A photovoltaic inverter fault detection system based on electrical signals

Abstract

The present invention relates to the technical field of electrical fault detection, and specifically to a photovoltaic inverter fault detection system based on electrical signals, the system comprising: a signal acquisition module, which continuously collects the output electrical signals of the inverter, records the amplitude, frequency and phase of the electrical signals, and generates original signal records; based on the original signal records, the electrical signals are screened, noise and non-feature data are eliminated, and screened signal data are generated. In the present invention, the time-frequency characteristics of the signal are extracted in real time by continuous wavelet transform, thereby improving the sensitivity and accuracy of fault detection. This meticulous feature recognition method allows even tiny abnormal waveforms or frequency offsets to be captured, thereby achieving early warning of potential faults. At the same time, the fault detection threshold is dynamically adjusted through environmental parameters, allowing flexible response according to changes in the actual operating environment, reducing false alarms and missed alarms, and improving adaptability and reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical fault detection, and in particular to a photovoltaic inverter fault detection system based on electrical signals. Background Art

[0002] Photovoltaic inverters can convert the variable DC voltage generated by photovoltaic (PV) solar panels into AC power at mains frequency, which can be fed back to the commercial power transmission system or used by off-grid power grids. Photovoltaic inverters are one of the important balance of systems (BOS) in photovoltaic array systems. They can be used with general AC-powered equipment and have an important impact on the efficiency and utilization of photovoltaic power generation. However, since photovoltaic inverters are long-term affected by environmental and operating conditions such as temperature, humidity, and acidic corrosive gases, they are prone to various faults. Therefore, fault detection of photovoltaic inverters is an important part of their daily operation and maintenance.

[0003] However, fault detection in the prior art mostly relies on static thresholds and simple data processing methods, which often lead to inaccurate fault diagnosis under changing environmental conditions. Static thresholds are difficult to adapt to fluctuations in equipment parameters caused by changes in ambient temperature and humidity, and are prone to ignoring changes in equipment behavior caused by environmental factors, resulting in unreliable fault detection results. Therefore, improvements are needed. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a photovoltaic inverter fault detection system based on electrical signals.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: A photovoltaic inverter fault detection system based on electrical signals comprises:

[0006] The signal acquisition module continuously collects the output electrical signal of the inverter, records the amplitude, frequency and phase of the electrical signal, and generates an original signal record; based on the original signal record, the electrical signal is screened to remove noise and non-characteristic data, and the screened signal data is generated;

[0007] A feature analysis module applies continuous wavelet transform to the screened signal data, extracts the time-frequency features of the signal in real time, and generates feature extraction results; based on the feature extraction results, identifies abnormal waveforms and frequency offsets in the signal, and generates abnormal signal features;

[0008] A threshold adjustment module adjusts the threshold of fault detection according to the operating environment of the inverter to obtain an adjusted threshold; based on the adjusted threshold, classifies the abnormal signal features, verifies whether it is a fault signal, and generates a classified fault result;

[0009] The early warning and recording module, based on the classified fault result, determines that it is a fault signal, activates the real-time early warning, sends an alarm to the maintenance team, and generates a real-time fault alarm.

[0010] Preferably, the steps of obtaining the original signal record are:

[0011] Continuously collect the output electrical signal of the inverter, record the amplitude, frequency and phase information of the electrical signal, and obtain the characteristic data of the electrical signal;

[0012] Based on the electrical signal characteristic data, integrating the recorded information of amplitude, frequency and phase to form a comprehensive signal record;

[0013] Based on the comprehensive signal record, data formatting processing is performed to obtain the original signal record.

[0014] Preferably, the steps of obtaining the screening signal data are:

[0015] Based on the original signal record, a bandpass filter is used to perform frequency domain screening on the electrical signal to filter out high-frequency and low-frequency noises exceeding a set frequency range, and high-frequency fluctuations are eliminated by time smoothing to obtain denoised signal data;

[0016] Based on the denoised signal data, abnormal peaks and troughs in the data are identified and separated through statistical analysis, and the abnormal peaks and troughs are excluded to obtain pure signal data;

[0017] Based on the clean signal data, data standardization is performed to generate screening signal data.

[0018] Preferably, the steps of obtaining the feature extraction result are:

[0019] Selecting Morlet wavelet as a mother waveform, adjusting scale parameters and displacement parameters of wavelet transform according to the variability of the screening signal data, and obtaining wavelet transform parameters;

[0020] Based on the wavelet transform parameters, continuous wavelet transform is performed to obtain wavelet transform features, the formula is:

[0021] ;

[0022] in, Represents signal data in the time domain, represents the scale transformation, represents the time displacement, is the Morlet wavelet function, is the wavelet transform feature, which characterizes the frequency characteristics of the signal at scale a and displacement b;

[0023] Based on the wavelet transform characteristics, the energy distribution of the signal at different times and frequencies is analyzed, and the time-frequency characteristics of the signal, including the peak energy and the width of the main frequency band, are extracted to obtain a feature extraction result.

