Economy Mode Setting Device for Wind-Diesel Power Plants

: The article is devoted to the problem of reducing fuel consumption in a diesel generator set (DGS) as a part of a wind-diesel power plant (WDPP). The object of the research is a variable speed DGS. The goal is to develop the WDPP intelligent control system, providing an optimal shaft speed of an internal combustion engine (ICE). The basis of the intelligent control system is an economy mode setting device (EMSD), which controls the fuel supply to the ICE. The functional chart of EMSD has been presented. The main EMSD blocks contain a controller and an associative memory block. The associative memory block is a software model of an artiﬁcial neural network that determines the optimal shaft speed of the ICE. An algorithm for the WDPP intelligent control system has been developed and tested using the WDPP Simulink model. The EMSD prototype has been created, and its research has been conducted. Dependences of the change in speciﬁc and absolute fuel consumption on the load power have been obtained for two 4 kW DGS: with constant rotation speed and variable rotation speed DGS with EMSD. It has been established that the use of EMSD in the mode of low loads allow one to reduce fuel consumption by almost 30%. The error in determining the optimal engine speed using EMSD prototype is not more than 15%.


Introduction
Wind-diesel power plants (WDPP) are widely used for power autonomous consumers all round the world [1]. The main components of the WDPP are a wind turbine and a diesel generator set (DGS). DGS provides guaranteed power supply to consumers, and the wind turbine can reduce fuel consumption and harmful emissions into the environment [2]. However, commercially available WDPP have certain disadvantages like an excessive diesel fuel consumption during DGS operation, low efficiency, an incomplete use of a wind turbine at low wind speeds [3].
The studies aimed at solving these problems can be divided into two main groups: • Optimization of the WDPP structure • Controlling WDPP operating modes and ensuring its stability The first group includes studies devoted to determining an optimal quantity and capacity of the power units in the system of the WDPP. The article [4] proposes a specific methodology for substantiating the parameters, placing and evaluating the WDPP effectiveness based on a multi-level system assessment of the resource potential and climate information. The article [5] considers the influence of the share of the wind turbine generator capacity in the installed capacity in the WDPP and the composition of consumers on the amount of replacement of the DGS energy output. The article [6]   EMSD is an adaptive controller that uses the principle of artificial neural networks. EMSD consists of four main elements: a management controller, a learning controller, an associative memory block, and nonvolatile memory. A common information bus, providing bi-directional exchange of information and expansion of the system in the case of controlling multiple ICEs, combines the elements.
EMSD optimizes the speed of the engine shaft by controlling the ICE fuel supply depending on the internal and external working conditions. The control signal is generated using the associative memory block, which is a software model of an artificial neural network. The optimal ICE shaft rotation speed is determined in real time. The principle of EMSD operation is as follows.
The sensors record the following parameters: DGS power (P), ICE fuel consumption (g e ), and ICE shaft speed (ω). These parameters are transferred to the learning controller, where they are compared with similar parameters available in the data memory block.
If the parameters of the current mode match to the parameters of one of the modes stored in the data memory block, the learning controller reads information about the position of the fuel pump rail (value h) and transmits it to the management controller. The management controller forms a control signal in the ICE fuel supply system. As a result, the fuel pump rail moves in the right direction by the required number of steps.
If the input data of the learning controller does not match to any mode stored in the data memory block, then the learning controller performs several learning cycles of the memory block. After training, a new optimal position of the fuel pump rail is formed, which is transmitted to the management controller. Additionally, parameters of a new mode are formed from the input (P, g e, and ω) and output (h) parameters, which are stored in the data memory block.
A block scheme of the neural network of the EMSD associative memory block is presented in Figure 2.
The neural network of the associative memory block consists of three layers: input, hidden, and output. The input layer distributes the input data. The hidden layer consists of two neurons whose activation function is hyperbolic tangent. The signals of the neurons of the hidden layer are displayed linearly in the output layer. The output layer sums up the signals of the hidden layer and normalizes them. Neural network learning is implemented according to the algorithm of back error propagation, and it is realized by the learning controller.

