Device for determining performance and generating performance patterns

Abstract

The invention relates to a device for determining power in power grids and / or for generating power patterns, comprising a current measuring device for measuring a mains current present at a measuring point of the power grid, a voltage measuring device for measuring a mains voltage present at the measuring point of the power grid, a frequency-based A / D converter module configured such that the mains current and / or the mains voltage are sampled at predetermined frequency intervals to produce digital frequency component current data and frequency component voltage data, a wavelet module configured such that frequency component current data and frequency component voltage data are multiplied by a predetermined wavelet function to produce wavelet frequency component current data and wavelet frequency component voltage data, a runtime correction module configured toshifting the frequency component current data and the frequency component voltage data as a function of a frequency by a latency time so that the frequency component current data and the frequency component voltage data of the frequency intervals are time-synchronized, a calculation module for calculating power data and / or energy data from the frequency component voltage data and frequency component current data.

Description

[0001] The invention relates to a device for determining power in power grids according to patent claim 1.

[0002] Power measurement in multi-conductor circuits is defined in Germany according to DIN 40110-2 and worldwide according to IEE-1459. The calculation methods of these standards form the basis for power calculations using modern measuring instruments. According to the standard, the specified operating range is + / - 1 Hz relative to a nominal frequency of 50 Hz and + / - 1.2 Hz relative to a nominal frequency of 60 Hz. Harmonics and overtones are taken into account. In the low-frequency range, they are only taken into account from 15 Hz upwards. Nonlinear system behavior of the grid is therefore not taken into account. Since in the future, systems based on decentralized converters, which influence system behavior in the low frequency range, will increasingly be put into operation, the power result will be distorted if a low-frequency range of the power-relevant variables, namely the grid current and the grid voltage, is not taken into account.The Fourier analysis-based algorithms used today generate errors for low-frequency modulated nominal frequencies that, depending on the technical design (sampling rate, window length), far exceed permissible tolerance values. It should be noted that a fast Fourier transform (FFT) delivers information in the frequency domain with a resolution that depends on the time window (measurement interval). No information about the signal is reproducible. To overcome this limitation, a "windowed" Fourier transform was introduced to generate a frequency-time spectrum. This variant of the FFT is also known as an SFT short-time Fourier transform. However, this short-time Fourier transform has the disadvantage that a fixed window width must be used. This property has a significant impact on the frequency / time resolution and thus on the accuracy with which a signal can be identified.

[0003] US Pat. No. 9,170,986 B2 discloses a power analysis method that determines information on harmonic components based on wavelets, specifically the reconstruction of waveforms that are analyzed and processed in other devices and applications. However, this method eliminates low-frequency components (sag / swell cycles), which fall under the term "flicker" according to power quality standards (DIN-EN 50160).

[0004] DE 10 2017 106 839 B3 discloses a device for recording and analyzing electrical energy consumption in a building, which device comprises a current measuring device and a voltage measuring device. Furthermore, the device comprises an A / D converter module. Furthermore, the device comprises a central data processing device for storing and processing consumption values ​​from a plurality of buildings. The central data processing device comprises means for recognizing recurring consumption patterns in the functions formed from the consumption values. The central data processing device comprises a learning system so that known consumption patterns can be separated from unknown consumption patterns. This allows a larger number of different consumption patterns or consumers to be quickly recognized. A wavelet transformation is not used.

[0005] From CN 1 01 055 299 A, a device for determining power in power grids is known, comprising a power grid device, a voltage measuring device, and an A / D converter module. Furthermore, the device comprises a wavelet module, by means of which a transformation of wavelet frequency component current data and wavelet frequency component voltage data is provided. A runtime correction module for shifting the wavelet frequency component current data and the wavelet frequency component voltage data depending on a frequency and a time value is not provided. Thus, an accurate power measurement and a time-accurate power measurement are not possible.

[0006] The object of the present invention is to provide a device for determining power in power grids that also takes low-frequency signal components into account when determining power. In particular, the device should enable the evaluation of transient, nonlinear processes and their active use for the precise, parallel, and time-correct evaluation of power and energy signals.

[0007] To solve this problem, the invention has the features of patent claim 1.

