Medium-voltage power distribution network fault positioning identification method and system

Abstract

The invention relates to the technical field of power systems, in particular to a medium-voltage power distribution network fault positioning identification method and system.The voltage value and the current value of a target power distribution line are collected through a detection unit, and a fault positioning mathematical model obtains the position of a medium-voltage information fault point according to the voltage value and the current value recorded when a fault happens to the target power distribution line; comprising the steps that a fault positioning mathematical model completes fault positioning of a target distribution line based on a single-ended impedance method; and fault type identification is carried out by using a convolutional neural network in deep learning. The method has the advantages that the cloud server transfers the acquired data of the target distribution line into the set mathematical model based on the single-ended impedance method, and the fault position is judged based on the mathematical model based on the single-ended impedance method; and the finally determined fault position, current, voltage, fault current data and overcurrent alarm information are sent to a WeChat applet together for remote checking by operation and maintenance personnel, so that the working pressure of the operation and maintenance personnel is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for locating and identifying faults in a medium-voltage distribution network. Background Art

[0002] With the development of society, people are becoming more and more dependent on electricity. As an important part of the power system, the safe and stable operation of the medium-voltage distribution network is of vital importance. When a fault occurs in the medium-voltage distribution network, the ability to quickly and accurately locate the fault and reduce power outage time and losses is the key to improving power supply reliability.

[0003] Many medium-voltage users lack telecontrol devices, preventing remote transmission of fault information. This makes fault inspection and troubleshooting for maintenance personnel arduous and challenging. Short-circuit current data is missing, making it impossible to obtain distribution network short-circuit current data, determine the transition resistance of the distribution network fault type, and use short-circuit current to aid in fault location determination.

[0004] At present, the mainstream distribution terminal equipment is assembled from three types of power devices, namely DTU-microcomputer remote terminal unit, FTU-feeder remote terminal system and TTU-distribution transformer remote terminal system.

[0005] DTU - microcomputer telecontrol terminal unit is usually installed in distribution substations. It can collect various electrical quantity data in the substation in real time, such as voltage, current, power, etc., and transmit the collected data to the dispatching center through the communication network. At the same time, it can also receive control instructions issued by the dispatching center to realize remote control of equipment in the substation and ensure the stable operation of the substation.

[0006] The FTU (Federal Terminal Unit) is an interface that bridges primary equipment and automation systems. FTUs are widely used in power ancillary facilities such as circuit breakers, sectionalizers, distribution system transformers, and ring main units. Their primary function is to comprehensively control and monitor the operating conditions of these external devices. By monitoring equipment operating parameters in real time, the FTU can promptly detect abnormalities and take swift action, such as automatically isolating the faulty area to ensure normal power supply to non-faulty areas. Furthermore, the FTU maintains a close communication relationship with the feeder's master station. During distribution system circuit operation, when relevant information about circuit conditions is generated, the FTU transmits this information to the feeder's master station. The feeder's master station provides data related to circuit operation control and management. Simultaneously, the master station's control and adjustment commands for terminal distribution equipment are smoothly transmitted through the FTU, enabling precise control of the entire feeder system.

[0007] The TTU-Distribution Transformer Remote Terminal System focuses on the monitoring and management of distribution transformers. It can monitor parameters such as oil temperature, winding temperature, and load conditions of distribution transformers in real time. Through data analysis, it can promptly identify potential faults of the transformer, provide early warning and maintenance, and effectively improve the operational reliability and service life of the distribution transformer.

[0008] However, although feeder automation terminals play an important role in power distribution network automation, there are also some problems that need to be solved urgently:

[0009] 1) The high cost of mainstream power distribution terminal equipment has limited its large-scale promotion and application to a certain extent;

[0010] 2) Due to the large number of existing users in the early stages of construction, the project of renovating these existing users is very large, requiring not only a large investment of manpower, material and financial resources, but also facing many challenges such as technical compatibility;

[0011] 3) Currently, power distribution terminal equipment (including DTU, FTU, and TTU) only serves as a fault diagnosis device and lacks fault location function. When a fault occurs, the fault location cannot be quickly and accurately determined, which affects the efficiency of fault repair.

[0012] 4) There are deficiencies in mobile data display, which hinders operation and maintenance personnel from obtaining equipment operation information anytime and anywhere, and cannot meet the convenience and efficiency requirements of modern power operation and maintenance. Summary of the Invention

[0013] The purpose of the present invention is to provide a method and system for locating and identifying faults in a medium-voltage distribution network. The method and system have a simple structure, low cost, and do not require manual visual line inspection. The method can quickly and accurately determine the fault location and fault type, and display the grid operation status and fault location information in a WeChat applet, so that operation and maintenance personnel can quickly and promptly handle the fault problem.

