On the ships of the merchant and passenger fleet, it is relevant to use powerful ship steam boilers of a wide design class. Marine boilers, as objects of automatic control systems, are subject to the influence of a significant number of internal and external disturbing factors. Such influences often lead to self-oscillatory processes of the controlled parameters of a ship's boiler with significant nonlinearities. For the optimal tuning of automatic control systems for the working processes of ship boilers, exact knowledge of mathematical models of controlled processes is required. Due to the presence of significant nonlinear characteristics, it is proposed to use neural networks in modeling processes. As shown by the modeling processes in the MatLab (System Identification Toolbox) program, the use of nonlinear ARX models with a built-in neural network apparatus makes it possible to display the experimental working processes of ship parameters with a high degree of adequacy. Obtaining nonlinear mathematical models with high adequacy will improve the process of adaptation of automatic control systems for ship boilers and optimize environmental parameters.
On the ships of the passenger and tanker fleet,
the technological scheme of operation of two
auxiliary steam boilers (ASB) and one utilization
boiler (USB) for a common steam line has found
wide application (Fig. 1). With such a design
solution, auxiliary boilers, performing the
function of generating steam of high temperature
and pressure, are subject to the influence of deep
external disturbances associated with the mode of
operation of the steam turbine and cargo
operations on ships [
Experimental transient processes of two ASBs
of Mitsubishi MAC 35 t / h, installed on the oil
tanker "Minerva Roxanne" and obtained using the
monitoring system "ACONIS-2000", are shown
in Fig. 2 [
Analysis of the type of transient processes
(Fig. 2) allows us to conclude that with a sharp
increase in the electric and steam load, the turbine
control system immediately increases steam
consumption, however, the combustion mode of
the ASB has not yet been built and an imbalance
occurs in the production and consumption of
steam, as a result of which the pressure drops.
steam in the main line and in the path of the
working medium of the boiler. An oscillatory
mode is formed, characterized by significant
nonlinearity. The ability of the ASB to change the
steam production in accordance with the change
in the external (electrical) load is called the
maneuverability of the boiler [
There are two classes of nonlinear regressions
[
- replacing the weighted sum of linear regressors with a nonlinear ARX model, which has a more flexible nonlinear display function: F(y(t −1), y(t − 2), y(t −3), …, u(t),
u(t −1),u(t − 2), …), the arguments forF are the y and u regressor models. For clarity, the nonlinear model of the ARX structure can be displayed in the block diagram in Fig. 3.
Using the System Identification Toolbox
application (Fig. 3 - 4), a discrete ARX [
A(z)y(t) = B(z)u(t) + e(t) A(z) = 1 + z^-1 + 0.5 z^-2; B(z) = 0.415;
е(t) –discrete white noise, where z-1 = e-sT is the delay operator; T - sampling interval. Nonlinear models are used to compile a
The process of determining the degree of adequacy of the selected model is shown in Fig. 5. According to the analysis of the degree of adequacy of the analyzed models, calculated in the software application (see Fig. 5), it was found that the ARX model demonstrates the highest degree of convergence with the experimental data.
For the process under study - two auxiliary boilers installed on the tanker "Minerva Roxanne", which operate on the common steam line of the turbine, using the SCADA monitoring system ACONIS2000E, installed in the central control room of the ship, the experimental characteristics obtained (Fig. 6 - 7).
It should be noted that control objects demonstrate significant nonlinear characteristics, therefore, to obtain a model of the system under consideration, a nonlinear ARX model was used.
The process of identification and validation on an independent data set in the System Identification Toolbox (SIT) is shown in Fig. 8 9.
In fig. 10-11 show the view of the nonlinear model and its three-dimensional surface, as determined using the SIT application.
It should be noted that the System
Identification Toolbox provides several
nonlinear estimates of g (x) for nonlinear ARX
models. Nonlinearity is formed in the form of a
wavelet, a sigmoid network (sigmoidnet), a
binary tree (treepartition), a multilevel neural
network (neuralnet), and a linear estimation
(linear) [
Based on the study, the results were obtained that make it possible to improve the toolkit for using the nonlinear ARX model in the form of a multilevel neural network and a linear assessment for parametric identification of the oxygen content characteristic in the exhaust gases from the thermal load of the ASB, which makes it possible to display the process under study with a degree of adequacy equal to 95%, and use the obtained model of a high degree of adequacy for analyzing the process of the appearance of oxygen corrosion in the equipment of a ship's boiler.