EP4639706A1

EP4639706A1 - System and method for smoothing fluctuations in the power of renewable sources of electrical energy

Info

Publication number: EP4639706A1
Application number: EP22847161.1A
Authority: EP
Inventors: Milan HRONSKÝ
Original assignee: Ing Milan Hronsky Phd
Current assignee: Ing Milan Hronsky Phd
Priority date: 2022-12-21
Filing date: 2022-12-22
Publication date: 2025-10-29
Also published as: SK289320B6; SK500682022A3; WO2024136763A1

Abstract

A system for smoothing fluctuations in electric power in the grid from connected terminal devices with a random time course of active power, having a differencer (5) with an output terminal (A) for controlling the power of a bidirectional AC/DC inverter (3), and having at least one input terminal (Bj) of the measured active power pj(t) from the connected terminal devices (1j) by wattmeters (2j) and superimposed on a non-inverting (positive) input terminal of the differencer (5), while the input terminals (Bj) carrying the active power signals pj(t) are individually connected to multi -predictors (9j). The outputs of multi -predictors (9j) with signals approximating the future active power pj(t+ Δτi) are superimposed on the input terminal of the numerical model (6) of low-pass filter, whose output terminal is connected to the inverting (negative) input terminal of the differencer (5). Multipredictors (9j) have their input terminals (Cj) of meteorological data time series fj(t), where the positive integer index (j) is from a set {1,...m}.

Description

SYSTEM AND METHOD FOR SMOOTHING FLUCTUATIONS IN THE POWER OF RENEWABLE SOURCES OF ELECTRICAL ENERGY

Field of the invention

The invention is related to a system applying a corresponding method for smoothing fluctuations in the output power of renewable sources of electricity and other terminal devices connected to the grid (grid-termination devices), showing a random time course of power. The invention solves the problem of electric power generators being excited by renewable energy sources (hereafter RES) solar radiation or wind. Since their primary energy source has a random time course, the electrical output of these generators is unstable, subject to random time changes. The invention allows to filter or to smooth out the RES power fluctuations by accumulating their excess power and compensating it in case of shortage, loss-free (except for losses arising from transformation and accumulation), while minimizing the necessary accumulation of energy. The invention brings a profitable smoothing of RES performance at the current costs for production and accumulation of electrical energy. Its application will make it possible to better utilize the energy potential of the installed RES without endangering the stability of the grid during strong solar or wind intermittency. Smoothing the output power of RES will eventually enable a substantial increase of the RES share in the overall production of electricity. The invention falls into the field of energy.

Background of the invention

Due to the destabilizing impact of RES on the grid, the maximum RES power share is technically limited in the grid. In countries with a higher penetration of RES, it is possible to supply their electric power to the grid only with additional regulatory measures. Additional regulation of RES performance due to its intermittency can be done according to three principles:

The first principle regulates generators of other types in a complementary way, provided that RES power fluctuations are not too dynamic (gas or hydropower plants compensate for the intermittency of wind sources). The HV/LV transformers use tap changers to keep the low- voltage grid within tolerance. However, the voltage regulation by tap changers is limited by the maximum power of the transformer and low-voltage grid.

Another principle supplements the complementary regulation with a so-called curtailment: The grid operator remotely sets off-grid those RES whose power fluctuations exceed the possibilities of the complementary regulation. This additional measure causes that energy from renewable sources is not used to the maximum extent possible. The costs of regulatory measures of the first and second principles have a negative impact on the feed-in tariff for PV energy, which in turn reduces the interest of investors in new PV installations or increases financial losses from such investments into PV power plants in countries mandating to install RES on newly built buildings. All of this restricts the share of RES infeed in the grid which is why fossil and nuclear fuels still dominate in the production of electricity.

According to the third principle, the complementary regulation is accomplished by the accumulation and compensation of electrical energy by its two-way flow between the grid and the accumulator (battery energy storage system BESS). This allows to smooth the power fluctuations quickly enough and in theory loss-free (except for losses by AC/DC conversion and accumulation), while the grid’s complete equilibrium state is achieved by the complementary regulation of the "slower" energy sources using the first principle. The practical use of such a technically-most advantageous combined balancing of the grid’s equilibrium state is conditioned by the financial costs for the operation of accumulators, which at the time exceed the production cost of equivalent fossil or nuclear electricity. This cost is closely related to the social cost-assessment of 1 ton of CO2 emission, or radioactive waste. The price of the accumulation smoothing depends on the production technology of BESS and the amount of accumulation required to smooth out the RES performance. The specific amount of stored energy depends on two factors:

• intensity of RES power fluctuations and the respective level of smoothing requested,

• management of accumulation and compensation of electric power flowing between the accumulator and the grid.