[0008] According to the invention, predefined wavelets are used for fixed, predetermined frequency intervals, with the frequency range extending from 0 Hz up to a maximum frequency. The frequency range to be examined is divided into frequency intervals in such a way that the frequency intervals currently used in standards are respected, and the corresponding current and voltage values ​​are each subjected to the wavelet function (wavelet pattern). The invention takes advantage of the fact that the length of the wavelets depends on the frequency interval or frequency level to be examined. By dividing the current or voltage signals into predefined frequency intervals, a frequency-based parallel analysis or calculation of the power can be carried out. This ensures highly accurate measurement of individual frequencies for all power components, in particular active and reactive power, and in particular for the nominal frequency.

[0009] The invention enables, on the one hand, precise power determination, whereby the exact amplitudes of the frequency components are determined in a first measuring path. On the other hand, grid disturbances can be evaluated based on power patterns, whereby the power components are determined in a second measuring path at a specific, predetermined time across a predetermined frequency range. The superposition of the transmitted power components forms power patterns that can characterize specific grid disturbances. This makes it possible to provide energy efficiency services that subsequently lead to the detection of an ongoing disturbance in the grid. According to the invention, the first measuring path can be used for precise power measurement, while the second measuring path can be used for time-accurate power pattern determination. Alternatively, a combination of both measuring paths can be used to determine exact and time-accurate voltage and current quantities.According to the invention, a time-based delay correction module is provided, which enables precise measurement. This allows patterns and time-correct assignments of frequency components to be determined. For example, rapidly changing dynamics can be detected, such as aging processes, fault detection of machine components, and changes in the operating behavior of systems and devices. The time-based delay correction module ensures that all wavelet values ​​are measured with precise time. This makes it possible to reconstruct the original signal from the individual components.

[0010] According to a further development of the invention, the signals, namely the mains current and the mains voltage, are sampled at equidistant frequency intervals. Advantageously, equal amplitude / frequency component values ​​are available.

[0011] According to a further development of the invention, a predetermined latency, with which the wavelet values ​​calculated in different frequency ranges (frequency intervals) are shifted, depends on a minimum frequency. Only after the latency period has elapsed can the signal be displayed and used for further processing. It is taken into account that wavelets have a changing time and frequency resolution. In order to display or measure non-stationary signals, the observation period and the frequency must be adapted to the signal to be observed. Therefore, low-frequency signal components must be observed for longer, or a longer wavelet must be used. To determine the precise power, larger time intervals must therefore be used, usually in the range of minutes. This latency period specifies the maximum runtime correction for synchronizing the respective frequency-dependent or frequency-interval-dependent wavelet values.Using the frequency-based delay correction module, a "backward transformation" of the time to the observation time takes place, i.e., from when the exact shift of the wavelet signal should be taken into account in relation to the useful signal (e.g., 50 Hz mains frequency). If the frequency interval is set to 1 Hz, for example, the lowest frequency value can be set to 0.01 Hz, which results in a latency of 100 seconds. After these 100 seconds have elapsed, the current and voltage signals present at the observation time in the respective frequency intervals (at the beginning of the 100 seconds) can be summed up, and the power values ​​at the observation time can be determined by shifting the time back accordingly.

[0012] According to a further development of the invention, an AI module is provided in which error pattern recognition algorithms are collected in order to be able to detect current errors by comparing them with the currently measured values.

[0013] Embodiments of the invention are explained in more detail below with reference to the drawings.

[0014] They show: Fig. 1 a block diagram of a measuring device according to the invention, Fig. 2 processing of measured power and energy values ​​broken down by frequency components, Fig. 3 a graph with a function of frequency or time intervals for different time windows based on a Heisenberg uncertainty principle, Fig. 4 a representation of wavelets of different resolution, Fig. 5a a representation of two derived test signals, Fig. 5b an analysis of the first test signal using Fast Fourier Transform (FFIT) and Wavelets (WT), Fig. 5c an analysis of the second test signal using Fast Fourier Transform (FFIT) and Wavelets (WT), Fig. 6 a representation of the frequency ranges to be examined, Fig. 7a a representation of the determined frequency components of the wavelets for a specific disturbance in the power grid, Fig. 7b an enlarged view of a low-frequency portion of the wavelet amplitudes according to Fig. 7a and Fig. 8 an exemplary representation of wavelet values ​​at different frequencies (frequency components).

[0015] A device according to the invention for determining a power in a power grid 1 essentially consists of a current measuring device 2 for measuring a grid current at a measuring point of the power grid 1, a voltage measuring device 3 for measuring a grid voltage at the measuring point of the power grid 1, a sampling device 4 for sampling the measured grid current and the grid voltage, and a processing device 5 for processing the sampled grid current and grid voltage values.

