A reactive power compensation device fault automatic early warning system based on intelligent algorithm

Abstract

The application discloses a kind of reactive power compensation equipment fault automatic early warning system based on intelligent algorithm, belong to the technical field of smart grid, including fault data acquisition module, feature processing module, fault diagnosis evaluation module and visual module, combine Matlab software, Simulink software, configurable point number fast Fourier transform algorithm, fluorescence imaging technology, capacitive detection technology and relevant intelligent algorithm, from three angles of surge arrester, capacitor and electric reactor, the analysis of reactive power compensation equipment fault is carried out, and the visual output of reactive power compensation equipment fault diagnosis model output result is realized in the form of chart.The application realizes the intelligent, precision diagnosis and visual early warning of reactive power compensation equipment fault by comprehensively collecting surge arrester, capacitor, electric reactor and overall compensation multidimensional data.

Description

An automatic early warning system for reactive power compensation equipment based on intelligent algorithms Technical Field

[0001] This invention belongs to the field of smart grid technology, specifically relating to an automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms. Background Technology

[0002] Reactive power compensation devices are a core guarantee for the stable operation of distribution networks, effectively preventing problems such as decreased active power output of distribution transformers, increased line losses in the distribution network, and excessively low voltage at the end of lines. With economic development, the demand for reactive power in distribution networks continues to grow, leading to a significant increase in the number of reactive power compensation devices deployed. Currently, most traditional reactive power compensation cabinets on the market are assembled from discrete components, exhibiting inherent defects such as complex internal nodes, lengthy wiring, and high internal losses. This not only causes significant inconvenience for wiring and daily maintenance but also results in low protection performance and operational reliability; most devices may fail to operate after about a year due to malfunctions. Furthermore, these devices have low integration and complex structures, making it difficult to meet the development needs of distribution networks for equipment miniaturization, standardization, and capacity expansion upgrades.

[0003] In the field of online monitoring and fault early warning, a mature online fault monitoring system for reactive power compensation equipment in power distribution has not yet been established. Traditional centralized reactive power compensation devices lack targeted capacitor protection design, fault status interlocking functions, and fault information uploading mechanisms, making it impossible to upload the equipment's operating status to the main station in a timely manner. This obstructs information flow and consequently hinders the device's ability to function properly in voltage regulation. Furthermore, a large number of reactive power compensation devices in the power grid are generally not equipped with online monitoring and fault diagnosis functions. Since compensation points are widely distributed in the distribution network, it is difficult to locate and repair core components such as capacitors in a timely manner once they fail. This not only affects users' normal power consumption but also results in unnecessary waste of human and material resources.

[0004] Traditional fault monitoring and early warning methods for reactive power compensation equipment have significant limitations, relying heavily on manual inspections and single-dimensional data analysis, resulting in low sensitivity and high false alarm rates. With the rapid advancement of smart grid construction and the increasingly widespread application of reactive power compensation devices in the power grid, the technical shortcomings in real-time monitoring and fault handling are becoming increasingly apparent. Failure to detect and address equipment faults promptly will prolong power outages and increase economic losses. Manual inspections are insufficient to cover all compensation points and cannot provide real-time fault warnings, making them inadequate for the high demands of smart grids on power reliability and quality. Therefore, there is an urgent need to establish an efficient and accurate automatic fault early warning system. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide an automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms. By comprehensively collecting multi-dimensional data of surge arresters, capacitors, reactors and overall compensation, the system realizes intelligent, accurate diagnosis and visual early warning of reactive power compensation equipment faults.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms, comprising:

[0007] The fault data acquisition module is used to collect fault monitoring data;

[0008] The feature processing module, based on fault monitoring data, combines Matlab software, Simulink software, configurable point-based fast Fourier transform algorithm, fluorescence imaging technology, and capacitive detection technology to acquire capacitor feature data, reactor feature data, and fault analysis data.

[0009] The fault diagnosis and assessment module, based on fault monitoring data, capacitor characteristic data, reactor characteristic data, and fault analysis data, uses a weighted summation method to calculate the fault state diagnosis coefficient of the zinc oxide surge arrester. Then, it constructs a capacitor fault assessment model using a decision tree algorithm to obtain the capacitor fault type. Next, it uses a random forest algorithm to construct a reactor fault assessment model to obtain the reactor fault type. Finally, it uses a neural network algorithm to construct a reactive power compensation equipment fault diagnosis model to obtain a comprehensive fault type diagnosis number.

[0010] The visualization module uses charts to visualize the output results of the fault diagnosis model for reactive power compensation equipment.

[0011] Preferably, in the fault data acquisition module, the fault monitoring data acquisition process includes:

[0012] Set up an online monitoring device and deploy different types of data acquisition equipment on the online monitoring device to collect fault monitoring data, including surge arrester data, capacitor data, reactor data and overall compensation monitoring data;

[0013] The surge arrester data includes the leakage current waveform, number of lightning strikes, valve plate temperature, and lightning absorbed energy of the zinc oxide surge arrester; the capacitor data includes steady-state electrical quantity data, transient electrical quantity data, and fault electrical quantity data of the power capacitor. Among them, the steady-state electrical quantity data includes the three-phase current, three-phase voltage, active power, and apparent power of the power capacitor; the transient electrical quantity data is the circuit breaker over-operating voltage of the power capacitor; and the fault electrical quantity data includes the internal pressure and partial discharge signal waveform of the power capacitor; the reactor data includes the winding current, winding voltage, solid insulation failure sound, inter-turn short-circuit discharge sound, and current waveform during inter-turn short circuit of the dry-type air-core reactor; the overall compensation monitoring data includes the total current, power factor, and bus voltage of the compensation circuit, the output voltage of the power module in the online monitoring device, the data upload rate of the online monitoring device, and the output voltage of the inverter;

[0014] For the leakage current waveform of zinc oxide surge arrester, the partial discharge signal waveform of power capacitor and the current waveform of dry air reactor during inter-turn short circuit, wavelet threshold noise reduction algorithm is used to eliminate random noise in the electromagnetic environment. Short-time Fourier transform is used to filter out the environmental background noise in the solid insulation damage sound of dry air reactor and the inter-turn short circuit discharge sound. Data cleaning and standardization processing are performed on the lightning strike number of zinc oxide surge arrester, valve plate temperature and lightning strike absorbed energy, three-phase current and three-phase voltage of power capacitor, circuit breaker over-operating voltage and internal pressure, winding current of dry air reactor and overall compensation monitoring data.

[0015] Based on the clock synchronization function of the FPGA module, the acquisition time of surge arrester data, capacitor data, reactor data and overall compensation monitoring data is synchronized, and the pre-processed reactive power compensation equipment fault monitoring data is integrated to generate a reactive power compensation equipment fault monitoring dataset.

[0016] Preferably, in the feature processing module, the process of acquiring capacitor feature data and reactor feature data includes:

[0017] Capacitor characteristic data includes the capacitance change rate and loss factor of power capacitors; reactor characteristic data includes the equivalent resistance, equivalent reactance, and loss factor of dry-type air-core reactors.

[0018] The capacitance of the power capacitor is calculated using the three-phase current and three-phase voltage. The loss factor of the power capacitor is calculated using the ratio of its active power to its apparent power.

