Abstract
This paper proposes an approach to one‐day‐through seven‐day‐ahead electrical load forecasting based on a realistic problem formulation which should contribute to more reliable and economic weekly power stations operation. Generally, the load forecasting has the following problems: (1) although the load is affected by various factors, such as temperatures, in the load forecasting, it is impossible to consider all of them; (2) the relationships between the load and some factors are not clear, and often vary with time; and (3) uncertainties in forecasts of the temperatures sometimes make the results of load forecasting worse. They are very influential in the power station operation. While a number of methods have been proposed to solve the problems (1) and (2), there have been few attempts to solve the problem (3). The following approach is proposed in this paper, taking these problems into consideration. First, concerning the problem (1), the focus is on such factors that have major influence on the load and whose values are obtainable on a weekly basis. The other factors are all regarded as stochastic and are not included in the forecasting model. Second, regarding the problem (2), a self‐organizing approach is used where the algorithm itself finds the optimal model structure or the optimal set of factors to be included in the model day by day. Finally, addressing the problem (3), a new performance index of model structures is proposed which can measure the balance between: i) improvement of the load‐forecasting accuracy due to inclusion of a factor in the model; and ii) degradation caused by uncertainty or error in the factor included. Using this index, a model is constructed which does not yield a large error in spite of errors in the temperature forecasts. Examples show that this approach improves the forecasted results when erroneous temperature forecasts are fed into the model, and verifies its effectiveness.
| Original language | English |
|---|---|
| Pages (from-to) | 22-32 |
| Number of pages | 11 |
| Journal | Electrical Engineering in Japan |
| Volume | 115 |
| Issue number | 8 |
| DOIs | |
| Publication status | Published - Jan 1 1995 |
Fingerprint
All Science Journal Classification (ASJC) codes
- Energy Engineering and Power Technology
- Electrical and Electronic Engineering
Cite this
One‐day‐ through seven‐day‐ahead electrical load forecasting in consideration of uncertainties of weather information. / Miyake, Takashi; Murata, Junichi; Hirasawa, Koutaro.
In: Electrical Engineering in Japan, Vol. 115, No. 8, 01.01.1995, p. 22-32.Research output: Contribution to journal › Article
}
TY - JOUR
T1 - One‐day‐ through seven‐day‐ahead electrical load forecasting in consideration of uncertainties of weather information
AU - Miyake, Takashi
AU - Murata, Junichi
AU - Hirasawa, Koutaro
PY - 1995/1/1
Y1 - 1995/1/1
N2 - This paper proposes an approach to one‐day‐through seven‐day‐ahead electrical load forecasting based on a realistic problem formulation which should contribute to more reliable and economic weekly power stations operation. Generally, the load forecasting has the following problems: (1) although the load is affected by various factors, such as temperatures, in the load forecasting, it is impossible to consider all of them; (2) the relationships between the load and some factors are not clear, and often vary with time; and (3) uncertainties in forecasts of the temperatures sometimes make the results of load forecasting worse. They are very influential in the power station operation. While a number of methods have been proposed to solve the problems (1) and (2), there have been few attempts to solve the problem (3). The following approach is proposed in this paper, taking these problems into consideration. First, concerning the problem (1), the focus is on such factors that have major influence on the load and whose values are obtainable on a weekly basis. The other factors are all regarded as stochastic and are not included in the forecasting model. Second, regarding the problem (2), a self‐organizing approach is used where the algorithm itself finds the optimal model structure or the optimal set of factors to be included in the model day by day. Finally, addressing the problem (3), a new performance index of model structures is proposed which can measure the balance between: i) improvement of the load‐forecasting accuracy due to inclusion of a factor in the model; and ii) degradation caused by uncertainty or error in the factor included. Using this index, a model is constructed which does not yield a large error in spite of errors in the temperature forecasts. Examples show that this approach improves the forecasted results when erroneous temperature forecasts are fed into the model, and verifies its effectiveness.
AB - This paper proposes an approach to one‐day‐through seven‐day‐ahead electrical load forecasting based on a realistic problem formulation which should contribute to more reliable and economic weekly power stations operation. Generally, the load forecasting has the following problems: (1) although the load is affected by various factors, such as temperatures, in the load forecasting, it is impossible to consider all of them; (2) the relationships between the load and some factors are not clear, and often vary with time; and (3) uncertainties in forecasts of the temperatures sometimes make the results of load forecasting worse. They are very influential in the power station operation. While a number of methods have been proposed to solve the problems (1) and (2), there have been few attempts to solve the problem (3). The following approach is proposed in this paper, taking these problems into consideration. First, concerning the problem (1), the focus is on such factors that have major influence on the load and whose values are obtainable on a weekly basis. The other factors are all regarded as stochastic and are not included in the forecasting model. Second, regarding the problem (2), a self‐organizing approach is used where the algorithm itself finds the optimal model structure or the optimal set of factors to be included in the model day by day. Finally, addressing the problem (3), a new performance index of model structures is proposed which can measure the balance between: i) improvement of the load‐forecasting accuracy due to inclusion of a factor in the model; and ii) degradation caused by uncertainty or error in the factor included. Using this index, a model is constructed which does not yield a large error in spite of errors in the temperature forecasts. Examples show that this approach improves the forecasted results when erroneous temperature forecasts are fed into the model, and verifies its effectiveness.
UR - http://www.scopus.com/inward/record.url?scp=0029521832&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0029521832&partnerID=8YFLogxK
U2 - 10.1002/eej.4391150803
DO - 10.1002/eej.4391150803
M3 - Article
AN - SCOPUS:0029521832
VL - 115
SP - 22
EP - 32
JO - Electrical Engineering in Japan (English translation of Denki Gakkai Ronbunshi)
JF - Electrical Engineering in Japan (English translation of Denki Gakkai Ronbunshi)
SN - 0424-7760
IS - 8
ER -