Contaminant current live detection method for non-contact power transformation equipment

Abstract

According to the non-contact live-line detection method for the pollution current of the transformer equipment, a live-line detection device with a novel current sensor based on the TMR technology is used for carrying out live-line non-contact detection, measurement data are transmitted to a monitoring host, and the external insulation pollution condition of the transformer equipment is judged by integrating background analysis of a deep learning algorithm into a host platform. The special array type measurement system comprises a power supply circuit, a TMR sensor array, a signal conditioning circuit, an ADC analog-to-digital converter, a microcontroller, a timer unit and a wireless communication module. The method is suitable for live detection of surface dirt leakage current of various types of capacitive equipment in a station, and has good practical value and popularization and application prospects. The device can be used in a transformer substation, and can be used for detecting the insulator of a power transmission line. The method can achieve the active early warning of potential operation risks, greatly improves the detection efficiency, effectively prevents the pollution flashover accident, and improves the power supply reliability of the system.

Description

Technical Field

[0001] The invention relates to the technical field of flashover tripping matching for power systems, and in particular to a method for detecting live contamination current of non-contact substation equipment. Background Art

[0002] Pollution flashover accidents of electrical equipment occur frequently in the national power grid, which bring great losses to power production and have a serious impact on the image of power companies. Research data analysis by experts in the field of power pollution flashover prevention shows that there are three main reasons for pollution flashover: First, with the rapid development of industrial enterprises, atmospheric pollution has intensified, especially in industrially developed areas; Second, the insulation performance of domestic electrical equipment cannot meet the requirements of local pollution levels for equipment insulation performance. Although power companies have taken various positive measures to improve the pollution prevention level of electrical equipment, there is a certain degree of blindness in the actual real-time process, and there is a lack of quantitative prevention and control measures based on specific pollution levels; Third, the current pollution monitoring level of equipment is relatively backward, and the pollution level varies with meteorological conditions such as seasons, temperature, and humidity. Pollution flashover tripping has always been one of the important issues affecting the safe operation of power systems. Affected by factors such as the degree of atmospheric pollution and geographical location (such as coastal areas or chemical parks), the pollution layer deposited on the surface of the insulator is damp, which often causes pollution flashover accidents in substations. After a pollution flashover accident occurs, it often causes the relay protection device to operate, causing the equipment to trip. Once a pollution flashover accident occurs, it will affect a large area and last for a long time. In serious cases, it may even cause electrical personal safety accidents. The economic losses and social impacts caused by this are enormous. Figure 1 and Figure 2 The figures show the dirt on the surface of the porcelain insulators on site and the flashover test results of the substation post insulators.

[0003] The existing technology currently mainly installs special insulators in substations to accumulate pollution together with the substation equipment in the station, and then uses the equivalent salt density method to regularly detect the pollution level of the insulator to indirectly reflect the pollution level of the entire station. There is also a method of installing a collector ring at the grounding flange of the insulator to measure the pollution leakage current to evaluate the pollution level of the insulator, but long-term operation experience shows that the measurement results of this method have a large deviation from the actual situation.

[0004] Both IEEE and IEE recommend the equivalent salt density method. my country has used this method for decades and has long divided the pollution level according to the maximum pollution value in one year and formulated pollution zone maps. According to the current salt density monitoring and processing method, it is required to clean the insulation equipment once a year to reduce the possibility of pollution flashover as much as possible. Then, three large-scale pollution flashover accidents occurred in the power grid in 1990, 1996 and 2001. Facts show that the previous method of manually cleaning insulators once a year cannot meet the requirements of insulation equipment management in the era of large power grids in my country. The equivalent salt density method is a commonly used method for monitoring the pollution of insulation equipment. This method is intuitive and easy to understand, but there are equivalence problems in operation. This method is a destructive test of insulation equipment. This method requires a large cost to obtain pollution data display. If a method that can remotely monitor the pollution status of on-site insulation equipment in real time can be studied, and data collection does not need to be destroyed, it will bring great progress to the pollution detection of power transmission and transformation equipment.

[0005] Another method of detecting contamination is leakage current detection technology, which reflects the insulation performance of insulation equipment through leakage current. Leakage current detection is to install leakage current collection equipment on the surface of insulation equipment, and collect leakage current data of insulation equipment in real time through the collection equipment. The collection equipment transmits the collected leakage current to the background system through wireless or wired and wireless network technology, calculates the collected leakage current, and calculates the maximum, minimum, average and current pulse number of the leakage current within a certain period. The status of insulation equipment can be evaluated through leakage current. Later, with the continuous development of technology, a large number of scientific researchers have successively developed a variety of methods for detecting the degree of contamination of power equipment, such as infrared temperature measurement and electric field measurement.

[0006] At present, the live detection methods of insulators of power transmission and transformation equipment are divided into two categories, including electrical quantity detection method and non-electrical quantity detection method. Among them, the electrical quantity detection method includes surface leakage current method, insulation resistance method, electric field measurement method and pulse current method, etc. The non-electrical quantity detection method includes observation method, infrared temperature measurement method, laser Doppler vibration method, etc. Among them, the electric field measurement method is relatively mature.

[0007] Since the insulators of power transmission and transformation equipment work in high voltage and strong magnetic field environment all year round and are exposed to the air, the particularity of this working environment and the particularity of the installation location make it difficult to detect the live insulators. In addition, there are a huge number of insulators, and the workload of checking them one by one is huge, which is also a huge test for manpower and material resources. The faults of insulators are divided into external surface defects and defects according to the location of occurrence. External surface defects include surface damage, cracks, etc., which are easier to find, while the changes in the electrical characteristics of the defects in the early stage are not obvious, not easy to detect, but once they occur, they are the most dangerous. The defects of insulators are manifested as brittle fracture of core rods, reduced insulation resistance and cracking of sheds. The defect manifestation is also related to the type of insulator, and different characteristics are presented according to the material and electrical characteristics of the insulator. The insulation problem of high-voltage equipment is a hot topic of research at home and abroad, and remarkable results have been achieved. There are many studies on high-voltage insulators at home and abroad, and a large number of experimental studies have been carried out. While certain scientific research results have been achieved, there are still some problems that have not been solved, such as: the detection equipment is expensive, the risk factors cannot be eliminated, the measurement error cannot be ignored, and the anti-interference ability is weak. Therefore, it is urgent to develop a set of software and hardware equipment for detecting insulator defects, which is also the focus of the power department. At present, the live detection methods of insulator defects include non-electrical detection methods represented by observation method and infrared temperature measurement method and electrical detection methods represented by electric field measurement method and leakage current method.

[0008] Non-electrical detection methods for insulator defects include: (1) Observation method: Observation method is usually the earliest, most widely used and most commonly used detection method, because this method does not require too much equipment. Its principle is that the staff stands on the ground at a certain distance from the equipment to be tested, and observes it with a telescope or other auxiliary tools to see if there are any damage defects on the surface of the insulator. The specific contents observed are: the roughness of the insulator surface, whether the core rod is exposed, whether the insulating shed is cracked and corroded, etc. However, this method cannot find defects inside the insulator, and the detection effect will also be affected by objective factors such as the angle chosen by the staff or the dirt covering the surface of the insulator. Therefore, this method has a certain scope of use, that is, it is not very reliable. It is precisely because of various reasons that the current observation method is usually only used as an auxiliary method.