[0019] Matlab software was used to preprocess the winding current and winding voltage of the dry-type air-core reactor, and then the effective values ​​of the voltage and current of the dry-type air-core reactor were calculated. The fundamental component was extracted by Fourier transform, and the phase difference between the voltage and current of the dry-type air-core reactor was determined.

[0020] A series equivalent circuit model of a dry-type air-core reactor was built in Simulink software. The winding current of the dry-type air-core reactor was used as the input signal to drive the series equivalent circuit model. Through simulation and actual measurement, the resistance and inductance of the dry-type air-core reactor in the series equivalent circuit model were adjusted and obtained. The effective voltage value, effective current value, and phase difference between voltage and current of the dry-type air-core reactor output in the series equivalent circuit model were fitted with the actual effective voltage value, effective current value, and phase difference between voltage and current of the dry-type air-core reactor to obtain the resistance and inductance of the fitted dry-type air-core reactor.

[0021] Based on the fitted resistance and inductance of the dry-type air-core reactor, calculate the equivalent resistance and equivalent reactance of the dry-type air-core reactor;

[0022] The loss factor of a dry-type air-core reactor can be calculated by the ratio of its equivalent resistance to its equivalent reactance.

[0023] The characteristic data of capacitors and reactors are integrated into the fault monitoring dataset of reactive power compensation equipment.

[0024] Preferably, in the feature processing module, the process of acquiring fault analysis data includes:

[0025] The fault analysis data includes the fundamental and third harmonic amplitudes of the resistive component of the leakage current of the zinc oxide surge arrester, the partial discharge of the power capacitor, the oil leakage in the power capacitor tank, and the fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit of the dry-type air-core reactor.

[0026] Using a configurable-point Fast Fourier Transform algorithm, the leakage current waveform of a zinc oxide surge arrester is analyzed to obtain the fundamental amplitude and third harmonic amplitude of the resistive component of the leakage current. The partial discharge signal waveform of a power capacitor is analyzed to obtain the partial discharge quantity. The current waveform of a dry-type air-core reactor during an inter-turn short circuit is analyzed to obtain the fundamental frequency, spectral amplitude, and power spectral density of the current. The obtained fault analysis data is then integrated into the reactive power compensation equipment fault monitoring dataset.

[0027] Preferably, in the feature processing module, the process of obtaining the oil leakage amount in the power capacitor tank includes:

[0028] Using fluorescence imaging technology, ultraviolet light pulses are generated and emitted by an ultraviolet light pulse emission unit. After reflection and focusing, the ultraviolet light pulses illuminate the area to be monitored in the power capacitor oil tank. During the fluorescence lifetime, the location of the leaking oil that produces fluorescence is photographed to obtain the fluorescence intensity signal of the power capacitor leaking oil. The fluorescence intensity signal of the power capacitor leaking oil is then stitched together and the location is restored. A fluorescence intensity threshold is set, and the fluorescence intensity signal of the power capacitor leaking oil that exceeds the threshold is analyzed to identify the leakage situation in the area to be monitored in the power capacitor oil tank.

[0029] Based on the identification results of leakage in the monitored area of ​​the power capacitor tank, capacitive detection technology is adopted, and a capacitive detection sensor is configured. Based on the dielectric constant of insulating oil and air, an oil leakage signal is generated. The amount of oil leakage in the power capacitor tank is obtained by referring to the database of the correspondence between capacitance and leakage amount.

[0030] The oil leakage data in the power capacitor oil tank is integrated into the reactive power compensation equipment fault monitoring data set.

[0031] Preferably, in the fault diagnosis and assessment module, the process of calculating the fault state diagnosis coefficient of the zinc oxide surge arrester using the weighted summation method includes:

[0032] Extract the number of lightning strikes, valve plate temperature, and lightning strike absorbed energy of zinc oxide surge arresters from the surge arrester data, as well as the resistive component of leakage current of zinc oxide surge arresters from the fault analysis data;

[0033] The fault state diagnosis coefficient of the zinc oxide surge arrester is calculated by assigning weights to the number of lightning strikes, valve plate temperature, lightning strike absorbed energy, and the resistive component of leakage current, respectively, and then using the weighted summation method.

[0034] Preferably, in the fault diagnosis and assessment module, the process of constructing a capacitor fault assessment model and obtaining the capacitor fault type using a decision tree algorithm includes:

[0035] Capacitor fault types include no fault in power capacitor, leakage, bulging, explosion, and fuse blowing. Each capacitor fault type is assigned a number, and the capacitor fault type number is obtained.

[0036] Extract the capacitance change rate, loss factor, circuit breaker over-operating voltage, internal pressure, partial discharge, and oil leakage in the tank from the reactive power compensation equipment fault monitoring dataset, and convert the extracted data into the first training set and the first test set.

[0037] The decision tree algorithm is used, with the first training set data as input data and the capacitor fault type number as output data. The nonlinear relationship between each item of the first training set data and the capacitor fault type number is learned to obtain the trained capacitor fault assessment model.

[0038] The first test set data is input into the trained capacitor fault assessment model, the parameters of the capacitor fault assessment model are adjusted, the performance of the capacitor fault assessment model is optimized, and the final capacitor fault assessment model is obtained.

[0039] Based on the current capacitance change rate, loss factor, circuit breaker over-operation voltage, internal pressure, partial discharge, and oil leakage in the tank, the corresponding capacitor fault type number is output, and the corresponding capacitor fault type is matched by the capacitor fault type number.

[0040] Preferably, in the fault diagnosis and assessment module, the process of constructing a reactor fault assessment model using the random forest algorithm to obtain the reactor fault type includes:

[0041] The types of reactor faults include no faults in dry-type air-core reactors, inter-turn short circuits, winding-to-ground short circuits, winding open circuits, and insulation aging. Each type of reactor fault is assigned a number, and the reactor fault type number is obtained.

[0042] The fault monitoring dataset of reactive power compensation equipment is extracted to obtain the solid insulation damage sound, inter-turn short circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and fundamental frequency, spectral amplitude and power spectral density of the current during inter-turn short circuit of the dry air reactor, and converted into the second training set and the second test set.

[0043] Using the random forest algorithm, the second training set data is used as input and the reactor fault type number is used as output. The nonlinear relationship between each item of the second training set data and the reactor fault type number is learned to obtain the trained reactor fault assessment model.

[0044] The second test set data is input into the trained reactive fault assessment model. Bayesian optimization is used to adjust the parameters of the reactive fault assessment model, optimize the performance of the reactive fault assessment model, and obtain the final reactive fault assessment model.

[0045] The solid insulation damage sound, inter-turn short-circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and the fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit of the current dry-type air-core reactor are input into the final reactor fault assessment model to obtain the reactor fault type number, and the corresponding reactor fault type is obtained by matching.

[0046] Preferably, in the fault diagnosis and assessment module, the process of constructing a fault diagnosis model for reactive power compensation equipment using a neural network algorithm to obtain a comprehensive fault type diagnosis number includes:

[0047] Set the threshold for the fault state diagnosis coefficient of zinc oxide surge arrester, determine the fault diagnosis type of zinc oxide surge arrester, assign a number to each fault diagnosis type of zinc oxide surge arrester, and obtain the fault type number of zinc oxide surge arrester.

