holiday effect and summer vacation effect were added to explain the phenomenon that the maximum power of each week is small during the week and summer intensive vacation periods where New Year and Chuseok belong.

The weekly prediction model of the present invention uses a cointegration model for estimating the cointegration relationship between power demand and special day effects such as standardized weekly sales volume, temperature effects, summer holidays, and holidays. After estimating the cointegration relationship that represents the long-term equilibrium relationship, the final prediction is obtained by using an error correction model that takes into account the short-term dynamic relationship.

Among the weekly models, the transformer maximum power tends to increase across all regions, with distinct seasonality. In addition, the maximum power of the state where the holiday or summer vacation belongs is decreasing. To reflect this in the model, the weekly transformer peak power prediction model uses standardized weekly sales volume, temperature effects, summer vacation, and holiday effects.

In particular, the holiday effect is analyzed by dividing the day of the holiday on Tuesday or Wednesday and other cases. Irrespective of the officially determined holiday days, the holiday effect appears every two days before the holiday and three days after the holiday. Over and therefore the maximum power among them drops significantly. On the other hand, if the holiday day is other than Tuesday or Wednesday, the holiday effect is divided into two weeks, so the maximum power decrease is relatively small.

To estimate the weekly transformer maximum power in 27 regions

은 예측 대상이 되는 각 지역을 나타낸다.

는 각 지역의 주별 변압기 최대전력이고,

는 표준화된 주별 판매량이다.

와

는 명절 효과를 나타내는 변수이며, 각각 명절 당일이 화요일이거나 수요일일 때와 그 이외의 날일 때의 변수이다.

는 하계휴가 더미변수이다. 일부 지역의 경우, 이 모형의 균형 관계가 시간에 따라서 변하는 것으로 보아

의 계수

를 시간변동계수로 설정한다.We establish a static long-term equilibrium relationship with a cointegration model such as here

Represents each region to be predicted.

Is the maximum power of the state transformer in each region,

Is the standardized weekly sales volume.

Wow

Are variables representing the holiday effect, respectively, when the holiday day is on Tuesday or Wednesday and on other days.

Is a summer vacation dummy variable. For some regions, the balance of this model changes over time.

Coefficient of

Is set as the time variation coefficient.

The maximum weekly direct power, as with the monthly direct direct power, varies significantly from region to region. This is due to the demand characteristics of direct deals, where the maximum power is determined by a few consumers. Therefore, it is reasonable to forecast the weekly direct trading maximum power forecasting model by selectively reflecting variables suitable for the characteristics of each region in the model as in the monthly direct trading forecasting model.

Step 107 is an out-of-sample prediction process that can confirm the predictive power in advance.

For example, if you want to forecast the monthly power demand in 2004 based on the data from December 2003, you can estimate the power demand forecast from July 2003 to December based on the data from June 2003 only. It is a step to confirm the accuracy of the forecast in advance by obtaining the revised model and comparing it with the performance value of the electric power demand. The predictive power can be confirmed in advance by using the cointegration model showing the long-term equilibrium relationship and the error correction model showing the dynamic relationship of the short-term.

The asynchronous maximum power forecast model predicts regional asynchronous maximum power using the national total sales data and quarterly dummy. In particular, the present invention assumes a cointegration model having a time variation coefficient in order to estimate the elasticity of asynchronous maximum power for each region for the total national sales. In contrast, existing power demand forecasting models use multiple regression models. To demonstrate the effect of improving the predictive power through the cointegration model, the mean error rate of the out-of-sample predictions from the asynchronous maximum power prediction model is compared with the mean error rate of the out-of-sample predictions from the multiple regression model.

The multiple regression model used to compare the predictive power

는 전국 총 전력 판매량이며

는 분기별 더미이다. 이 모형은 비동시 최대전력 예측 모형이 사용하는 설명 변수를 그대로 사용하되 단위근 시계열임을 감안하여 각 변수를 차분하여 사용한다. 파라미터의 추정을 위해서는 최소 자승 추정법 (OLS:Ordinary Least Squares)을 사용한다.It is expressed as here

Is the national total electricity sales

Is a quarterly pile. In this model, the explanatory variables used by the asynchronous maximum power prediction model are used as they are, but each variable is differentially used in consideration of the unit root time series. To estimate the parameters, least squares estimation (OLS) is used.