[0009] (2) Ultrasonic detection method: With the increasing maturity of ultrasonic technology and its wide application field, the response obtained by emitting ultrasonic waves onto the surface of insulators can be used to detect whether the insulators are cracked. The specific principle is as follows: If there are tiny cracks in the insulator, when the surface of the insulator is irradiated with ultrasonic waves, physical phenomena such as reflection and refraction will occur at the cracks of the insulator. The received ultrasonic waves are then processed by the receiving device, and the degree of deterioration of the insulator can be determined through analysis, and then it can be determined whether it needs to be repaired or replaced. However, this method has disadvantages. It can only be applied to cracked insulators. There is no way to judge insulators that are not cracked. Therefore, this method also has certain limitations.

[0010] (3) Laser Doppler vibration method: The principle of this detection method is to make a judgment by comparing the vibration frequency of the cracked insulator with the vibration frequency of the normal insulator. The specific measures are as follows: By using a sound wave generator or with the help of external force to make the insulator vibrate, during this process, the Doppler instrument is placed face to face with the insulator, and the laser irradiates the insulator to produce reflection or refraction. The oscillation frequency of the center of the insulator is judged by the returned signal, so that it can be concluded that there are defects inside the insulator being tested. This detection method is similar to the principle of using ultrasonic detection method. Its defect is that it cannot detect insulators that have defects inside but have not cracked on the surface. In addition, since the Doppler instrument occupies a large area and is relatively complicated to operate and maintain, it cannot be widely promoted and applied.

[0011] (4) Radio wave detection method: The radio wave detection method is to analyze and judge the electromagnetic waves emitted by the insulator through a receiving device. Because the intensity of the electromagnetic waves emitted by normal insulators and cracked insulators is different, the insulator can be judged whether it is deteriorated according to the intensity of the electromagnetic waves. The advantages of this method are simple principles and easy operation of the equipment. The disadvantages are insufficient anti-interference ability, low sensitivity and accuracy, and large workload.

[0012] (5) Corona camera method: This method is based on the fact that when a bad insulator appears, the normal insulators around the bad insulator will have an enhanced corona phenomenon. The corona camera can clearly capture this phenomenon and is not affected by the natural environment such as sunlight radiation. The principle of this method is to use the corona camera's ability to capture corona. When there is a bad insulator, the corona phenomenon will occur. The corona camera method can observe this phenomenon and can work in different natural environments without affecting performance. However, the insulator defects cannot be observed, and the equipment is expensive and is not suitable for promotion. In addition, there are fiber optic measurement methods and hyperspectral analysis methods.

[0013] The non-electrical detection method does not need to consider insulation and safety issues because it is at a large distance from the insulator being tested. However, it requires the addition of auxiliary equipment, and the equipment cost and maintenance cost are high. The detection equipment itself also has a certain impact on the test results, resulting in low detection accuracy.

[0014] The electrical quantity detection method can be divided into the following three categories: (1) Voltage distribution measurement method: The principle of the voltage distribution measurement method is: when the insulator string is working normally, the voltage distribution curve is displayed as a smooth non-standard saddle shape. The voltage values ​​borne by different positions of the insulator string are different. The closer to the conductor side, the higher the voltage value, which is about 1.7-3 times the value near the ground end. The middle part of the insulator string bears the lowest voltage. According to statistics, when working normally, the voltage ratio between the two ends of two adjacent insulators in the insulator string is 1.1 to 1.3. The adjacent comparison method can be used to determine whether there is a degraded insulator. The adjacent comparison method is widely used, and usually selects the adjacent voltage ratio lower than 50% as the judgment standard. That is, if it is lower than this value, there is a deterioration phenomenon. Near this value, the specific location of the degraded insulator can be found. There is also a longitudinal comparison method, which compares the changes in the voltage distribution of the same insulator at different measurement times to make a judgment.

[0015] The advantages of the voltage distribution measurement method are simplicity, intuitiveness, accurate measurement, and easy reading. At present, a self-climbing insulator live detector has been developed to reduce the workload of staff. The disadvantage is that it requires high-altitude measurement and there are certain unsafe factors. Therefore, this method is still difficult to promote widely.

[0016] (2) Pulse current detection method: Insulators work in extremely non-uniform electric fields and produce corona discharge. The pulse current detection method measures the insulation condition of the insulator string by measuring the pulse current when corona occurs. The principle of this method is: when a bad insulator appears, the insulation strength is reduced and the insulation resistance drops sharply. Since the insulator string is connected in series in the system, the resistance of the deteriorated insulator decreases, and the voltage at both ends of it also becomes relatively small, which causes the normal insulator to withstand a much larger voltage than under normal operation, posing a threat to the working condition of the normal insulator. As the total resistance decreases, the current increases while the working voltage remains unchanged, and the corona phenomenon becomes more severe. Based on the important feature that the amplitude and number of corona pulses appearing in cracked insulators change, this paper uses a pulse current sensor to obtain the pulse current signal and performs signal processing to determine the presence of deteriorated insulators for troubleshooting.

[0017] (3) Insulation resistance method: The essence of the insulation resistance method is to measure the leakage current. Generally, the insulators in high-voltage transmission lines are synthetic insulators. The insulators have a simple structure, high mechanical strength, and a wide range of applications. In addition, they have a wider range of applications.

[0018] Under normal working conditions, there will be a small leakage current on the surface of most insulators. When an insulator fails, the current on its surface will increase. At this time, the degree of cracking of the insulator can be evaluated by measuring the change in leakage current. In general, the degree and number of cracking of the insulator will affect the increase in current. At present, the research on current sensors has become increasingly mature, and the leakage current method is an effective method.

[0019] Research situation of foreign research institutions or companies: Abroad, Japan was the first to use the equivalent salt density method (ESDD method) for pollution measurement. This method uses the unique conductive properties of ionizable substances for measurement. The specific method is to dissolve the pollutants in distilled water with a certain volume and a certain temperature, and then dissolve NaCl in distilled water with the same temperature and volume as the former, and then measure the conductivity of the two solutions respectively. During the measurement process, if the conductivity of the NaCl solution with a certain volume and temperature is exactly the same as that of the pollutant solution with a certain volume, it can be concluded that the number of conductive items is equal to the amount of NaCl. Finally, we divide the weight of NaCl (mg) by the surface area of ​​the insulating equipment to obtain the equivalent salt density.

[0020] Domestic research situation: In the 1990s, Wuhan High Voltage Research Institute and Wuhan Compu Evergreen Software Technology Co., Ltd. conducted research on the use of optical sensors to measure salt density, in order to obtain a new method for real-time and accurate salt density measurement. Optical sensors measure salt density based on the theory of light field distribution and light energy loss mechanism in dielectric optical waveguides. The low-loss quartz rod placed in the atmosphere is a multimode dielectric optical waveguide with the rod as the core and the atmosphere as the cladding. When there is no pollution on the quartz rod, the fundamental mode and higher-order mode in the optical waveguide transmit the energy of light together, most of which is transmitted in the core of the optical waveguide, but a small part of the light energy will be transmitted along the cladding of the core-cladding interface, and the light loss during the light wave transmission process is very small. When there is pollution on the quartz glass rod, the transmission conditions of the higher-order mode and the fundamental mode are changed due to the pollutants; at the same time, the absorption and scattering of light energy by the polluting particles will cause light energy loss; the amount of salt on the sensor surface can be calculated by detecting the light energy parameters. Since the sensor and the insulator string are in the same environment, the salt density value of the insulator surface can be obtained by calculation. In recent years, the research team of Southwest Jiaotong University in China proposed a method to collect the spectral line data of the pollution level of insulating sheets of different materials through hyperspectral technology and perform relevant preprocessing, establish the main detection model of support vector machine with glass material test data, and directly standardize the main detection model in sections to realize the pollution level detection of samples of different materials under the same detection model. However, this method is currently in the stage of laboratory theoretical research and has not been verified on the field and related products.