[0048] The fault type numbers of zinc oxide surge arresters, capacitors, reactors, and overall compensation monitoring data are integrated and transformed into the third training set and the third test set.

[0049] Using a neural network algorithm, the third training set data is set as input and the comprehensive fault type diagnosis number is set as output. The nonlinear relationship between each third training set data and the comprehensive fault type diagnosis number is learned to obtain a trained fault diagnosis model for reactive power compensation equipment.

[0050] The third test set data is input into the trained reactive power compensation equipment fault diagnosis model. The parameters of the reactive power compensation equipment fault diagnosis model are adjusted to optimize its performance and obtain the final reactive power compensation equipment fault diagnosis model. Combining the current fault type numbers of zinc oxide surge arresters, capacitors, reactors, and overall compensation monitoring data, the corresponding comprehensive fault type diagnosis number is output.

[0051] Preferably, in the visualization module, the process of visualizing the output results of the reactive power compensation equipment fault diagnosis model in the form of charts includes:

[0052] The actual fault types of reactive power compensation equipment include fault diagnosis types of zinc oxide surge arresters, capacitor fault types, and reactor fault types;

[0053] Based on the comprehensive fault type diagnosis number, the fault diagnosis types of zinc oxide surge arresters, capacitors and reactors are analyzed respectively.

[0054] The number of occurrences of fault diagnosis types of zinc oxide surge arresters, capacitor fault types, and reactor fault types were statistically analyzed, along with the total number of occurrences of actual fault types of reactive power compensation equipment. The proportions of the number of faults of zinc oxide surge arresters, power capacitors, and dry-type air-core reactors in the total number of occurrences of actual fault types of reactive power compensation equipment were calculated.

[0055] Using the Origin plotting tool, a pie chart of the actual fault types of reactive power compensation equipment is drawn. Combined with the OriginScript script, the actual fault types of reactive power compensation equipment are marked in real time, realizing the visualization output of the fault diagnosis model of reactive power compensation equipment.

[0056] The beneficial effects of this invention are:

[0057] This invention integrates Matlab and Simulink software with configurable point-based Fast Fourier Transform, fluorescence imaging, and capacitive detection technologies to achieve comprehensive acquisition and precise processing of fault monitoring data. The system not only covers multi-dimensional operational data from surge arresters, capacitors, reactors, and the overall compensation circuit, but also eliminates environmental interference through wavelet threshold denoising and short-time Fourier transform. Combined with FPGA clock synchronization, it ensures data consistency and ultimately accurately extracts key features such as leakage current resistive component harmonics, capacitance change rate, oil leakage, and equivalent reactance. This completely solves the problems of traditional early warning data being single-dimensional and lacking accuracy, achieving full-scenario, high-precision monitoring of the operating status of reactive power compensation equipment.

[0058] This invention utilizes multi-level intelligent algorithms such as weighted summation, decision trees, random forests, and neural networks to construct a complete system from single-device fault identification to multi-component collaborative diagnosis. By using targeted models to determine the fault types of zinc oxide surge arresters, power capacitors, and dry-type air-core reactors, and then outputting accurate fault numbers through a comprehensive diagnostic model, the intelligence level of fault identification is significantly improved, effectively reducing the false positive and false negative rates of traditional manual judgment. Simultaneously, the visualization module generates and marks fault type pie charts in real time using the Origin plotting tool, providing a clear visual representation of fault distribution and proportion, offering efficient support for operation and maintenance decisions, significantly reducing manpower and material resource consumption, shortening fault handling time, and effectively ensuring the stable operation of the distribution network and power supply quality. Attached Figure Description

[0059] Figure 1 is a schematic diagram of the system of the present invention. Detailed Implementation

[0060] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0061] As shown in Figure 1, an automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms includes a fault data acquisition module, a feature processing module, a fault diagnosis and evaluation module, and a visualization module, wherein the modules are interconnected.

[0062] The fault data acquisition module is used to collect fault monitoring data, including surge arrester data, capacitor data, reactor data, and overall compensation monitoring data, providing comprehensive and accurate raw data support for subsequent fault diagnosis.

[0063] The feature processing module uses capacitor data to calculate capacitor feature data, obtains reactor feature data through Matlab software, Simulink software and reactor data, and obtains fault analysis data based on reactive power compensation equipment fault monitoring data, combined with configurable point-number fast Fourier transform algorithm, fluorescence imaging technology and capacitive detection technology, thereby achieving data dimension enhancement and accurate extraction of key features.

[0064] The fault diagnosis and assessment module first uses a weighted summation method, combining surge arrester data and fault analysis data, to calculate the fault state diagnosis coefficient of the zinc oxide surge arrester. Then, through decision tree algorithm, capacitor data, capacitor feature data, and fault analysis data, a capacitor fault assessment model is constructed to obtain the capacitor fault type. Next, combining random forest algorithm, reactor data, reactor feature data, and fault analysis data, a reactor fault assessment model is constructed to obtain the reactor fault type. Finally, using a neural network algorithm, the fault state diagnosis coefficient of the zinc oxide surge arrester, the capacitor fault type, the reactor fault type, and the overall compensation monitoring data are comprehensively analyzed to construct a reactive power compensation equipment fault diagnosis model and obtain a comprehensive fault type diagnosis number. This achieves comprehensive fault diagnosis of reactive power compensation equipment from three perspectives: zinc oxide surge arrester, power capacitor, and dry-type air-core reactor, and achieves collaborative diagnosis of multiple equipment fault types.

[0065] The visualization module, based on the output of the reactive power compensation equipment fault diagnosis model, obtains the actual fault type of the reactive power compensation equipment and visualizes the actual fault type of the reactive power compensation equipment in the form of charts, which significantly enhances the intuitiveness of fault information and the efficiency of reactive power compensation equipment fault decision support.

[0066] In the fault data acquisition module, the fault monitoring data acquisition process includes:

[0067] An online monitoring device is installed, which has strong customizable configuration capabilities and deploys different types of data acquisition equipment on the device to collect fault monitoring data. The data acquisition equipment includes oscilloscopes, lightning strike counters, temperature sensors, energy meters, current transformers, voltage transformers, power analyzers, pressure sensors, partial discharge detectors, ultrasonic sensors, ultrasonic detectors, power quality monitoring devices, and network performance testers. The fault monitoring data includes surge arrester data, capacitor data, reactor data, and overall compensation monitoring data.

