L.-L. Cui et al.
for thermal power plant unit, but realize optimal load distribution the key step is accurate fitting of the unit coal
consumption curve.
At present, coal consumption curve of the thermal power plant is usually obtained by the performance para-
meters which are provided by the manufacturer, or thermal test data, and these curves remain unchanged for a
long time. However, the unit in the actual operation will be affected by the mode of operation, coal quality, de-
vice status, the technical level of operators and other factors, and make these curves have a great differences
with the actual operation situation of the unit. According to this situation, it needs to refit the coal consumption
curve in actual operation. The method of curve fitting most are using least squares on currently, but it is difficult
to solve complex nonlinear problems by this method, sometimes these curves cannot meet requirements of the
actual applications.
Genetic algorithm, or GA for short [1]-[4], is a global optimization algorithm based on selection and natural
genetic, which developed from evolution theory and genetic theory. Compared with the least square method, the
main characteristic of genetic algorithm is not depend on gradient information, especially suited to be used deal
with complex and nonlinear problems which are difficult to be solved by traditional search methods, it makes up
for its shortcomings of the least square method. However, the data of the thermal power plant is huge, strong
nonlinear, this method is used to fit coal consumption curve of the thermal power plant in this paper, and com-
paring the proposed method with least squares method.
2. Data Processing
Since this paper is aimed at nonlinear static system of the coal consumption curve, that is to say it needs to the
unit work on stable state, and use the static data of the unit to model. All the tested unit data are derived from the
bottom of the DCS control system, and because these date are effected by unit operating mode and environmen-
tal conditions, there are errors and noise for the collected data in DCS, if modeling directly use the data in DCS,
it will cause great interference for coal consumption curve, meanwhile it is meaningless if study on coal con-
sumption of the unit under the unit operating exist fluctuation. Therefore, it needs to process original data before
modeling.
There are two main kinds of interference for the data from the DCS: One is the random interference at the
time of data collection, it can be removed by filtering method; the other is jump point that fluctuation is very
conspicuous. In this paper, the method of data detection is residual analysis. Collected data are set to
, the da-
ta after process is
,
, is the limit value of data changing, under the same sample period, the rate of change
can be judged by the absolute value of the difference between two successive sampling data.
is the kth
sampling value, to calculate
, if
, there is not outlier,
; if
, there may be outliers, we can take another point
, if
, and
there are same changing trend, we think there is disturbance. Let
, where
and
are weights, and
; if there are opposite trend, we think
is jump point. Let
( )(1)(1)y kax kbx k=−+ +
; if
, we think there are outliers, they should be removed.
3. Coal Consumption Curve Fitting of the Thermal Power Based on Genetic
Algorithm
Now it needs to fit coal consumption curve of the unit for a thermal power plant with 4 × 328.5 MW, operation
energy consumption data of the each unit, as in Table 1 below [5]. In fact, in reality project, just need quadratic
curve then it can meet requirements of the accuracy. Therefore, t he objective function set to the following qua-
dratic function:
, where
is the parameters that needs to estimate.
Program object function for the 1# unit, and be saved with the filename ga_curfit.m to Matlab directory.
1# unit fitting results as shown in Figure 1.
The same method can be used to fit coal consumption curve of the other three units, and the fitting results as
shown in Figures 2-4.
Coal consumption curve equations of each unit are fitted by genetic algorithm, which are:
2
1 11
0.0002857929064713220.15229546322378449.7469230797674F PP= ++
2
2 22
0.0004716869196219790.17577929964833743.0063408707241F PP=++