In the near future, the price relationship between the average electricity production tariff and the tariff for its accumulation will very likely turn in favour of accumulation smoothing, using a low-pass filter (hereinafter LPF) driven by the RES performance predictor. The cost of accumulation smoothing is gradually reduced mainly by:

• technological development of redox accumulators, supercapacitors and hybrid supercapacitors and the related decrease in the price of BESS,

• development of LPFs achieving effective filtration of the RES output power with little need for energy accumulation,

• increasing accuracy of the RES power nowcasting up to /i hour advance by predictors using an artificial intelligence technology. Documents such as e. g. patent CN105552969A have been published in this field. The patent smooths out fluctuations in RES power with a technique like Fig.1 using an accumulator connected to the grid via a bidirectionally controlled inverter. The power of inverter in the AC— >DC direction is determined by a difference between the RES output power signal p(t) and the output signal s(t+At) of the LPF. Unlike Fig.1, however, the LPF is driven by the predicted power signal p(t+At). The accumulating-compensating energy flow between the grid and the accumulator smooths the output power of RES. Also, the relaxation time constant of the used Ist-order LPF and its corresponding optimal advance time interval At are being dynamically controlled, both according to the predicted intensity of RES power fluctuations, as well as regarding the current state of charge (SOC) of the accumulator. During weak intermittency of the primary energy source, the unnecessary accumulation of energy by the filter is thus limited due to a reduced error of the predicted power with less advance. During strong intermittency of solar irradiance, the LPF time constant and the necessary advance interval At are enlarged accordingly. Thus, the prediction error is increased, raising excessive energy accumulated by the filter.

Patent CN109995076A is also known in this field, which predicts the future development trend of solar irradiance, and at the same time it monitors SOC of the accumulator coupled with the grid by a bidirectionally controlled inverter. Based on the monitored conditions, the algorithm defines the accumulation or compensation power of the accumulator, which apparently approximates the function of a LPF driven by a predicted power of RES, while the cut-off frequency of LPF and the advance interval At are dynamically controlled.

Excessive accumulation by smoothing of photovoltaic power is also reduced by filters with shortened time-lag, e. g., moving average, moving median. Such filters reduce the required advance At of the predictor compared to the application of a standard LPF. By reducing the required advance interval, the prediction error is also reduced, ultimately reducing the excess accumulation by smoothing. Excess accumulation can also be effectively reduced by applying regression polynomials in the predictor function, calibrated based on recent history of the RES performance, or by a fuzzy -logic type predictor, driving the input of LPF. Under stable conditions, smoothing the RES output power by so-called zero group-delay filters (e.g., Kalman filter) eliminates the filter’s time-lag, so only the energy of short-term RES power fluctuations is accumulated. However, if the global trend of the mean RES power changes, then these filters will extremely deviate their output signal from the real future RES power. Such a temporary deviation increases the energy accumulation into such extent, that the filter eventually loses its advantage gained during the steady trend of the mean RES power.

Summary of the invention

Deficiencies in the state of the art are eliminated by the present invention which manifests itself that the accumulation smoothing is performed by the electric power flowing between the grid and accumulator, being controlled in such a manner that minimum energy is accumulated by the smoothing, given a limited accuracy of the predicted RES power.

The present invention is expressed by a system and by the corresponding method for smoothing fluctuations in electrical power in the grid from connected terminal devices with a random time course of active power according to the invention, having at least one input terminal with the active power signal pj(t) measured by wattmeters connecting the terminal devices with the grid, whereas the signals pj(t) are further superimposed as p(t) on a noninverting (positive) input terminal of a differencer (difference amplifier with a unit signal gain). In parallel, the signals pj(t) are individually coupled to multi -predictors estimating the future power signals pj(t+Aii) which are further superimposed into signals p(t+Ar,) exciting the numerical model of LPF. The multi -predictors are trained by the measured signals pj(t) and by time series of meteorological data fj(t). The LPF numerical model’s output is coupled to the inverting (negative) input terminal of the differencer. The positive integer index j is from a set { 1, ...m}.

When several terminal devices with a random time course of active power are connected to the system, a summator of the measured power signals drives the non-inverting (positive) input terminal of the differencer, and a vector summator superimposes the multi -predicted power signals on a multi-input terminal of the LPF numerical model.

At least one input terminal of the measured power signal pj(t) is coupled to the wattmeter of the grid-termination device with a random time course of active power, and at least one input terminal of the time series of meteorological data fj(t) is coupled to a meteorological sensor corresponding to the coupled wattmeter of the grid-termination device. The output terminal A of the differencer is coupled to the control input of the bidirectional AC/DC inverter which drives the electric power flowing between the accumulator and the grid.

According to the first aspect of the present invention, the grid-termination device with a random time course of active power is a photovoltaic power plant, whereas the meteorological sensor is a camera for capturing sky imagery from the Earth’s surface or from the Earth’s orbit. According to the second aspect of the invention, the grid-termination device with a random time course of active power is a wind power plant, while the meteorological data sensor is a combination of sensors relevant for prediction of the wind turbine power.

According to the third aspect of the invention, the grid-termination device with a random time course of active power is a network of electrical appliances, while a combination of meteorological data sensors and appliances’ status parameters provide the training data fj(t) relevant for prediction of the consumption of appliances.