[0024] Preferably, the steps of acquiring the abnormal signal characteristics are:

[0025] Based on the feature extraction results, analyzing the energy distribution of each frequency band, identifying different energy peaks or valleys, and obtaining a candidate list of abnormal frequencies;

[0026] Based on the abnormal frequency candidate list, the energy difference measure between each candidate frequency and the adjacent frequency is calculated, and the calculation formula is:

[0027] ;

[0028] in, is the energy of the ith frequency region, is the average value of energy in all frequency regions, is the standard deviation of energy, is the weighted power exponent, is the energy difference measure, N is the total number of candidate frequency regions;

[0029] Based on the energy difference metric, abnormal candidates exceeding an energy difference threshold are screened to obtain abnormal signal features.

[0030] Preferably, the step of obtaining the adjusted threshold value is:

[0031] Collect inverter operation data, including temperature, humidity and power;

[0032] Based on the temperature, humidity and power, the adjusted threshold is calculated using the formula:

[0033] ;

[0034] in, is the basic fault detection threshold, is the overall adjustment factor, is the adjustment coefficient, represents the square of humidity, is the adjusted threshold, P is power, and T is temperature.

[0035] Preferably, the steps of obtaining the classification fault result are:

[0036] Based on the abnormal signal characteristics, each signal feature is scored, and the calculation formula is:

[0037] ;

[0038] in, is the weight of the ith feature, is the observed value of the ith feature, is the mean value of the ith feature, is the standard deviation of the ith feature, is the total number of features, is the rating result;

[0039] The scoring result is compared with the adjusted threshold value. If the scoring result exceeds the adjusted threshold value, it is confirmed as a fault signal. Otherwise, it is considered to be a normal operation, and a classified fault result is obtained.

[0040] Preferably, the steps of obtaining the real-time fault alarm are:

[0041] Based on the classified fault result, determining whether a fault state is indicated, and obtaining a fault confirmation state;

[0042] Based on the fault confirmation state, if the result shows a fault, a real-time early warning is activated to obtain an activated alarm signal;

[0043] Based on the activated alarm signal, an alarm message is sent to the maintenance team including the type and location of the fault.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are:

[0045] In the present invention, the time-frequency characteristics of the signal are extracted in real time through continuous wavelet transform, thereby improving the sensitivity and accuracy of fault detection. This meticulous feature recognition method allows even tiny abnormal waveforms or frequency offsets to be captured, thereby achieving early warning of potential faults. At the same time, the fault detection threshold is dynamically adjusted through environmental parameters, allowing flexible response according to changes in the actual operating environment, reducing false alarms and missed alarms, and improving adaptability and reliability. This adaptive adjustment mechanism ensures that the threshold setting not only reflects the real-time environmental impact, but also adapts to different operating conditions, ensuring the effectiveness and accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] See also Figure 1The present invention provides a technical solution: a photovoltaic inverter fault detection system based on electrical signals comprises:

[0049] The signal acquisition module continuously collects the output electrical signals of the inverter, records the amplitude, frequency and phase of the electrical signals, and generates original signal records; based on the original signal records, the electrical signals are screened to remove noise and non-characteristic data, and screened signal data is generated;

[0050] The feature analysis module applies continuous wavelet transform to the filtered signal data, extracts the time-frequency characteristics of the signal in real time, and generates feature extraction results; based on the feature extraction results, it identifies abnormal waveforms and frequency offsets in the signal and generates abnormal signal features;

[0051] The threshold adjustment module adjusts the threshold of fault detection according to the operating environment of the inverter to obtain an adjusted threshold; based on the adjusted threshold, classifies the abnormal signal features, verifies whether it is a fault signal, and generates a classified fault result;

[0052] The early warning and recording module, based on the classified fault results, activates the real-time early warning when it is determined to be a fault signal, sends an alarm to the maintenance team, and generates a real-time fault alarm.

[0053] The steps to obtain the original signal record are:

[0054] Continuously collect the output electrical signal of the inverter, record the amplitude, frequency and phase information of the electrical signal, and obtain the characteristic data of the electrical signal;

[0055] Based on the electrical signal characteristic data, the recorded information of amplitude, frequency and phase is integrated to form a comprehensive signal record;

[0056] Based on the comprehensive signal record, data formatting processing is performed to obtain the original signal record.

