RU2465716C1 - Device to control excitation of synchronous generator - Google Patents

Abstract

FIELD: electricity.

SUBSTANCE: device to control excitation of a synchronous generator comprises a metre of rotor current, a metre of stator current and voltage, a unit of feedback by excitation current, a metre of active power, a sensor of generator voltage frequency, a unit of analog-to-digital conversion, an operational unit, a unit of digital-to-analog conversion and a unit of parameters adaptation. The operational unit and the unit of parameters adaptation are arranged as specified in the application materials.

EFFECT: considerable reduction of value and duration of transition processes, higher speed of their suppression due to automatic turning of controller parameters as power system condition changes.

3 dwg

Description

The invention relates to a class of devices for controlling electric generators in order to obtain the required value of the output parameters, in particular for controlling the excitation of the generator in order to mitigate the harmful effects of overloads or transients, for example, when a load is suddenly connected, removed or changed, and can be used as part of equipment for controlling synchronous generators in enterprises generating electric energy.

At power generating enterprises, there are problems with ensuring the required level of quality and stability of supply of electric energy, which is confirmed by incidents and failures in the operation of power equipment. Achieve high quality indicators of electricity, increase efficiency and lowering the cost of putting equipment into operation is possible only if the stability of operation and the risk of breakdown of power plants, power plants and power systems as a whole are reduced.

As studies have shown, one of the main reasons for the violation of the stability of electricity supply and a decrease in its quality on a local and global scale is the electromechanical transients that occur in electric generators due to local disturbances in energy systems. The electrical characteristics of modern energy systems are in the process of constant fluctuation: new consumers are introduced or permanent consumers are switched off, capacities are transferred to adjacent networks, capacities are introduced and local accidents occur.

The stability and efficiency of the use of equipment of power generating enterprises, as well as the quality of the generated electric energy, are largely determined by the perfection of the excitation control systems of synchronous generators generating electricity.

The magnitude and duration of transients in a synchronous generator is determined by the accuracy of tuning the parameters (coefficients) of the excitation controller.

In existing control systems for the excitation of synchronous generators, the optimization of the parameters of the excitation controller is based on the state of the power system at the time of tuning. When changing the electrical characteristics of energy systems, the parameters of the excitation controller lose their optimality, the parameters of transients worsen, the range of regulation stability is narrowed down to a complete loss of stability and the need to stop equipment.

The present invention is aimed at solving the problem of optimizing the characteristics of transients occurring in electric generators by adapting the parameters of the excitation controller to the current state of the power system.

A device is known "Excitation controller of a synchronous generator" [1] for the same purpose as the present invention, comprising a voltage sensor, analog-to-digital converter, fuzzy controller, digital-to-analog converter, power amplifier, reactive current sensor, averaging unit, digital signal adder and neural network a controller trained to provide the necessary signals depending on the input values to control the excitation of the generator with uneven loading of parallel-running generators, p In this case, the voltage sensor and the reactive current sensor are connected to the output terminals of the synchronous generator, the outputs of the voltage sensor through an analog-to-digital converter are connected to the inputs of the fuzzy controller, the output of the reactive current sensor is connected to one of the inputs of the reactive current averaging unit and to the first input of the neural network controller, the second the input of the neural network controller is connected to the output of the averaging block of reactive currents, the outputs of the neural network and fuzzy controllers are connected to the inputs of the adder of digital signals, the output of which is connected to the input of a power amplifier through a digital-to-analog converter, and its output is connected to the terminals of the excitation winding of a synchronous generator.

The disadvantage of this device is the increased duration and magnitude of transients when changing the parameters of the power system due to the lack of adaptation of the parameters of the voltage regulation channel (fuzzy controller) to the parameters of the synchronous generator and power system. The second disadvantage of this device is the low stability of the excitation mode of a synchronous generator due to the absence of a control channel in its composition with respect to the derivative of the change in the excitation winding current.

A device "Automatic voltage regulator" ([2], p. 93, Fig. 4.15) is known, which contains a rotor current meter with two inputs for supplying generator current and voltage, a stator current and voltage meter with inputs combined with the same inputs of the rotor current meter , and a feedback block for the excitation current connected to the operating unit, including the setpoint shaper, the output of which is connected to the first input of the first adder, the second input of which serves to supply a signal for the voltage of the current and voltage meters the stator, the output of the specified adder is connected to the first input of the integrator, the first input of the second adder and to the input of the voltage differentiator, and the second input of the integrator serves to supply a signal for the current meter of the current and voltage of the stator, the output of the integrator is connected to the second input of the second adder, the output of which is connected to the input of the control voltage forcing unit, the output of this unit is connected to the first input of the amplifier-adder, to the second and third inputs of which the outputs of the second voltage and voltage differentiator, the fourth input of the amplifier-adder is used to supply a signal from the output of the feedback block on the excitation current, to the fifth input is the output of the current differentiator, the input of which is used to supply the signal of the rotor current meter to the operation unit, and the output of the amplifier-adder is the output of the operating unit.