[0021] In the prior art, operation and maintenance personnel, based on the results of on-site pollution level analysis, adopt measures such as increasing the number of insulators, applying hydrophobic anti-fouling coatings on the surface of insulators or using pollution-resistant insulators, and regularly cleaning insulators to reduce pollution flashover accidents, but the effect is still not ideal. The main reason is that there are many devices in the substation and the distribution area is wide. The traditional method of monitoring the pollution level of external insulation only monitors some points in the station and does not achieve full coverage of the equipment in the station. It is impossible to fully and comprehensively reflect the pollution level of the external insulation of the equipment in the station; moreover, regular pollution detection work is cumbersome and the cycle is long, which leads to the operation and maintenance personnel's lack of timely and accurate understanding of the pollution level, and the inability to perceive and effectively deal with the potential pollution flashover hazards of the substation in advance. Summary of the invention

[0022] The present invention provides a contactless live detection method for contaminated current in substation equipment. It can solve the time-consuming and laborious problems caused by regular contamination detection of traditional substation equipment. By non-contact measuring the contamination current on the surface of the equipment, it can realize the rapid identification of equipment contamination without power outage, and can fully cover the degree of contamination on the external insulation surface of the equipment, while improving the safety and efficiency of the detection process. Domestic and foreign researchers have conducted a lot of research on the live detection of insulator defects, and have achieved certain research results and proposed many detection methods, but these methods have advantages and some problems, and lack relatively mature complete sets of equipment to bring convenience to on-site work. Therefore, the work done by the present invention is to integrate the current insulator detection methods, put forward its own ideas and verify the feasibility, and it is urgent to ensure the safe and reliable operation of the power system.

[0023] The technical solution of the present invention is: a non-contact substation equipment contaminated current live detection method, characterized in that: the method uses a live detection device with a new current sensor based on TMR technology for live detection, the sensor is fixed on an insulating rod, and the tester can perform non-contact detection of the contaminated current and discharge pulse at the grounding flange of the substation high-voltage electrical equipment by holding the insulating rod, and the measurement data is transmitted to the monitoring host through a wireless communication network, and the background software analysis of the host platform integrated with the deep learning algorithm can judge the external insulation contamination of the substation equipment; The host can collect data from 2 or more current sensors at the same time to improve the efficiency of field tests; the power supply mode adopts AC and DC dual-use; The non-contact substation equipment dirty current live detection method uses a dedicated array measurement system. The circuit structure diagram of the array measurement system is as follows: Figure 1 As shown; the dedicated array measurement system includes a power supply circuit, a TMR sensor array, a signal conditioning circuit, an ADC analog-to-digital converter, a microcontroller, a timer unit, and a wireless communication module; wherein: The TMR sensor array includes at least two TMR sensors, the TMR sensor array is connected to the microcontroller through the signal conditioning circuit and the ADC analog-to-digital converter in sequence, and the wireless communication module is also connected to the microcontroller; the power supply circuit is respectively connected to the TMR sensor array, the signal conditioning circuit, the ADC analog-to-digital converter, the microcontroller, the timer unit, and the wireless communication module; the timer unit is also connected to the microcontroller; The power supply circuit adopts a single 5 V power supply, which is converted by the power module to generate multi-level voltages of ±5V, +3.3V, +3V, and +1.5V to meet the power supply requirements of various components; The TMR sensor array uses the linear magnetic field sensor chip TMR2001 array; the linear magnetic field sensor chip TMR2001 array consists of 4 unshielded high-sensitivity TMRs to form a Wheatstone full-bridge structure, and the output has good temperature stability; the linear magnetic field sensor chip TMR2001 adopts SOT23-5 package. Under the action of an external magnetic field parallel to the sensitive direction, the bridge provides a differential voltage output. Figure 2 One of the typical output curves of TMR2001 when powered by 1 V, Figure 3 The second typical output curve of TMR2001 when powered by 1V; the linear magnetic field range of TMR2001 is ±5 Oe, the sensitivity is 8 mV / V / Oe, and the saturation magnetic field is -25Oe and +40 Oe.

[0024] The microcontroller is MSP430F149; the wireless communication module is a single-chip wireless transceiver chip nRF24L01 operating in the ISM frequency band of 2.4GHz~2.5GHz; the circuit diagram of the microcontroller MSP430F149 can be found in Figure 4 ; The timer unit is a 16-bit timer; The dedicated array measurement system uses software methods to realize the numerical integration function. The differential output signal of the TMR sensor is conditioned by the instrument amplifier and sent to the ADC for sampling and conversion. The definite integral is calculated by the microcontroller MSP430F149.

[0025] The non-contact substation equipment dirty current live detection method of the present invention preferably requires protection of the technical content: the non-contact substation equipment dirty current live detection method also meets the following requirements: In order to make full use of the 0-3V conversion range of the ADC, Rg selects a 1kΩ precision resistor with an accuracy of 0.1%, and the actual gain is 10.9 times; in order to enable the ADC to convert bipolar signals, a 1.5V forward bias is provided to the REF pin of the AD8421 to raise the output voltage; the output TMR sensor array of the AD8421 is designed on a circular PCB with an inner diameter of 60mm and an outer diameter of 100mm, and 8 TMR2001 chips are evenly arranged on a circle with a radius of 35mm along the tangent of the sensitive axis.

[0026] Table 1 TMR2001 performance parameters (VCC = 1.0 V, TA = 25°C)

[0027] The non-contact substation equipment dirty current charged detection method is characterized in that: the non-contact substation equipment dirty current charged detection method also meets the following requirements: The TMR sensor and signal amplification circuit also meet the following requirements: In order to improve the measurement accuracy of the system, TMR2001 and ADC external reference voltage VeREF+ both use the 3V reference voltage provided by the high-precision voltage source chip ADR4530, so the maximum output of TMR2001 in the linear range is ±120mV; because the output differential signal is weak, it is amplified by AD8421; this is a low-power, low-noise, high-speed instrumentation amplifier with a high common-mode rejection ratio, and the sensor offset voltage and system zero point accurately measure the current on the premise that the sensor can accurately reflect the magnetic field strength at the location. However, differences in manufacturing processes make it impossible for the TMR characteristics to be completely consistent, resulting in the output voltage of the bridge not being zero when there is no magnetic field, which is called the offset voltage V0. According to the data provided in Table 1, when powered by 1V, the V0 of TMR2001 is as high as ±30mV, which almost accounts for 3 / 4 of the linear output range of ±40mV, so the impact of the chip offset voltage on the system zero point cannot be ignored;

[0028] When an external magnetic field acts on the sensor, the output is superimposed with a magnetic induction intensity on the basis of the offset voltage: ωk is determined by the product formula, VTMR(k) and V0(k) represent the output and offset voltage of the kth sensor respectively. When there is no current in the loop, the above formula should be equal to zero. In any environment, the AD8421 output Vout(k) is collected using ADC. The output voltage VTMR(k) of the TMR sensor is obtained by formula (3), and the integral of V0 can be inferred. During the measurement process, correction is made according to formula (7), which can eliminate the system zero point caused by the chip offset voltage. Even if the external magnetic field environment changes, its integral is always zero, which will not affect the system zero point. This solution does not require magnetic field shielding and avoids repeated and tedious manual measurement operations. The microprocessor can achieve zero point compensation of the current sensor through simple calculations.