[0014] To achieve the above object, the present invention is implemented through the following technical solutions:

[0015] A method for locating and identifying faults in a medium-voltage distribution network uses a detection unit to collect voltage and current values of a target distribution line. A fault location mathematical model obtains the location of the medium-voltage information fault point based on the voltage and current values recorded when the target distribution line fault occurs. The method specifically includes:

[0016] S1. The fault location mathematical model is based on the single-ended impedance method to locate the target distribution line fault;

[0017] S2. Use convolutional neural networks in deep learning to identify fault types.

[0018] In S1, the fault location mathematical model is based on the single-ended impedance method, and the calculation formula is as follows:

[0019] a-phase voltage U A The calculation formula is as follows:

[0020] U A =Z s I a +Z m (I b +I c ) (1)

[0021] In formula (1), Z S is the self-impedance of the conductor, Z M is the mutual impedance of the conductor, I a is the a-phase current, I b is the b-phase current, I c is the c-phase current;

[0022] I r The sum of the three-phase currents a, b, and c is I r The calculation formula is as follows:

[0023] I r =I a +I b +I c (2)

[0024] The calculation formulas for the self-impedance and mutual impedance of the conductor using sequence components are as follows:

[0025]

[0026] In formulas (3) and (4), Z1 is the positive sequence impedance; Z0 is the zero sequence impedance;

[0027] Substituting equations (2) to (4) into equation (1), we can obtain:

[0028] U a =Z1(I a +kI r ) (5)

[0029]

[0030] In formulas (5) and (6), k is a coefficient;

[0031] The calculation formula of phase A voltage when a one-way ground fault occurs on phase A is as follows:

[0032] U a =mZ L (I a +kI r )+I F RF (7)

[0033] Assuming that the transition resistance is purely resistive, the measured current I is in phase with the short-circuit current at the fault point. Multiply both sides of formula (7) by I* and take only the imaginary part. By measuring the ratio of the reactance to the line reactance, the position of the fault point can be obtained. The calculation formula is as follows:

[0034] When a unidirectional ground fault occurs on phase a, the percentage of the fault position to the total line length is as follows:

[0035]

[0036] In formulas (7) and (8), Z L is the total line impedance, U a is the phase voltage of phase a; I a is the phase current of phase a; Z L is the total line impedance; I F is the fault current; R F is the transition resistance; For I r The conjugate complex number of , m is the percentage of the fault position to the total length of the line; I mg Performs conjugate multiplication on a complex expression and extracts the imaginary part of the result.

[0037] In S2, the convolutional neural network includes a convolutional layer, an activation function layer, a pooling layer, a fully connected layer, and a softmax classification layer. The convolutional neural network uses the convolutional layer and the pooling layer to extract the core features of the target distribution line status and identify the fault type.

[0038] Fault type identification uses the current waveform as the initial input. Features are gradually extracted through the convolution layer, activation function layer, and pooling layer to capture spatial features in the image. Finally, the softmax classification layer is used to effectively identify the target distribution line status. The formula for the problem classification model is as follows:

[0039]

[0040] In formula (9), x is the input current waveform, W j and b j They correspond to the relevant weights and bias values of the jth state, K is the total number of categories of different states, W k For each state, b k is the bias corresponding to the kth state, and P(y=j|x) is the probability that the model predicts that the input x belongs to category j;

[0041] In the target distribution line fault detection, the problem classification model is used to classify the input current waveform data into a fault state or a non-fault state:

[0042] If the tag is set to 1, it means that the current target distribution line is in a fault state;

[0043] If the tag is set to 0, it means that the current target distribution line is in a non-fault state;

[0044] Each sample data is evaluated using cross entropy loss, and the calculation formula is as follows:

[0045] Loss=-(y*log(py)+(1-y)*log(1-p)) (10)

[0046] In formula (10), y is the true label of the sample, which has a value of 0 or 1; p is the predicted probability of the sample by the question classification model, which is used to narrow the gap between the predicted probability of the question classification model and the true label.

[0047] The cross entropy loss function is used to measure the gap between the predicted probability and the true label. Back propagation is used to continuously correct the relevant weights W and b corresponding to a certain state, and the results are passed back from the output layer by layer to continuously adjust the relevant weights W and bias b.

[0048] The cross entropy loss function is used to measure the gap between the predicted probability and the true label. The cross entropy loss function is passed back layer by layer from the output of the neural network. The internal parameters of the problem classification model are adjusted according to the gap between the predicted probability and the true label, and the cross entropy loss function value is continuously reduced.

[0049] The internal parameters of the question classification model include the relevant weight W and bias b.

[0050] The medium voltage information fault location and identification system includes a detection unit, which includes a CPU controller, a data acquisition module, a data storage module, and a communication module. The output end of the data acquisition module is connected to the CPU controller, the data storage module is connected to the CPU controller, and the CPU controller is connected to the cloud server through the communication module.

[0051] The detection unit also includes a power supply module, which is used to provide working power to the CPU controller, data acquisition module, data storage module, and communication module.