According to the invention, the method for smoothing fluctuations in electrical power in the grid from connected terminal devices with a random time course of active power is functionally manifested in that the difference between the summed RES power p(t) and the output s(t+At) of the low-pass filter defines the inverter’s power flowing in AC->DC direction (or vice versa), where the advance At results from the LPF group delay in its pass-band frequency. According to the method, the non-inverting (positive) input of the differencer is driven by the signal p(t) superimposed from all active power signals pj(t) measured by wattmeters of the grid-termination devices indexed by j from a positive integer set { 1, ...m}, whereas the inverting (negative) input of the differencer is driven by the response s(t+At) of a numerical model of LPF to a set of n+1 input signals p(t+Ar,) superimposed from a set of m x (n+1) signals pj(t+Aii) estimated by multi -predictors. The integer index i is from a nonnegative set {0, ...n} and the relation 0 < Aii< At holds for each i, and it is assumed that the LPF response is calculated by its numerical model in a negligible time compared to the advance At. The multi -predictors are excited and trained by the measured random power pj(t) of the gridtermination devices with a random time course of active power and by the time series of meteorological data fj(t) relevant for the power of the corresponding terminal devices. The prediction error of each i-th signal component pft+Ari) is minimized with respect to its specific advance An by the state-of-art power prediction technology.

The numerical model of LPF can be preferably implemented in the spectral domain by a product of the input signal spectra and the frequency response of LPF, while the output signal of LPF is calculated by inverse Fourier transformation of the resulting spectral product. The numerical model of LPF can be alternatively implemented as a convolution of the input signal and the impulse response of the low-pass filter.

This section explains theoretical principles of the system and method claims according to the invention. To simplify the explanation, let us first assume that there is only one RES power plant (having a random time course of power) connected to the grid via a wattmeter. The schematic diagram (Fig. 1) shows a simplified single-phase connection to the grid, hiding the neutral wire. There is an accumulator (BESS) connected to the grid via a bidirectional AC/DC inverter. For the sake of simplicity of the analysis, let us assume 100% efficiency of storagecompensation and AC/DC power conversion. If the inverter’s positive power flows in the AC— >DC direction, its power is defined by the output signal p(t)-s(t+At) of the differencer. The total active power superimposed to the grid by both wattmeter and inverter is thus equal to s(t+At) which is actually the filter’s output signal driving the inverting input terminal of the differencer. The non-inverting input terminal of the differencer is driven by the signal p(t) which is the measured active power of the RES power plant. The filter with its "smooth" output signal thus determines the total power superimposed by the RES and BESS to the grid. The time course of accumulated energy by battery gives the difference in state-of-charge SOC [Wh] since time 0 until time t, while in theory, the SOC acquires both positive and negative values. The time advance At reduces absolute values of the integrated function. It is known in the state of the art, that always-positive SOC can be achieved in practice, if the accumulator consists of two separate batteries.

Assume now, that the smooth signal s(t+At) is identical with the LPF response to the input signal p(t+At). The optimal time advance At of the input signal is given by the group delay T_g of LPF at frequencies f « f_c (LPF attenuates by 3dB at the cut-off frequency f_c and r_g is a function of f_c.) The optimal advance At ensures that the rate of accumulation during one working cycle T (24 hours in PV case) drops to the minimum, while the following technical criteria are satisfied:

(II) mean value of SOC is zero:

(III) mean quadratic deviation of SOC is near its minimum: -f_Q SOC²(t)dt « min, therefore:

(IV) Near-minimum difference max(SOC) - min(SOC) is acquired per cycle, having

(V) near-minimum total energy throughput - f_Q |p(r) — s(r + At) |dr accumulated per cycle.

With the non-optimal time advance At, greater capacity (IV) of the accumulator would be needed and its greater energy throughput (V) and thus its faster wear. In summary: the accumulation rate would increase.

In theory, the minimum accumulation rate is satisfied if the LPF is driven by the future signal p(t+At). In technical practice, however, the signal p(t+At) is not available, but only its forecast pt(t+At) by a predictor. In practice, the rate of accumulation by the filter depends not only on the choice of At, but also on the accuracy of estimated future values pf at the optimal advance At. In general: The larger the desired smoothing effect, the larger the filter group delay T_g, the larger the required advance At, and the larger error of pf values. To sufficiently smooth out the photovoltaic intermittency, an optimum advance At in the order of tens of minutes is required. Even more advanced prediction is needed to smooth out fluctuations in wind power. As we will show in the 1st embodiment example of this invention, with the advance At = 30 minutes of solar irradiance, even a relatively accurate predictor will cause such a large increase in the accumulation rate that, with the optimal choice of At, the above-mentioned accumulation aggregates (IV) and (V) will acquire larger values than they would have acquired without any prediction at At=0. This means that driving the LPF with a real predictor may be less profitable than smoothing the RES power with a low-pass filter alone without a predictor.

If we were to smooth out the RES performance by an analogue LPF implemented, e. g., from elements of electrical inductance and capacity, then the excess accumulation of energy due to its group delay cannot be technically eliminated other than by continuously driving the filter’s input by the signal pt(t+At). Unlike the analogue LPF, its numerical model calculates the left-shifted output signal s(t+At) immediately (if we neglect duration of its computation) at any time t, if currently, the entire time series of input values p(r) is available in the interval (0, t+At], The numerical model of LPF can be given input values from the near future more precisely, than they are predictable at the maximum advance At. Simultaneous excitation of the LPF model by multiple predictors (that is, by a multi-predictor) with different advance times improves the accuracy of the aggregated signal s(t+At) which is reflected in a smaller rate of accumulation (or this one will get closer to its theoretical minimum), in contrast to the LPF driven by a single input signal with an advance of At, approximated by a single predictor based on equivalent technology.