[0029] The steps and contents of the non-contact type substation equipment dirty current live detection method are as follows: After the hardware circuit is completed, signal acquisition and processing, numerical integration, zero compensation and other operations are all implemented by the internal program of the microcontroller. The specific process is as follows (see Figure 5 ): (1) After the system is powered on, the registers of the microcontroller MSP430F149's internal clock, I / O port, timer Timer_A, ADC12, serial port USART0 and other peripheral modules are initialized and set, and interrupts are enabled; the microcontroller can respond to the USART0RX_VECTOR interrupt to receive control instructions sent by the host computer and perform corresponding operations according to the instructions; the basic instructions include: reset (0x01), return the quadrature formula (0x02), return the ADC conversion result (0x03), switch the quadrature formula (0x04), zero adjustment (0x05), send measurement data (0x06), stop sending measurement data (0x07) and return parameter information (0xFF); (2) Before measurement, a zero adjustment operation must be performed. When there is no current in the loop, the host computer sends a zero adjustment command (0x05). After receiving it, the microcontroller MSP430F149 starts the ADC to measure the voltage signals of the eight channels and calculate the offset voltage integral. (3) The 16-bit hardware timer Timer_A inside the microcontroller MSP430F149 is set to generate a timer interrupt every 1ms. In the interrupt processing function, the ADC is started to perform serial channel conversion, collect the voltage of 8 channels for numerical integration, correct the result, and calculate and store the instantaneous value of the current; (4) Every 1 second, the 1000 stored instantaneous current values ​​are averaged to obtain the DC component, and the RMS calculation is performed to obtain the AC component. The measurement data is sent to the host computer using the wireless communication module. The project develops a non-contact substation equipment dirty current detection device. The required materials include TMR sensor (current measurement 0-100uA) for non-contact measurement, wireless communication module using 430MHZ for data communication, host shell using aluminum shell, current protector (capable of circuit protection against impact and interference, with a maximum withstand of 20kA), current acquisition module (capable of I / V conversion, forming a voltage range of 0-3.3V for the back-end available signal.

[0030] The non-contact type substation equipment dirty current live detection method also meets the following requirements: (1) The array measurement system specifically uses a non-shielded structure with good sensitivity: the four tunnel magnetoresistance TMR resistors in different sensitive directions in the TMR sensor are divided into two groups and connected into a Wheatstone bridge. Every two TMR resistors in opposite positions are divided into one group, and the magnetoresistance characteristics are consistent. The directions of TMR resistors in different groups are opposite. Under the same magnetic field conditions, the resistance of one group of TMR resistors increases, while the resistance of the other group decreases. A monitoring voltage that follows the change of the magnetic field can be generated at the monitoring outlet of the TMR sensor, and a corresponding monitoring current can be generated in the post-processing circuit according to the monitoring voltage. Application of Gaussian algorithm in measurement: Since the data will be interfered by many noises such as Gaussian white noise in the process of measuring data, the size and frequency of these noises are different, and the interference of these noises will affect the accuracy of the measured data. In order to obtain more accurate measurement data, the data needs to be denoised. Filtering is also called denoising. The filter in the system often cannot remove all the noise, and the filtering function is required to strengthen the filtering. White noise is a common noise in the power field. Since the power spectrum density width in practice is limited and its average power cannot be estimated, there is no white noise that strictly obeys the uniform distribution in practice. Gaussian white noise is ubiquitous, and the one-dimensional probability density of Gaussian white noise is also easy to calculate. Removing these noises is conducive to extracting load characteristics. Convolving the Gaussian function with the original signal is the key to the Gaussian algorithm. Compared with traditional algorithms, the Gaussian algorithm has the advantages of filtering window and phase shift. It not only retains the sudden signal, does not weaken and eliminate the sudden signal, but also can smooth the sudden data;

[0031] Gaussian filter formula: ,in, is the output value of the original load signal after the k+1th iteration; is the input value of the original load signal after the k+1th iteration; is the sum of the weighted coefficients of the filter at the kth iteration; is the weighting coefficient at the kth iteration point a+x; Set the initial iteration number k to 0 and set the number of iterations; bring the collected initial current and voltage data into the formula and iterate to the set number of iterations, and output the iteration result, which is the filtered value.

[0032] (2) The non-contact substation equipment dirty current detection method also meets the requirements of wireless communication system design principles. The detection data is transmitted by wireless network. At present, the wireless communication used in the online detection system of the domestic power system mainly includes LORA-433MHz wireless technology. These are all short-range wireless communication technologies and use the ISM license-free frequency band, but they each have their own characteristics. LoRaWAN is a highly integrated, low-power wireless control module that mainly uses LoRa's patented remote modulation and demodulation technology. It has a built-in STM32L series CPU for ultra-long-distance spread spectrum communication. It minimizes current consumption while greatly improving anti-interference and sensitivity. LWM supports three protocol modes, LoRaWAN Class A, B, and C. LoRa supports high-performance (G)FSK mode for systems such as WM-Bus. Compared with similar devices, LWM uses RF solutions to significantly reduce current consumption while significantly optimizing various performance such as phase noise, selectivity, receiver linearity, and third-order input intercept point (IIP3). The LoRa network structure is as follows: Figure 6 shown.

[0033] The wavelength of the 433MHz frequency band is nearly one meter. According to the relationship between wavelength and propagation distance, we know that the longer the wavelength, the better the diffraction of the electromagnetic wave, the better it can bypass obstacles and propagate farther. Therefore, compared with WiFi and Zigbee, the significant advantage of LORA-433MHz is that the wireless signal has strong penetration and can propagate farther. Such advantages are very suitable for wireless communication in cable online detection systems. This is mainly because most urban cables are buried underground, and wireless signals require high penetration to transmit information to receivers on the ground. At the same time, due to the dense distribution of high-rise buildings in cities and the frequent bending changes in the layout, wireless signals can only transmit signals to receivers if they have a high ability to bypass obstacles. Although its data transmission rate is only 9600bps, which is much lower than the data rate of WiFi and Zigbee, it can fully meet the wireless communication requirements of cable online monitoring systems with small data volumes. In addition, any low-voltage equipment in power system applications must consider the strong electromagnetic interference generated by the power system itself due to its operating characteristics, which is often called EMC problems. For example, when the cable is in operation, partial discharge problems may occur due to the process. The electromagnetic signals generated by partial discharge are often frequency signals ranging from 1 GHz to hundreds of GHz. The interference signals in this frequency band will affect the wireless communication system running at 2.4 GHz, causing system transmission errors. The operating frequency of 433MHz communication technology is much lower than this frequency band, so it can well avoid the impact of related interference on system communication. The advantage of using LoRa-WAN wireless sensor network is that it will not be affected by local GPRS signals, and it can be networked over a large area without increasing the subsequent maintenance costs. After each LoRa wireless node is networked, it can realize the reliable transmission of data between multiple non-contact sensors and detection hosts on site.