[0052] The data acquisition module includes a current data acquisition module and a voltage data acquisition module. The input end of the current data acquisition module is connected to the current transformer of the target distribution line for collecting the current value of the target distribution line. The input end of the voltage data acquisition module is connected to the voltage transformer of the target distribution line for collecting the voltage value of the target distribution line. The output end of the current data acquisition module is connected to the CPU controller, and the output end of the voltage data acquisition module is connected to the CPU controller.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The detection unit collects the voltage and current values of the target distribution line and transmits the collected data to the cloud server. The cloud server then enters the collected data into a pre-set mathematical model based on the single-ended impedance method. The cloud server uses the single-ended impedance method mathematical model to determine the fault location and sends the final fault location, current, voltage, fault current data, and overcurrent alarm information to the WeChat mini-program for remote viewing by operation and maintenance personnel, thereby reducing the workload of operation and maintenance personnel and lowering operating costs.

[0055] 2. The state recognition and problem classification model for distribution network fault types based on deep learning is a key link in ensuring the safe and stable operation of the distribution network system. State recognition mainly involves accurately judging the current operating status of the distribution network system and can accurately and quickly determine the fault type. Combined with the fault location model based on the single-ended impedance method, it can determine the fault type and predict the fault location in a short time, assisting operation and maintenance personnel to handle faults in a timely manner, greatly improving the fault handling efficiency of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is the main control circuit diagram of the CPU controller.

[0057] Figure 2 The power supply module principle Figure 1 .

[0058] Figure 3 The power supply module principle Figure 2 .

[0059] Figure 4 The power supply module principle Figure 3 .

[0060] Figure 5 The principle of current data acquisition module Figure 1 .

[0061] Figure 6 The principle of current data acquisition module Figure 2 .

[0062] Figure 7The principle of current data acquisition module Figure 3 .

[0063] Figure 8 The principle of voltage data acquisition module Figure 1 .

[0064] Figure 9 The principle of voltage data acquisition module Figure 2 .

[0065] Figure 10 The principle of voltage data acquisition module Figure 3 .

[0066] Figure 11 This is the schematic diagram of the data storage module.

[0067] Figure 12 This is the schematic diagram of the communication module.

[0068] Figure 13 It is a three-phase power distribution circuit diagram.

[0069] Figure 14 This is a single-phase ground fault circuit diagram.

[0070] Figure 15 It is the basic structure diagram of deep convolutional network.

[0071] Figure 16 It is the main interface of the WeChat mini program.

[0072] Figure 17 This is the WeChat mini program device details page.

[0073] Figure 18 This is a model diagram of a faulty distribution line.

[0074] Figure 19 It is a schematic diagram of the single-phase ground fault current waveform.

[0075] Figure 20 It is a schematic diagram of the voltage waveform of a single-line ground fault.

[0076] Figure 21 This is the working flow chart of the medium voltage distribution network fault online monitoring system. DETAILED DESCRIPTION

[0077] The present invention will be described in detail below with reference to the accompanying drawings, but it should be noted that the implementation of the present invention is not limited to the following embodiments.

[0078] The following examples are implemented under the premise of the technical solution of the present invention, and provide detailed implementation methods and specific operating processes, but the scope of protection of the present invention is not limited to the following examples. The methods used in the following examples are conventional methods unless otherwise specified.

[0079] Example 1

[0080] The target distribution line is equipped with a voltage transformer and a current transformer. The voltage transformer and current transformer collect current and voltage data information of the target distribution line and transmit it to the data acquisition module. The CPU controller is connected to the human-machine interface screen through communication module 2. The CPU controller compares the current and voltage data information collected by the data acquisition module with the overcurrent threshold set on the human-machine interface on the human-machine interface screen. When the current and voltage collected values of the target distribution line exceed the set overcurrent threshold, an alarm signal is triggered. The microcontroller uploads the overcurrent value (fault current) and alarm signals (current alarm, voltage alarm) to the communication module via serial communication. The communication module uploads the data processed by the microcontroller to the cloud server. The cloud server moves the collected data into the set mathematical model based on the single-ended impedance method. The cloud server determines the fault location based on the mathematical model of the single-ended impedance method and sends the final fault location, as well as the current, voltage, fault current data and overcurrent alarm information, to the WeChat applet for remote viewing by operation and maintenance personnel, thereby reducing the workload of operation and maintenance personnel and lowering operating costs.

[0081] A medium-voltage distribution network fault location and identification system includes a detection unit based on an STM32 single-chip microcomputer. The detection unit includes a CPU controller, a data acquisition module, a data storage module, a WIFI communication module, and a power supply module. The power supply module is used to provide working power to the CPU controller, the data acquisition module, the data storage module, and the communication module. The output end of the data acquisition module is connected to the CPU controller, the data storage module is connected to the CPU controller, and the CPU controller is connected to the cloud server through the communication module. Figures 1-12 .