Assuming the linearity of the filter and the valid principle of superposition, we will theoretically proof the possibility for an immediate calculation of the signal s(t+At) at time t using the mathematical model of LPF excited by a multi -predictor: Let us calculate the response of a linear, time-invariant system (which also includes our filter) to the input signal p(t+At) as a convolution of the input signal and the filter’s impulse response h(t): while in the T area where At-r<0 applies, the measured power signal is integrated, while in the T area where At-r>0 applies, the predicted power values are integrated. The convolution (VI) integrates the entire input signal of the filter over infinitesimally small intervals dr, and within each differential dr for At-r >0 the future signal p(t+At-r) may be estimated by a specific predictor producing minimized error with respect to its particular advance At-r.

In theory, the advantage of smoothing the random fluctuations in RES power by means of a numerical model of LPF excited by the multi -predictor is based on a synergy of three fundamental features: 1) The greater the advance At, the greater the prediction error and vice versa, 2) position of the first extremum in the time course of LPF impulse response, 3) the fact that two opposing signals are integrated by convolution. The synergistic effect suppresses the error in the filter-aggregated signal s(t+At) in the sense that, for a given RES intermittency, the filter-model excited by a multi -predictor exhibits much less accumulation rate than a physical instance of the same filter driven by a single predicted signal advancing by At exhibits. This relative advantage is independent on the technological implementation of the RES power predictor.

The advantages of the system and the corresponding method for smoothing fluctuations in electrical power from the grid-termination devices with a random time course of active power according to the invention are apparent from its external effects. For a given RES intermittency and RES power prediction technology, the minimum required battery capacity and the required total energy flow through the battery (in summary: accumulation rate) are determined by the required quality of the RES power (in terms of its fluctuations) and its smoothing method. So far known methods for smoothing RES power fluctuations minimize the accumulation rate mainly by dynamically tuning the cut-off frequency f_c of the filter with appropriate corrections of the advance At according to the forecasted solar intermittency. Known smoothing methods use LPF excited by a single predicted signal pi(t+At) with a relative error Ap corresponding to the minimum required advance At. However, it should be noted that this error is the largest value of interval [0, Ap], Other known smoothing methods are more or less close in quality to patent CN105552969A.

Regardless of the above-mentioned optimizing of f_c and At, the present invention applies a numerical model of LPF concurrently excited by a set of predicted signals, estimating their real counterparts p(t+Ar,) with evenly-distributed advances An from the interval [0, At], and with accordingly smaller errors distributed in interval [0, Ap], Aggregation of the signal s(t+At) by the numerical LPF model from the set of input signals with distributed error will significantly reduce the rate of energy accumulated by the filtering, compared with the LPF driven by a single input signal approximating p(t+At) with the error of Ap. In practice, batteries with a smaller capacity can be used for the smoothing, and the lower energy throughput will extend their lifetime. The invention can be combined with the dynamic optimization of f_c and corresponding advance At, according to the predicted RES intermittency. A quantitative analysis of the accumulation rate by the invention vs. by existing solutions is present in the conclusion of the 1st embodiment example in this application.

Brief description of the drawings

The system for smoothing fluctuations of electrical power in the grid from the connected terminal devices with a random time course of active power and the method according to the invention will be further explained by drawings in which:

Fig. 1 shows a block diagram of RES power smoothing by a numerical model of a low-pass filter driven by a multi-predictor.

Fig. 2 shows an analogous block diagram of the central smoothing of power aggregated from several RES.

Fig. 3 shows a block diagram of a hybrid PV plant.

Fig. 4 shows a block diagram of a hybrid PV plant, smoothing its power overflow to the grid.

Fig. 5 shows a block diagram of a wind power plant with power smoothing.

Fig. 6 shows the simulated time course of GI prediction from the 11th hour onwards with "better accuracy" of the prediction.

Fig. 7 shows the simulated time course of GI prediction from the 11th hour onwards with "worse accuracy" of the prediction.

Fig. 8 shows the specific power GI(t): measured, and filtered by the methods “LP”, “Ideal LP” and “Sim PLP with better prediction accuracy”.

Fig. 9 shows the specific power GI(t): measured, and filtered by the methods “LP”, “Ideal LP” and “Sim MPLP with better prediction accuracy”.

Fig. 10 shows the specific accumulated energy GX(t) by the smoothing methods “LP”, “Ideal PLP” and “Sim PLP with better prediction accuracy”.

Fig. 11 shows the specific accumulated energy GX(t) by the smoothing methods “LP”, “Ideal PLP” and “Sim MPLP with better prediction accuracy”.

Fig. 12 shows the specific power GI(t): measured, and filtered by the methods “LP”, “Ideal LP” and “Sim PLP with worse prediction accuracy”. Fig. 13 shows the specific power GI(t): measured, and filtered by the methods “LP”, “Ideal LP” and “Sim MPLP with worse prediction accuracy”.

Fig. 14 shows the specific accumulated energy GX(t) by the smoothing methods “LP”, “Ideal PLP” and “Sim PLP with worse prediction accuracy”.

Fig. 15 shows the specific accumulated energy GX(t) by the smoothing methods “LP”, “Ideal PLP” and “Sim MPLP with worse prediction accuracy”.