To evaluate the predictive power of the asynchronous transformer peak power prediction model, the model was estimated from the first quarter of 1991 to the fourth quarter of 2001 from the first quarter of 1991 to the fourth quarter of 2003, and then the first quarter of 2015 to 2015. Forecast by the fourth quarter of year and compare the average forecast error rates for 2002 and 2003.

FIG. 11 is a cointegration model of the Seoul Gangbuk area according to an embodiment of the present invention, and FIG. 12 is a multiple regression model of the Seoul Gangbuk area according to an embodiment of the present invention.

In order to evaluate the predictive power of the Asynchronous Direct Selling Power Prediction Model, the model was estimated from the first quarter of 1991 to the fourth quarter of 2001 from the first quarter of 1991 to the fourth quarter of 2003. Forecast by the fourth quarter of 2015 and compare the average forecast error rates for 2002 and 2003.

FIG. 13 is a cointegration model of a direct trade Seoul Gangbuk region according to an embodiment of the present invention, and FIG. 14 is a multiple regression model of a direct trade Seoul Gangbuk region.

In order to verify the predictive power of the simultaneous maximum power prediction model, the asynchronous maximum power prediction value calculated through short and long term linkage is converted into the simultaneous maximum power prediction value.

To convert the asynchronous transformer peak power estimates to the simultaneous transformer peak power estimates, use the 2001 simultaneous / asynchronous ratio. The reason for using only 2001 data to calculate the concurrent / asynchronous ratios is that since 1999, some regions have experienced rapid changes in the simultaneous / asynchronous ratios due to the increase in midnight power.

When estimating the simultaneous direct peak power, the simultaneous / asynchronous average ratio since 1999 is used, unlike when estimating the simultaneous transformer maximum power. This is because the simultaneous / asynchronous ratio of direct power peaks is different from that of transformer maximum power.

The timing of simultaneous peak power generation is generally in the summer of July or August, but the timing of non-simultaneous peak power of direct deals is very different depending on the production activity of local direct customers and local economic conditions. Thus, if we look at the concurrent / asynchronous ratios by region, we find that the average concurrent / asynchronous ratio and the pattern of change are not constant.

In the present invention, the short-term prediction model sets the long-term balance relationship of the power time series as the cointegration model, and uses the cointegration and error correction model to model the short-term dynamic relationship with the long-term balance relationship. In contrast, the existing maximum power prediction model uses an autoregressive model or a dynamic regression model. To demonstrate the effect of improving the predictive power through the cointegration and error correction model, the average error rate of the out-of-sample predictions from the short-term prediction model of the present invention is compared with the mean error rate of the out-of-sample predictions from the autoregressive and dynamic regression models.

The autoregressive model used to compare predictive power

를 설명하기 위해서 최대전력의 시차변수

를 이용하여 회귀 분석하는 모형이므로, 자기 회귀 모형이라고 부르며 시차변수의 차수인 p를 이용하여 AR(p)라고 표현한다.It is expressed as This model is the maximum power at time t

Parallax variable of maximum power to explain

Since it is a model for regression analysis using, it is called autoregressive model and is expressed as AR (p) using p, which is the order of parallax variables.

Dynamic regression model

를 설명하기 위해서 최대전력 의 시차변수

뿐만 아니라 t시점의 설명 변수

를 이용하여 회귀 분석하는 모형이다. 여기서 설명 변수

는 본 발명의 예측 모형에서 사용하는 변수를 그대로 사용하고 시차변수의 차수 p는 자기 회귀 모형의 p을 이용한다.It is expressed as This model is the maximum power at time t

Parallax variable of maximum power to explain

As well as the explanatory variable at time t

Regression analysis model using. Description variable here

Denotes the variable used in the predictive model of the present invention as it is, and the order p of the parallax variable uses the p of the autoregressive model.

Data from January 1998 to December 2003 are used to verify the predictive power of the monthly asynchronous transformer peak power prediction model. After estimating the model from January 1998 to December 2002, the model predicts the monthly maximum power of the transformer for 12 months from January 2003 and compares the predicted error rate.