[0082] The CPU controller includes a single-chip microcomputer U1, a crystal oscillator Y1, a capacitor C26, a capacitor C27, a resistor R19, a magnetic bead FB1, and a magnetic bead FB2; the PH0 interface of the single-chip microcomputer U1 is respectively connected to one end of the crystal oscillator Y1, one end of the capacitor C26, and one end of the resistor R19, the PH1 interface of the single-chip microcomputer U1 is respectively connected to the other end of the crystal oscillator Y1, one end of the capacitor C27, and the other end of the resistor R19, the other ends of the capacitors C26 and C27 are grounded, the magnetic beads FB1 and FB2 are respectively connected to the 1 and 13 interfaces of the single-chip microcomputer U1, and the other sides are respectively connected to the 3.3V and 3V DC voltage sources, see Figure 1 .

[0083] The CPU controller serves as the data receiving and processing center, and controls the division of labor and cooperation among various modules. The CPU controller's core is based on the ARM Cortex-M, which has the characteristics of high performance and low power consumption. It is equipped with a high-speed processor and memory, and can easily handle complex applications. The CPU controller is an STM32 microcontroller, which supports multiple peripheral interfaces and communication protocols, such as USB, CAN, SPI, I2C, etc., to facilitate communication with other devices.

[0084] The power supply module includes a 5V power supply, capacitors C9, C11, C13, C15, and a low-voltage regulator U6; the 5V power supply is connected to the in interface of the low-voltage regulator U6, and connected to one end of the capacitors C9 and C15, and the other end of the capacitors C9 and C15 is grounded; the out interface of the low-voltage regulator U6 is connected to one end of the capacitors C11 and C13, and outputs a stable 3.3V voltage, the other end of the capacitors C11 and C13 is grounded, and the GND end of the low-voltage regulator U6 is grounded, see Figure 2-Figure 4 .

[0085] The power supply module not only provides a stable 3.3V voltage to the CPU controller and Wi-Fi communication module, but also provides 5V voltage to the serial port screen (i.e., the human-machine interface screen) and the voltage and current transformers. The external power supply inputs a 4-35V voltage through port P1. This voltage passes through an LM2596 low-voltage regulator diode U3, which stabilizes the output voltage at 5V. This voltage then passes through TLV1117-3.3 linear regulator chips, resulting in low-voltage regulator diodes U6 and U5, which stabilize the output voltage at 3.3V. Finally, LEDs LED5 and LED3 verify the voltage output.

[0086] The data acquisition module includes a current data acquisition module and a voltage data acquisition module. The input end of the current data acquisition module is connected to the current transformer of the target distribution line for collecting the current value of the target distribution line. The input end of the voltage data acquisition module is connected to the voltage transformer of the target distribution line for collecting the voltage value of the target distribution line. The output end of the current data acquisition module is connected to the CPU controller, and the output end of the voltage data acquisition module is connected to the CPU controller.

[0087] See Figure 5-Figure 7 The circuit of the current data acquisition module includes: interface ADC_IA, interface ADC_IB, interface ADC_IC, resistor R33, resistor R34, resistor R35, operational amplifier U14B, operational amplifier U14A, operational amplifier U12B; the 5th interface of the operational amplifier U14B is connected to IA, the 3rd interface of the operational amplifier U14A is connected to IB, and the 5th interface of the operational amplifier U12B is connected to IC; the 6th, 2nd, and 6th interfaces of the operational amplifier U14B, operational amplifier U14A, and operational amplifier U12B are connected to Figure 3The corresponding ports 7, 1, and 7 of the operational amplifier are connected to one end of resistors R33, R34, and R35, and the other ends of resistors R33, R34, and R35 are connected to the interface ADC_IA, the interface ADC_IB, and the interface ADC_IC respectively.

[0088] See Figures 8-10 The circuit of the voltage data acquisition module includes interfaces UAin, UBinn, UCInn, operational amplifiers U18A, U18B, U15A, resistors R63, R64, and R65; interface 3 of the operational amplifier U18A is connected to UA, interface 5 of the operational amplifier U18B is connected to UB, and interface 3 of the operational amplifier U15A is connected to UC; ports 2, 6, and 2 of the operational amplifier U18A, the operational amplifier U18B, and the operational amplifier U15A are respectively connected to ports 1, 7, and 1 of the corresponding operational amplifiers and one end of the resistors R63, R64, and R65, and the other ends of the resistors R63, R64, and R65 are respectively connected to interfaces UAin, UBinn, and UCin.

[0089] The data acquisition module uses capacitors for filtering when using current transformers to collect phase A, B, and C currents, or uses voltage transformers to collect UA, UB, and UC. The LM358 operational amplifier processes the collected data and transmits it to the STM32 chip for data processing. The human-computer interaction interface is used to set the overcurrent threshold. The CPU controller compares the collected current value of the target distribution line with the threshold set through the human-computer interaction interface. When the collected current value is greater than the current threshold set through the human-computer interaction interface, an overcurrent alarm signal is triggered.