[0016] Preferably, the device additionally comprises a communication / control device 6, which is connected to remotely located communication participants via a communication interface 7.

[0017] Communication interface 7 can be used, in particular, to transmit the determined power patterns that were determined at a specific point in time. The determined power patterns can be used for control / protection applications, for example, via an algorithm to control a protective device or trigger protective devices.

[0018] Preferably, the device has a display 22 for displaying the determined performance results.

[0019] The processing device 5 is divided into two processing sub-devices 5a and 5b, whereby the respective processing sub-devices 5a, 5b can be implemented independently of one another. The first processing sub-device 5a is used to determine precise power and energy measurements. A second processing sub-device 5b is used to quickly detect changes and the associated time of the individual frequency components, for example, for use in methods for pattern recognition and error analysis (NILM Non-Intrusive Load Monitoring) or predictive maintenance. The power or power components that can be determined using the device according to the invention are shown below. Apparent power S: S=P2+Q2 Active power P: P2=(PG+PO+PM+∑m=1∞(PWm))2 Reactive power Q: Q2=(QG+QO+Qm+∑m=1∞(QWm))2+DO2+DM2+DMO2 +∑m=1∞(DWmO2)+∑m=1∞(DMWm2)+∑m=1∞(DWm2)+U2 • Total active power P in watts ◯ P G : Active component of fundamental oscillation υ = 1 ◯ P O : Effective component distortion υ > 1 ◯ P M : Active component modulation 0 < υ < 1 ◯ P W : Effective part of interactions m ∈ (1... ∞) 0 < υ < 1 • Total reactive power Q in Var consists of the components ◯ Q G : Displacement reactive power fundamental mode υ = 1 ◯ Q O : Displacement reactive power distortion υ > 1 ◯ Q M : Displacement reactive power modulation 0 < υ < 1 ◯ O Wm : Displacement reactive power interactions m ∈ (1...∞) 0 < υ < 1 ◯ D0: Deformation reactive power distortion υ > 1 ◯ D M : Deformation reactive power modulation 0 < υ < 1 ◯ D Wm : Deformation reactive power interactions m ∈ (1...∞) 0 < υ < 1 ◯ D MO : Interaction of D O and DM m ∈ (1...∞) υ > 1 ◯ D MWm :Interaction of D M and D Wm m ∈ (1...∞) υ > 1 ◯ D OWm Interaction of D O and D Wm m ∈ (1...∞) υ > 1 ◯ U: Unbalanced reactive power

[0020] The underlined power components are not determined by conventional power measuring devices, but are reserved exclusively for the measuring device according to the invention.

[0021] In the Fig. 5a the determination of test signals S T1 and S T2 from two different signals S1, S2 of different frequencies. In the Fig. 5b and Fig. Figure 5c shows the analysis based on a Fast Fourier Transform (FFT) and a Wavelet Transform. After the Fourier Transform, frequency values ​​of different magnitudes are determined, while after the Wavelet Transform, identical wavelet values ​​are determined, regardless of the composition of the frequency components of the test signal S. T1 or S T2 .

[0022] The signals S1 and S2 are harmonic oscillations, with the frequency of signal S1 being lower than the frequency of signal S2. The first test signal S T1 is obtained by superimposing or summing the first signal S1 and the second signal S2. The second test signal S T2 results from the temporal sequence of the first signal S1 and the second signal S2. As can be seen from Fig. As can be seen in Figure 5a, the second test signal S T2in a first time period exclusively from the first signal S1 of lower frequency and in a second period exclusively from the signal S2 of higher frequency.

[0023] If you apply to the first test signal S T1 the Fast Fourier Transformation FFT, the result is Fig. 5b, that the test signal S T1 consists of two frequencies, namely a low frequency f1 and a higher frequency f2. If the wavelet transformation is applied to the first test signal S T1 , two values ​​W1 and W2 result, with the first value W1 corresponding to the lower frequency f1 and the second value W2 corresponding to the higher frequency f2. These two constant values ​​W1 and W2 are generated over the entire time window.

[0024] It should be noted that wavelets are wave patterns of a specific duration. For example, Morlet wavelets and Daubechies wavelets are well-known and are suitable as pattern functions for power measurement. Fig. Figure 4, for example, shows a wavelet (wavelet function) WL. In a time window of + / - 20 time units, it exhibits oscillations with an amplitude that increases in the first half and decreases in the second half. With respect to time unit zero, the wavelet is symmetric, meaning that the increasing first half is symmetrical to the decreasing second half of the wavelet and has its maximum amplitude in the middle of the time window.