[0057] Specifically, a continuous monitoring device is deployed at the output end of the inverter to obtain real-time electrical signals and an internal sampling control function is called to cyclically obtain multiple continuous sampling values ​​at fixed time intervals. During the sampling process, pre-set parameters are used to judge the high-level state and the low-level state and summarize the instantaneous amplitude information of each sampling. The frequency change and phase offset between adjacent samples are obtained by recording the electrical signal fluctuations at each sampling point. In order to accurately record the amplitude, an adjustable resolution can be pre-set in the monitoring device and the corresponding measurement gear can be determined according to the rated output range of the inverter. If the amplitude rises rapidly in multiple consecutive samplings, the situation is marked as a high-amplitude fluctuation and the time of occurrence and the associated frequency value are recorded at the same time. In order to take into account the range from low frequency to high frequency bands, frequency scale information covering the target frequency band will be written into the monitoring device to ensure accurate estimation of the phase difference. After each sampling is completed, the amplitude curve and phase offset data recorded at that time will be temporarily stored and frequency analysis will be performed. The sampling values ​​​​are rearranged in sequence through the built-in software timer to prevent deviations caused by disorder. Subsequently, new observation values ​​are continuously superimposed on the complete sampling sequence after timing rearrangement to form a sampling set extended by time. The specific time points of the peaks and troughs can be further marked in this sampling set to facilitate subsequent processing. By repeating the above-mentioned acquisition and recording operations, the electrical signal observation samples in the same period are continuously enriched, and finally the electrical signal characteristic data are obtained.

[0058] Based on the electrical signal characteristic data obtained previously, a special data integration logic is used to merge the amplitude reference values ​​at each moment and synchronously import the record entries of phase offset and frequency change. First, the timestamp index in the electrical signal characteristic data is read and arranged in ascending order as a unified benchmark. The amplitude elements under each time index are retrieved one by one and the phase and frequency entries are matched according to the same time index. After the matching is completed, the corresponding relationship between amplitude and phase and the corresponding relationship between amplitude and frequency are summarized in the same data structure. If there are missing frequency records under certain time indexes, the data integration logic will interpolate the frequency elements of adjacent time indexes. Combined with the phase value, a temporary interpolation item is obtained. After the interpolation is completed, a new complete data record can be created at the corresponding index and marked as the data row generated by the interpolation. All time indexes and their corresponding amplitude, phase and frequency results are merged into a unified multi-column data table structure and a memory-level loop check is performed to ensure that there are no repeated timestamps and to ensure the consistency of the order. Subsequently, an overall scan is performed on the merged data table to determine whether the interpolation part exceeds the set maximum interpolation step limit. This limit value is generally inferred by experience and pre-set by the operation and maintenance team before the system is started. Finally, the amplitude, phase and frequency data obtained after matching and interpolation are organized into a comprehensive signal record.

[0059] Based on the comprehensive signal record obtained above, the internal data formatting module is called to rearrange and normalize the time index field and signal attribute field involved in the record content. In order to enable the subsequent processing stage to perform unified analysis for the same format, the order of each field will be set in the data formatting module and the field type will be clarified. For example, the amplitude and frequency data are defined as floating point types and the phase record is defined as a signed floating point type. Then the batch processing mechanism is started to read each line of the comprehensive signal record in sequence according to the time index and build a structured list in the memory. For abnormal records that appear in the formatting process, the exception handler inside the data formatting module identifies and eliminates them one by one. For records that lack key field content, filtering operations are performed before formatting and their time index source is marked in the task log of the data formatting module. After completing the type alignment and field normalization of all records, a new data structure is generated to clarify the correspondence between the time index and the amplitude, phase and frequency and ensure the consistency of the field type. This data structure is associated with the final indication written in the data formatting module to finally obtain the original signal record.

[0060] The steps to obtain the screening signal data are:

[0061] Based on the original signal record, a bandpass filter is applied to perform frequency domain screening on the electrical signal to filter out high-frequency and low-frequency noise that exceeds the set frequency range, and high-frequency fluctuations are eliminated through time smoothing to obtain denoised signal data;

[0062] Based on the denoised signal data, the abnormal peaks and troughs in the data are identified and separated through statistical analysis, and the abnormal peaks and troughs are eliminated to obtain pure signal data;

[0063] Based on the pure signal data, data standardization is performed to generate screening signal data.

[0064] Specifically, based on the original signal record obtained above, first set the required bandpass filter boundary frequency in the internal configuration and clarify a reference range. The upper and lower limits of the reference range can be derived from the rated operating frequency band of the equipment and the historical operation data summarized by the operation and maintenance personnel. If the equipment mainly works between 50Hz and 60Hz, it can be set and As a preliminary range, the bandpass filter routine is then called at the software level to operate on the recorded electrical signal. During this period, the signal frequency value at each sampling moment is compared point by point to determine whether it is within the above range. The part that falls outside the range is determined to be noise and attenuated during the filtering process. At the same time, time smoothing operations are required for sampling segments with obvious time domain fluctuations. Specifically, a sliding window length can be set within the program and the amplitudes of several adjacent sampling points can be calculated as the arithmetic mean or median. If the sliding window length is set too large, distortion may occur. Therefore, the operation and maintenance personnel generally give an empirical value after checking the past data fluctuations. For example, a window size that matches the fluctuation period is selected according to the sampling rate. When all sampling points have completed frequency domain judgment and time smoothing, the filtered, attenuated and smoothed data can be merged and output to finally obtain denoised signal data.