[0034] (3) Deep learning methods Deep learning is based on machine learning. The focus of deep learning is to build a neural network that imitates the human brain and understand images, videos, texts, etc. based on the human brain's thinking. Deep learning and machine learning are both branches of the AI ​​field. Deep learning is a type of machine learning that uses artificial neural networks with multiple hidden layers to solve complex problems.

[0035] The advantages of deep learning models are that they have extremely high predictive capabilities, are suitable for complex data and fault modes, and only require a small amount of feature engineering to process data formats. The disadvantages are that they have high requirements for data quality, are difficult to train models, require a lot of computing resources and time during training, and require high requirements for model hyperparameter adjustment. After years of development, more typical deep learning models include deep trust network models, autoencoder network models, and convolutional neural network (CNN) models. Convolutional neural networks are a feedforward neural network with a deep structure. CNN has strong feature learning potential and can perform translation invariant classification operations on input data based on the internal structure of each layer. Its research started before the 1980s and 1990s of the last century. Since the 21st century, with the extensive research of deep learning theory, convolutional neural networks have also been widely favored by researchers. And it has been applied to abnormal power consumption detection, image processing, speech recognition and other fields. Table 2 shows four relatively classic convolutional neural networks.

[0036] Convolutional neural network consists of the following three parts: input layer, hidden layer and output layer, where the hidden layer is divided into convolution layer, pooling layer and fully connected layer; Using deep learning models to predict power grid equipment failures usually involves the following steps: The first step is data collection and preprocessing: collecting historical data related to the operation and maintenance of power grid equipment, and preprocessing the data to make it suitable for deep learning model training; Step 2, feature engineering: Feature engineering refers to identifying relevant features that can help predict equipment failures. These features can be generated using time series analysis, statistical methods, or domain-specific knowledge;

[0037] Step 3: Model building: Use deep learning frameworks such as TensorFlow and Keras to build a deep learning model and train it using training data; Step 4: Model evaluation: Use the model to predict the data and use indicators such as accuracy and recall to evaluate the performance of the model; Step 5: Model optimization.

[0038] The beneficial effects of the non-contact type substation equipment dirty current live detection method of the present invention are: The present invention is applicable to the live detection of contamination leakage current on the surface of various capacitive equipment in the station, and has good practical value and promotion and application prospects. The promotion of the results is through the production of instruments and equipment. The new instrument will be applicable to the live detection of contamination current in power transmission and transformation equipment at home and abroad. The technology described in the present invention can not only be used in substations, but also can be used to detect insulators of transmission lines.

[0039] Aiming at the impact of pollution of substation equipment on its operation safety and the shortcomings of traditional detection methods, the present invention conducts research on the detection technology of external insulation pollution current of electrical equipment in substations based on advanced non-contact measurement technology and deep learning methods. It combines wireless communication technology to simultaneously perform live detection and analysis of parameters such as pollution current and discharge pulses of multiple electrical equipment, which can realize active early warning of potential operation risks, greatly improve detection efficiency, effectively prevent the occurrence of pollution flashover accidents, and improve the power supply reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 There is a contamination pattern on the surface of the porcelain insulator; Figure 2 Flashover test for post insulators in substations; Figure 3 This is the circuit block diagram of the array measurement system; Figure 4 One of the typical output curves of TMR2001 when powered by 1V; Figure 5 The second typical output curve of TMR2001 when powered by 1V; Figure 6 This is the typical output curve of TMR2001; Figure 7 It is a program flow chart; Figure 8 It is the LoRa network structure; Fig. 9 Simplified diagram for a convolutional neural network. DETAILED DESCRIPTION

[0041] The present invention will be further described below in conjunction with the embodiments and the accompanying drawings, but is not limited thereto.

[0042] Example 1 A non-contact method for detecting live pollution current in substation equipment; it can solve the time-consuming and laborious problem caused by regular pollution detection of traditional substation equipment. By non-contactly measuring the pollution current on the surface of the equipment, it can realize the rapid identification of equipment pollution without power outage, and can fully cover the pollution degree of the external insulation surface of the equipment, while improving the safety and efficiency of the detection process. What this embodiment does is to integrate the current insulator detection method, put forward its own ideas and verify the feasibility, and it is urgent to ensure the safe and reliable operation of the power system.

[0043] The technical solution of this embodiment is: a non-contact substation equipment contaminated current live detection method, the key technology of which is: this method uses a live detection device with a new current sensor based on TMR technology for live detection, the sensor is fixed on an insulating rod, and the tester can perform non-contact detection of the contaminated current and discharge pulse at the grounding flange of the substation high-voltage electrical equipment by holding the insulating rod. The measurement data is transmitted to the monitoring host through a wireless communication network, and the background software analysis of the host platform integrated with the deep learning algorithm can judge the external insulation contamination of the substation equipment; The host can collect data from 2 or more current sensors at the same time to improve the efficiency of field tests; the power supply mode adopts AC and DC dual-use; The non-contact substation equipment dirty current live detection method uses a dedicated array measurement system. The circuit structure diagram of the array measurement system is as follows: Figure 1 As shown; the dedicated array measurement system includes a power supply circuit, a TMR sensor array, a signal conditioning circuit, an ADC analog-to-digital converter, a microcontroller, a timer unit, and a wireless communication module; wherein: The TMR sensor array includes at least two TMR sensors, the TMR sensor array is connected to the microcontroller through the signal conditioning circuit and the ADC analog-to-digital converter in sequence, and the wireless communication module is also connected to the microcontroller; the power supply circuit is respectively connected to the TMR sensor array, the signal conditioning circuit, the ADC analog-to-digital converter, the microcontroller, the timer unit, and the wireless communication module; the timer unit is also connected to the microcontroller; The power supply circuit adopts a single 5 V power supply, which is converted by the power module to generate multi-level voltages of ±5V, +3.3V, +3V, and +1.5V to meet the power supply requirements of various components; The TMR sensor array uses the linear magnetic field sensor chip TMR2001 array; the linear magnetic field sensor chip TMR2001 array consists of 4 unshielded high-sensitivity TMRs to form a Wheatstone full-bridge structure, and the output has good temperature stability; the linear magnetic field sensor chip TMR2001 adopts SOT23-5 package. Under the action of an external magnetic field parallel to the sensitive direction, the bridge provides a differential voltage output. Figure 2 One of the typical output curves of TMR2001 when powered by 1 V, Figure 3 The second typical output curve of TMR2001 when powered by 1V; the linear magnetic field range of TMR2001 is ±5 Oe, the sensitivity is 8 mV / V / Oe, and the saturation magnetic field is -25Oe and +40 Oe.

[0044] The microcontroller is MSP430F149; the wireless communication module is a single-chip wireless transceiver chip nRF24L01 operating in the ISM frequency band of 2.4GHz~2.5GHz; the circuit diagram of the microcontroller MSP430F149 can be found in Figure 4 ; The timer unit is a 16-bit timer; The dedicated array measurement system uses software methods to realize the numerical integration function. The differential output signal of the TMR sensor is conditioned by the instrument amplifier and sent to the ADC for sampling and conversion. The definite integral is calculated by the microcontroller MSP430F149.

[0045] The non-contact type substation equipment dirty current live detection method described in this embodiment also meets the following requirements: In order to make full use of the 0-3V conversion range of the ADC, Rg selects a 1kΩ precision resistor with an accuracy of 0.1%, and the actual gain is 10.9 times; in order to enable the ADC to convert bipolar signals, a 1.5V forward bias is provided to the REF pin of the AD8421 to raise the output voltage; the output TMR sensor array of the AD8421 is designed on a circular PCB with an inner diameter of 60mm and an outer diameter of 100mm, and 8 TMR2001 chips are evenly arranged on a circle with a radius of 35mm along the tangent of the sensitive axis.