By the similarity of most features, this device is taken as a prototype.

The disadvantage of the prototype is the deterioration of the regulation parameters of the excitation of the synchronous generator when the deviation of the parameters of the power system from the original values at which the controller was tuned; the task of adapting the parameters of the regulator to the parameters of the generator and the power system is not posed or solved.

The essence of the invention is a device for controlling the excitation of a synchronous generator, adapting to the parameters of the power system, operating on the basis of the method of neural network identification and used for a wide class of synchronous electric generators.

A comparable analysis with the prototype shows that the proposed device differs from the prototype in that the elements of the operating unit are digital, the integrator, current differentiator and voltage differentiator of the operating unit additionally each contain two parametric inputs for supplying time constant and gain signals, and the device additionally contains active power meter, including two inputs for supplying current and voltage, combined with the same inputs of the rotor current meter, generator voltage frequency counter, analog-to-digital conversion unit, the inputs of which are connected to the outputs of the rotor current meter, stator current and voltage meter, feedback unit and active power meter, digital-to-analog conversion unit, the input of which is connected to the output of the operating unit, and the output is the output devices, and a parameter adaptation unit with six inputs, the first output of the analog-to-digital conversion unit connected to the input of the current differentiator of the operating unit and to the first input parameter adaptation unit, the second output is to the second input of the first adder of the operation unit and to the second input of the parameter adaptation unit, the third output is to the second input of the integrator of the operation unit and to the third input of the parameter adaptation unit, the fourth output is to the fourth input of the operational unit adder-adder and the fifth output is to the fifth input of the parameter adaptation unit, the output of the generator voltage frequency sensor and the output of the operation unit are connected respectively to the fourth and sixth inputs of the pair adaptation unit meters, and the parameter adaptation block is made in the form of a delay block with six inputs and six outputs, a neural network identification block with nine signal inputs and an input for supplying weight coefficients and with three outputs, a coefficient memory block, a neural network prediction error calculation block with three reference inputs and three signal inputs, a weight correction block with ten signal inputs, an input for supplying weight coefficients and a resolution input, a controller coefficient calculation block with six outputs, an averaging unit with six inputs and six outputs, and a transient registration unit, the first, second, and third signal inputs of the neural network prediction error calculation unit being the first, third, and fifth inputs of the parameter adaptation block, outputs of the first to sixth delay units connected to the same signal inputs of the neural network identification unit and the weight correction block, and the outputs of the first to third neural network identification unit are connected to the same name to the input inputs of the delay unit and the same reference inputs of the neural network prediction error calculation unit, the fourth and sixth inputs of the delay unit are respectively combined with the seventh and ninth signal inputs of the neural network identification unit and the weighting correction block and are respectively the fourth and sixth inputs of the parameter adaptation block, the fifth input of the block delays combined with the eighth signal inputs of the neural network identification block and the correction block of the weight coefficients and the input of the register block and the transient is the second input of the parameter adaptation block, the output of the transient registration block is connected to the resolution input of the weighting correction block, the output of the neural network prediction error calculation block is connected to the tenth signal input of the weighting correction block, the output of which is connected to the input of the coefficient memory block, the output of which is connected to the input of the unit for calculating the coefficient of the regulator, the inputs for supplying the weight coefficients of the block correction weight coefficients sources and the neural network identification block, the outputs of the first to sixth coefficients of the controller coefficient calculation unit are connected to the inputs of the homogenization unit of the same name, the output pairs of which the first and second, third, fourth, fifth and sixth are connected to the parametric inputs for supplying the time constant and gain signals, respectively, of the integrator , a current differentiator and a voltage differentiator of the operating unit, and the neural network identification unit is made in the form of a multilayer perceptron, the calculation unit about the errors of the neural network prediction are in the form of a calculator of the vector difference between the signals at the reference and signal inputs, and the weight correction block is in the form of a processor device operating according to the algorithm for the back propagation of learning errors.

Achievable technical result consists in a significant reduction in the magnitude and duration of transients, as well as an increase in the rate of their extinction due to automatic adjustment of the regulator's parameters when the state of the power system changes.

Therefore, the device meets the criterion of "novelty."

Comparison with other technical solutions shows that the proposed device has features that allow the adaptation of the parameters of the operating unit (controller) to the parameters of the power grid through continuous training during operation. The operating unit at each moment of time has parameters that are close to optimal in terms of the magnitude and duration of the transition processes occurring in the power system.

The invention is illustrated by the following drawings, where:

figure 1 is a functional diagram of a device for controlling the excitation of a synchronous generator;

figure 2 is a functional diagram of an operating unit;

figure 3 is a functional block diagram of the adaptation of parameters.