In order to measure the effect of improving the predictive power obtained by using the cointegrating and error correcting models, the mean error rate of the predicted values obtained from the cointegrated and error correcting models is compared with the mean error rate of the predicted values obtained from the autoregressive and dynamic regression models. The order of parallax is set to 13 by AIC (Akaike Information Criterion).

FIG. 15 is a comparison model of monthly transformer maximum power predictive power in Gangbuk, Seoul, according to an embodiment of the present invention. FIG.

The data from January 1998 to December 2003 are used to verify the predictive power of the monthly asynchronous direct power forecast model. After estimating the model from January 1998 to December 2002, we estimate the monthly direct sales peak power for 12 months from January 2003 and compare the predicted error rate.

To measure the effect of improving the predictive power by using the cointegration and error correction model, the mean error rate of the predictions obtained from the cointegration and error correction model is compared with the mean error rate of the predictions obtained from the autoregressive and dynamic regression models. Here, the order of the parallax variable is 13 as in the transformer.

Data from the first week of January 1999 to the fourth week of December 2002 are used to verify the predictive power of the weekly asynchronous transformer peak power prediction model. After estimating the model from the first week of January 1999 to the fourth week of May 2002, we estimate the maximum transformer weekly power from the first week of June 2002 to 31 weeks and calculate the prediction error rate.

Data from the first week of January 1999 to the fourth week of December 2002 are used to verify the predictive power of the weekly non-concurrent direct-current peak power forecasting model. After estimating the model from the first week of January 1999 to the fourth week of May 2002, we estimate the maximum weekly direct power for the first week of June 2002 and 31 weeks and calculate the forecast error rate.

In step 108, prediction is performed using a cointegration model and an error correction model. It is a step of estimating long-term equilibrium and short-term dynamic relations based on the results using the cointegration model and error correction model introduced above, and predicting future electric power demand. In this case, in the case of the temperature effect, the average temperature effect is estimated based on the temperature data so far and used as an explanatory variable.

FIG. 16 is a long-term power sales forecast result according to an embodiment of the present invention, FIG. 17 is a long-term maximum power prediction result according to an embodiment of the present invention, and FIG. 18 is a Seoul Gangbuk ratio according to an embodiment of the present invention. Simultaneous transformer maximum power prediction result, Figure 19 is a monthly non-synchronous transformer maximum power prediction result according to an embodiment of the present invention, Figure 20 is the weekly transformer maximum power prediction result of the Gangbuk area of Seoul according to an embodiment of the present invention to be.

21 is a block diagram showing the overall demand forecasting system according to an embodiment of the present invention.

In the above embodiment, the overall demand forecasting system includes a server for storing performance values and data, a server for storing various data required for demand forecasting and performing a preparation step necessary for demand forecasting, a client for performing demand forecasting using the data, and the Includes the vendor's server that receives the demand forecast results.

The above embodiment schematically illustrates a system in which a client receives various data necessary for demand forecast from various servers and transmits the data to a related company server after the demand forecast.

Referring to FIG. 21, a server storing various data necessary for the demand forecast and performing a preparation step necessary for the demand forecast includes a engine required for the demand forecast as shown in FIG. 22.

The engine includes a library unit, a module unit, a data manager and an arithmetic unit. The library includes the co-integration library and the error correction library described above, and the module unit includes a temperature distribution function module, a temperature response function module, a temperature effect module, and a time variation coefficient module. Contains logic

The server for storing the various data necessary for the demand forecast and performing the preparation steps required for the demand forecast, the entire demand forecast system receives power consumption and related data from several years ago from the server storing the performance value and data Various functions necessary for prediction and preliminary operations necessary for it are performed. That is, the function of temperature distribution function, temperature response function, temperature effect, and time variation coefficient is preset using the cointegrating library and the error correction library, and the database manager uses the arithmetic logic device and the calculation logic device of the library and the arithmetic unit. It is linked with.

The client receives various prepared functions and data from a server that stores various data necessary for the demand forecast and performs a preparation step required for the demand forecast, and inputs additional data to perform the demand forecast. The additional data is preferably included in the meteorological office temperature data, national economy and power, gas data.