[0065] Based on the denoised signal data obtained above, a statistical analysis module is first pre-loaded in the software and adjacent comparisons are made for the amplitude of each sampling point. In order to identify the peak, the local maximum detection method can be used to determine whether the current amplitude is larger than several previous and subsequent points. If the conditions are met, it is determined to be a peak. When identifying the trough, it is necessary to check whether the current amplitude is smaller than several previous and subsequent points. In order to reduce repeated triggering or noise interference, a threshold can be set in the statistical analysis module as the minimum change in amplitude comparison. The threshold can be set by historical observations or field experience. For example, a value is selected based on the amplitude fluctuation during stable operation to determine whether the adjacent point can be used as a peak or trough. Once an abnormal peak or trough is identified, the corresponding time index together with its amplitude is marked as abnormal and removed from the current sequence. If the number of peak and trough detection results is too large, it is necessary to check again whether there is peak accumulation caused by multiple repeated sampling. After completing the above statistical analysis and excluding abnormal peaks and troughs, pure signal data is obtained.

[0066] Based on the previously obtained clean signal data, data standardization is performed so that the subsequent processes can use the same numerical range on the same scale. Specifically, the overall average value of the batch of clean signal data can be calculated first. and standard deviation And the amplitude of each sampling point is Normalization is performed. For scenarios with uneven distribution or extreme values, a more robust normalization method can be selected, but the calculation still needs to be performed based on the mean and standard deviation obtained previously. If the standard deviation is zero or very small in some sections, a preventive mechanism must be set in the software, such as skipping normalization or using an alternative value, to avoid division by zero in the calculation. After all samples have completed the corresponding normalization operations, they can be summarized and recorded at one time, and the filtered signal data is finally generated.

[0067] The steps to obtain feature extraction results are:

[0068] Select Morlet wavelet as the mother waveform, adjust the scale parameter and displacement parameter of wavelet transform according to the variability of the screening signal data, and obtain the wavelet transform parameters;

[0069] Based on the wavelet transform parameters, continuous wavelet transform is performed to obtain the wavelet transform features. The formula is:

[0070] ;

[0071] in, Represents signal data in the time domain, represents the scale transformation, represents the time displacement, is the Morlet wavelet function, is the wavelet transform feature, which characterizes the frequency characteristics of the signal at scale a and displacement b;

[0072] Based on the wavelet transform characteristics, the energy distribution of the signal at different times and frequencies is analyzed, and the time-frequency characteristics of the signal, including the peak energy and the width of the main frequency band, are extracted to obtain the feature extraction results.

[0073] Specifically, based on the previously obtained screening signal data, first set a numerical indicator for recording signal variability internally and compare the indicator with the amplitude fluctuation degree of the same batch of data, read the frequency fluctuation and amplitude difference degree of each sampling segment in the screening signal data, and combine the comparison result to determine whether it is necessary to modify the selected mother waveform type. If the Morlet wavelet is continued to be used, load the corresponding basic definition and initialization parameters in the software, including the reference center frequency and the normalization factor, and then calibrate the scale parameter and displacement parameter of the wavelet transform according to the variability of the screening signal data. The scale parameter is usually obtained by the same batch of signals in different time intervals. The frequency span shown is calculated. If the distribution range is wide, a larger value is assigned to cover multiple frequency bands. If the distribution is relatively concentrated, the value is appropriately reduced to focus on detecting key frequency bands. The displacement parameter is indexed and extracted by checking the position where the amplitude suddenly changes significantly on the sampling time axis and compared with the moving step table in the software. The moving step table is usually summarized by maintenance personnel based on previous test data and sets several reference scales, so that each time the difference between adjacent time periods reaches a certain ratio, a displacement update can be triggered. After completing the batch calibration of the scale parameters and the displacement parameters, a set of wavelet transform parameters are formed, and finally the wavelet transform parameters are obtained.

[0074] The benefit of the formula is that it uses variable scales and variable displacements to perform multi-scale analysis of the signal, thereby extracting detailed features in different time periods and different frequency ranges; represents the time domain signal obtained after the previous screening process, Represents the scale factor obtained by calculating the maximum frequency and minimum frequency distribution segments in the signal. For example, according to actual monitoring, the frequency range is between 48Hz and 62Hz. The operation and maintenance personnel use the empirical method to , It represents the displacement obtained by combining the time index position and continuous fluctuation difference statistics. For example, if the sampling time is 6 seconds and the amplitude fluctuates significantly between 2.5 seconds and 3.5 seconds, it can be determined based on the difference statistics. ;

[0075] Calculation process:

[0076] If you will Considered as arrive The amplitude of the time series can be expressed as A measured electrical signal function is stored in a sampling sequence, the Morlet wavelet function It can be regarded as a sine waveform with a Gaussian envelope, which is digitized by the center frequency and attenuation parameters obtained earlier. and Substitute into the integral formula one by one and Integrating from 0 seconds to 6 seconds, with the integration step length set to 0.001 seconds, the multi-level formula can be written as:

[0077] ;

[0078] In the integration phase, the measured sequence is multiplied point by point by the corresponding wavelet function value and accumulated. If the amplitude of the sampling sequence changes strongly in many places, the product with the wavelet function in the relevant section is relatively larger, and the final accumulation can be obtained (This value depends on the specific signal shape and wavelet function shape, 3.62 is an example measurement result). This result shows that the signal matches the Morlet wavelet function to a high degree at a given scale and displacement. If this value is significantly higher than other time periods in the same batch, it means that this time period has prominent frequency characteristics.