[0090] See Figure 11 The data storage module includes: register U8, interface RESET, CLK_273, CTR1_RLY, CTR2_RLY, CTR3_RLY, RELAY1, RELAY2, RELAY3, CLK_273; interface 1 of register U8 is connected to the RESET interface of the STM32 microcontroller U1, interfaces 19, 16, and 15 of register U8 are respectively connected to the CTR1_RLY, CTR2_RLY, and CTR3_RLY interfaces of the composite transistor array U9, and interfaces 18, 17, and 14 of register U8 are respectively connected to the RELAY1, RELAY2, and RELAY3 interfaces of the microcontroller U1; interface 11 of register U8 is connected to CLK_273 of the microcontroller U1; the register is composed of multiple latches, each latch is responsible for storing a binary number; by controlling different input signals, the read and write operations of each bit in the register can be realized.

[0091] See Figure 12The circuit of the WIFI communication module includes: a WiFi module U17, a 3.3V DC power supply, a switch S1, a resistor R62, a resistor R58, a resistor R59, an interface WiFi_RX, and an interface WiFi_TX; one end of the switch S1 is grounded and the other end is connected to one end of the resistor R62, the other end of which is connected to the EN / NC interface of the WiFi module U17; the interface WiFi_RX and the interface WiFi_TX of the microcontroller U1 are respectively connected to the IO16_TXD and IO7_RXD interfaces of the chip; one end of the resistors R58 and R59 is commonly connected to the 3.3V DC voltage source, and the other side is respectively connected to the IO16_TXD and IO7_RXD interfaces of the WiFi module U17; the WIFI communication module adopts the FS704U module, and the WIFI communication module uploads the data processed by the data processing module to the cloud server. The WIFI communication module adopts a baud rate of 115200 baud.

[0092] The cloud server uses Alibaba Cloud Server ECS (Elastic Compute Service), which is a cloud computing service based on the IaaS (Infrastructure as a Service) level, providing excellent performance, stable reliability and elastic scalability. Users do not need to build their own computer rooms, and can use server resources as conveniently as using water and electricity; it can provide diversified computing capabilities and support multiple processor architectures such as x86 and Arm; it provides multiple server types such as CPU, GPU, elastic bare metal and supercomputing clusters, and has hundreds of instance specification families to meet the needs of users of different sizes and types. The patent of this invention mainly uses its computing power to analyze and calculate the data collected by the equipment, runs the fault location mathematical model based on the single-ended impedance method set for it, and uses its data flow function to send the equipment collected data and fault location results to the WeChat applet.

[0093] Example 2

[0094] In this embodiment, a method and system for locating and identifying a fault in a medium-voltage distribution network are the same as those in Example 1, with a method for locating and identifying a fault in a medium-voltage distribution network being added thereto.

[0095] Distribution lines are exposed to the elements year-round and are affected by factors such as tree branches, lightning strikes, birds and animals. Compared to other parts of the power system, they are more prone to failure. To prevent the expansion of faults and ensure the stable operation of the power system, it is crucial to quickly repair distribution line faults. According to relevant statistics, single-phase grounding faults account for as much as 90% of all distribution line faults. Currently, the main methods for fault location are impedance and traveling wave methods. These methods can be divided into single-ended and two-terminal methods based on how the fault information is obtained. The single-ended impedance method uses the voltage and current values recorded at the time of the line fault to determine the location of the fault point through analysis and calculation. This method is simple to operate, low-cost, and easy to implement, and has therefore been widely used in the field of fault location.

[0096] In the fault location measurement of distribution lines, the prerequisite for accurate distance measurement is to clearly identify the fault type of the line. This is because different fault types correspond to different fault circuits. Taking a three-phase distribution line with a grounded neutral point as an example, its fault characteristics will vary depending on the fault type; see Figure 13 The impedance diagram of a three-phase distribution line with a grounded neutral point is given. By analyzing and understanding this impedance diagram, the line condition when a single-phase grounding fault occurs is simulated, thus providing a more accurate basis for fault location. Figure 13 .

[0097] A method for fault location and identification in medium voltage distribution network, see Figure 21 , the detection unit is used to collect the voltage and current values of the target distribution line. The fault location mathematical model obtains the location of the medium voltage information fault point based on the voltage and current values recorded when the target distribution line fault occurs. Specifically, it includes:

[0098] S1. The fault location mathematical model is based on the single-ended impedance method to locate the target distribution line fault;

[0099] The fault location mathematical model is based on the single-ended impedance method, and the calculation formula is as follows:

[0100] a-phase voltage U A The calculation formula is as follows:

[0101] U A =Z s I a +Z m (I b +I c ) (1)

[0102] In formula (1), Z S is the self-impedance of the conductor, Z M is the mutual impedance of the conductor, I a is the a-phase current, I b is the b-phase current, I cis the c-phase current;

[0103] I r The sum of the three-phase currents a, b, and c is I r The calculation formula is as follows:

[0104] I r =I a +I b +I c (2)

[0105] The calculation formulas for the self-impedance and mutual impedance of the conductor using sequence components are as follows:

[0106]

[0107]