The demand prediction result derived by the client is transmitted to the related company server, the performance value and the data server.

The related company server is preferably a company that produces power. Relevant companies that receive the demand forecast results use it as basic data for power generation.

The performance value and the data server which received the demand forecasting result perform a demand forecast several years before the current time to be predicted as described above, and then compare the result with the performance value. The comparison result is used for error correction. In other words, assuming that the current year is 2004, the power demand from 2001 to 2003 is predicted using the performance values and data from 2000, and the error is measured by comparing with the actual power consumption from 2001 to 2003.

As described above, the result of the correction is transmitted to a server that stores various data necessary for the demand forecast and performs a preparation step required for the demand forecast.

The demand forecasting system does not end with obtaining a demand forecasting result, but analyzes the result with a demand forecasting person in the field, and reflects the result of the analysis to derive an improved result. Instead of unilaterally producing results, two-way communication between demand forecasting systems and field personnel is possible.

The derived forecast results can be confirmed by modifying the forecast results by the ratio of the forecast forecaster by the forecast range. This means to reflect more on-site opinions of the results, and by doing this, more accurate prediction results can be obtained.

Energy demand forecasting can form an independent system by itself as a result of demand forecasting, but it may need to be linked to existing systems in the industry. In this case, since only the D / B table that stores the demand forecast result needs to be shared, it can be easily and efficiently linked with other systems.

When unexpected events occur, such as sudden price spikes in power and gas, or sudden changes in temperature, the forecasting results may not be correct if the existing models are applied. Such unexpected events can be made into scenarios, put into model explanatory variables, and predicted.

As described above, the present invention has been described with reference to the preferred embodiments, but those skilled in the art can variously modify and modify the present invention without departing from the spirit and scope of the present invention as set forth in the claims below. It will be appreciated that it can be changed.

As described above, according to the present invention, by designing with an automated prediction system and providing a convenient user environment, it is possible to reduce errors that can be caused by conventional manual prediction, and to greatly reduce the time and manpower resources required for prediction. .

Claims (10)

a) Review the trends and seasonality of the regional maximum power using previous non-synchronous maximum power data, review the rate of increase in concurrent maximum power and the occupancy rates in the region using previous simultaneous maximum power data, and Examining whether or not a factor that causes a structural change has occurred in the time series characteristic of? b) modeling the volatility of energy demand as the seasons change; c) adapting energy demand based on performance to long-term and short-term forecasting models with a covariance model with time-varying coefficients through appropriate explanatory variables; d) an out-of-sample prediction step to confirm the predictive power in advance; And e) an energy demand prediction method comprising performing prediction using a cointegration model and an error correction model. 2. The method of claim 1, wherein the asynchronous and simultaneous maximum power data of step a) includes transformer maximum power by region and direct current maximum power by region. The method of claim 1, wherein the step of modeling the variability of energy demand according to the seasonal change of step b) b1) determining the relative frequency of how often the temperature for a given time period appears at each temperature; b2) determining the sensitivity of power demand to each temperature; b4) modeling the total nonlinear temperature effect. delete According to claim 1, wherein c1) long-term prediction model is c11) predicting asynchronous maximum power per region; c12) incorporating short-term asynchronous peak power prediction and long-term asynchronous peak power prediction; And c13) A method for predicting energy demand comprising the step of predicting the simultaneous maximum power for each region. The method of claim 1, wherein c2) the short-term prediction model c21) predicting monthly asynchronous maximum power; And c22) A method for predicting energy demand comprising estimating weekly asynchronous maximum power. The power demand of claim 1, wherein Energy demand prediction method, characterized in that the long-term maximum power demand and the short-term maximum power demand. 8. The method of claim 7, wherein the long-term maximum power demand is An energy demand prediction method comprising a long-term non-synchronous transformer maximum power, a long-term non-simultaneous direct sales maximum power, a long-term simultaneous transformer maximum power amount and a long-term simultaneous direct sales maximum power amount. 8. The method of claim 7, wherein the short term maximum power demand is An energy demand prediction method comprising a short-term monthly transformer maximum power, a short-term monthly direct trade power, a short-term weekly transformer maximum power, and a short-term weekly direct power maximum power. delete

Images