[0079] Based on the wavelet transform features obtained above, we first establish an energy distribution analysis program and write the corresponding time and frequency index mapping table internally. Then, we perform frequency-by-frequency calculation for each time interval in the wavelet transform features to find the corresponding energy value. The energy value can be obtained by traversing the absolute value of the wavelet coefficients and performing weighted summation. If the accumulated energy of a certain frequency range is much higher than other frequency segments in the same batch, it is marked as peak energy in the software. If the energy at the high-frequency or low-frequency end is too low and deviates too much from the overall mean, it can be marked as an edge segment in the energy distribution analysis program. During this period, a special numerical threshold will be set up to identify anomalies. Peak value: This threshold is calculated by adding a certain multiple of the standard deviation to the operating energy mean of the same type of equipment in historical monitoring. For example, the mean peak energy of the same type of equipment in the past is defined as E0 and its standard deviation is determined as Es. The operation and maintenance personnel have calculated that E0+1.2×Es often represents a significant and drastic fluctuation in energy. The analysis program sets the threshold accordingly and compares each item. When the energy corresponding to certain sampling time points exceeds the threshold, its frequency bandwidth is marked in the result set. Finally, after completing the traversal and judgment of all time indexes, the peak energy interval and the width of the main frequency band are summarized to obtain the feature extraction result.

[0080] The steps for obtaining abnormal signal features are:

[0081] Based on the feature extraction results, the energy distribution of each frequency band is analyzed, different energy peaks or valleys are identified, and a candidate list of abnormal frequencies is obtained;

[0082] Based on the abnormal frequency candidate list, the energy difference measure between each candidate frequency and the adjacent frequency is calculated. The calculation formula is:

[0083] ;

[0084] in, is the energy of the ith frequency region, is the average value of energy in all frequency regions, is the standard deviation of energy, is the weighted power exponent, is the energy difference measure, N is the total number of candidate frequency regions;

[0085] Based on the energy difference metric, the abnormal candidates exceeding the energy difference threshold are screened to obtain the abnormal signal features.

[0086] Specifically, based on the feature extraction results obtained above, a frequency band index list is first established internally and the energy levels of the corresponding time periods are read one by one. The energy values ​​of each frequency band are compared horizontally and stacked vertically to extract the frequency band range that may have obvious peaks or troughs. Then, multiple empirical thresholds are introduced in the software to determine whether a frequency band has a large energy fluctuation compared with other frequency bands. These thresholds are mainly derived from the comprehensive statistics of the previous operation monitoring data of similar equipment, and the operation and maintenance personnel select the specific energy difference dividing line when analyzing the historical data curve. After the energy value of each frequency band is compared with other frequency bands, if the difference exceeds the corresponding The corresponding threshold will be marked as a potential peak or trough. These potential frequency bands will be marked in the program in a list manner and associated with the corresponding time index and energy value. In order to distinguish between peaks and troughs, the degree of deviation of the relative energy value greater than or less than the global mean will be recorded while creating the mark. Then an aggregate matching operation will be performed to determine whether multiple peak segments with continuous time or similar energy values ​​can be merged into a candidate group. If the frequency difference and energy difference of adjacent peak segments are lower than the parallel threshold set by the operation and maintenance personnel in the historical statistics, they will be merged into a larger range of peak segments, and finally a candidate list of abnormal frequencies will be obtained.

[0087] The formula is beneficial in that it analyzes the difference between the candidate frequency energy and the global energy average and standard deviation, thereby highlighting the part that is too different from the mainstream energy level in a multi-band environment; It is The energy data of each frequency region in a specific time period is calculated and accumulated in the frequency domain. It is the arithmetic mean of the energy of all the candidate frequency regions at present. and divided by get, is the standard deviation of these energy values ​​obtained by distributed computing, The weighted power index is set by the operation and maintenance personnel within the corresponding frequency band according to the sensitivity of energy differences in previous observations. For example, To highlight the impact of larger deviations on the overall results;

[0088] Calculation process:

[0089] When analyzing a When there are examples of candidate frequency regions, first obtain them through monitoring records. , all energy values ​​are measured in the same power unit, and the calculation is , and then further calculate the four energy values ​​and the average value ,Will After bringing it in, we get:

[0090] ;

[0091] After step-by-step calculation, we can get , which shows that in this set of candidate frequency regions, the overall energy level has a certain degree of dispersion. When the difference exceeds the threshold set by the operation and maintenance personnel based on historical data, it can be regarded as an abnormality.