[0108] In formulas (3) and (4), Z1 is the positive sequence impedance; Z0 is the zero sequence impedance;

[0109] Substituting equations (2) to (4) into equation (1), we can obtain:

[0110] U a =Z1(I a +kI r ) (5)

[0111]

[0112] In formulas (5) and (6), k is a coefficient, which is substituted into formulas 7 and 8 to reduce the influence of the uncertain ground resistance on the results;

[0113] See Figure 14 , m is the percentage of the fault position to the total line length. When a unidirectional ground fault occurs on phase a, the current will flow from the power supply through the fault point and then return to the power supply through the ground. The calculation formula for the phase a voltage is as follows:

[0114] U a =mZ L (I a +kI r )+I F R F (7)

[0115] Assuming that the transition resistance is purely resistive, the measured current I is in phase with the short-circuit current at the fault point. Multiply both sides of formula (7) by I* and take only the imaginary part. By measuring the ratio of the reactance to the line reactance, the position of the fault point can be obtained. The calculation formula is as follows:

[0116] When a unidirectional ground fault occurs on phase a, the percentage of the fault position to the total line length is as follows:

[0117]

[0118] In formulas (7) and (8), Z L is the total line impedance, U a is the phase voltage of phase a; I a is the phase current of phase a; Z L is the total line impedance; I F is the fault current; R F is the transition resistance; For I r The complex conjugate of mg After performing conjugate multiplication on a complex expression, extract the imaginary part of the result. For example, It means multiplying the complex voltage Ua by the conjugate of the current Ir and taking the imaginary part.

[0119] Furthermore, deep learning-based state recognition and problem classification models for distribution network fault types are crucial for ensuring the safe and stable operation of distribution network systems. State recognition primarily involves accurately determining the current operating state of the distribution network system, including the identification of normal and various abnormal states. This requires deep learning models to process and analyze real-time distribution network data to extract characteristic information that reflects the system status. During the training process, deep learning models utilize extensive historical data to enable the model to learn the data characteristics of various states, enabling it to accurately identify the current system status during real-time operation.

[0120] S2. Use convolutional neural networks in deep learning to identify fault types;

[0121] Convolutional neural network includes convolution layer, activation function layer, pooling layer, fully connected layer, and softmax classification layer. Figure 15 ,The convolutional neural network uses the convolution layer and the pooling layer to extract the ,core features of the target distribution line state and perform fault ,type identification;

[0122] Taking current waveform recognition as an example, the current waveform can be used directly as input data. It is processed through multiple convolutional and pooling layers to capture the spatial features in the image. The data then passes through a fully connected layer and is classified using the Softmax function to accurately identify the distribution network status, including normal operation and fault conditions.

[0123] The core features of the target distribution line state are extracted by CNN through a large amount of historical data, and each state is assigned a relevant weight w and bias b. The core features of the target distribution line state include six states, as follows:

[0124] 1. Operating status: At this time, the voltage and current fluctuate normally;

[0125] 2. Single-phase grounding fault: A phase conductor is in direct contact with the ground (or neutral line), resulting in an increase in grounding current;

[0126] 3. Phase-to-phase short circuit fault: including two-phase short circuit and three-phase short circuit, in which the short circuit current will be much greater than the normal load current;

[0127] 4. Ground fault: At this time, the current of the grounded phase will increase and the voltage will decrease, while the voltage and current of the ungrounded phase will remain unchanged;

[0128] 5. Line break fault: The line break phase current is 0;

[0129] 6. Arc fault: accompanied by unstable arc discharge, the current amplitude becomes low.

[0130] Fault type identification uses the current waveform as the initial input. Features are gradually extracted through the convolution layer, activation function layer, and pooling layer to capture spatial features in the image. Finally, the softmax classification layer is used to effectively identify the target distribution line status. The formula for the problem classification model is as follows:

[0131]

[0132] In formula (9), x is the input current waveform, W j and b j They correspond to the relevant weights and bias values of the jth state, K is the total number of categories of different states, W k In the equation, k is the total number of categories for judging the distribution network status, k is the relevant weight corresponding to the kth state, and W j is a specific item of prediction, and W k is a state in each state; b k Where k is the total number of categories for judging the distribution network status, and k is the bias corresponding to the kth state; b k is the bias corresponding to the kth state, and P(y=j|x) is the probability that the model predicts that the input x belongs to category j;

[0133] In the target distribution line fault detection, the problem classification model (Equation 9) is used to classify the input current waveform data into a fault state or a non-fault state:

[0134] If the tag is set to 1, it means that the current target distribution line is in a fault state;

[0135] If the tag is set to 0, it means that the current target distribution line is in a non-fault state;

[0136] Among them, the label is the predicted distribution network state corresponding to the above input if it belongs to an abnormal operating state;

[0137] Each sample data is evaluated using cross entropy loss, and the calculation formula is as follows:

[0138] Loss=-(y*log(py)+(1-y)*log(1-p)) (10)

[0139] In formula (10), y is the true label of the sample, which has a value of 0 or 1; p is the predicted probability of the sample by the problem classification model (formula 9), which is used to narrow the gap between the predicted probability of the problem classification model and the true label;

[0140] The sample data is the input and related weights or biases in each prediction. A sample is a prediction of an input.