[0068] The surge arrester data includes the leakage current waveform, number of lightning strikes, valve plate temperature, and lightning absorbed energy of the zinc oxide surge arrester; the capacitor data includes steady-state electrical quantity data, transient electrical quantity data, and fault electrical quantity data of the power capacitor. Among them, the steady-state electrical quantity data includes the three-phase current, three-phase voltage, active power, and apparent power of the power capacitor; the transient electrical quantity data is the circuit breaker over-operating voltage of the power capacitor; and the fault electrical quantity data includes the internal pressure and partial discharge signal waveform of the power capacitor; the reactor data includes the winding current, winding voltage, solid insulation failure sound, inter-turn short-circuit discharge sound, and current waveform during inter-turn short circuit of the dry-type air-core reactor; the overall compensation monitoring data includes the total current, power factor, and bus voltage of the compensation circuit, the output voltage of the power module in the online monitoring device, the data upload rate of the online monitoring device, and the output voltage of the inverter. Among them, the power module is the built-in core component of the online monitoring device, and the inverter is the external auxiliary equipment of the online monitoring device, whose function is to solve the matching problem between the DC power supply environment and the AC power demand of the online monitoring device;

[0069] Using an oscilloscope, lightning strike counter, temperature sensor, and energy meter, the leakage current waveform, number of lightning strikes, valve plate temperature, and lightning strike absorbed energy of the zinc oxide surge arrester were collected. Current transformers were used to collect the three-phase current of the power capacitor, the winding current of the dry-type air-core reactor, and the total current of the compensation circuit. Voltage transformers were used to collect the three-phase voltage of the power capacitor, the over-operating voltage of the circuit breaker, and the winding voltage of the dry-type air-core reactor. A power analyzer was used to collect the active power and apparent power of the power capacitor. Combined with a pressure sensor and partial discharge detector, the internal pressure and partial discharge signal waveforms of the power capacitor were collected. Ultrasonic sensors, ultrasonic detectors, and an oscilloscope were used to collect the solid insulation breakage sound, inter-turn short-circuit discharge sound, and current waveform during inter-turn short circuits of the dry-type air-core reactor. Voltage transformers were used to collect the bus voltage of the compensation circuit, the output voltage of the power module in the online monitoring device, and the output voltage of the inverter. A power quality monitoring device was used to collect the power factor of the compensation circuit. A network performance tester was used to collect the data upload rate of the online monitoring device.

[0070] For the leakage current waveform of zinc oxide surge arrester, the partial discharge signal waveform of power capacitor and the current waveform of dry air reactor during inter-turn short circuit, wavelet threshold noise reduction algorithm is used to eliminate random noise in the electromagnetic environment of these waveforms. Short-time Fourier transform is used to filter out the environmental background noise in the solid insulation damage sound of dry air reactor and the inter-turn short circuit discharge sound. Data cleaning and standardization processing are performed on the lightning strike number of zinc oxide surge arrester, valve plate temperature and lightning strike absorbed energy, three-phase current and three-phase voltage of power capacitor, circuit breaker over-operating voltage and internal pressure, winding current of dry air reactor and overall compensation monitoring data.

[0071] Based on the clock synchronization function of the FPGA module, the acquisition time of surge arrester data, capacitor data, reactor data and overall compensation monitoring data is synchronized, and the pre-processed reactive power compensation equipment fault monitoring data is integrated to generate a reactive power compensation equipment fault monitoring dataset.

[0072] In the feature processing module, the process of acquiring capacitor feature data and reactor feature data includes:

[0073] Capacitor characteristic data includes the capacitance change rate and loss factor of power capacitors; reactor characteristic data includes the equivalent resistance, equivalent reactance, and loss factor of dry-type air-core reactors.

[0074] The capacitance of a power capacitor is calculated using its three-phase current and three-phase voltage. The loss factor is calculated using the ratio of active power to apparent power. The capacitance C is related to the capacitance change rate. The calculation formula is as follows:

[0075] ;

[0076] ;

[0077] in, and These represent the three-phase current and three-phase voltage of the power capacitor, respectively, where f is the power frequency (grid frequency), usually taken as 50Hz. Take 3.14, and select the capacitance values ​​of the power capacitor at times t1 and t2 respectively, that is... and These are the capacitance values ​​of the power capacitor at times t1 and t2, respectively.

[0078] Matlab software was used to preprocess the winding current and winding voltage of the dry-type air-core reactor, and then the effective values ​​of the voltage and current of the dry-type air-core reactor were calculated. The fundamental component was extracted by Fourier transform, and the phase difference between the voltage and current of the dry-type air-core reactor was determined.

[0079] Building a dry air-core reactor in Simulink software - Series equivalent circuit model of the dry-type air-core reactor - The equivalent circuit model typically consists of resistors and inductors, used to simulate the electrical characteristics of a reactor. Simulate the winding losses of a dry-type air-core reactor, in order to The inductive characteristics of a dry-type air-core reactor are simulated. The winding current of the dry-type air-core reactor is used as the input signal to drive the series equivalent circuit model of the dry-type air-core reactor. Through simulation and actual measurement, the resistance and inductance of the dry-type air-core reactor in the series equivalent circuit model are adjusted and obtained. The effective voltage and current values ​​of the dry-type air-core reactor output in the series equivalent circuit model are fitted to the actual effective voltage and current values ​​and phase differences between the actual dry-type air-core reactor. The fitted resistance and inductance of the dry-type air-core reactor are then obtained.

[0080] Based on the fitted resistance and inductance of the dry-type air-core reactor, the equivalent resistance and equivalent reactance of the dry-type air-core reactor are calculated using the following formulas:

[0081] ;

[0082] ;

[0083] in, and These are the equivalent resistance and equivalent reactance of a dry-type air-core reactor, respectively. The resistance of the fitted dry-type air-core reactor; The inductance of the fitted dry-type air-core reactor; The angular frequency of the power grid;

[0084] The loss factor of a dry-type air-core reactor can be calculated by the ratio of its equivalent resistance to its equivalent reactance. If the loss factor of the dry-type air-core reactor is too large, the loss of the dry-type air-core reactor can be reduced and its efficiency improved by improving the materials, optimizing the winding structure, or adjusting the inductance value.

[0085] Integrate capacitor characteristic data and reactor characteristic data into the reactive power compensation equipment fault monitoring dataset;

[0086] In summary, the equivalent circuit model established by combining Matlab and Simulink software helps to evaluate and improve the performance of dry-type reactors, and through parameter analysis of dry-type reactors, automatic early warning of faults in subsequent reactive power compensation equipment can be achieved.

[0087] In the feature processing module, the process of acquiring fault analysis data includes:

[0088] The fault analysis data includes the fundamental and third harmonic amplitudes of the resistive component of the leakage current of the zinc oxide surge arrester, the partial discharge of the power capacitor, the oil leakage in the power capacitor tank, and the fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit of the dry-type air-core reactor.

[0089] Using a configurable-point Fast Fourier Transform algorithm, the leakage current waveform of a zinc oxide surge arrester is analyzed to obtain the fundamental amplitude and third harmonic amplitude of the resistive component of the leakage current. The partial discharge signal waveform of a power capacitor is analyzed to obtain the partial discharge quantity. The current waveform of a dry-type air-core reactor during an inter-turn short circuit is analyzed to obtain the fundamental frequency, spectral amplitude, and power spectral density of the current. The obtained fault analysis data is then integrated into the reactive power compensation equipment fault monitoring dataset.

[0090] The configurable-point Fast Fourier Transform algorithm described above is simply referred to as the configurable-point FFT algorithm.