[0092] Based on the energy difference measurement results obtained above, a screening module is first loaded into the program to receive the energy difference of each candidate frequency. The value is compared with the energy difference threshold written by the operation and maintenance personnel in the system configuration. This threshold is usually obtained by statistically analyzing the data of multiple runs of the same type of equipment, and is finally fixed at a certain energy difference limit through multiple corrections and comparisons. The screening module compares the energy difference measurement value of each candidate frequency with the energy difference threshold one by one. If a candidate frequency is found to correspond to If the frequency is greater than the threshold, the candidate frequency is marked as an abnormal type and included in the output list. The remaining candidates that do not exceed the threshold are not further marked. After completing the batch comparison, these marking results are summarized to obtain the final abnormal signal characteristics.

[0093] The steps to obtain the adjusted threshold are:

[0094] Collect inverter operation data, including temperature, humidity and power;

[0095] Based on temperature, humidity and power, the adjusted threshold is calculated using the formula:

[0096] ;

[0097] in, is the basic fault detection threshold, is the overall adjustment factor, is the adjustment coefficient, represents the square of humidity, is the adjusted threshold, P is power, and T is temperature.

[0098] Specifically, by placing temperature measurement components and humidity measurement components on the inverter body and reading power information on the power output side, three types of operating data, namely temperature, humidity and power, are obtained respectively. The temperature measurement component generally adopts a digital temperature measurement sensor device and records the real-time degree in the main control module in degrees Celsius. The humidity measurement component obtains the relative humidity value through a capacitive humidity sensor and converts the measured percentage into a decimal form for storage according to a pre-defined quantization formula. The power data is usually read in real time through the inverter's own output power metering module and presented in kilowatts. In order to ensure data synchronization, the temperature, humidity and power acquisition timestamps will be compared in the internal system and matched with the closest time index. If any of the temperature, humidity or power data exceeds the normal range summarized by the operation and maintenance personnel in the past maintenance process, an identification will be added to the current record for subsequent verification. These abnormal identifications refer to the fluctuation standards in the operation and maintenance historical data and form an upper and lower threshold based on experience. If the above three types of data are observed to be within the pre-defined normal range at the same time index, it can be regarded as a valid measurement operation and stored in the system's operation data list. Subsequently, a simple correction is performed on the data collected in the list to avoid single spikes caused by occasional sensor jumps. After the correction is completed, the temperature, humidity and power are stored in the subsequent available data set, and finally the operation data containing temperature, humidity and power are obtained.

[0099] The benefit of the formula is that it simultaneously incorporates the three core operating data of temperature, humidity, and power into the calculation of the adjustment factor, thereby adaptively correcting the fault detection threshold under multiple environmental influences; It is the basic fault detection threshold, which is obtained by operation and maintenance personnel through statistical analysis based on a large amount of historical detection experience of similar systems. It is the total adjustment factor used to enlarge or reduce the overall correction amount. The adjustment coefficients represent the weights of the impact on the temperature index, denominator, and logarithmic terms, and the specific values ​​can be determined by segmented regression analysis of the large sample data collected in the early stage. The Celsius data is continuously read from the temperature measurement component and saved in the system. The relative humidity is measured by a digital sensor and converted into a decimal and then squared. , The inverter power output value is in kilowatts and is connected to the logarithmic operation;

[0100] Calculation process:

[0101] In a practical example, , observed by the temperature measurement component Celsius, and relative humidity from the humidity sensor , read by the power metering module Kilowatts, substitute these values ​​into the following formula:

[0102] Calculate first (rounding);

[0103] Then calculate ;

[0104] Find the exponential term divided by the denominator: ;

[0105] Recalculate ;

[0106] Add the two parts above: ;

[0107] Multiply by the total adjustment factor ;

[0108] Finally, the basic fault detection threshold Add together: , record this result as ;

[0109] The results show that under the current temperature, humidity and power conditions, the fault detection threshold has been dynamically increased to about 5.94. When the operation and maintenance system subsequently detects that the fault score result or related indicators exceed this value, it will further determine that the equipment is in a fault state.

[0110] The steps to obtain the classification fault results are:

[0111] Based on the abnormal signal characteristics, each signal feature is scored, and the calculation formula is:

[0112] ;

[0113] in, is the weight of the ith feature, is the observed value of the ith feature, is the mean value of the ith feature, is the standard deviation of the ith feature, is the total number of features, is the rating result;

[0114] The scoring result is compared with the adjusted threshold value. If the scoring result exceeds the adjusted threshold value, it is confirmed as a fault signal. Otherwise, it is considered to be normal operation and a classified fault result is obtained.