The device for controlling the excitation of a synchronous generator (Fig. 1) contains a rotor current meter 1 with two inputs, which are inputs of a device for supplying generator current and voltage, a stator current and voltage meter 2 with inputs combined with the inputs of the rotor 1 current meter of the same name, and two outputs , feedback block on the excitation current 3 with an input being the input of the device for supplying the generator excitation current, an active power meter 4 with two inputs combined with the inputs of the roto current meter a 1, a generator frequency voltage sensor 5 with an input combined with an input of a device for supplying a generator voltage, an analog-to-digital conversion unit 6 with five inputs and five outputs, an operation unit 7, a digital-to-analog conversion unit 8 and a parameter adaptation unit 9 with six inputs and six outputs, the first input of the analog-to-digital conversion unit 6 is connected to the output of the current meter of the rotor 1, the second and third inputs to the first and second outputs of the current and voltage meter of the stator 2, the fourth input to the output of the inverse unit 3, and the fifth input is to the output of the active power meter 4, the input of the digital-to-analog conversion unit 8 is connected to the output of the operating unit 7, and the output is the output of the device, and the first output of the analog-to-digital conversion unit 6 is connected to the input of the current differentiator of the operating unit 7 and to the first input of the adaptation block of parameters 9, the second output to the second input of the first adder of the operating unit 7 and to the second input of the adaptation block of parameters 9, the third output to the second input of the integrator of the operating unit 7 to the third input of the adaptation block of parameters 9, the fourth output to the fourth input of the adder-amplifier of the operating unit 7, and the fifth output to the fifth input of the adaptation block of parameters 9, the output of the voltage frequency generator of generator 5 and the output of the operating unit 7 are connected to the fourth and sixth, respectively the inputs of the adaptation block of parameters 9, the pairs of outputs of the adaptation block of parameters 9, the first and second, third and fourth, fifth and sixth are connected to the parametric inputs for supplying signals of a time constant and gain with tvetstvenno integrator and differentiator differentiator current voltage operating unit.

A device for controlling the excitation of a synchronous generator operates as follows.

The meter 1 generates an analog signal I _f proportional to the rotor current of the synchronous generator. At its input, sinusoidal signals of linear voltage and stator current are received [2]. The current and voltage meter of the stator 2 generates analog signals U _f and I _R , respectively proportional to the voltage and reactive component of the stator current of the synchronous generator [2]. Feedback block 3 generates an analog signal I _OC proportional to the exciter current for realizing hard feedback, and also performs the functions of galvanic isolation of the exciter circuit of the generator and device [2]. The active power meter 4 generates an analog signal P proportional to the active power generated by the generator. The sensor 5 is designed to measure the instantaneous frequency of the voltage of the generator and generates a digital signal f proportional to the specified value.

The signals from the rotor current meter 1, the current meter and stator voltage 2, the excitation current feedback block 3, and the active power meter 4 (respectively, I _f , U _f , I _R , I _OC , P) are fed to the analog-to-digital conversion unit 6 where they are converted to digital form (respectively, to IF, UG, IR, IOC, PP signals). Next, the digital signals of the rotor current IF, voltage UG, the reactive component of the stator current IR and the excitation current of the pathogen IOC from the output of the analog-to-digital conversion unit 6 are supplied to the operation unit 7, which controls the stator voltage, to change the first derivative of the rotor current and to change the first derivative of the voltage of the stator of the generator with internal control parameters specified by the adaptation unit 9, and also generates a digital signal of the excitation voltage UF, which is converted into the log form by the digital-to-analog conversion unit 8, from the output of which an analog signal U _f is supplied to the exciter of the synchronous generator.

The digital signals of the rotor current IF, voltage UG, the reactive component of the stator current IR, signals of the active power of the PP generator and the instantaneous voltage frequency f, as well as the output signal of the operation unit 7 (UF) are supplied to the adaptation unit 9, which, based on their neural network processing, produces computation of control parameters of the operation unit 7: the time constant T _U and the gain K _U for regulating the stator voltage, the time constant and T _i1 K _i1 gain adjustment for changing the first pro a rotor current, time constant T _U1 and the gain K _U1 regulation to change the first derivative of the stator voltage.

The regulation parameters of the operating unit 7 (T _U , K _U , T _i1 , K _i1 , T _U1 , K _U1 ) are continuously adjusted based on the criterion of the minimum value and duration of transients through neural network identification. Thus, the operating unit 7 contains regulation parameters close to optimal for the current state of the power system and provides the minimum duration and magnitude of the electromechanical transients occurring in a controlled synchronous generator.

The rotor current meter 1 is a known device and can be implemented, for example, as a rotor current sensor described in [2], p. 96 and including a voltage filter containing an input for supplying voltage to the generator, a current filter containing an input for supplying generator current third an adder, a rectifier, a nonlinear block, a product block, a fourth adder and a frequency-dependent filter, the output of which is the output of the rotor current meter, connected in series with the output of the voltage filter connected to the first input of the third Matora, the output of which is connected to the second input work unit and to an input of the rectifier, a current filter output is connected to the second input of the third adder and the second input of the fourth adder.