[0046] Table 1 TMR2001 performance parameters (VCC = 1.0 V, TA = 25°C)

[0047] The non-contact type substation equipment dirty current live detection method also meets the following requirements: The TMR sensor and signal amplification circuit also meet the following requirements: In order to improve the measurement accuracy of the system, TMR2001 and ADC external reference voltage VeREF+ both use the 3V reference voltage provided by the high-precision voltage source chip ADR4530, so the maximum output of TMR2001 in the linear range is ±120mV; because the output differential signal is weak, it is amplified by AD8421; this is a low-power, low-noise, high-speed instrumentation amplifier with a high common-mode rejection ratio, and the sensor offset voltage and system zero point accurately measure the current on the premise that the sensor can accurately reflect the magnetic field strength at the location. However, differences in manufacturing processes make it impossible for the TMR characteristics to be completely consistent, resulting in the output voltage of the bridge not being zero when there is no magnetic field, which is called the offset voltage V0. According to the data provided in Table 1, when powered by 1V, the V0 of TMR2001 is as high as ±30mV, which almost accounts for 3 / 4 of the linear output range of ±40mV, so the impact of the chip offset voltage on the system zero point cannot be ignored;

[0048] When an external magnetic field acts on the sensor, the output is superimposed with a magnetic induction intensity on the basis of the offset voltage: ωk is determined by the product formula, VTMR(k) and V0(k) represent the output and offset voltage of the kth sensor respectively. When there is no current in the loop, the above formula should be equal to zero. In any environment, the AD8421 output Vout(k) is collected using ADC. The output voltage VTMR(k) of the TMR sensor is obtained by formula (3), and the integral of V0 can be inferred. During the measurement process, correction is made according to formula (7), which can eliminate the system zero point caused by the chip offset voltage. Even if the external magnetic field environment changes, its integral is always zero, which will not affect the system zero point. This solution does not require magnetic field shielding and avoids repeated and tedious manual measurement operations. The microprocessor can achieve zero point compensation of the current sensor through simple calculations.

[0049] The steps and contents of the non-contact type substation equipment dirty current live detection method are as follows: After the hardware circuit is completed, signal acquisition and processing, numerical integration, zero compensation and other operations are all implemented by the internal program of the microcontroller. The specific process is as follows (see Figure 5 ): (1) After the system is powered on, the registers of the microcontroller MSP430F149's internal clock, I / O port, timer Timer_A, ADC12, serial port USART0 and other peripheral modules are initialized and set, and interrupts are enabled; the microcontroller can respond to the USART0RX_VECTOR interrupt to receive control instructions sent by the host computer and perform corresponding operations according to the instructions; the basic instructions include: reset (0x01), return the quadrature formula (0x02), return the ADC conversion result (0x03), switch the quadrature formula (0x04), zero adjustment (0x05), send measurement data (0x06), stop sending measurement data (0x07) and return parameter information (0xFF); (2) Before measurement, a zero adjustment operation must be performed. When there is no current in the loop, the host computer sends a zero adjustment command (0x05). After receiving it, the microcontroller MSP430F149 starts the ADC to measure the voltage signals of the eight channels and calculate the offset voltage integral. (3) The 16-bit hardware timer Timer_A inside the microcontroller MSP430F149 is set to generate a timer interrupt every 1ms. In the interrupt processing function, the ADC is started to perform serial channel conversion, collect the voltage of 8 channels for numerical integration, correct the result, and calculate and store the instantaneous value of the current; (4) Every 1 second, the 1000 stored instantaneous current values ​​are averaged to obtain the DC component, and the RMS calculation is performed to obtain the AC component. The measurement data is sent to the host computer using the wireless communication module. The project develops a non-contact substation equipment dirty current detection device. The required materials include TMR sensor (current measurement 0-100uA) for non-contact measurement, wireless communication module using 430MHZ for data communication, host shell using aluminum shell, current protector (capable of circuit protection against impact and interference, with a maximum withstand of 20kA), current acquisition module (capable of I / V conversion, forming a voltage range of 0-3.3V for the back-end available signal.

[0050] The non-contact type substation equipment dirty current live detection method also meets the following requirements: (1) The array measurement system specifically uses a non-shielded structure with good sensitivity: the four tunnel magnetoresistance TMR resistors in different sensitive directions in the TMR sensor are divided into two groups and connected into a Wheatstone bridge. Every two TMR resistors in opposite positions are divided into one group, and the magnetoresistance characteristics are consistent. The directions of TMR resistors in different groups are opposite. Under the same magnetic field conditions, the resistance of one group of TMR resistors increases, while the resistance of the other group decreases. A monitoring voltage that follows the change of the magnetic field can be generated at the monitoring outlet of the TMR sensor, and a corresponding monitoring current can be generated in the post-processing circuit according to the monitoring voltage. Application of Gaussian algorithm in measurement: Since the data will be interfered by many noises such as Gaussian white noise in the process of measuring data, the size and frequency of these noises are different, and the interference of these noises will affect the accuracy of the measured data. In order to obtain more accurate measurement data, the data needs to be denoised. Filtering is also called denoising. The filter in the system often cannot remove all the noise, and the filtering function is required to strengthen the filtering. White noise is a common noise in the power field. Since the power spectrum density width in practice is limited and its average power cannot be estimated, there is no white noise that strictly obeys the uniform distribution in practice. Gaussian white noise is ubiquitous, and the one-dimensional probability density of Gaussian white noise is also easy to calculate. Removing these noises is conducive to extracting load characteristics. Convolving the Gaussian function with the original signal is the key to the Gaussian algorithm. Compared with traditional algorithms, the Gaussian algorithm has the advantages of filtering window and phase shift. It not only retains the sudden signal, does not weaken and eliminate the sudden signal, but also can smooth the sudden data;

[0051] Gaussian filter formula: ,in, is the output value of the original load signal after the k+1th iteration; is the input value of the original load signal after the k+1th iteration; is the sum of the weighted coefficients of the filter at the kth iteration; is the weighting coefficient at the kth iteration point a+x; Set the initial iteration number k to 0 and set the number of iterations; bring the collected initial current and voltage data into the formula and iterate to the set number of iterations, and output the iteration result, which is the filtered value.

[0052] (2) The non-contact substation equipment dirty current detection method also meets the requirements of wireless communication system design principles. The detection data is transmitted by wireless network. At present, the wireless communication used in the online detection system of the domestic power system mainly includes LORA-433MHz wireless technology. These are all short-range wireless communication technologies and use the ISM license-free frequency band, but they each have their own characteristics. LoRaWAN is a highly integrated, low-power wireless control module that mainly uses LoRa's patented remote modulation and demodulation technology. It has a built-in STM32L series CPU for ultra-long-distance spread spectrum communication. It minimizes current consumption while greatly improving anti-interference and sensitivity. LWM supports three protocol modes, LoRaWAN Class A, B, and C. LoRa supports high-performance (G)FSK mode for systems such as WM-Bus. Compared with similar devices, LWM uses RF solutions to significantly reduce current consumption while significantly optimizing various performance such as phase noise, selectivity, receiver linearity, and third-order input intercept point (IIP3). The LoRa network structure is as follows: Figure 6 shown.