[0115] Specifically, the formula is beneficial in that it integrates the degree of difference of each signal feature relative to its distribution mean and standard deviation, and introduces weight factors on this basis to strengthen more important or more sensitive features, thereby obtaining a more accurate scoring result; It is used to measure the The weight of the importance of each feature is obtained by weighted regression analysis of historical operating condition records and multiple amplitude fluctuation monitoring data. is extracted from the abnormal signal features. observations, collected in real time by sensors or monitoring components and obtained after the previous data cleaning process, It is the first accumulation of the same type of equipment in similar operating environment over a long period of time. The average value of the features is calculated through a large number of samples and stored in the internal parameter library. It is The standard deviation of each feature is obtained by calculating the statistical variance. The total number of features depends on the fault identification requirements and the number of signal types. It is the scoring result generated at the end;

[0116] Calculation process:

[0117] In one example, let Indicates that there are three main features. The following parameters will be set based on the historical data and local monitoring information obtained previously. They are obtained by segmented fitting by operation and maintenance personnel based on the previous analysis of the importance of amplitude and frequency characteristics. The average is calculated from thousands of sample records. Correspondingly, it is obtained by taking the square root of the variance of each feature, Abnormal signal characteristics from the current observation period; Substitute the above values ​​one by one into:

[0118] Calculate item 1: ;

[0119] Calculate the second term: ;

[0120] Calculate item 3: ;

[0121] Adding the above results together ;

[0122] The results show that when the scoring results When it reaches 1.7352, combined with the current feature weight and feature fluctuation, it has shown a certain degree of abnormal amplitude. Whether to determine a fault still needs to be compared with the adjusted threshold obtained earlier.

[0123] According to the above scoring results , a comparison logic module is loaded in the software to read the score value and compare it one-to-one with the adjusted threshold value obtained earlier. The comparison logic module will pre-store the threshold value refined by the operation and maintenance team based on historical fault records. , and do a difference analysis between the two. When the score result is greater than When the fault type is output in the program, the subsequent fault reporting link will further process it. If the scoring result is lower than or equal to the threshold, the normal operation label is maintained and the fault warning operation is not performed. The comparison logic module analyzes the scoring results generated in each sampling period separately and retrieves the record information of the same batch for visual presentation after identifying the fault sign and reminds the maintenance personnel. After all comparison operations are completed, the classified fault results are obtained.

[0124] The steps to obtain real-time fault alarms are:

[0125] Based on the classification fault result, determine whether a fault state is indicated and obtain a fault confirmation state;

[0126] Based on the fault confirmation status, if the result shows a fault, a real-time warning is activated to obtain an activated alarm signal;

[0127] Based on the activated alarm signal, an alarm message is sent to the maintenance team including the type and location of the fault.

[0128] Specifically, according to the previously obtained classification fault results, a fault status judgment logic is set in the program to read the generated fault score and the fault classification result obtained after comparison with the adjusted threshold, and then verify the fault classification information corresponding to each sampling point one by one. If the classification result shows a fault type, an alarm entry is created in the internal temporary buffer area. If the classification result shows normal, no additional identification is made. At the same time, the program will load a matching abnormal record table to store the fault type code detected each time and attach the current time index to the record entry. The operation and maintenance personnel will formulate a set of identification code tables in the early stage to distinguish different fault phenomena. The type names and value ranges in these fault type code tables are determined in multiple field tests and historical data summaries. The type code of each alarm entry is cross-checked with the record table. If it matches any item in the record table, the corresponding fault type will be indicated. After completing this series of verifications, the program will regard the fault classification result as having further confirmation significance and generate a fault confirmation state.

[0129] Based on the fault confirmation status obtained previously, a trigger module is first loaded into the monitoring main control program. The trigger module polls the previously stored fault confirmation status in real time. Once a significant fault entry is found, the real-time warning process is started. To implement this process, it is necessary to first parse the fault type and occurrence time as well as the corresponding possible impact range from the fault confirmation status. Then the trigger module will call the internal alarm emission interface to generate an activation instruction. The generation of the activation instruction needs to be based on the trigger threshold established by the operation and maintenance personnel. When the degree of the fault confirmation status is greater than a certain empirical threshold, the highest level alarm will be triggered. If it is at a lower level, a medium or low level alarm will be triggered. The setting of these empirical thresholds is based on the historical fault frequency and the degree of impact on equipment performance. After activation, the system will create a tag field in the monitoring main control program, set its value to the activated alarm signal, and continue to track and associate the handling of the alarm signal with subsequent maintenance in subsequent processes.

[0130] Based on the activated alarm signal obtained earlier, the message management module is started to write the fault type and location information into the alarm message content. The message management module contains a communication interface that can be connected to the maintenance team's duty system. First, the fault type code is extracted from the activated alarm signal and the internal maintenance dictionary is retrieved to obtain the text description corresponding to the type code. Then, the fault location information is integrated into the message subject according to the description. In order to accurately locate the specific component location of the inverter, it is necessary to compile the component number and record the installation location coordinates of each component during daily equipment registration. Therefore, when the early warning is triggered, the component number and actual coordinate information can be summarized into the alarm message. Then, the alarm message is pushed to the monitoring terminal of the maintenance team through the communication interface to facilitate the team to know in time and carry out necessary processing. For multiple fault alarms, they are output in a hierarchical list in chronological order in the system, and finally an alarm message with the fault type and location is sent to the maintenance team.