The current and voltage meter of the stator 2 is a known device and can be implemented, for example, as a measuring unit described in [2], p. 95, and includes three synchronous filters, the signal inputs of which are combined with the input for supplying the input voltage of the current and voltage meter a stator, a phase-sensitive rectifier, the signal input of which is an input for supplying the input current of a current meter and a stator voltage, three comparators, the signal inputs of each of which are connected to sources of three phase voltages, about the pore inputs are to the sources of zero signals, and the outputs are to the control inputs of each of the synchronous filters, the outputs of which are connected to the inputs of the fifth adder, the output of which is the output of the voltage and current meter of the stator, the output of the first comparator connected to the control input of the phase-sensitive rectifier, the output of which is the output of the current meter and stator voltage current.

The excitation current feedback block 3 is a known device and can be implemented, for example, as a feedback block described in [2], p. 97, and including a generator and a series-connected modulator, the signal input of which is the input of the current feedback block excitation, amplifier, demodulator and filter, the output of which is the output of the feedback unit for the excitation current, the output of the generator is connected to the reference inputs of the modulator and demodulator.

The active power meter 4 is a known device and can be implemented, for example, as a device containing a series-connected vector multiplier, the first input of which is an input of an active power meter for supplying a generator current, the second input of which is an input of an active power meter for supplying a generator voltage, and an integrator whose output is the output of an active power meter.

The voltage meter of the generator 5 is a known device and can consist, for example, of a 90 degree phase shifter, the input of which is the input of the voltage frequency meter and is combined with the input of the first differentiator and the second input of the second multiplier, the output of which is connected to the first input of the first multiplier, the output of which connected to the inverting input of the adder, and the output of the phase shifter is connected to the second input of the first multiplier and the input of the second differentiator, the output of which is connected to the first input to the second multiplier, the output of which is connected to the non-inverting input of the adder, the output of which is the output of the voltage frequency meter.

The digital-to-analog conversion unit 8 is a known device.

The analog-to-digital conversion unit 6 includes five identical analog-to-digital converters, which are known devices and operating in parallel with the same sampling frequency, the input of the Qth analog-to-digital converter is the Qth input, and the output is the Qth output of the analog-to-digital block transformations.

The operating unit of the device (figure 2) is digital and is designed to generate a control signal for the excitation of a synchronous generator by processing signals from the current meter of the rotor 1, the current meter and voltage of the stator 2, the feedback unit 3 using the parameters determined by block 9, and consists of a shaper settings 10, the first adder 11 containing the first and second inputs, an integrator 12 containing the first and second signal and first and second parametric inputs, the second adder 13, containing the first and second inputs, a voltage differentiator 14 and a current differentiator 15, each containing a signal and first and second parametric inputs, a boost unit 16 and an adder amplifier 17 containing first to fifth inputs, the signal input of the current differentiator 15 is the first input of the operation unit 7, the setpoint shaper 10 is connected to the first input of the first adder 11, the second input of which is the second input of the operation unit 7, and the output of the first adder 11 is connected to the first signal input of the integrator 12, the first in by the second adder 13 and the signal input of the voltage differentiator 14, the second signal input of the integrator 12 is the third input of the operating unit 7, and the output of the integrator 12 is connected to the second input of the second adder 13, the output of which is connected to the input of the boost unit 16 and to the first input of the amplifier-adder 17, the outputs of the blocks: boost 16, the voltage differentiator 14 and the current differentiator 15 are respectively connected to the second, third and fourth inputs of the amplifier-adder 17, the fifth input of the amplifier-adder 17 is a fourth m input, the first and second parametric inputs of the integrator 12 are respectively the sixth and fifth, the first and second parametric inputs of the current differentiator 15, respectively, the seventh and eighth, the first and second parametric inputs of the voltage differentiator 14, respectively, the ninth and tenth inputs of the operating unit 7, and the output amplifier-adder - the output of the operating unit 7.

The operation unit 7 (figure 2) works as follows. Digital signals of the rotor current IF, voltage UG, the reactive component of the stator current IR and the excitation current of the exciter IOC, generated by the analog-to-digital conversion unit 6. The setpoint generator 10 generates the voltage setpoint of the generator stator. The first adder 11 performs algebraic addition of the setpoint signal with the stator voltage signal UG [2]. A voltage proportional to the deviation of the generator voltage UG from the setpoint is generated at the output of the adder, which is then fed to a group of devices consisting of a second adder 13, an integrator 12 and a voltage differentiator 14 that implements a standard widely known PID controller for voltage deviation and for changing the first derivative voltage the stator [2].

The regulation of the change in the first derivative of the rotor current is carried out by means of a current differentiator 15. Force unit 16 is designed to generate excitation voltage during accidents in the power system that cause a decrease in the voltage on the generator buses [2]. When decreasing the voltage of the stator UG of the generator relative to the set value, the boost unit 16 generates a boost signal independently of other channels. Amplifier-adder 17 performs the function of forming a total control channel and large-scale amplification of the excitation signal UF.

Integrator 12, differentiators: voltages 14 of current 15 (unlike similar nodes of similar purpose for the prototype) have tunable control parameters - gain and time constants, respectively (T _U , K _U ) - for integrator 12, (T _i1 , K _i1 ) - for the current differentiator 15, (T _U1 , K _U1 ) - for the voltage differentiator 14. Regulation parameters are generated in the adaptation block of parameters 9 (Fig. 1).