Description of the embodiments

It is understood that the individual embodiments of the invention are presented for illustration and not as limitations of the solutions. Regardless of the specific preferred embodiments of the invention described below, those skilled in the art will be able to suggest many variations and modifications, using no more than routine experimentation, without departing from the scope of the invention as defined in the appended claims.

Example 1

In this particular embodiment of the system and the corresponding method for smoothing fluctuations in electric power of the grid-termination devices with a random time course of active power according to the invention, a fundamental application of the numerical model of a low-pass filter excited by the multi-predictor is disclosed, as shown in Fig. 1. In this case, the grid-termination device 1 with a random time course of active power is a photovoltaic power plant (PVPP). The system has the input terminal B with active power p(t) measured by the wattmeter 2 which is coupled to the non-inverting (positive) input of differencer 5. The multi -predictor 9 is also excited with signal p(t) from the input terminal B. The power output of PVPP is connected to phase L of the grid via wattmeter 2. The multi -predictor 9 is trained via input terminal C by a time series of meteorological data f(t) from the meteorological sensor 10 which is a sky -imagery camera on the Earth's surface, or alternatively a camera on the Earth’s orbit. The multi -predictor’s output, approximating the future tuples p(t+Ar,), excites the numerical LPF model 6, whose output signal s(t+At) drives the differencer 5 via its inverting (negative) input. The differencer 5 drives the bidirectional AC/DC inverter 3 by the signal p(t)- s(t+At), defining the inverter’s power flowing from AC (grid) to DC (accumulator), or vice versa.

Based on the long-term measurement of the time course of solar irradiance on the earth's surface, it was possible to empirically analyse the technical contribution of the invention to the smoothing of PV power. This example assumes a single PVPP connected to the grid, whose output power is to be smoothed out. Let us assume that the output power of PVPP is proportional to the global solar irradiance GI [W/m²] measured at incident plane (20 cm x 16 cm) on the earth’s surface (latitude=48° and longitude=17°), where the plane is banked by 60° and is south-oriented. Based on the presumption, we substitute the signals p(t), s(t+At) in the expression (I) for SOC(t) with the measured signal GI(t) and its shifted smoothed counterpart. After the substitution, (I) computes a time course of the specific accumulated energy GX(t) [Wh/m²] by the filter.

The prediction error was simulated into GI(t) signal in such a way as to respect the fundamental properties of predictors:

• as the advance At increases, the effect of smoothing the predicted signal becomes stronger,

• as the advance At increases, the random error of prediction increases (statistically cumulates). Both errors were applied and superimposed on the measured signal p(t) in its "future" time interval (t, t+At] and thus the predicted signal pr(t+Ar) was simulated at each time t. Subsequently, the smoothing of PV power during a day with strong solar intermittency was numerically evaluated by 4 methods:

• “LP”: Input of LPF is driven by the measured PV power signal p(t)

• “Ideal PLP”: Input of LPF is driven by the measured, optimal left-shifted PV power signal p(t+At)

• “Sim PLP”: Input of LPF is driven by the measured, optimal left-shifted PV power signal p(t+At) having the simulated prediction error in its future part

• “Sim MPLP”: Exciting the numerical model of LPF by a multi -predictor of PV power with an equivalent simulated prediction error as in the Sim PLP method.

The simulated smoothing by the above-described 4 methods is graphically displayed in Fig. 8 to Fig. 15. In all four methods, the quality of PV power smoothing and the rate of energy accumulation by the filter were evaluated. The simulation of prediction error and the subsequent comparative analysis of individual smoothing methods are defined by three quantitative (OLAP) dimensions:

• Prediction time interval after which the smoothing effect of the predicted signal is always strengthened, is represented by the parameter smooth step [minute]. The shorter this interval, the more often or the steeper the effect of smoothing the predictor rises towards future.

• Standard deviation of the random prediction error which changes (and statistically cumulates) after passing every 6 minutes of prediction, is represented by the parameter SE (dimensionless). • Order of low-pass filter (1 to 4) used for smoothing (for comparative analysis only). The filters of different orders used in the simulation were tuned so that the quality of the power smoothing by a filter of any order is equivalent in terms of energy infeed, provided that the filters are driven by a measured, always optimally shifted input signal. When fulfilling this tuning condition, increasing the order of the filter increased its cut-off frequency, while the optimal advance At increased slightly.

The simulation showed the following empirical dependencies along the partial OLAP dimensions:

Effect of smoothing the future signal by a predictor:

• Changing the value of the smooth step [minute] parameter does not affect the smoothing quality of both “Sim MPLP” and “Sim PLP” methods, with the “Sim MPLP” method having worse smoothing at all parameter values than the “Sim PLP” method.

• Decreasing the value of smooth step significantly increases the accumulation rate by the “Sim PLP” method, which for all parameters is much larger than that of the “Sim MPLP” method. For some parameter values, even the “Sim PLP” method exceeds the accumulation rate of the “LP” method. The accumulation rate of the "Sim MPLP" method increases only slightly, in any case much less than that of the “Sim PLP” method.

Random error of the predictor:

• Increasing the value of the SE parameter does not degrade the smoothing quality of the "Sim PLP" method, but it does degrade the smoothing of the "Sim MPLP" method, in which the smoothing performance with standard error SE>0 is always worse than that of the "Sim PLP" method.