[0025] If a Fast Fourier Transform (FFT) and a Wavelet Transform are applied to the second test signal ST2, the result is Fig. 5c. After the Fast Fourier Transformation (FFT), a value corresponding to the lower frequency f1 is obtained in a first time window in which the first signal S1 with a lower frequency f1 is present, and a value corresponding to the higher frequency f2 is obtained in a second time interval in which the signal S2 with a higher frequency f2 is present. According to the wavelet transformation, the value W1 corresponding to the lower frequency f1 is obtained in the first time period T1, and the value W2 corresponding to the higher frequency f2 is obtained in the second time period T2. In addition, the wavelet transformation provides information about when the frequency change takes place, namely at time t1, when the value W1 jumps to the higher value W2. This time t1 of the change from one wavelet value W1 to the other wavelet value W2 is used in the following time-correct arrangements of the determined power values ​​for different frequencies.

[0026] In Fig. 2 shows the processing device 5 in more detail. A first processing sub-device 5a of the processing device 5 is frequency-based, while a second processing sub-device 5b of the processing device 5 is time-based.

[0027] The first processing subunit 5a processes data from a frequency-based A / D converter module 8 of the sampling device 4, which samples the measured mains current and mains voltage L1, L2, L3, N, and ground - 4-phase in parallel at predetermined frequency intervals A, B, C, D, E, and F. The frequency ranges are divided as follows: Frequency range 0: Direct current (DC) Frequency range A: Sampling rate > 2.4 kHz in the range approx. 0 to 55 Hz, or approx. 0 to 66 Hz - depending on the nominal frequency Frequency range B: Sampling rate > 24 kHz in the range 47 Hz < 50 Hz < 53 Hz, or 55 Hz < 60 Hz < 66 Hz Frequency range C, D: Sampling rate > 40 kHz - typically: 40.96 kHz in the nominal frequency range 50 Hz to 2.5 kHz, or 60 Hz to 3.0 kHz, in range D in intervals of 200 Hz Frequency range E: Sampling rate > 400 kHz - typical: 409.6 kHz, in the range from 9 kHz to 150 kHz Frequency range F: Sampling rate > 1.5 MHz, in the range 150 kHz to 500 kHz. Preferably, the frequency intervals in the individual ranges are equidistant, for example, with an interval width of 1 Hz in frequency range A. A lowest frequency interval can, for example, run from 0.005 Hz to 1 Hz. The next second-higher frequency interval can run from 1 Hz to 2 Hz, and so on, up to frequencies in the kHz range, for example. In addition, special frequency ranges can be defined if particularly interesting frequency components need to be measured with very high accuracy.

[0028] In Fig. Figure 6 shows an overview of the different frequency ranges O, A, B, C, D, E, F. As already described above, the sampling rate in the frequency range B is higher than in the frequency range A. Within the respective frequency ranges A, B, C, D, E, the sampling rate is preferably constant, so that corresponding wavelet values ​​can be determined for equal time intervals. Fig. 7a is the result of a Fig. 2 is presented. In addition to frequency components and a relatively high frequency component in the operating band B between 47.5 Hz and 52.5 Hz, low-frequency components with a frequency of 25 Hz and less were detected, which lie in the first frequency band A between 0 Hz and 50 Hz. The magnitude and distribution of the frequency components, i.e. the frequency pattern determined based on the wavelet transformation, allow conclusions to be drawn about the type of damage. In addition, after a minimum latency period, the power at the beginning of the latency period can be determined. A more detailed examination of the frequency components in the low-frequency range is shown in Fig. 7b. It can be seen that a low-frequency resonance with sidebands is present, with the resonant frequency being 25 Hz. The sidebands are located at 12.5 Hz and 37.5 Hz (see hatched bar). Such a low-frequency resonance with sidebands can be detected by modulating electric vehicles and other power electronic loads connected to the power grid.

[0029] The first processing sub-device 5a further comprises a first wavelet module 9. The digitized current and voltage values ​​are multiplied for each frequency interval by a predetermined wavelet function and then transferred to a first wavelet propagation time correction module 10 on the signal flow side. Fig. 4 are exemplary wavelet functions WL0, WL 5a and WL 5b, which have a different signal width but the same signal height at time t0 of maximum amplitude. For example, it can be a Morlet wavelet or a Daubechies wavelet, which is particularly suitable for measuring frequency intervals. The wavelet function can be scaled according to requirements, namely with regard to the signal width and signal height. The signal width depends on the frequency and is adapted to the frequency ranges so that the required accuracy can be achieved.