The setpoint generator 10, the first adder 11, the second adder 13, the boost unit 16 and the amplifier-adder 17 are standard devices and can be performed by analogy with the devices of the same name described in

[2], p. 97, 98, and performing similar functions.

Integrator 12 is a standard radio link - a filter with controlled coefficients and can be implemented, for example, as a recursive digital filter with the equation y _U [n] = a ₁ (x _U [n] + x _U [n-1]) - b ₁ y _U [n-1], where x _U [n], y _U [n] are the input and output signals of the filter at the nth step; T _S - signal sampling period; a ₁ , b ₁ - filter coefficients calculated by the formulas: a ₁ = K _U T _S / (T _S + 2T _U ); b ₁ = (T _S -2T _U ) / (T _S + 2T _U ) depending on the regulation parameters that are generated in the adaptation block of parameters 9; Ts is the sampling period of quantities.

Differentiators: voltages 14 of current 15 are made in the same way, they are standard digital radio engineering devices - filters with controlled coefficients and can be implemented, for example, in the form of recursive digital filters with the equation: y _I1 , _U1 [n] = a ₀₂ (x _{I1, U1} [ n] -x _{I1, U1} [n-1]) - b ₁₂ y _{I1, U1} [n-1], where x _{I1, U1} [n], y _{I1, U1} [n] are the input and output signals of the filter by n m step; a ₀₂ = (2K _{I1, U1} T _S ) / (T _S + 2T _{I1, U1} ), b ₁₂ = (T _S -2T _{I1, U1} ) / (T _S + 2T _{I1, U1} ) are the filter coefficients calculated in depending on the regulation parameters that are generated in the adaptation block of parameters 9.

The adaptation block of parameters 9 (Fig. 3) consists of a delay block 18 containing inputs and outputs from the first to the sixth, a neural network identification block 19 containing signal inputs from the first to the ninth and an input for supplying weight coefficients, outputs from the first to the third, block the coefficients memory 20, the neural network prediction error calculation block 21, containing the first to third reference inputs and the first to third signal inputs, the weight coefficient correction block 22, which contains the first to tenth signal inputs, an input for supplying weight coefficients and the permission input, the coefficient calculation block of the controller 23, containing the first to sixth outputs, the averaging block 24, the first to sixth inputs and outputs, and the transient recording unit 25, the signal first, second and third inputs of the neural network error calculation block predictions 21 are, respectively, the first, third and fifth inputs of the adaptation block of parameters 9, the outputs of the delay unit 18 from the first to the sixth are connected to the same signal inputs of the neural network identification block 19 the weighting correction block 22, and the outputs of the neural network identification unit 19 from the first to the third are connected to the same inputs of the delay unit 18 and the same reference inputs of the neural network prediction error calculation unit 21, the fourth and sixth inputs of the delay unit 18 are combined with the seventh and ninth signal inputs of the block neural network identification 19 and the correction block of the weighting factors 22 and, respectively, are the fourth and sixth inputs of the adaptation block of parameters 9, the fifth input of the delay block 18 is combined with the eight signal inputs of the neural network identification block 19 and the weighting correction block 22, the input of the transient registration unit 25 and is the second input of the adaptation unit 9, the output of the transient registration unit 25 is connected to the resolution input of the weighting correction block 22, the output of the block computing errors of the neural network prediction 21 is connected to the tenth signal input of the weighting correction block 22, the output of which is connected to the input of the coefficient memory block 20, the output of which is connected to the input of the coefficient calculation block of the regulator 23, the input for supplying the weight coefficients of the weight correction block 22 and the input for supplying the weight coefficients of the neural network identification block 19, the outputs of the coefficient calculation block 23 of the controller from the first to the sixth are connected to the inputs of the same name averaging 24, the outputs of which from the first to the sixth are the same outputs of the adaptation block of parameters 9.

The adaptation block of parameters 9 operates on the principle of neural network identification as follows. The neural network identification unit 19, the delay unit 18, the neural network prediction error calculation unit 21, the weight coefficient correction unit 22, and the coefficient memory unit 20 form a neural network identification system for controlling a dynamic object with external inputs described in [3], pp. 108-110. The neural network identification unit 19, together with the delay unit 18, is a parametric model of a dynamic object including a synchronous generator, a primary motor, an excitation system, and a power grid. The model parameters are the weights of the neural network W [n].

The sequence of operations in the block adaptation of the parameters 9 is as follows (figure 3).

The inputs of the adaptation block of parameters 9 are the following data:

on the first - a discrete sample of the signal evaluation of the rotor current of the generator IF '[n];

on the third - a discrete count of the signal evaluation of the reactive component of the stator current IR '[n];

on the fifth, a delayed discrete sample of the signal of estimating the active power of the generator PP '[n];

on the fourth - a discrete sample of the generator voltage frequency signal f = f [n];

on the second - a discrete readout of the voltage signal of the stator UG-UG [n]; on the sixth, a discrete sample of the generator excitation voltage signal UF = UF [n], n is the measure number.