121
The neural network of the associative memory block consists of three layers: input, hidden, and 122 output. The input layer distributes the input data. The hidden layer consists of two neurons whose  (1) xi2 = th(xi1 · Wi1 + Bi); (2) where n ij x is the input i of the neuron (j) of the layer n;   At the beginning of learning process, the load power vector P load is applied to the inputs of the first layer of the network. At the outputs of the first layer, the following values are calculated: (1) where x n ij is the input i of the neuron (j) of the layer n; W (n) ij is the weight ratio of the corresponding neuron input; K norm is the normalization coefficient required to bring the load power value to the unit basis of the neural network; B i is the weight of additional input for the i-th neuron initialization; M is the number of neurons in the layer n; y n i( j−1) is the output i of the neuron (j − 1) of the layer n; P q is the component q of the input vector P; and ω q is the component q of the output speed array ω.
For the output layer and the hidden layer n, the error value (δ) and the weight change (∆W ij ) are calculated: where d j is an ideal (desired) state of the neuron and β is the learning coefficient. Next, the values of the network link weights are corrected: Energies 2020, 13, 1274

of 14
where t is the number of the learning era of the associative memory block. The value of the optimal rotation speed (ω opt ) is determined in the output: where x i1 and x i2 are the inputs of the i-th neurons of layers 1 and 2 of the neural network, respectively, and K norm is the normalization coefficient needed to reduce the load power value P to the unit basis of the neural network. The calculation continues until the network operation error (E ω ) becomes less than permissible. Network operation error is determined by: where ω t is an actual value of the rotation speed at the output of the neural network at the era t; ω opt is an optimal value of the rotation speed at the output of the learning controller; and p is the number of the learning era corresponding to the permissible error.
A necessary condition for the correct learning of the neural network is the presence of a learning set; that is, a set of logical pairs ω opt = f (P). A table array of speed values ω opt = f (P) for discrete values of power P is stored in a data memory block.

Algorithm of the WDPP Intelligent Control System
The WDPP intelligent control system consists of four modules:

•
A network learning management module • A module for determining the minimum fuel consumption of ICE • A control module for the optimal position of the ICE fuel pump rail • A module for managing associative memory and data memory of the neural network A block-chart of the algorithm of the WDPP intelligent control system is shown in Figure 3.

172
Based on the artificial neural network of the associative memory block, the optimal value of the 173 ICE speed is determined. In accordance with the optimal value of the ICE speed, a control signal for 174 the fuel pump rail h is generated. This provides an efficient mode of WDPP operation. The input data of the algorithm are an actual value of the DGS power (P DGS ), an ICE fuel consumption (g e ), a rotational speed of the ICE shaft (ω ICE ), and an actual value of the output power of the wind generator (P w ).
The management controller monitors change in the value of the load power during the WDPP operation. If the load power changes by more than a predetermined value P step , then a new value P new is fixed. Further, the management controller determines that the new power value belongs to the range of values stored in the associative memory block.
Based on the artificial neural network of the associative memory block, the optimal value of the ICE speed is determined. In accordance with the optimal value of the ICE speed, a control signal for the fuel pump rail h is generated. This provides an efficient mode of WDPP operation. The developed algorithm significantly expands the limits of the efficient operation of the WDPP and makes the intelligent control system adaptive towards to the external and internal operating conditions of the ICE.