[0091] During long-term operation, the varistors of zinc oxide surge arresters may experience a decline in insulation performance due to moisture or aging. This degradation is directly reflected in the resistive component of the leakage current. Therefore, accurately separating and monitoring the resistive component of the leakage current of zinc oxide surge arresters is a core means of assessing their health status and achieving early warning of reactive power compensation equipment faults. The leakage current waveform of the zinc oxide surge arrester is input into a configurable point-based FFT algorithm module. This module transforms the leakage current waveform signal from the time domain to the frequency domain, obtaining the spectrum of the leakage current waveform. Through the FFT algorithm, the fundamental and third harmonic waves in the resistive component of the leakage current of the zinc oxide surge arrester are identified. The resistive component of the leakage current of the zinc oxide surge arrester is then used to... The frequency characteristics of the resistive and capacitive components differ. The fundamental and third harmonic amplitudes of the resistive component of the leakage current in a zinc oxide arrester are separated from the spectrum of the leakage current waveform. It should be noted that the capacitive component mainly manifests as the fundamental wave in the resistive component of the 50Hz leakage current of the zinc oxide arrester, with a relatively large amplitude, primarily determined by the inherent capacitance of the zinc oxide arrester. Due to the nonlinear characteristics of the zinc oxide varistor, its resistive current contains not only the fundamental component but also odd harmonics, such as the third harmonic in the resistive component of the 150Hz leakage current of the zinc oxide arrester. When the varistor is damp, the fundamental component of the resistive current increases significantly, while the increase in the third harmonic component is even more pronounced when the varistor ages.

[0092] For the partial discharge signal waveform of power capacitors, a configurable point-number FFT algorithm is used to convert the partial discharge signal of power capacitors into a frequency domain spectrum. Based on the characteristic spectrum range of the partial discharge signal, the corresponding spectrum components are identified and extracted. Combined with the correlation between the partial discharge amount and the spectrum amplitude, the partial discharge amount of the power capacitor is calculated, thereby realizing the monitoring of the insulation status of the power capacitor.

[0093] When a dry-type air-core reactor experiences an inter-turn short circuit, a configurable-point FFT algorithm is applied to perform spectral analysis on the acquired current waveform during the inter-turn short circuit. This identifies the fundamental frequency and harmonics of the current, extracts the spectral amplitudes corresponding to the fundamental and harmonic frequencies, and then calculates the power spectral density of the current during the inter-turn short circuit. The calculation formula is as follows:

[0094] ;

[0095] in, This represents the power spectral density of the current during an inter-turn short circuit in a dry-type air-core reactor. This refers to the fundamental frequency and amplitude of each harmonic of the current in a dry-type air-core reactor, such as... This is the fundamental amplitude of the current in a dry-type air-core reactor. , corresponding to the second harmonic amplitude and third harmonic amplitude of the dry-type air-core reactor current at 100Hz and 150Hz respectively; N is the number of samples involved in the calculation.

[0096] In the feature processing module, the process of obtaining the oil leakage amount in the power capacitor tank includes:

[0097] Using fluorescence imaging technology, a synchronous control unit controls an ultraviolet pulse emission unit to generate and emit ultraviolet pulses. After reflection and focusing, the ultraviolet pulses illuminate the area to be monitored in the power capacitor tank. If there is oil leakage in the power capacitor tank, the insulating oil will fluoresce under ultraviolet light due to the fluorescence effect. During the fluorescence lifetime, the location of the fluorescent oil leakage is photographed to obtain the fluorescence intensity signal of the power capacitor leakage oil. This imaging process is completed by a fluorescence intensity signal acquisition unit, which is equipped with a narrow-band filter and an attenuator on one side to filter out ambient light and attenuate strong light, thereby improving the signal-to-noise ratio of the fluorescence intensity signal. The fluorescence intensity signal of the power capacitor leakage oil is spliced ​​and the location is restored. A fluorescence intensity threshold is set, and the fluorescence intensity signal of the power capacitor leakage oil exceeding the threshold is analyzed to identify the leakage situation in the area to be monitored in the power capacitor tank. The leakage situation in the area to be monitored in the power capacitor tank includes whether leakage has occurred in the area to be monitored and the location of the leakage oil.

[0098] Based on the identification results of leakage in the monitored area of ​​the power capacitor tank, capacitive detection technology is adopted. Capacitive detection sensors are installed at the location of oil leakage in the monitored area of ​​the power capacitor tank. The initial capacitance value is recorded and the current capacitance value is collected. According to the dielectric constant of insulating oil and air, when oil leakage occurs and flows into the capacitive detection sensor, the capacitance of the capacitive detection sensor changes due to the difference in dielectric constant between insulating oil and air, forming an oil leakage signal. Referring to the database of the correspondence between capacitance and leakage oil volume (pre-marked capacitance-oil volume relationship model), the change in capacitance value is calculated by the absolute value of the difference between the current capacitance value and the initial capacitance value. Then, the change in capacitance value is converted into the volume of leakage oil to obtain the amount of oil leakage in the power capacitor tank. The amount of oil leakage in the power capacitor tank is then integrated into the reactive power compensation equipment fault monitoring dataset.

[0099] In the fault diagnosis and assessment module, the process of calculating the fault state diagnosis coefficient of zinc oxide surge arrester using the weighted summation method includes:

[0100] Extract the number of lightning strikes, valve plate temperature, and lightning strike absorbed energy of zinc oxide surge arresters from the surge arrester data, as well as the resistive component of leakage current of zinc oxide surge arresters from the fault analysis data;

[0101] The fault state diagnosis coefficient of the zinc oxide surge arrester is calculated by assigning weights to the number of lightning strikes, valve plate temperature, lightning strike absorbed energy and leakage current resistive component of the zinc oxide surge arrester, and using the weighted summation method.

[0102] The formula for calculating the fault condition diagnostic coefficient of zinc oxide surge arrester is as follows:

[0103] ;

[0104] in, This refers to the fault condition diagnostic coefficient for zinc oxide surge arresters. , , and These are the weights of the number of lightning strikes, valve temperature, lightning strike absorbed energy, and the resistive component of the leakage current for the zinc oxide surge arrester. , , and These are the number of lightning strikes, valve plate temperature, lightning strike absorbed energy, and resistive component of leakage current for the zinc oxide surge arrester.

[0105] In the fault diagnosis and assessment module, the process of constructing a capacitor fault assessment model and obtaining the capacitor fault type using a decision tree algorithm includes:

[0106] Capacitor fault types include no fault, leakage, bulging, explosion, and fuse blown. Each capacitor fault type is assigned a number, and the capacitor fault type number (power capacitor fault type number) is obtained. No fault, leakage, bulging, explosion, and fuse blown correspond to x1, x2, x3, x4, and x5, respectively.

[0107] Extract the capacitance change rate, loss factor, circuit breaker over-operating voltage, internal pressure, partial discharge, and oil leakage in the tank from the reactive power compensation equipment fault monitoring dataset. Then, convert the extracted data into the first training set and the first test set in a 6:4 ratio.

[0108] The decision tree algorithm is used to set the capacitance change rate, loss factor, circuit breaker over-operating voltage, internal pressure, partial discharge, and oil leakage in the tank of the power capacitor as input data, and the capacitor fault type number as output data. The algorithm learns the nonlinear relationship between the capacitance change rate, loss factor, circuit breaker over-operating voltage, internal pressure, partial discharge, oil leakage in the tank, and capacitor fault type number of the power capacitor, so as to realize the classification of capacitor fault types and obtain a trained capacitor fault assessment model.