The signal inputs of the neural network identification block 19 are fed with digital input signals forming the input vector of the neural network X, and

X = {IF '[n-1], IR' [n-1], PP '[n-1], f [n-1], UG [n-1], UF [n-1], f [ n], UG [n], UF [n]}, where

IF '[n-1] - delayed discrete sample signal evaluation of the current of the rotor of the generator from the first output of the delay unit 18;

IR '[n-1] - delayed discrete sample signal evaluation of the reactive component of the stator current from the second output of the delay unit 18;

PP '[n-1] - delayed discrete sample signal evaluation of the active power of the generator from the third output of the delay unit 18;

f = f [n-1] is the delayed discrete sample of the generator voltage frequency signal from the fourth output of the delay unit 18;

UG = UG [n-1] - delayed discrete counting of the stator voltage signal from the fifth output of the delay unit 18;

UF = UF [n-1] is the delayed discrete sample of the generator excitation voltage signal from the sixth output of the delay unit 18.

The neural network identification unit 19 performs continuous calculation of discrete samples of signals: estimates of the generator rotor current IF '= IF' [n] at the first output, the reactive component of the stator current IR '= IR' [n] at the second output and the active power of the PP generator = PP '[n] - at the third output of the block, which is close in magnitude to the generator signals: the rotor current IF, the reactive component of the stator current IR and the assessment of the active power PP by calculating the response of a multilayer neural network with weight coefficients W [n] extracted from the memory block weight ratios nt 20.

Block 18 serves to delay the input data received at the signal inputs of the neural network identification unit from the first to the sixth by one clock cycle of the sampling frequency.

The output signals of the neural network identification unit 19 (IF '[n], IR' [n], PP '[n]) - in the calculation unit of the error of the neural network prediction 21 are compared with the actual values of the output signals of the generator (respectively, IF = IF [n] - the first signal input, IR = IR [n] - at the second signal input and PP = PP [n] - at the third signal input of the neural network prediction error calculation unit 21), at the output of which an error vector E (mismatch) of the latter with the input signals is generated, calculated by the neural network in block 19 and supplied to the reference inputs of the block Nia neural network prediction error 21: IF '[n] - at first, IR [n] - the second and RR' [n] - the third.

The error vector E then goes to the weight correction block 22, which simultaneously feeds the weight coefficients W [n] stored in block 20 and the input neural network vector X.

When registering the beginning of the transient voltage of the generator unit 25 at the output of the latter generates a logical signal Enable, designed to enable and start operations to correct the weight coefficients in the same unit 22. After the termination of the transient process, the Enable signal is removed.

The block of correction of weighting coefficients 22, in the presence of an Enable signal at the resolution input, changes the values of weighting factors W [n] in the direction of decreasing the value of the error vector E in proportion to its absolute value by means of linear computational operations embedded in it, which are an algorithm for training the neural network. After that, the newly calculated weighting coefficients W [n + 1] are supplied to the coefficient memory 20 and at the next iteration will be used to calculate the values IF '[n + 1], IR' [n + 1] and PP '[n + 1] neural network identification unit 19. Next, the process is continuously repeated for subsequent samples of the input signals IF, IR and PP. When the error vector E is close to zero, and in the absence of a transient process on the generator buses, the values of the weighting coefficients of the neural network W [n] remain constant.

The unit for calculating the coefficients of the controller 23 for each set of weight coefficients W [n] arriving at its input selects pre-calculated control parameters (T _U , K _U , T _i1 , K _i1 , T _U1 , K _U1 ) that are close to optimal and provide a minimum value and duration of transients.

,

,

,

,

,

), которые далее подаются на вход блока адаптации параметров 9.The averaging unit 24 is designed to suppress noise and sudden changes in control parameters (T _U , K _U , T _i1 , K _i1 , T _U1 , K _U1 ) by integrating each of them. The average values of the control parameters (

,

,

,

,

,

), which are then fed to the input of parameter adaptation block 9.

The delay unit 18 consists of a set of six identical digital registers, which are known devices.

The neural network identification block 19 is a multilayer perceptron — a well-known device described in [3], p. 26 with the number of layers of at least three.

The neural network prediction error calculation unit 21 performs component-wise subtraction of the signals at the signal inputs (IF [n], IR [n] and PP [n]) from the signals at the reference inputs (IF [n], IR '[n] and PP [n] ) and can be implemented as a known device - a vector subtractor. The error vector at the output of the device is E = {IF '[n] -IF [n]; IR '[n] -IR [n]; PP '[n] -PP [n]}.

The block of correction of weighting coefficients can be implemented in the form of a processor operating according to the well-known algorithm for the backward propagation of learning errors given in [2], p. 32, and in the accepted terms and notation of the present description, can be implemented as follows.