Simulation
A Simulink model of the WDPP has been created to test the developed algorithm. The Simulink model allows one to study the static and dynamic modes of WDPP taking into account the mutual influence of WDPP units on each other.
The model is built on the basis of differential equations describing the functioning of the main units of the WDPP: • A wind turbine, consisting of a turbine and a synchronous generator • DGS, consisting of a diesel engine and a synchronous generator • Energy storage systems • Three-phase active-inductive load The block scheme of the Simulink model is in Figure 4.
where Uabc is the stator voltage vector;

197
Ψabc is the stator flux distribution vector;

198
Rabc is the stator windings resistance vector;

199
Iabc is the stator current vector;

200
HS is the matrix of independent circuits [17]; Mathematical models have been developed to simulate the operation of each WDPP unit. The models are differential equations that describe the processes of change of physical quantities such as current, voltage, frequency, torque, etc.
So, the operation of a synchronous generator as part of a wind turbine and DGS is described by a system of equations based on the law of electromagnetic induction and Ohm's law [16]: where U abc is the stator voltage vector; Ψ abc is the stator flux distribution vector; R abc is the stator windings resistance vector; I abc is the stator current vector; H s is the matrix of independent circuits [17]; U x is the vector of the voltage across stator clamps; Energies 2020, 13, 1274 8 of 14 ψ f is the excitation winding flux distribution; e f is the electromotive force of the excitation winding; r f is the resistance of the excitation winding; and i f is the current of the excitation winding.
Differential equations allow one to determine the phase currents and voltages for static and dynamic modes of operation based on the initial parameters for a specific model of a synchronous generator. The detailed descriptions of the mathematical models of WDPP units and their study results are given in [16].
A system of matrix differential equations has been developed to study WDPP operation modes, taking into account the mutual influence of WDPP units on each other: Vectors Y with corresponding indices represent variable values describing the operation of the diesel engine (D), the synchronous generator (G), the wind turbine (W), and the load (E). Vectors Y are determined for each WDPP unit with appropriate mathematical models. So, the variables for the synchronous generator are the stator voltage and current vectors, which are determined by Equations (10). The variables for the diesel engine are the engine speed and the position of the fuel pump rail. The variables for the wind turbine are wind turbine power and torque.
All equations have been compiled in relative units that facilitate their joint integration. The basic values correspond to the nominal passport data of the elements [16].
The algorithm of the WDPP intelligent control system for WDPP consisting of a 4 kW wind turbine and a 4 kW diesel generator has been tested on the Simulink model. Modeling was carried out for relative values of the load power in the range from 0.00 to 1.00 in increments of 0.05. Figure 5 shows the obtained dependence of the ICE rotation speed on the load power P. results are given in [16].

210
A system of matrix differential equations has been developed to study WDPP operation modes,

217
The variables for the wind turbine are wind turbine power and torque.

218
All equations have been compiled in relative units that facilitate their joint integration. The basic 219 values correspond to the nominal passport data of the elements [16].

232
The research has shown that the developed algorithm of the WDPP intelligent control system 233 provides a minimum error in determining the optimal ICE shaft speed.  The research has shown that the developed algorithm of the WDPP intelligent control system provides a minimum error in determining the optimal ICE shaft speed.

Economy Mode Setting Device (EMSD) Prototype
An EMSD prototype has been developed. In addition to the learning and management controller, associative and data memory blocks (for generating a control signal), the prototype includes a block of parameters from measuring sensors (for obtaining initial data) and a stepper motor control unit (for changing the position of the fuel pump rail).
The appearance of the prototype is shown in Figure 6.

235
An EMSD prototype has been developed. In addition to the learning and management 236 controller, associative and data memory blocks (for generating a control signal), the prototype 237 includes a block of parameters from measuring sensors (for obtaining initial data) and a stepper 238 motor control unit (for changing the position of the fuel pump rail).

239
The appearance of the prototype is shown in Figure 6.

250
The appearance of the test bench is shown in Figure 7. The test bench consists of the following  Learning and management controllers are combined in one device (3), which is based on the TM4C123GH6 microcontroller. Its memory capacity is 256 kB, which makes it possible to store the control algorithm, process sensor signals, and also store intermediate data during the functioning of the neural network. To control the position of the fuel pump rail, a linear stepper motor (2) is used, which is connected to a programmable control unit (1). The control unit generates control pulses with a frequency of up to 10 kHz and an accuracy of 0.2%.