[0109] The first test set data is input into the trained capacitor fault assessment model, the parameters of the capacitor fault assessment model are adjusted, the performance of the capacitor fault assessment model is optimized, and the final capacitor fault assessment model is obtained.

[0110] Based on the current capacitance change rate, loss factor, circuit breaker over-operating voltage, internal pressure, partial discharge, and oil leakage in the tank, the corresponding capacitor fault type number is output, and the corresponding capacitor fault type is matched by the capacitor fault type number.

[0111] The specific matching process is as follows: when the capacitor fault type number is x1, the corresponding capacitor fault type is no fault in the power capacitor; when the capacitor fault type number is x2, the corresponding capacitor fault type is leakage; when the capacitor fault type number is x3, the corresponding capacitor fault type is bulging; when the capacitor fault type number is x4, the corresponding capacitor fault type is explosion; when the capacitor fault type number is x5, the corresponding capacitor fault type is fuse blown.

[0112] In the fault diagnosis and assessment module, the random forest algorithm is used to construct a reactor fault assessment model. The process of obtaining the reactor fault type includes:

[0113] The fault types of reactors include no fault in dry-type air-core reactors, inter-turn short circuit, winding-to-ground short circuit, winding open circuit, and insulation aging. Each fault type of reactor is assigned a number, and the reactor fault type number (dry-type air-core reactor fault type number) is obtained. No fault in dry-type air-core reactors, inter-turn short circuit, winding-to-ground short circuit, winding open circuit, and insulation aging correspond to y1, y2, y3, y4, and y5, respectively.

[0114] The fault monitoring dataset of reactive power compensation equipment is extracted to obtain the solid insulation damage sound, inter-turn short circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and fundamental frequency, spectral amplitude and power spectral density of the inter-turn short circuit current of the dry air reactor. The dataset is then converted into a second training set and a second test set, with the ratio of the second training set data to the second test set data being 7:3.

[0115] Using the random forest algorithm, the solid insulation damage sound, inter-turn short-circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and the fundamental frequency, spectral amplitude, and power spectral density of the inter-turn short-circuit current of the dry-type air-core reactor are taken as inputs, and the reactor fault type number is taken as the output. The algorithm learns the nonlinear relationship between the solid insulation damage sound, inter-turn short-circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and the fundamental frequency, spectral amplitude, and power spectral density of the inter-turn short-circuit current of the dry-type air-core reactor and the reactor fault type number, so as to realize the classification of reactor fault types and obtain a trained reactor fault assessment model.

[0116] The second test set data is input into the trained reactive fault assessment model. Bayesian optimization is used to adjust the parameters of the reactive fault assessment model, optimize the performance of the reactive fault assessment model, and obtain the final reactive fault assessment model.

[0117] The solid insulation damage sound, inter-turn short-circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit of the current dry-type air-core reactor are input into the final reactor fault assessment model to obtain the reactor fault type number, and the corresponding reactor fault type is obtained by matching.

[0118] The specific matching process is as follows: when the reactor fault type number is y1, the corresponding reactor fault type is no fault in the dry-type air-core reactor; when the reactor fault type number is y2, the corresponding reactor fault type is inter-turn short circuit; when the reactor fault type number is y3, the corresponding reactor fault type is winding to ground short circuit; when the reactor fault type number is y4, the corresponding reactor fault type is winding open circuit; when the reactor fault type number is y5, the corresponding reactor fault type is insulation aging.

[0119] In the fault diagnosis and assessment module, the process of constructing a fault diagnosis model for reactive power compensation equipment using a neural network algorithm and obtaining a comprehensive fault type diagnosis number includes:

[0120] Set the fault state diagnosis coefficient threshold for zinc oxide surge arresters, determine the fault diagnosis types for zinc oxide surge arresters, assign a number to each fault diagnosis type for zinc oxide surge arresters, and obtain the fault type number for zinc oxide surge arresters. The fault type number for zinc oxide surge arresters consists of z1, z2, z3, z4, and z5, which correspond to the fault diagnosis types for zinc oxide surge arresters as follows: no fault in zinc oxide surge arrester, moisture fault in valve plate, aging fault in valve plate, insulation defect fault, and abnormal energy absorption fault, respectively.

[0121] The fault type numbers of zinc oxide surge arresters, capacitors, reactors, and overall compensation monitoring data are integrated and converted into a third training set and a third test set, with a ratio of 7:3.

[0122] Using a neural network algorithm, the fault type numbers of zinc oxide surge arresters, capacitors, reactors, and overall compensation monitoring data are set as inputs, and the comprehensive fault type diagnosis number is set as output. The nonlinear relationship between the fault type numbers of zinc oxide surge arresters, capacitors, reactors, overall compensation monitoring data, and comprehensive fault type diagnosis number is learned to obtain a trained reactive power compensation equipment fault diagnosis model.

[0123] The third test set data is input into the trained reactive power compensation equipment fault diagnosis model. The parameters of the reactive power compensation equipment fault diagnosis model are adjusted and its performance is optimized to obtain the final reactive power compensation equipment fault diagnosis model. Combining the current zinc oxide surge arrester fault type number, capacitor fault type number, reactor fault type number and overall compensation monitoring data, the corresponding comprehensive fault type diagnosis number is output.

[0124] Among them, the comprehensive fault type diagnosis number corresponds to the fault type of the reactive power compensation equipment. It comprehensively analyzes the possible fault factors of the compensation circuit and online monitoring device during operation, and makes a final diagnosis of the fault of the reactive power compensation equipment, which greatly improves the accuracy of the fault diagnosis of the reactive power compensation equipment. During this process, a variety of intelligent algorithms are used to enhance the intelligence level of the automatic early warning process of reactive power compensation equipment faults in this invention, and solve the problems of high cost and low accuracy of manual fault monitoring and early warning.

[0125] The comprehensive fault type diagnosis number consists of the fault type numbers for zinc oxide surge arresters, capacitors, and reactors. Since there are five types of fault type numbers for each of these three types, the comprehensive fault type diagnosis number includes 125 types. Different comprehensive fault type diagnosis numbers correspond to different fault types of reactive power compensation equipment. For example, if the comprehensive fault type diagnosis number is x1-y1-z1, the corresponding reactive power compensation equipment is fault-free. If the comprehensive fault type diagnosis number is x2-y1-z1, the corresponding reactive power compensation equipment has a leakage fault in its power capacitors during operation, but the dry-type air... If the electrocardiogram reactor and zinc oxide surge arrester are fault-free, and the comprehensive fault type diagnosis number is x2-y2-z3, then the corresponding reactive power compensation equipment has a leakage fault in the power capacitor, an inter-turn short circuit fault in the dry-type air-core reactor, and a valve plate moisture fault in the zinc oxide surge arrester. It should be explained that the reactive power compensation equipment includes power capacitors and dry-type air-core reactors, which are the core functional components of the reactive power compensation equipment. The zinc oxide surge arrester is not part of the reactive power compensation equipment itself, but is an auxiliary device that provides protection for the safe operation of the reactive power compensation equipment. Therefore, all three are of great reference value for the fault monitoring of the reactive power compensation equipment.