,

, причем

где -

,

- весовые коэффициенты первого, …, M-1-го (скрытого) и M-го (выходного) слоя многослойного персептрона соответственно; k=1, …, N_M, j=1, …, N_M-1, i=1, …, N_M-2, p=1, …, N₁, q=1, …, N; N_M - количество нейронов в M-м слое, N - количество входов нейросети; M - количество слоев нейросети.Step 1. Assign initial random values to

,

, and

where -

,

- the weights of the first, ..., M-1st (hidden) and Mth (output) layer of the multilayer perceptron, respectively; k = 1, ..., N _M , j = 1, ..., N _M-1 , i = 1, ..., N _M-2 , p = 1, ..., N ₁ , q = 1, ..., N; N _M is the number of neurons in the Mth layer, N is the number of neural network inputs; M is the number of layers of the neural network.

Step 2. Calculate the responses of the hidden layers of the neural network from the first to the M-1st:

, где E_k - компонент вектора ошибки нейросетевого предсказания E;

- k-й компонент M-го (выходного) вектора нейросети; F(x)=1/(1+exp(-x)) - активационная функция нейросети, одинаковая для всех нейронов M слоев; X_q - элементы входного вектора нейросети X.and the correction value of the weight coefficients of the output layer of the neural network:

where E _k is the component of the neural network prediction error vector E;

- k-th component of the M-th (output) vector of the neural network; F (x) = 1 / (1 + exp (-x)) is the activation function of the neural network, the same for all neurons of the M layers; X _q - elements of the input vector of the neural network X.

и откорректировать весовые коэффициенты выходного слоя нейросети:

, где η<1 - коэффициент скорости обучения; η<1 - фильтрующий коэффициент; n - номер итерации.Step 3. Calculate the correction values of the coefficients of the Mth layer of the neural network:

and adjust the weights of the output layer of the neural network:

where η <1 is the coefficient of learning speed; η <1 - filtering coefficient; n is the iteration number.

Step 4. Calculate the correction coefficient of the weight coefficients of the M-1st (hidden) layer of the neural network in accordance with the formula:

Step 5. Calculate the correction values:

and adjust the weights of the M-1st (hidden) layer of the neural network in accordance with the formula:

.Next, repeat steps 4 and 5 for all remaining layers of the neural network from the first to the M-2, using recurrence formulas (1) - (3) and taking into account that

.

Step 6. Go to step 2 for the next, n + 1th input vector.

The coefficient calculation block of the controller 23 can be implemented, for example, in the form of a permanent storage device, the input values of which are supplied with the values of the neural coefficients W [n], and the control parameters (T _U , K _U , T _i1 , K _i1 , T _U1 , K _U1 ), which are calculated in advance.

The averaging unit 24 can be implemented from a set of six identical digital integrators, which are well-known devices, with the inputs of the averaging unit one through six being the inputs, and the outputs the outputs of each of the integrators.

The transient registration unit can be implemented, for example, in the form of a series-connected differentiator, the input of which is the input of the unit, the device for calculating the module and the comparator, the output of which is the output of the unit, and the reference input of which is connected to the reference value source. The reference value source generates a constant positive signal, the value of which determines the sensitivity of the block. The block works as follows. When a transient occurs, the derivative of the generator stator voltage, calculated by the differentiator, will become non-zero and at the output of the module calculation device will take a positive value, which goes to the comparator, which serves to generate the enable enable signal and filter slow fluctuations of the stator voltage, which are not transients.

When changing the parameters of the power system, for example, the voltage or impedance of the power supply network, which occurs when connecting new or disconnecting existing consumers, short circuits, overvoltages, and other disturbances, the value of the error vector E will become nonzero due to the fact that, with the current parameters of the neural network, weights W [n] the parameters of the power system simulated by it will change. The discrepancy between the estimated signals of the neural network model and the operating signals taken from the generator will be compensated according to the described sequence of operations.

Thus, for each current state of the power system, weights can be determined W [n] containing information about the state of the power system for which the regulation parameters (T _U , K _U , T _i1 , K _i1 , T _U1 , K _U1 ) are calculated, close to optimal and providing a minimum value and duration of transients in the power system.

The process of correcting the weighting factors (training) of the multilayer perceptron in the neural network identification unit 19 and, therefore, occurs only during the action of even insignificant transients, by the signal Enable. In this case, the correction of the weight coefficients is made by small quantities proportional to the learning speed 77, as a result of which a gradual iterative refinement of the controller coefficients occurs, tracking slow changes in the parameters of the power system.

Therefore, in the proposed device, the parameters of the operating unit (controller) are adapted to the parameters of the energy network during operation, and the latter at each moment of time has parameters that are close to optimal in terms of the magnitude and duration of transients occurring in the energy system. Due to this, a significant decrease in the magnitude and duration of transients, as well as an increase in the rate of their extinction when the state of the power system changes, is achieved.

In the presented implementation of the synchronous generator excitation control device, the neural network identification unit 19 is a three-layer perceptron with the number of neurons in the first layer - 50, in the second layer - 50, in the third (output) layer - 9, there is a reduction in the magnitude and duration of the transient process when the network voltage changes in the form of a step function more than 2 times.