• Increasing the value of the SE parameter significantly increases the accumulation rate of the “Sim PLP” method, which for all values of SE >0 is much higher than that of the “Sim MPLP” method. For some values of SE, even the “Sim PLP” method exceeds the “LP” method in its accumulation rate. The accumulation rate of the "Sim MPLP" method increases only slightly, and less so the higher the filter order.

Order of the filter:

• The filter order does not affect the quality of the RES power smoothing by the "Sim PLP" method and for small filter orders, it is better than the smoothing quality by the "Sim MPLP" method.

• Increasing the filter order improves the smoothing quality of the "Sim MPLP" method, which is equivalent to the "Sim PLP" method from the 4th order onwards. • Increasing the filter order only slightly decreases the accumulation rate of the "Sim PLP" method, which is always significantly larger than the accumulation rate of the "Sim MPLP" method.

• Increasing the filter order appreciably reduces the accumulation rate by the "Sim MPLP" method. With a relatively small prediction error, this method gradually converges to the accumulation rate of the "Ideal PLP" method.

Simulations corresponding to two qualitative levels of prediction are displayed graphically and numerically:

• “Better prediction accuracy”: smooth step = 6 min., SE = 0.05

• "Worse prediction accuracy": smooth step = 3 min., SE = 0.1.

The signal GI(t) is graphically displayed as measured, smoothed, and finally the specific accumulated energy GX [Wh/m²] corresponding to the applied smoothing method. In the simulations shown, a 3^rd-order low-pass filter is used. A day with high solar exposure and strong solar intermittency was selected for the analysis. In order to achieve the desired smoothing of the GI(t) signal, it was necessary to use the advance of At = 30 minutes on this day. The accumulation rate by 4 different smoothing methods was calculated, based on the measurement of GI(t) during the selected day in Table 1, where: In the column "A GX" is the difference between the maximum and minimum specific accumulated energy according to the expression (IV), in the column "Throughput" is the total daily flow of specific energy through the accumulator according to expression (V). For comparison, the daily specific global exposure of the incidence area is shown in the last column.

Table 1 : Specific daily accumulation by smoothing, and global exposure

With "better prediction accuracy" of GI, the "Sim MPLP" smoothing method requires 22% of the energy capacity of the accumulator (A GX) required by the "LP" method, or 30% of the capacity required by the “Sim PLP” method, and finally requires 1.4 times the capacity required by the “Ideal PLP” method. The "Sim MPLP" method by smoothing causes 62% of the daily energy flow through the accumulator by the "LP" method, or 74% of the throughput by the “Sim PLP” method and finally induces 1.1 times the throughput by the “Ideal PLP” method.

With "worse prediction accuracy" of GI, the "Sim MPLP" smoothing method requires 32% of the energy capacity of the accumulator (A GX) required by the "LP" method, or 21% of the capacity required by the “Sim PLP” method, and finally requires 2 times the capacity required by the “Ideal PLP” method. The "Sim MPLP" method by smoothing causes 68% of the daily energy flow through the accumulator by the "LP" method, or 58% of the throughput by the “Sim PLP” method and finally induces 1.2 times the throughput by the “Ideal PLP” method.

Example 2

This example of a particular embodiment of the system and the corresponding method for smoothing fluctuations in electric power of the grid-termination devices with a random time course of active power according to the invention presents the smoothing of several terminal devices lj of different types of RES, connected to the grid through their own wattmeters 2j as shown in Fig. 2. For the sake of clarity, the block diagram shows only one functional block of the power plant, wattmeter and predictor, with the integer index j from a positive set { 1, ...m}.

The central device of the system is in the block diagram to the right of the mains phase conductor L, while the accumulator 4 is connected to the grid by means of the bidirectional AC/DC inverter 3, e. g. at the location of the HV/LV transformer station. The central device is connected to the remaining part of the scheme by the following signal terminals: the output terminal A of the differencer 5, input terminals Bj carrying the active power pj(t) and input terminals Cj carrying the meteorological data time series fj(t). For each j-th RES power plant there is a separate multi -predictor 9j of its active power. As the power from individual terminal devices lj is superimposed into the grid, their measured pj(t) and multi -predicted active powers Pj(t+Aii) are superimposed by the summator 11 and by the vector summator 12, while analogously to expression (VI) holds: An = At-n, where the integer index i is from a nonnegative set {0, ...n} and the relation At-ri >0 holds for the time position of future signal p. The output of the summator 11 excites the non-inverting input of the differencer 5, while the vector summator 12 aggregates n+1 signals p(t+An) which in parallel (simultaneously) excite the numerical model 6 of LPF. The capacity of accumulator 4 is calculated by the expression (IV) and its maximum power results from the sum of installed RES power and the maximum intermittency of the aggregated RES power. When smoothing the aggregated power from several RES, the BESS capacity and the relation to its maximum power are affected by the number, mutual distance and type of installed RES in the network.

Example 3

In this example of a specific embodiment of the system and the corresponding method for smoothing fluctuations in electric power of the grid-termination devices with a random time course of active power according to the invention, a utilization of one accumulator integrates the both BESS and power smoothing functions in the same accumulator, as shown in Fig. 4. In this case, two terminal devices with a random time course of the active power are connected to the grid: a photovoltaic power system h, and a network h consisting of passive appliances. Fig. 3 shows a block diagram of a conventional hybrid PV power system (hereinafter HPVS).