[0053] The wavelength of the 433MHz frequency band is nearly one meter. According to the relationship between wavelength and propagation distance, we know that the longer the wavelength, the better the diffraction of the electromagnetic wave, the better it can bypass obstacles and propagate farther. Therefore, compared with WiFi and Zigbee, the significant advantage of LORA-433MHz is that the wireless signal has strong penetration and can propagate farther. Such advantages are very suitable for wireless communication in cable online detection systems. This is mainly because most urban cables are buried underground, and wireless signals require high penetration to transmit information to receivers on the ground. At the same time, due to the dense distribution of high-rise buildings in cities and the frequent bending changes in the layout, wireless signals can only transmit signals to receivers if they have a high ability to bypass obstacles. Although its data transmission rate is only 9600bps, which is much lower than the data rate of WiFi and Zigbee, it can fully meet the wireless communication requirements of cable online monitoring systems with small data volumes. In addition, any low-voltage equipment in power system applications must consider the strong electromagnetic interference generated by the power system itself due to its operating characteristics, which is often called EMC problems. For example, when the cable is in operation, partial discharge problems may occur due to the process. The electromagnetic signals generated by partial discharge are often frequency signals ranging from 1 GHz to hundreds of GHz. The interference signals in this frequency band will affect the wireless communication system running at 2.4 GHz, causing system transmission errors. The operating frequency of 433MHz communication technology is much lower than this frequency band, so it can well avoid the impact of related interference on system communication. The advantage of using LoRa-WAN wireless sensor network is that it will not be affected by local GPRS signals, and it can be networked over a large area without increasing the subsequent maintenance costs. After each LoRa wireless node is networked, it can realize the reliable transmission of data between multiple non-contact sensors and detection hosts on site.

[0054] (3) Deep learning methods Deep learning is based on machine learning. The focus of deep learning is to build a neural network that imitates the human brain and understand images, videos, texts, etc. based on the human brain's thinking. Deep learning and machine learning are both branches of the AI ​​field. Deep learning is a type of machine learning that uses artificial neural networks with multiple hidden layers to solve complex problems.

[0055] The advantages of deep learning models are that they have extremely high predictive capabilities, are suitable for complex data and fault modes, and only require a small amount of feature engineering to process data formats. The disadvantages are that they have high requirements for data quality, are difficult to train models, require a lot of computing resources and time during training, and require high requirements for model hyperparameter adjustment. After years of development, more typical deep learning models include deep trust network models, autoencoder network models, and convolutional neural network (CNN) models. Convolutional neural networks are feedforward neural networks with deep structures. CNN has strong feature learning potential and can perform translation invariant classification operations on input data based on the internal structure of each layer. Its research started before the 1980s and 1990s of the last century. Since the 21st century, with the extensive research of deep learning theory, convolutional neural networks have also been widely favored by researchers. And it has been applied to abnormal power consumption detection, image processing, speech recognition and other fields. Table 2 shows four relatively classic convolutional neural networks.

[0056] Convolutional neural network consists of the following three parts: input layer, hidden layer and output layer, where the hidden layer is divided into convolution layer, pooling layer and fully connected layer; Using deep learning models to predict power grid equipment failures usually involves the following steps: The first step is data collection and preprocessing: collecting historical data related to the operation and maintenance of power grid equipment, and preprocessing the data to make it suitable for deep learning model training; Step 2, feature engineering: Feature engineering refers to identifying relevant features that can help predict equipment failures. These features can be generated using time series analysis, statistical methods, or domain-specific knowledge;

[0057] Step 3: Model building: Use deep learning frameworks such as TensorFlow and Keras to build a deep learning model and train it using training data; Step 4: Model evaluation: Use the model to predict the data and use indicators such as accuracy and recall to evaluate the performance of the model; Step 5: Model optimization.

[0058] The beneficial effects of the non-contact type substation equipment dirty current live detection method of the present invention are: The present invention is applicable to the live detection of contamination leakage current on the surface of various capacitive equipment in the station, and has good practical value and promotion and application prospects. The promotion of the results is through the production of instruments and equipment. The new instrument will be applicable to the live detection of contamination current in power transmission and transformation equipment at home and abroad. The technology described in the present invention can not only be used in substations, but also can be used to detect insulators of transmission lines.

[0059] Aiming at the impact of pollution of substation equipment on its operation safety and the shortcomings of traditional detection methods, the present invention conducts research on the detection technology of external insulation pollution current of electrical equipment in substations based on advanced non-contact measurement technology and deep learning methods. It combines wireless communication technology to simultaneously perform live detection and analysis of parameters such as pollution current and discharge pulses of multiple electrical equipment, which can realize active early warning of potential operation risks, greatly improve detection efficiency, effectively prevent the occurrence of pollution flashover accidents, and improve the power supply reliability of the system.

Claims

1. A non-contact type detection method for dirty current in transformer substation, characterized in that: This method uses a live detection device with a new type of current sensor based on TMR technology for live detection. The sensor is fixed on an insulating rod. The tester can perform non-contact detection of the contamination current and discharge pulse at the grounding flange of the high-voltage electrical equipment in the substation by holding the insulating rod. The measurement data is transmitted to the monitoring host through a wireless communication network. The background software analysis of the host platform integrated with the deep learning algorithm can then determine the external insulation contamination of the substation equipment. The host can collect data from 2 or more current sensors at the same time to improve the efficiency of field tests; The power supply mode adopts AC and DC dual-use; The contactless substation equipment dirty current live detection method uses a dedicated array measurement system, which includes a power supply circuit, a TMR sensor array, a signal conditioning circuit, an ADC analog-to-digital converter, a microcontroller, a timer unit, and a wireless communication module; wherein: The TMR sensor array includes at least two TMR sensors, the TMR sensor array is connected to the microcontroller through the signal conditioning circuit and the ADC analog-to-digital converter in sequence, and the wireless communication module is also connected to the microcontroller; the power supply circuit is respectively connected to the TMR sensor array, the signal conditioning circuit, the ADC analog-to-digital converter, the microcontroller, the timer unit, and the wireless communication module; the timer unit is also connected to the microcontroller; The power supply circuit adopts a single 5 V power supply, which is converted by the power module to generate multi-level voltages of ±5V, +3.3V, +3V, and +1.5V to meet the power supply requirements of various components; The TMR sensor array uses the linear magnetic field sensor chip TMR2001 array; the linear magnetic field sensor chip TMR2001 array is composed of 4 unshielded high-sensitivity TMRs to form a Wheatstone full-bridge structure, and the output has good temperature stability; the linear magnetic field sensor chip TMR2001 adopts SOT23-5 package. Under the action of an external magnetic field parallel to the sensitive direction, the bridge provides a differential voltage output. The linear magnetic field range of TMR2001 is ±5 Oe, the sensitivity is 8 mV / V / Oe, and the saturation magnetic field is -25 Oe and +40 Oe. The microcontroller is MSP430F149; the wireless communication module is a single-chip wireless transceiver chip nRF24L01 operating in the ISM frequency band of 2.4GHz~2.5GHz; The timer unit is a 16-bit timer; The dedicated array measurement system can realize the numerical integration function. The differential output signal of the TMR sensor is conditioned by the instrument amplifier and sent to the ADC for sampling and conversion. The definite integral is calculated by the microcontroller MSP430F149.