[0126] In the visualization module, the process of visualizing the output results of the reactive power compensation equipment fault diagnosis model in the form of charts includes:

[0127] The actual fault types of reactive power compensation equipment include fault diagnosis types of zinc oxide surge arresters, capacitor fault types, and reactor fault types;

[0128] Based on the comprehensive fault type diagnosis number, the fault type numbers of zinc oxide surge arresters, capacitors, and reactors are obtained respectively. By analyzing the correspondence between the fault type numbers of zinc oxide surge arresters and the fault diagnosis types of zinc oxide surge arresters, the fault type numbers of capacitors and the fault types of capacitors, and the fault type numbers of reactors and the fault types of reactors, the fault diagnosis types of zinc oxide surge arresters, the fault types of capacitors, and the fault types of reactors are analyzed respectively.

[0129] The number of occurrences of fault diagnosis types of zinc oxide surge arresters, capacitor fault types, and reactor fault types were statistically analyzed, along with the total number of occurrences of actual fault types of reactive power compensation equipment. The proportions of the number of faults of zinc oxide surge arresters, power capacitors, and dry-type air-core reactors in the total number of occurrences of actual fault types of reactive power compensation equipment were calculated.

[0130] Using the Origin plotting tool, a pie chart of the actual fault types of reactive power compensation equipment is drawn. Combined with the OriginScript script, the actual fault types of reactive power compensation equipment are marked in real time, realizing the visualization output of the fault diagnosis model of reactive power compensation equipment.

Claims

1. An automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms, characterized in that, include: The fault data acquisition module is used to collect fault monitoring data; The feature processing module, based on fault monitoring data, combines Matlab software, Simulink software, a configurable-point Fast Fourier Transform algorithm, fluorescence imaging technology, and capacitive detection technology to acquire capacitor feature data, reactor feature data, and fault analysis data. The fault diagnosis and assessment module, based on fault monitoring data, capacitor feature data, reactor feature data, and fault analysis data, uses a weighted summation method to calculate the fault state diagnosis coefficient of the zinc oxide surge arrester. Then, a decision tree algorithm is used to construct a capacitor fault assessment model to obtain the capacitor fault type. A random forest algorithm is used to construct a reactor fault assessment model to obtain the reactor fault type. Finally, a neural network algorithm is used to construct a reactive power compensation equipment fault diagnosis model to obtain a comprehensive fault type diagnosis number. The visualization module visualizes the output results of the reactive power compensation equipment fault diagnosis model in the form of charts. In the fault data acquisition module, the fault monitoring data acquisition process includes: setting up an online monitoring device and deploying different types of acquisition equipment on the online monitoring device to collect fault monitoring data, including surge arrester data, capacitor data, reactor data, and overall compensation monitoring data; surge arrester data includes the leakage current waveform, number of lightning strikes, valve plate temperature, and lightning absorbed energy of the zinc oxide surge arrester; capacitor data includes steady-state electrical quantity data, transient electrical quantity data, and fault electrical quantity data of the power capacitor, wherein the steady-state electrical quantity data includes the three-phase current, three-phase voltage, active power, and apparent power of the power capacitor, the transient electrical quantity data is the circuit breaker over-operation voltage of the power capacitor, and the fault electrical quantity... The data includes the internal pressure and partial discharge signal waveforms of power capacitors; reactor data includes winding current, winding voltage, solid insulation failure sound, inter-turn short-circuit discharge sound, and current waveform during inter-turn short circuits of dry-type air-core reactors; overall compensation monitoring data includes the total current, power factor, and bus voltage of the compensation circuit, the output voltage of the power module in the online monitoring device, the data upload rate of the online monitoring device, and the output voltage of the inverter; for the leakage current waveform of zinc oxide surge arresters, the partial discharge signal waveform of power capacitors, and the current waveform during inter-turn short circuits of dry-type air-core reactors, wavelet threshold noise reduction algorithm is used to eliminate random noise in the electromagnetic environment, and short-time Fourier transform is used. The system filters out environmental background noise from the solid insulation damage sound and inter-turn short-circuit discharge sound of dry-type air-core reactors. It performs data cleaning and standardization on the lightning strike count, valve temperature and lightning absorbed energy of zinc oxide surge arresters, three-phase current and three-phase voltage of power capacitors, circuit breaker over-operating voltage and internal pressure, winding current of dry-type air-core reactors, and overall compensation monitoring data. Based on the FPGA module's clock synchronization function, it synchronizes the acquisition time of surge arrester data, capacitor data, reactor data, and overall compensation monitoring data. It also integrates the pre-processed reactive power compensation equipment fault monitoring data to generate a reactive power compensation equipment fault monitoring dataset. In the feature processing module, capacitor feature data... The process of acquiring reactor characteristic data includes: capacitor characteristic data including the capacitance change rate and loss factor of the power capacitor; reactor characteristic data including the equivalent resistance, equivalent reactance, and loss factor of the dry-type air-core reactor; calculating the capacitance value of the power capacitor using the three-phase current and three-phase voltage, and calculating the loss factor of the power capacitor using the ratio of active power to apparent power; using Matlab software to preprocess the winding current and winding voltage of the dry-type air-core reactor, and then calculating the effective voltage and effective current values ​​of the dry-type air-core reactor, extracting the fundamental component using Fourier transform, and determining the phase difference between the voltage and current of the dry-type air-core reactor;A series equivalent circuit model of a dry-type air-core reactor was built in Simulink software. The winding current of the dry-type air-core reactor was used as the input signal to drive the model. Through simulation and actual measurement, the resistance and inductance of the dry-type air-core reactor in the series equivalent circuit model were adjusted and obtained. The effective voltage and current values, as well as the phase difference between voltage and current, of the dry-type air-core reactor output from the series equivalent circuit model were fitted to the actual effective voltage and current values, as well as the phase difference between voltage and current, to obtain the fitted resistance and inductance of the dry-type air-core reactor. Based on the fitted resistance and inductance, the equivalent resistance and equivalent reactance of the dry-type air-core reactor were calculated. The loss factor of the dry-type air-core reactor was calculated by the ratio of its equivalent resistance to its equivalent reactance. The capacitor characteristic data and reactor characteristic data were integrated into the reactive power compensation equipment fault monitoring dataset.

2. The automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms according to claim 1, characterized in that, In the feature processing module, the process of acquiring fault analysis data includes: the fault analysis data includes the fundamental amplitude and third harmonic amplitude of the resistive component of the leakage current of the zinc oxide surge arrester, the partial discharge of the power capacitor, the oil leakage in the power capacitor tank, and the fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit of the dry-type air-core reactor; the leakage current waveform of the zinc oxide surge arrester is subjected to spectral analysis using a configurable point-number fast Fourier transform algorithm to obtain the fundamental amplitude and third harmonic amplitude of the resistive component of the leakage current of the zinc oxide surge arrester; the partial discharge signal waveform of the power capacitor is subjected to spectral analysis to obtain the partial discharge of the power capacitor; the current waveform of the dry-type air-core reactor during inter-turn short circuit is subjected to spectral analysis to obtain the fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit of the dry-type air-core reactor, and the acquired fault analysis data is integrated into the reactive power compensation equipment fault monitoring dataset.