The HPVS is typically installed in a building having its own appliances, altogether connected to the grid (so-called on-grid HPVS). In its hybrid version, the basic PVPP li is equipped with a wattmeter 7, another bidirectional AC/DC inverter 8, and with a BESS 4. During a solar day, the goal is to store the whole excess of photovoltaic energy (not consumed by the appliances h) in order to cover the consumption of object h during the rest of the day, when the PVPP h does not supply enough energy to operate the appliances. Since the accumulator 4 is assumed to be discharged (empty) at the beginning of the day, another bidirectional AC/DC inverter 8 regulates the energy flow to the accumulator 4 in such a way, that the total system power measured by the wattmeter 7 is po(t)=O until the accumulator 4 is fully charged, or po(t)< 0 at times when PV power is insufficient - then the battery charging is temporarily interrupted. The purpose of such a regulation is to prioritize the use of surplus solar energy to charge the accumulator 4. Once this is charged, the excess solar energy is transferred to the grid. Such an arrangement eliminates the energy flow to the grid (i. e. makes it smooth) at times t when po(t)=O applies. At other times, the power through the wattmeter 7 has a random time course.

A hybrid photovoltaic power system smoothing its power flow to the grid according to the invention is sketched in Fig. 4. The conventional HPVS is extended with a signal chain starting with the meteorological sensor 10i and the appliances’ sensor IO2, ending with the differencer 5, controlling the power of the bidirectional AC/DC inverter 3 which alternatively, after switching the accumulator 4 from D- to E-node, substitutes the conventional regulation of the zero hybrid-system power po(t). While the accumulator is connected to D-node at the beginning of the day, it is charged by another bidirectional AC/DC inverter 8 in the conventional HPVS mode and holds po(t)<O. When the SOC reaches its threshold value (e.g., 85 %), the accumulator 4 is switched to the E-node and from then on, its performance is controlled by the differencer 5. The output signal p(t)-s(t+At) from the differencer 5 determines the system power of HPVPS such that the smooth power s(t+At) flows through the wattmeter 7. From the moment of switching until the end of the solar day, the accumulator 4 filters the fluctuations in the difference between the PV power (h) and the consumption of appliances h, i.e., pi(t)-(-p2(t)) provided that the necessary part of the accumulator’s capacity remained free at the time of switching.

By setting-up a suitable imparity between the advance At and the filter’s group delay r (corresponding to frequency f « f_c), it is possible to control the SOC value of the accumulator even during the smoothing process. Before the noon, the average SOC value would increase (the battery would be charged) using the relation At<r (or vice versa would be discharged), while in the afternoon the battery would be charged using the relation r_g<At. Assuming that the intermittency of (h) is definitely decreasing in the afternoon, by increasing the cut-off frequency f_c we can reduce the value of r in relation to At: Hence the accumulator will be charged beyond the SOC threshold value, while still keeping on smoothing.

Based on the long-term measurement of the time course of solar irradiance, the numerical simulation in example 1 of the embodiment of the invention showed that, according to expression (IV), it is enough to reserve <10% of a BESS energy capacity of a HPVS, if its capacity equals to 2 hours x installed PV power. With such a capacity, the Li-Ion battery will never be overloaded by the bidirectional AC/DC inverter 3 actuating the smoothing power. The presumption for the calculation is the optimally controlled storage-compensation power between the battery and the grid, using the numerical filter model 6 driven by multi -predictors 9i and 92. The modified HPVS smooths the time course of its system power from the moment the SOC threshold value is reached until the end of the solar day, when the SOC can reach its maximum value of 100%. Modem Li-Ion technologies harmonize well with the requested rate power-to-capacity of BESS in HPVS. The same accumulator can fulfil both energy storage and smoothing functions.

Example 4

In this example of a specific embodiment of the system and the corresponding method for smoothing fluctuations in electric power of the grid-termination devices with a random time course of active power according to the invention, the smoothing of the wind power by a numerical model of a low-pass filter excited by a multi -predictor is applied, as shown in Fig. 5. In this case, the grid-termination device 1 with a random time course of active power is a wind power plant having its own energy accumulator 4, the parameters of which are set to provide sufficient power and energy to compensate for the intermittency of the wind power, and, unlike the hybrid PV power system, not to compensate for the day - night cycle. Other properties are analogous to the described photovoltaic power plants with power smoothing according to the invention.

Industrial applicability

The system and method for smoothing fluctuations in electrical power from the gridtermination devices with a random time course of the active power can be used in large solar power plants and wind energy farms, in transformer stations of the distribution network, as well as in small power plants utilizing the renewable energy sources. The invention will enable a substantial increase of the RES share in the overall production of electricity.

