A Study on the Applicability of Neural Networks on Unit Commitment
| Project/Area Number |
02650203
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
電力工学
|
|---|
| Research Institution | Hiroshima University |
|---|
Principal Investigator |
SASAKI Hiroshi Hiroshima Univ., Faculty of Engineering, Professor, 工学部, 教授 (10034349)
|
|---|
| Project Period (FY) |
1990 – 1991
|
| Project Status |
Completed (Fiscal Year 1991)
|
| Budget Amount *help |
¥1,900,000 (Direct Cost: ¥1,900,000)
Fiscal Year 1991: ¥800,000 (Direct Cost: ¥800,000)
Fiscal Year 1990: ¥1,100,000 (Direct Cost: ¥1,100,000)
|
|---|
| Keywords | Neural Network / Hopfield Network / Unit Commitment / Combinatorial Optimization Problem / ニュ-ラルネットワ-ク / 混合整数計画問題 / 不等式制約 |
|---|
| Research Abstract |
1. I have conducted theoretical studies on the convergence characteristics of Hopfield type neural networks(abbreviated as 'NN' hereafter)without definite conclusions. However, it has made clear that the NN with inequality constraint neural networks do not converge well unless the magnitudes of output coefficients of inequality neurons are selected adequately. Again, this is still in a stage of trial and error.
2. In order to handle optimization problems with a large number of inequality constraints, we have adopted the Hopfield NN with NN dedicated to equality constraints and demonstrated that the NN could solve a large scale unit commitment problem consisting 30 machines with 24 periods. We have improved the neural network simulator and could curtail the convergence time by half.
3. In order to confirm the effectiveness of the proposed approach, it has been applied to generation planning problem and thermal plant maintenance scheduling problem, both of which belong to a class of integer programming problems and could be modeled quite accurately. The proposed NN could successfully have solved such problems of practical size which are intractable by the branch and bound(B-B)method. For a small problem that can be solved by B-B method, the minimum cost obtained by the NN is about 3 % larger than that by the B-B method. However, the difference could be several million dollars in electric utilities. Therefore, I am currently studying the possibility of coupling the NN and B-B method, that is, the best solution by the NN should be used as a starting value of the latter.
|
Report
(3 results)
Research Products
(20 results)