3. The automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms according to claim 2, characterized in that, In the feature processing module, the process of acquiring the oil leakage amount in the power capacitor tank includes: using fluorescence imaging technology, ultraviolet light pulses are generated and emitted by an ultraviolet light pulse emission unit. After reflection and focusing, the ultraviolet light pulses illuminate the area to be monitored in the power capacitor tank. During the fluorescence lifetime, the location of the leaking oil that generates fluorescence is photographed to obtain the fluorescence intensity signal of the power capacitor leaking oil. The fluorescence intensity signal of the power capacitor leaking oil is then stitched together and the location is restored. A fluorescence intensity threshold is set, and the fluorescence intensity signals of the power capacitor leaking oil that exceed the threshold are analyzed to identify the leakage situation in the area to be monitored in the power capacitor tank. Based on the identification result of the leakage situation in the area to be monitored in the power capacitor tank, capacitive detection technology is used, and a capacitive detection sensor is configured. Based on the dielectric constant of insulating oil and air, an oil leakage signal is generated. Referring to the database of the correspondence between capacitance and leakage amount, the oil leakage amount in the power capacitor tank is obtained. The oil leakage amount in the power capacitor tank is integrated into the reactive power compensation equipment fault monitoring dataset.

4. The automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms according to claim 1, characterized in that, In the fault diagnosis and assessment module, the process of calculating the fault state diagnosis coefficient of zinc oxide surge arrester using the weighted summation method includes: extracting the number of lightning strikes, valve temperature, and lightning absorbed energy of zinc oxide surge arrester from the surge arrester data, as well as the resistive component of leakage current of zinc oxide surge arrester from the fault analysis data; assigning weights to the number of lightning strikes, valve temperature, lightning absorbed energy, and resistive component of leakage current of zinc oxide surge arrester respectively, and calculating the fault state diagnosis coefficient of zinc oxide surge arrester using the weighted summation method.

5. The automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms according to claim 2, characterized in that, In the fault diagnosis and assessment module, a capacitor fault assessment model is constructed using a decision tree algorithm. The process of obtaining capacitor fault types includes: capacitor fault types such as no fault, leakage, bulging, explosion, and fuse blowing. Each capacitor fault type is assigned a number to obtain the capacitor fault type number; the capacitance change rate, loss factor, circuit breaker over-operation voltage, internal pressure, partial discharge, and oil leakage in the tank of the power capacitor are extracted from the reactive power compensation equipment fault monitoring dataset, and the extracted data are converted into a first training set and a first test set; the decision tree algorithm is used to set the first training set data as the input data, and... The capacitor fault type number is set as the output data. The nonlinear relationship between the first training set data and the capacitor fault type number is learned to obtain the trained capacitor fault assessment model. The first test set data is input into the trained capacitor fault assessment model, the capacitor fault assessment model parameters are adjusted, the performance of the capacitor fault assessment model is optimized, and the final capacitor fault assessment model is obtained. Combining the current capacitance change rate, loss factor, circuit breaker over-operating voltage, internal pressure, partial discharge, and oil leakage in the tank of the power capacitor, the corresponding capacitor fault type number is output. The corresponding capacitor fault type is matched by the capacitor fault type number.

6. The automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms according to claim 5, characterized in that, In the fault diagnosis and assessment module, a random forest algorithm is used to construct a reactor fault assessment model. The process of obtaining reactor fault types includes: reactor fault types such as no fault in dry-type air-core reactors, inter-turn short circuit, winding-to-ground short circuit, winding open circuit, and insulation aging. Each reactor fault type is assigned a number, and the reactor fault type number is obtained. The reactive power compensation equipment fault monitoring dataset is extracted to obtain the solid insulation damage sound, inter-turn short circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and the fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit in dry-type air-core reactors. This data is then converted into a second training set and a second test set. Using the random forest algorithm, the second training set data is used as input to analyze the reactor... The fault type number is used as the output. The nonlinear relationship between the second training set data and the reactor fault type number is learned to obtain the trained reactor fault assessment model. The second test set data is input into the trained reactor fault assessment model. Bayesian optimization is used to adjust the reactor fault assessment model parameters and optimize the performance of the reactor fault assessment model to obtain the final reactor fault assessment model. The solid insulation failure sound, inter-turn short circuit discharge sound, equivalent resistance, equivalent reactance, loss factor, and the fundamental frequency, spectral amplitude, and power spectral density of the current during inter-turn short circuit of the current dry-type air-core reactor are input into the final reactor fault assessment model to obtain the reactor fault type number. The corresponding reactor fault type is obtained by matching.

7. The automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms according to claim 6, characterized in that, In the fault diagnosis and assessment module, the process of constructing a fault diagnosis model for reactive power compensation equipment using a neural network algorithm to obtain a comprehensive fault type diagnosis number includes: setting a threshold for the fault state diagnosis coefficient of the zinc oxide surge arrester, determining the fault diagnosis type of the zinc oxide surge arrester, and assigning a number to each fault diagnosis type of the zinc oxide surge arrester to obtain the fault type number of the zinc oxide surge arrester; integrating the fault type numbers of the zinc oxide surge arrester, capacitor fault type numbers, reactor fault type numbers, and overall compensation monitoring data, and converting them into a third training set and a third test set; using a neural network algorithm, setting the third training set data as input, and then... The comprehensive fault type diagnosis number is set as the output. The nonlinear relationship between the third training set data and the comprehensive fault type diagnosis number is learned to obtain the trained reactive power compensation equipment fault diagnosis model. The third test set data is input into the trained reactive power compensation equipment fault diagnosis model. The parameters of the reactive power compensation equipment fault diagnosis model are adjusted to optimize the performance of the reactive power compensation equipment fault diagnosis model and obtain the final reactive power compensation equipment fault diagnosis model. Combining the current zinc oxide surge arrester fault type number, capacitor fault type number, reactor fault type number and overall compensation monitoring data, the corresponding comprehensive fault type diagnosis number is output.

8. The automatic early warning system for reactive power compensation equipment faults based on intelligent algorithms according to claim 7, characterized in that, In the visualization module, the process of visualizing the output results of the reactive power compensation equipment fault diagnosis model through charts includes: The actual fault types of the reactive power compensation equipment include zinc oxide surge arrester fault diagnosis types, capacitor fault types, and reactor fault types; based on the comprehensive fault type diagnosis number, the fault diagnosis types of zinc oxide surge arresters, capacitor fault types, and reactor fault types are analyzed separately; the occurrence frequency of each fault type is statistically analyzed relative to the total occurrence frequency of the actual fault types of the reactive power compensation equipment, calculating the proportion of zinc oxide surge arrester faults, power capacitor faults, and dry-type air-core reactor faults in the total occurrence frequency of the actual fault types of the reactive power compensation equipment; using the Origin plotting tool, a pie chart of the actual fault types of the reactive power compensation equipment is drawn, and combined with OriginScript, the actual fault types of the reactive power compensation equipment are marked in real time, thus achieving the visualization output results of the reactive power compensation equipment fault diagnosis model.

Images