Claims

1. A photovoltaic inverter fault detection system based on electrical signals, characterized in that: The system comprises: The signal acquisition module continuously collects the output electrical signal of the inverter, records the amplitude, frequency and phase of the electrical signal, and generates an original signal record; based on the original signal record, the electrical signal is screened to remove noise and non-characteristic data, and the screened signal data is generated; A feature analysis module applies continuous wavelet transform to the screened signal data, extracts the time-frequency features of the signal in real time, and generates feature extraction results; based on the feature extraction results, identifies abnormal waveforms and frequency offsets in the signal, and generates abnormal signal features; A threshold adjustment module adjusts the threshold of fault detection according to the operating environment of the inverter to obtain an adjusted threshold; based on the adjusted threshold, classifies the abnormal signal features, verifies whether it is a fault signal, and generates a classified fault result; An early warning and recording module, based on the classified fault result, activates a real-time early warning when a fault signal is determined, sends an alarm to the maintenance team, and generates a real-time fault alarm; The steps of acquiring the abnormal signal characteristics are: Based on the feature extraction results, analyzing the energy distribution of each frequency band, identifying different energy peaks or valleys, and obtaining a candidate list of abnormal frequencies; Based on the abnormal frequency candidate list, the energy difference measure between each candidate frequency and the adjacent frequency is calculated, and the calculation formula is: ； in, is the energy of the ith frequency region, is the average value of energy in all frequency regions, is the standard deviation of energy, is the weighted power exponent, is the energy difference measure, N is the total number of candidate frequency regions; Based on the energy difference metric, screening abnormal candidates exceeding an energy difference threshold to obtain abnormal signal features; The steps for obtaining the adjusted threshold value are: Collect inverter operation data, including temperature, humidity and power; Based on the temperature, humidity and power, the adjusted threshold is calculated using the formula: ； in, is the basic fault detection threshold, is the overall adjustment factor, is the adjustment coefficient, represents the square of humidity, is the adjusted threshold, P is power, and T is temperature.

2. The photovoltaic inverter fault detection system based on electrical signals according to claim 1, characterized in that: The steps of obtaining the original signal record are: Continuously collect the output electrical signal of the inverter, record the amplitude, frequency and phase information of the electrical signal, and obtain the characteristic data of the electrical signal; Based on the electrical signal characteristic data, integrating the recorded information of amplitude, frequency and phase to form a comprehensive signal record; Based on the comprehensive signal record, data formatting processing is performed to obtain the original signal record.

3. The photovoltaic inverter fault detection system based on electrical signals according to claim 1, characterized in that: The steps for obtaining the screening signal data are: Based on the original signal record, a bandpass filter is used to perform frequency domain screening on the electrical signal to filter out high-frequency and low-frequency noises that exceed the set frequency range, and high-frequency fluctuations are eliminated by time smoothing to obtain denoised signal data; Based on the denoised signal data, abnormal peaks and troughs in the data are identified and separated through statistical analysis, and the abnormal peaks and troughs are excluded to obtain pure signal data; Based on the clean signal data, data standardization is performed to generate screening signal data.

4. The photovoltaic inverter fault detection system based on electrical signals according to claim 1, characterized in that: The steps for obtaining the feature extraction results are: Selecting Morlet wavelet as a mother waveform, adjusting scale parameters and displacement parameters of wavelet transform according to the variability of the screening signal data, and obtaining wavelet transform parameters; Based on the wavelet transform parameters, continuous wavelet transform is performed to obtain wavelet transform features, the formula is: ； in, Represents signal data in the time domain, represents the scale transformation, represents the time displacement, is the Morlet wavelet function, is the wavelet transform feature, which characterizes the frequency characteristics of the signal at scale a and displacement b; Based on the wavelet transform characteristics, the energy distribution of the signal at different times and frequencies is analyzed, and the time-frequency characteristics of the signal, including the peak energy and the width of the main frequency band, are extracted to obtain a feature extraction result.

5. The photovoltaic inverter fault detection system based on electrical signals according to claim 1, characterized in that: The steps for obtaining the classification fault result are: Based on the abnormal signal characteristics, each signal feature is scored, and the calculation formula is: ； in, is the weight of the ith feature, is the observed value of the ith feature, is the mean value of the ith feature, is the standard deviation of the ith feature, is the total number of features, is the rating result; The scoring result is compared with the adjusted threshold value. If the scoring result exceeds the adjusted threshold value, it is confirmed as a fault signal. Otherwise, it is considered to be a normal operation, and a classified fault result is obtained.

6. The photovoltaic inverter fault detection system based on electrical signals according to claim 1, characterized in that: The steps of obtaining the real-time fault alarm are: Based on the classified fault result, determining whether a fault state is indicated, and obtaining a fault confirmation state; Based on the fault confirmation state, if the result shows a fault, a real-time early warning is activated to obtain an activated alarm signal; Based on the activated alarm signal, an alarm message is sent to the maintenance team including the type and location of the fault.

Images