2. The method for detecting the contaminated current of non-contact substation equipment according to claim 1, characterized in that: The non-contact type substation equipment dirty current live detection method also meets the following requirements: In order to make full use of the 0-3V conversion range of the ADC, Rg selects a 1kΩ precision resistor with an accuracy of 0.1%, and the actual gain is 10.9 times; in order to enable the ADC to convert bipolar signals, a 1.5V forward bias is provided to the REF pin of the AD8421 to raise the output voltage; the output TMR sensor array of the AD8421 is designed on a circular PCB with an inner diameter of 60mm and an outer diameter of 100mm, and 8 TMR2001 chips are evenly arranged on a circle with a radius of 35mm along the tangent of the sensitive axis.

3. The method for detecting the contaminated current of non-contact substation equipment according to claim 1 or 2, characterized in that: The non-contact type substation equipment dirty current live detection method also meets the following requirements: The TMR sensor and signal amplification circuit also meet the following requirements: In order to improve the measurement accuracy of the system, TMR2001 and ADC external reference voltage VeREF+ both use the 3V reference voltage provided by the high-precision voltage source chip ADR4530, so the maximum output of TMR2001 in the linear range is ±120mV; because the output differential signal is weak, it is amplified by AD8421; When an external magnetic field acts on the sensor, the output is superimposed with a magnetic induction intensity on the basis of the offset voltage: ωk is determined by the quadrature formula, VTMR(k) and V0(k) represent the output and offset voltage of the kth sensor respectively; when there is no current in the loop, the above formula should be equal to zero. In any environment, the ADC is used to collect the AD8421 output Vout(k), and the output voltage VTMR(k) of the TMR sensor is obtained, which can be inverted to deduce the V0 integral; Correction during the measurement process can eliminate the system zero point caused by chip offset voltage; Even if the external magnetic field environment changes, its integral is always zero and will not affect the system zero point.

4. The method for detecting the contaminated current of non-contact substation equipment according to claim 3 is characterized in that: The steps and contents of the non-contact type substation equipment dirty current live detection method are as follows: (1) After the system is powered on, the registers of the microcontroller MSP430F149's internal clock, I / O port, timer Timer_A, ADC12, and serial port USART0 and other peripheral modules are initialized and enabled, and the interrupt is enabled; the microcontroller can respond to the USART0RX_VECTOR interrupt to receive the control instructions sent by the host computer and perform corresponding operations according to the instructions; (2) Before measurement, a zero adjustment operation must be performed. When there is no current in the loop, the host computer sends a zero adjustment command (0x05). After receiving it, the microcontroller MSP430F149 starts the ADC to measure the voltage signals of the eight channels and calculate the offset voltage integral. (3) The 16-bit hardware timer Timer_A inside the microcontroller MSP430F149 is set to generate a timer interrupt every 1ms. In the interrupt processing function, the ADC is started to perform serial channel conversion, collect the voltage of 8 channels for numerical integration, correct the result, and calculate and store the instantaneous value of the current; (4) Every 1s, the 1000 stored instantaneous current values ​​are averaged to obtain the DC component, and the RMS calculation is performed to obtain the AC component, and the measurement data is sent to the host computer using the wireless communication module; in the non-contact substation equipment dirty current detection device, the current measurement range of the TMR sensor is 0~100uA, which is used for non-contact measurement; the wireless communication module uses 430MHz for data communication; the host shell uses an aluminum shell, and the current protector can protect the circuit against impact and interference, with a maximum withstand of 20kA; the current acquisition module can perform I / V conversion, and the voltage range of the back-end available signal is 0-3.3V.

5. The method for detecting the contaminated current of non-contact substation equipment according to claim 4, characterized in that: The non-contact type substation equipment dirty current live detection method also meets the following requirements: The array measurement system specifically uses a non-shielded structure with good sensitivity: the four tunnel magnetoresistance TMR resistors in different sensitive directions in the TMR sensor are divided into two groups and connected into a Wheatstone bridge, and each two TMR resistors in opposite positions are divided into one group, and the magnetoresistance characteristics are consistent, and the TMR resistors in different groups are in opposite directions; under the same magnetic field conditions, the resistance value of one group of TMR resistors increases, and the resistance value of the other group decreases; a monitoring voltage that follows the change of the magnetic field can be generated at the monitoring outlet of the TMR sensor, and a corresponding monitoring current can be generated in the post-processing circuit according to the monitoring voltage; Application of Gaussian algorithm in measurement: Since the data will be interfered by Gaussian white noise during measurement, the Gaussian function is convolved with the original signal to retain the mutation signal without weakening or eliminating the mutation signal, and the mutation data can be smoothed. Gaussian filter formula: ,in, is the output value of the original load signal after the k+1th iteration; is the input value of the original load signal after the k+1th iteration; is the sum of the weighted coefficients of the filter at the kth iteration; is the weighting coefficient at the kth iteration point a+x; Set the initial iteration number k to 0 and set the number of iterations; bring the collected initial current and voltage data into the formula and iterate to the set number of iterations, and output the iteration result, which is the filtered value.

6. The method for detecting the contaminated current of non-contact substation equipment according to claim 5, characterized in that: The non-contact type substation equipment dirty current live detection method also meets the following requirements: The contactless substation equipment dirty current live detection method also meets the requirements of the wireless communication system design principle: the detection data is transmitted by wireless network, LoRaWAN is a highly integrated low-power wireless control module, mainly using LoRa patented remote modulation and demodulation technology, built-in STM32L series CPU, for ultra-long distance spread spectrum communication, which minimizes current consumption while greatly improving anti-interference and sensitivity; LWM supports three protocol modes: LoRaWAN Class A, B, and C. LoRa supports the high-performance FSK mode of the WM-Bus system. Compared with similar devices, LWM uses RF solutions to significantly reduce current consumption while significantly optimizing phase noise, selectivity, receiver linearity, and third-order input intercept point performance. The wavelength of the 433MHz frequency band is nearly one meter, and its operating frequency can effectively avoid the impact of related interference on system communications; the LoRa-WAN wireless sensor network will not be affected by the local GPRS signal, and can be networked over a large area without increasing subsequent maintenance costs; after the LoRa wireless nodes are networked, they can realize reliable transmission of data between multiple non-contact sensors and detection hosts on site.

7. The method for detecting the contaminated current of non-contact substation equipment according to claim 5, characterized in that: The contactless substation equipment dirty current live detection method also meets the following requirements: Deep learning method: Deep learning is based on machine learning; Convolutional neural network is a feedforward neural network with a deep structure; Convolutional neural network has a strong feature learning potential and can perform translation invariant classification operations on input data according to the internal structure of each layer.

8. The method for detecting the contaminated current of non-contact substation equipment according to claim 7, characterized in that: The non-contact type substation equipment dirty current live detection method also meets the following requirements: Convolutional neural network consists of the following three parts: input layer, hidden layer and output layer, where the hidden layer is divided into convolution layer, pooling layer and fully connected layer; Using deep learning models to predict power grid equipment failures usually involves the following steps: The first step is data collection and preprocessing: collecting historical data related to the operation and maintenance of power grid equipment, and preprocessing the data to make it suitable for deep learning model training; Step 2: Feature Engineering: Feature engineering refers to identifying relevant features that can help predict equipment failures; these features can be generated by using time series analysis, statistical methods, or domain-specific knowledge; Step 3: Model building: Use the deep learning framework to build a deep learning model and train it using training data; Step 4: Model evaluation: Use the model to predict the data and use indicators to evaluate the performance of the model; Step 5: Model optimization.

Images