Information sources

1. Patent of the Russian Federation No. 65317 dated 03.20.2007 "Regulator of excitation of a synchronous generator."

2. Yurganov A.A., Kozhevnikov V.A. Regulation of the excitation of synchronous generators. - St. Petersburg: Nauka, 1996 .-- 138 p.

3. Shigeru Omata, Marzuki Khalid, Rubia Yusof. Neurocontrol and its applications. Book 2. Per. from English / Ed. A.I. Galushkina. V.A. Ptichkina. - M .: IPRZhR, 2000 .-- 272 p. (Neurocomputers and their application).

Claims (1)

A device for controlling the excitation of a synchronous generator, comprising a rotor current meter with two inputs for supplying generator current and voltage, a stator current and voltage meter with inputs combined with the inputs of the rotor current meter of the same name, an excitation current feedback unit and an operation unit including a setpoint shaper, the output of which is connected to the first input of the first adder, the second input of which serves to supply a signal for the voltage of the current meter and the stator voltage, the output of the specified sum pa is connected to the first input of the integrator, the first input of the second adder and to the input of the voltage differentiator, and the second input of the integrator serves to supply a signal for the current meter of the current and voltage of the stator, the output of the integrator is connected to the second input of the second adder, the output of which is connected to the input of the control unit voltage, the output of this unit is connected to the first input of the amplifier-adder, to the second and third inputs of which the outputs of the second adder and differentiator are respectively connected Ia, the fourth input of the amplifier-adder is used to supply a signal from the output of the feedback block according to the excitation current, the fifth input is the output of the current differentiator, the input of which serves to supply the signal of the rotor current meter to the operation unit, and the output of the amplifier-adder is the output of the operation unit characterized in that the elements of the operating unit are digital, the integrator, current differentiator and voltage differentiator of the operating unit additionally each contain two parametric inputs for signaling of time constant and gain, and the control device additionally contains an active power meter, including two inputs for supplying current and voltage, combined with the same inputs of the rotor current meter, a generator voltage frequency sensor, an analog-to-digital conversion unit, the inputs of which are connected to the meter outputs rotor current, current and voltage stator meter, feedback unit and active power meter, digital-to-analog conversion unit, the input of which is connected to the output o unit, and the output is the output of the device, and a parameter adaptation unit with six inputs, the first output of the analog-to-digital conversion unit connected to the input of the current differentiator of the operating unit and to the first input of the parameter adaptation unit, the second output to the second input of the first adder of the operation unit and to the second input of the parameter adaptation block, the third output to the second input of the integrator of the operating unit and to the third input of the parameter adaptation block, the fourth output to the fourth input of the amplifier -adder of the operating unit, and the fifth output is to the fifth input of the parameter adaptation unit, the output of the generator voltage frequency sensor and the output of the operating unit are connected respectively to the fourth and sixth inputs of the parameter adaptation unit, and the parameter adaptation unit is made in the form of a delay unit with six inputs and six outputs, a neural network identification unit with nine signal inputs and an input for supplying weighting factors and with three outputs, a coefficient memory unit, a neural network error calculation unit stories with three reference inputs and three signal inputs, a weighting correction block with ten signal inputs, a weight input and resolution input, a controller coefficient calculation block with six outputs, an averaging block with six inputs and six outputs, and a transient recording unit, moreover, the first, second, and third signal inputs of the neural network prediction error calculation unit are, respectively, the first, third, and fifth inputs of the parameter adaptation block The first to sixth delay blocks are connected to the same signal inputs of the neural network identification block and the weighting correction block, while the first to third outputs of the neural network identification block are connected to the same inputs of the delay block and the same reference inputs of the neural network prediction error calculation block, fourth and sixth inputs the delay unit, respectively, combined with the seventh and ninth signal inputs of the neural network identification unit and the correction block of the weight coefficients and, respectively They are the fourth and sixth inputs of the parameter adaptation block, the fifth input of the delay block is combined with the eight signal inputs of the neural network identification block and the weight correction block and the input of the transient registration block and is the second input of the parameter adaptation block, the output of the transient registration block is connected to the resolution input a weighting correction block, the output of the neural network prediction error calculation block is connected to the tenth signal input of the section of the weight coefficients, the output of which is connected to the input of the coefficient memory block, the output of which is connected to the input of the controller coefficient calculation unit, inputs for supplying the weight coefficients of the weight coefficient correction unit and the neural network identification unit, the outputs of the controller coefficient calculation unit from the first to the sixth are connected to the inputs of the same name an averaging unit, the output pairs of which the first and second, third and fourth, fifth and sixth are connected to the parametric inputs for supplying constant signals the time and gain, respectively, of the integrator, current differentiator and voltage differentiator of the operating unit, and the neural network identification unit is made in the form of a multilayer perceptron, the neural network prediction error calculation unit is in the form of a vector difference calculator between the signals at the reference and signal inputs, and the weight coefficient correction block - in the form of a processor device operating according to the algorithm for the back propagation of learning errors.

Images