[0030] The frequency-based first runtime correction module 10 is designed such that, depending on a frequency or frequency interval size, the determined wavelet value is adjusted by a latency time σ A , σ B , σ C , σ D , σ E , σ F is shifted back so that the current and voltage values ​​for each frequency interval are time-synchronized.

[0031] The first processing subunit 5a further comprises a processing module 11, by means of which the wavelet values ​​representing the voltage and current values ​​present for the respective frequency intervals are summed for the respective real and imaginary components and calculated to form a power value (active and reactive components). To determine the respective imaginary component, the wavelet values ​​are each shifted by 90°. Thus, first wavelet values ​​for a real component and second wavelet values ​​for an imaginary component of the current and voltage can be determined.

[0032] The specified latency depends on the minimum frequency, which is defined by the lowest frequency interval. According to the Heisenberg uncertainty principle, the lower the frequency, the longer the wait time is required until the complete measurement signal can be correctly captured.

[0033] Fig. Figure 3 shows the function of frequency over the time required to determine the current / voltage values ​​in the respective frequency intervals, above the abscissa. For example, a 0.1 Hz signal requires a duration of 10 seconds. A 0.01 Hz signal requires 100 seconds. These measured values ​​in the frequency domain are further processed in processing module 11 to determine the performance data.

[0034] The second processing subunit 5b comprises a second wavelet module 12, by means of which the digitized current and voltage data provided by the A / D converter module 8 in the corresponding frequency intervals A, B, C, D, E, F are calculated using time-based wavelet functions to form wavelet values ​​Δt, Bt, Ct, Dt, Et, Ft. These wavelet functions are dimensioned such that a relatively fast measurement can be determined at the expense of accuracy.

[0035] A second delay correction module 13 of the second processing sub-device 5b located downstream of the signal carries out the delay correction so that the time-correct values ​​can be superimposed or summed up for the respective frequency intervals.

[0036] At the output of the first runtime correction module 10 or the second runtime correction module 13, the resolved frequency components A, B, C, D, E, F are summed to yield power and energy quantities for the respective intervals, which can then be used to determine a total energy or power. For this purpose, the processing module 11 has a calculation module 15. The calculation module 15 has a corresponding power calculation routine.

[0037] In Fig. Figure 8 shows an example of wavelet values ​​A1, A2, A3 determined at different frequencies, namely 0.3 Hz, 29 Hz, and 35 Hz. To determine the correct power, the latency t L3for the frequency with the lowest frequency, namely 0.3 Hz, in order to then sum the wavelet values ​​A1, A2, A3 for all frequency ranges. The latency time for the 35 Hz component is T L1 smaller than for the 29 Hz component with T L2 . The largest latency in the present embodiment applies to the frequency component 0.3 Hz with T L3 .

[0038] The latency times can, for example, correspond to the fixed sampling rates in the respective frequency bands A, B, C, D, E, F.

[0039] Preferably, the amplitudes A1, A2, A3 ...A nThe frequency components are compared with a predefined threshold value S. If the determined amplitudes are smaller than the predefined threshold value S, they are not taken into account in the calculation. In the present exemplary embodiment, the amplitude A1 is smaller than the threshold value, so the 35 Hz component is not taken into account when calculating the power. To determine the power, only the amplitudes A2 and A3 are taken into account. This threshold comparison allows insignificant frequency components, which may extend the measurement time, to be neglected, as they are insignificant for determining the power.

[0040] Fig. Figure 8 shows an example of the course of three different wavelets WL1, WL2, and WL3 in a fault situation. All three wavelet responses are above a specified threshold S. Wavelets below the threshold S would not be considered in the calculation. The first wavelet WL1 refers to a measurement at a frequency of 35 Hz. The second wavelet WL2 refers to a measurement at 29 Hz, and the third wavelet WL3 refers to a measurement at a frequency of 0.3 Hz. Accordingly, the corresponding latency times t L1 , t L2 and t L3 , where for the performance measurement the latency t L3 the lowest frequency must be waited for in order to obtain the power result at time t0 by summing up the measured values ​​for the individual frequencies.

[0041] In addition, a database (not shown) can be provided in which a trainable AI module is arranged, which is intended to assign corresponding wavelet pattern functions for the characteristic fault constellation, so that the assigned wavelet functions are selected depending on a recorded parameter characterizing a given fault.