[0141] The cross entropy loss function is used to measure the gap between the predicted probability and the true label. Back propagation is used to continuously correct the relevant weights W and b corresponding to a certain state. The results are passed back from the output layer by layer to continuously adjust the relevant weights W and bias b. Among them, the gap between the true labels is the predicted state.

[0142] The cross entropy loss function is used to measure the gap between the predicted probability and the true label and pass it back layer by layer from the output of the neural network. The internal parameters of the problem classification model are adjusted according to the gap between the predicted probability and the true label, and the cross entropy loss function value is continuously reduced. The internal parameters of the problem classification model include the relevant weight W and bias b.

[0143] A state recognition and problem classification model for distribution network fault types based on deep learning is used to collect a large amount of distribution network system operation data, which covers key parameters such as current and voltage under various fault conditions. This data is used to train the deep learning model so that it can accurately identify the different states of the distribution network system. After the model training is completed, a series of experimental verifications are carried out to simulate different fault scenarios to observe the actual performance of the model in fault detection, location, etc. The experimental results show that the state recognition and problem classification model for distribution network fault types based on deep learning can significantly improve the accuracy and speed of fault detection. This algorithm has shown strong application potential and practical value. Figures 18-20 .

[0144] Deep learning models have demonstrated remarkable effectiveness in identifying power system states and classifying problems. They can not only accurately detect various fault conditions but also precisely locate the fault location, which is extremely important for the stable operation of distribution networks and the rapid resolution of faults.

[0145] Example 3

[0146] In this embodiment, a method and system for locating and identifying faults in a medium-voltage distribution network are the same as those in Example 1, with a monitoring module added on the basis of Example 1 and / or Example 2.

[0147] The detection unit also includes a monitoring module, which includes a display screen and a WeChat applet. The monitoring module receives the current, voltage, and fault status information from the WIFI communication module for real-time online monitoring. The device side provides a display screen as a human-computer interaction interface. The human-computer interaction interface can not only display the grid operation status online, but also the frequency of sending collected information can be adjusted according to customer customization, and customized time windows are used for data transmission. It also provides a voltage transformer and current transformer ratio input interface, which can convert the secondary value measured by the system into a primary value, so that operation and maintenance personnel can observe the grid operation status in real time; in addition, the human-computer interaction interface also provides an overcurrent alarm threshold setting function. Operation and maintenance personnel can input an appropriate overcurrent alarm threshold according to on-site requirements. When the current data detected by the equipment is greater than the threshold, an overcurrent alarm will be triggered. The overcurrent alarm signal will be displayed in the human-computer interaction interface. The signal will also be sent to the WeChat applet by the single-chip microcomputer, so that the equipment can achieve dual-end monitoring to ensure the normal operation of the distribution network. A monitoring WeChat applet is designed. The WeChat applet is divided into a main interface and a device interface. The main interface displays the added devices and the device operation status. See Figure 16 , Figure 17 , click on the added device in the main interface to enter the device details page, which displays the data collected by the device, overcurrent alarm and other information.

[0148] The present invention collects the voltage and current values of the target distribution line through a detection unit, and transmits the collected data information to the cloud server. The cloud server moves the collected data into a set mathematical model based on the single-ended impedance method. The cloud server determines the fault location through the single-ended impedance method mathematical model, and sends the final determined fault location as well as the current, voltage, fault current data and overcurrent alarm information to the WeChat applet for remote viewing by operation and maintenance personnel, thereby reducing the work pressure of operation and maintenance personnel and reducing operating costs; the state recognition and problem classification model of distribution network fault types based on deep learning is a key link to ensure the safe and stable operation of the distribution network system. State recognition mainly involves accurate judgment of the current operating status of the distribution network system, and can accurately and quickly determine the fault type. Combined with the fault location model based on the single-ended impedance method, it can determine the fault type and predict the fault location in a short time, assisting operation and maintenance personnel to deal with faults in a timely manner, greatly improving the fault handling efficiency of the distribution network.

Claims

1. A method for locating and identifying faults in a medium voltage distribution network, characterized in that: The detection unit collects the voltage and current values of the target distribution line. The fault location mathematical model obtains the location of the medium voltage fault point based on the voltage and current values recorded when the target distribution line fault occurs. Specifically, it includes: S1. The fault location mathematical model is based on the single-ended impedance method to locate the target distribution line fault; S2. Use convolutional neural networks in deep learning to identify fault types.