Experimental Studies and Discussion
The appearance of the test bench is shown in Figure 7. The test bench consists of the following parts: (1 and 2) DGS and wind turbine simulators based on an asynchronous variable frequency drive and an alternator, (3) a DGS based on a gasoline ICE and synchronous generator, (4) connecting terminals for connecting a wind turbine, (5) an accumulator battery, (6) a programmable logic controller,

267
The EMSD prototype is connected to the frequency converter as part of the DGS simulator.

268
Therefore, it is able to adjust the DGS speed. DGS simulator (1) and wind turbine simulator (2) are 269 used for testing and tuning the software of the WDPP intelligent control system.

270
The main purpose of the EMSD prototype is to control the DGS based on a gasoline ICE and Technical parameters of the test bench are shown in Table 1.

Parameter Value Units
Wind turbine rated power 4 kW DGS-rated power 4 kW Wind speed changes from 0 to 10 m/s Active load changes from 0 to 8 with step 0,1 kW The programmable logic controller (6) provides an overall control of the test bench. EMSD prototype is designed to control the operation of DGS simulator (1) and DGS with the gasoline ICE (3).
The EMSD prototype is connected to the frequency converter as part of the DGS simulator. Therefore, it is able to adjust the DGS speed. DGS simulator (1) and wind turbine simulator (2) are used for testing and tuning the software of the WDPP intelligent control system.
The main purpose of the EMSD prototype is to control the DGS based on a gasoline ICE and synchronous generator (3). The learning and management controller (9) as part of EMSD prototype receives information from the sensors about the DGS power, the ICE fuel consumption, and the rotation speed of the ICE shaft and determines the optimal value of the ICE speed according to the algorithm (Figure 3). Information is transmitted to a programmable control unit (10), which forms a control signal for a linear stepper motor. The linear stepper motor changes the position of the ICE fuel pump rail according to the optimal engine speed. The placement of the learning and management controller, the programmable control unit, and the linear stepper motor as the part of the EMSD prototype is shown in Figure 8.

284
A test bench has been used to study the parameters of the WDPP operation modes with EMSD.
285 Figure 9 shows the dependences of the specific ge (Figure 9 a) and absolute g (Figure 9

291
In Figure 9, the blue lines show the calculated characteristics, and the red lines show the 292 experimental characteristics. The obtained results show that the error in determining the optimal 293 engine speed using EMSD prototype is not more than 15%.  A test bench has been used to study the parameters of the WDPP operation modes with EMSD. Figure 9 shows the dependences of the specific g e (Figure 9a) and absolute g (Figure 9b) fuel consumption of 4 kW DGS with variable rotation speed. The results were obtained by calculation (using the multi-parameter characteristics of the ICE) and experimentally (using the EMSD prototype).
Energies 2020, 13, x FOR PEER REVIEW 3 of 15 management controller, the programmable control unit, and the linear stepper motor as the part of 278 the EMSD prototype is shown in Figure 8.

284
A test bench has been used to study the parameters of the WDPP operation modes with EMSD.
285 Figure 9 shows the dependences of the specific ge (Figure 9 a) and absolute g (Figure 9

291
In Figure 9, the blue lines show the calculated characteristics, and the red lines show the 292 experimental characteristics. The obtained results show that the error in determining the optimal 293 engine speed using EMSD prototype is not more than 15%.
294 Figure 9. Dependence of specific (a) and absolute (b) fuel consumption for DGS with variable rotation speed: (P) load power, (g e ) specific fuel consumption, (g) absolute fuel consumption.
In Figure 9, the blue lines show the calculated characteristics, and the red lines show the experimental characteristics. The obtained results show that the error in determining the optimal engine speed using EMSD prototype is not more than 15%. Figure 10 shows that the dependences of changes in specific ( Figure 10a) and absolute (Figure 10b) fuel consumptions on the load power have been studied for two types of 4 kW DGS: with a constant (blue lines) and variable speed when using EMSD (red lines).