[0042] The AI ​​module is trainable using an algorithm that processes a large number of learning data sets, each containing learning input values ​​for input variables and learning output values ​​for output variables. Using the AI ​​module, one or more input variables can be mapped to one or more output variables based on the parameters.

[0043] It should be noted that the frequency component current data and the frequency voltage data are each represented by wavelet function values ​​at time t0 according to Fig. 4. The time t0 is always the time at which the maximum amplitude of the wavelet occurs.

[0044] The length of a wavelet pattern is determined by the sampling rate, as shown in Fig. 4. If the sampling rate is relatively high, as in the frequency range C, D, the signal duration of the wavelet is relatively short. However, if the sampling rate is relatively low, as in the frequency range A, the signal duration of the wavelet is relatively long, ie, it is temporally stretched compared to the wavelet with a high sampling rate.

[0045] The invention utilizes a changing time and frequency resolution of wavelet to represent non-stationary signals, with the observation period and frequency being adapted to the signal to be observed.

[0046] The wavelet pattern functions stored in the database correspond to specific faults. If a match is found between the currently determined wavelet pattern or frequency spectrum of the wavelet function and a predefined wavelet pattern function, a characteristic fault can be easily identified. This fault can be caused by faults or changes in dynamics due to aging processes, fault detection of machine components, or changes in the operating behavior of power electronic systems and devices. This requires that the wavelet pattern functions for these faults or faults are available or have been empirically determined.

[0047] For example, these wavelet pattern functions can be specified empirically by using corresponding measurement data or learned using the AI ​​module.

Claims

[1] Device for determining a power in power grids (1) and / or for generating power patterns with - a current measuring device (2) for measuring a mains current present at a measuring point of the power grid (1), - a voltage measuring device (3) for measuring a mains voltage applied to the measuring point of the power grid (1), - a frequency-based A / D converter module (8) which is arranged such that the mains current and the mains voltage are sampled at predetermined frequency intervals (Δf) to produce digital frequency component current data and frequency component voltage data, - a first wavelet module (9) configured such that frequency component current data and frequency component voltage data are multiplied by a predetermined wavelet function to produce wavelet frequency component current data and wavelet frequency component voltage data, - a second wavelet module (12) which is configured such that the frequency component current data and the frequency component voltage data are shifted by a latency time as a function of a time value, so that the frequency component current data and frequency component voltage data of the respective frequency intervals are available at a predetermined time within a time interval after the occurrence of a fault in the network, wherein a length of the time interval is determined by a smallest frequency component of the current signal or voltage signal to be observed or evaluated, - a runtime correction module (10) configured to shift the wavelet frequency component current data and the wavelet frequency component voltage data by a latency time as a function of a frequency (f), so that the wavelet frequency component current data and the wavelet frequency component voltage data of the frequency intervals are time-synchronized, - a calculation module (15) for calculating power data and / or energy data from the frequency component voltage data and frequency component current data. [2] Device according to claim 1, characterized by that the A / D converter module (8) is designed to sample the mains current and the mains voltage at equidistant frequency intervals (Δf). [3] Device according to claim 1 or 2, characterized bythat the wavelet module (9) is configured to shift the predetermined wavelet function in the frequency plane by 90° to provide a first wavelet function for a real part and to provide a second wavelet function for an imaginary part of the current or voltage data. [4] Device according to one of claims 1 to 3, characterized by that the specified latency depends on the minimum frequency of the frequency component current data or frequency component voltage data. [5] Device according to one of claims 1 to 4, characterized by that the latency is determined based on a Heisenberg uncertainty principle. [6] Device according to one of claims 1 to 5, characterized by that the wavelet functions are weighted and / or validated depending on the frequency contents. [7] Device according to one of claims 1 to 6, characterized bythat only wavelet frequency component current data and wavelet frequency component voltage data that are above a specified threshold value (S) are taken into account in the calculation and / or evaluation. [8] Device according to one of claims 1 to 7, characterized by that a database is provided in which a trainable AI module is arranged, which is intended to assign corresponding wavelet pattern functions for characteristic fault constellations, so that the assigned wavelet function is selected depending on a recorded parameter characterizing a given fault. [9] Device according to one of claims 1 to 8, characterized bythat the AI ​​module is trainable by means of an algorithm that processes a plurality of read data sets, each containing learning input values ​​for input variables and learning output values ​​for output variables, and that is designed to map one or more input variables to one or more output variables based on the parameters.

Images