2. A method for locating and identifying faults in a medium voltage distribution network according to claim 1, characterized in that: In S1, the fault location mathematical model is based on the single-ended impedance method, and the calculation formula is as follows: a-phase voltage U A The calculation formula is as follows: U A =Z s I a +Z m (I b +I c ) (1) In formula (1), Z S is the self-impedance of the conductor, Z M is the mutual impedance of the conductor, I a is the a-phase current, I b is the b-phase current, I c is the c-phase current; I r The sum of the three-phase currents a, b, and c is I r The calculation formula is as follows: I r =I a +I b +I c (2) The calculation formulas for the self-impedance and mutual impedance of the conductor using sequence components are as follows: In formulas (3) and (4), Z1 is the positive sequence impedance; Z0 is the zero sequence impedance; Substituting equations (2) to (4) into equation (1), we can obtain: U a =Z1(I a +kI r ) (5) In formulas (5) and (6), k is a coefficient; The calculation formula of phase A voltage when a one-way ground fault occurs on phase A is as follows: U a =mZ L (I a +kI r )+I F R F (7) Assuming that the transition resistance is purely resistive, the measured current I is in phase with the short-circuit current at the fault point. Multiply both sides of formula (7) by I* and take only the imaginary part. By measuring the ratio of the reactance to the line reactance, the position of the fault point can be obtained. The calculation formula is as follows: When a unidirectional ground fault occurs on phase a, the percentage of the fault position to the total line length is as follows: In formulas (7) and (8), Z L is the total line impedance, U a is the phase voltage of phase a; I a is the phase current of phase a; Z L is the total line impedance; I F is the fault current; R F is the transition resistance; For I r The conjugate complex number of , m is the percentage of the fault position to the total length of the line; I mg Performs conjugate multiplication on a complex expression and extracts the imaginary part of the result.

3. A method for locating and identifying faults in a medium voltage distribution network according to claim 1, characterized in that: In S2, the convolutional neural network includes a convolutional layer, an activation function layer, a pooling layer, a fully connected layer, and a softmax classification layer. The convolutional neural network uses the convolutional layer and the pooling layer to extract the core features of the target distribution line state and identify the fault type.

4. A method for locating and identifying faults in a medium voltage distribution network according to claim 3, characterized in that: The fault type identification described above uses the current waveform as the initial input, and gradually extracts features through the convolution layer, activation function layer, and pooling layer to capture the spatial features in the image. Finally, the softmax classification layer is used to effectively identify the target distribution line status. The formula of the problem classification model is as follows: In formula (9), x is the input current waveform, W j and b j They correspond to the relevant weights and bias values of the jth state, K is the total number of categories of different states, W k For each state, b k is the bias corresponding to the kth state, and P(y=j|x) is the probability that the model predicts that the input x belongs to category j; In the target distribution line fault detection, the problem classification model is used to classify the input current waveform data into a fault state or a non-fault state: If the tag is set to 1, it means that the current target distribution line is in a fault state; If the tag is set to 0, it means that the current target distribution line is in a non-fault state; Each sample data is evaluated using cross entropy loss, and the calculation formula is as follows: Loss=-(y*log(py)+(1-y)*log(1-p)) (10) In formula (10), y is the true label of the sample, which has a value of 0 or 1; p is the predicted probability of the sample by the question classification model, which is used to narrow the gap between the predicted probability of the question classification model and the true label.

5. A method for locating and identifying faults in a medium voltage distribution network according to claim 4, characterized in that: The cross entropy loss function is used to measure the gap between the predicted probability and the true label. Back propagation is used to continuously correct the relevant weights W and b corresponding to a certain state, and the results are passed back from the output layer by layer to continuously adjust the relevant weights W and bias b. The cross entropy loss function is used to measure the gap between the predicted probability and the true label. The cross entropy loss function is passed back layer by layer from the output of the neural network. The internal parameters of the problem classification model are adjusted according to the gap between the predicted probability and the true label, and the cross entropy loss function value is continuously reduced.

6. A method for locating and identifying faults in a medium voltage distribution network according to claim 5, characterized in that: The internal parameters of the question classification model include the relevant weight W and bias b.

7. A medium voltage distribution network fault location and identification system for implementing the method according to any one of claims 1 to 6, characterized in that: It includes a detection unit, which includes a CPU controller, a data acquisition module, a data storage module, and a communication module. The output end of the data acquisition module is connected to the CPU controller, the data storage module is connected to the CPU controller, and the CPU controller is connected to the cloud server through the communication module.

8. A medium voltage distribution network fault location and identification system according to claim 7, characterized in that: The detection unit also includes a power supply module, which is used to provide working power to the CPU controller, data acquisition module, data storage module, and communication module.

9. A medium voltage distribution network fault location and identification system according to claim 7, characterized in that: The data acquisition module includes a current data acquisition module and a voltage data acquisition module. The input end of the current data acquisition module is connected to the current transformer of the target distribution line to collect the current value of the target distribution line. The input end of the voltage data acquisition module is connected to the voltage transformer of the target distribution line to collect the voltage value of the target distribution line. The output end of the current data acquisition module is connected to the CPU controller, and the output end of the voltage data acquisition module is connected to the CPU controller.

Images