Reference numbers

1 grid-termination device

2 wattmeter

3 bidirectional AC/DC inverter

4 accumulator, battery energy storage system (BESS)

5 differencer (difference amplifier with a unit signal gain)

6 numerical model of a low-pass filter

7 wattmeter

8 another bidirectional AC/DC inverter

9 multi-predictor

10 meteorological sensor

11 summator

12 vector summator

A output terminal

B input terminal of the active power signal pj(t)

C input terminal of the meteorological data time series fj(t)

D connecting node between accumulator (4) and another bidirectional AC/DC inverter (8)

E connecting node between accumulator (4) and bidirectional AC/DC inverter (3)

Claims

1. A system for smoothing fluctuations in electric power in a grid from connected terminal devices exhibiting a random time course of an active power, having a differencer (5) with an output terminal (A) controlling a power of a bidirectional AC/DC inverter (3), characterized in that it has at least one input terminal (Bj) of the active power pj(t) from a terminal device (lj) measured by a wattmeter (2j) and superimposed on a non-inverting (positive) input terminal of the differencer (5), while the input terminals (Bj) of the active power pj(t) are individually coupled to corresponding multi -predictors (9j); output terminals of the multi -predictors (9j) are superimposed onto an input terminal of a numerical model (6) of a low-pass filter, whose output terminal is coupled to an inverting (negative) input terminal of the differencer (5); and input terminals (Cj) carrying meteorological data time series fj(t) are coupled to the multi-predictors (9j), where the positive integer index (j) is from a set { 1, ...m}.

2. The system for smoothing fluctuations in electrical power in the grid from connected terminal devices with a random time course of the active power according to claim 1, characterized in that a summator (11) superimposing the active power signals pj(t) of the terminal devices (lj) is coupled to the non-inverting (positive) input terminal of the differencer (5), and a vector summator (12) superimposing multi-predicted power signals is coupled to the numerical model (6) of the low-pass filter.

3. The system for smoothing fluctuations in electrical power in the grid from connected terminal devices with a random time course of the active power according to claim 2, characterized in that at least one input terminal (Bj) of the active power pj(t) is connected to the wattmeter (2j) of the terminal device (lj) with a random time course of the active power connected to the grid (AC) and a corresponding meteorological sensor (10j) is coupled to at least one input terminal (Cj) with a time series of the meteorological data fj(t), while the output terminal (A) of the differencer (5) is coupled to a control input of the bidirectional AC/DC inverter (3), by which an accumulator (4) is connected to the grid (AC).

4. The system for smoothing fluctuations in electric power in the grid from connected terminal devices with a random time course of the active power according to any of the preceding claims, characterized in that the terminal device (lj) with a random time course of the active power is a photovoltaic power plant, where the meteorological sensor (10j) is a camera for capturing sky imagery from the Earth's surface or from the Earth's orbit.

5. The system for smoothing fluctuations in electric power in the grid from connected terminal devices with a random time course of the active power according to any of the preceding claims 1 to 3, characterized in that the terminal device (lj) with a random time course of the active power is a wind power plant, where the meteorological sensor (10j) is a combination of the meteorological sensors relevant for the power prediction of a wind turbine.

6. The system for smoothing fluctuations in electric power in the grid from connected terminal devices with a random time course of the active power according to any of the preceding claims, characterized in that the terminal device (lj) with a random time course of the active power is a network of an electrical appliances, where the meteorological sensor (1 Oj) is a combination of the meteorological sensors and appliance status parameters, relevant for the prediction of their power consumption.

7. A method of smoothing fluctuations in electric power in the grid from connected terminal devices with a random time course of the active power by an accumulation filtering using the system according to any of the preceding claims 1 to 6, where the power of a bidirectional AC/DC inverter (3) flowing from the grid (AC) to the accumulator (4) is defined as a difference between the aggregated power p(t) of the terminal devices (lj) and the output s(t+At) of a low-pass filter by means of the differencer (5), where the advance At is given by a group delay of the low-pass filter in its pass-band frequency range, characterized by the fact that the non-inverting (positive) input of the differencer (5) is driven by the signal p(t) superimposed from all inputs (Bj) of active power signals pj(t) measured by the wattmeters (2j) of the terminal devices (lj) indexed by a positive integer (j) from a set { 1, ...m}; where the inverting (negative) input of the differencer (5) is driven by the signal s(t+At), which at any time is the response of the numerical model (6) of low-pass filter to a set of n+1 signals p(t+Ar,) superimposed from a set of m x (n+1) signals pj(t+Aii) estimated by multi -predictors (9j), where the non-negative integer index (i) is from a set {0, ...n} and the relation 0 < Ari < At holds, and the calculation of a response of low-pass filter by its numerical model (6) lasts negligibly short compared to the advance At, while the multi -predictors (9j) are driven by the measured power pj(t) of the connected terminal devices (lj) and by the time series of relevant meteorological data fj(t), and the prediction error of each i-th component of the signals pj(t+Aii) is minimized with respect to its specific advance ATI by the state-of-art power prediction technology corresponding to the terminal devices (lj).

8. The method of smoothing fluctuations in electrical power in the grid from connected terminal devices with a random time course of active power according to claim 7, characterized in that the numerical model (6) of a low-pass filter is implemented by convolution of the input signal and the impulse response of the low-pass filter.

9. The method of smoothing fluctuations in electrical power in the grid from connected terminal devices with a random time course of active power according to claim 7, characterized in that the numerical model (6) of a low-pass filter is implemented in the spectral domain as a product of the input signal spectra and the frequency response of the low- pass filter, whereas the output signal is calculated by inverse Fourier transformation of the resulting spectral product.