A control method and system of a turbine air-cooled fan frequency converter fault automatic switching soft starter

Abstract

The application relates to a control method and system for automatic switching of a soft starter of a turbine air cooling fan frequency converter. The method comprises the following steps: obtaining a fault code and real-time operation data, and obtaining a fault level according to the fault code and the real-time operation data; when the fault level meets preset switching requirements, generating a frequency converter stop working instruction; extracting residual voltage change characteristics of the motor based on three-phase voltage instantaneous data of the motor, and obtaining a real-time residual voltage amplitude; when the real-time residual voltage amplitude is smaller than a preset safety threshold, recording a sampling time stamp corresponding to the real-time residual voltage amplitude as a condition meeting time stamp, obtaining a zero time stamp based on three-phase voltage instantaneous data of a power grid side, and calculating an instruction sending time stamp based on the condition meeting time stamp and the zero time stamp set; and generating a soft starter starting instruction at the instruction sending time stamp. The method can improve the switching success rate of the frequency converter and the soft starter.

Description

Technical Field

[0001] This invention belongs to the field of frequency converter control for wind turbines, and in particular relates to a control method and system for an automatic switching soft starter for a frequency converter fault in a turbine air-cooled fan. Background Technology

[0002] With the development of frequency conversion control technology for turbine air-cooled fans, frequency converters are widely used in fan speed regulation to achieve precise speed regulation and energy-saving operation. They can smoothly adjust the cooling air volume according to process requirements and have become the core control equipment to ensure the stable operation of large compressor units.

[0003] In traditional methods, when a frequency converter malfunctions, maintenance personnel usually manually determine the type of fault. If it is determined to be a serious fault, they manually disconnect the power supply to the frequency converter and close the power frequency bypass circuit to switch the fan to power frequency constant speed operation. Alternatively, a pre-configured simple bypass contactor can be directly closed after the frequency converter is disconnected to maintain the continuous rotation of the fan.

[0004] However, this approach has the following problems: First, it cannot automatically distinguish the severity of the fault, and even minor faults can trigger switching, causing unnecessary downtime. Second, the lack of synchronous control of the motor residual voltage and the grid phase at the switching moment results in an impact of several times the rated current at the moment of closing, which can easily damage the thyristor and cause grid fluctuations. In severe cases, it can exacerbate the risk of equipment damage, resulting in a low switching success rate and insufficient production continuity assurance capability. Summary of the Invention

[0005] Therefore, it is necessary to provide a control method and system for automatically switching the soft starter when the frequency converter fails in a turbine air-cooled fan, which can improve the success rate of switching between the frequency converter and the soft starter, in response to the above-mentioned technical problems.

[0006] In a first aspect, this application provides a control method for an automatic switching soft starter for a turbine air-cooled fan inverter fault, comprising:

[0007] Obtain the fault codes and real-time operating data of the frequency converter within the current control cycle, and determine the fault level based on the fault codes and real-time operating data to obtain the fault level of the current control cycle;

[0008] When the fault level meets the preset switching requirements, a frequency converter stop operation command is generated; the frequency converter stop operation command is used to instruct the frequency converter to stop output and acquire instantaneous data of the motor three-phase voltage and the grid side three-phase voltage in real time.

[0009] Based on the instantaneous data of the three-phase voltage of the motor, the residual voltage change characteristics of the motor are extracted to obtain the real-time residual voltage amplitude;

[0010] When the real-time residual voltage amplitude is less than the preset safety threshold, the sampling timestamp corresponding to the real-time residual voltage amplitude is recorded as the condition compliance timestamp. Based on the instantaneous data of the three-phase voltage on the grid side, the zero point time of the grid voltage is predicted to obtain the zero point timestamp set. The command issuance timestamp is calculated based on the condition compliance timestamp and the zero point timestamp set.

[0011] At the timestamp of the instruction issuance, a soft starter startup instruction is generated; the soft starter startup instruction is used to instruct the soft starter to start working.

[0012] Furthermore, the real-time operating data includes DC bus voltage, IGBT junction temperature, and output frequency. All three values ​​are accompanied by a sampling timestamp. Based on the fault codes and real-time operating data, a fault classification is determined to obtain the fault level for the current control cycle, including:

[0013] The average value of the DC bus voltage corresponding to each sampling timestamp is calculated to obtain the average DC bus voltage. Based on the normal DC bus voltage value and the average DC bus voltage, the DC bus voltage deviation rate is calculated.

[0014] Extract the maximum value of the IGBT junction temperature corresponding to the starting sampling timestamp within the current control cycle, and use it as the maximum IGBT junction temperature. Calculate the difference between the maximum IGBT junction temperature and the rated upper limit of the IGBT junction temperature to obtain the IGBT junction temperature difference.

[0015] Extract the output frequency value corresponding to the first sampling timestamp in the current control cycle as the starting output frequency value, and extract the output frequency value corresponding to the last sampling timestamp in the current control cycle as the ending output frequency value.

[0016] Based on the starting output frequency value, the ending output frequency value, and the control cycle duration parameter, the output frequency change rate is calculated, and a periodic change feature vector is constructed based on the DC bus voltage deviation rate, IGBT junction temperature difference value, and output frequency change rate.

[0017] Identify the intersection of fault codes and minor fault code sets to obtain the minor fault intersection result; identify the intersection of fault codes and major fault code sets to obtain the major fault intersection result.

[0018] Based on the fault code mapping rule set, fault level analysis is performed on the intersection results of minor faults, the intersection results of severe faults, and the periodic change feature vector to obtain the fault level.

[0019] Furthermore, based on the fault code mapping rule set, fault level analysis is performed on the intersection results of minor faults, the intersection results of severe faults, and the periodic change feature vectors to obtain the fault level, including:

[0020] When the intersection result of a minor fault is a fault code or the intersection result of a major fault is a fault code, the intersection result of a minor fault, the intersection result of a major fault, and the periodic change feature vector are mapped based on the fault code mapping rule set to obtain the fault level;

[0021] Alternatively, when the intersection results of minor faults and major faults are both empty sets, the difference between the periodic change feature vector and the normal operating condition mean vector is calculated to obtain the feature deviation degree.

[0022] The characteristic deviation degree is calculated and the characteristic deviation degree threshold difference is set to a preset deviation threshold. Based on the deviation level mapping rule set, the fault level is obtained by mapping the characteristic deviation degree threshold difference.

[0023] Furthermore, based on the instantaneous data of the motor's three-phase voltage, the residual voltage change characteristics of the motor are extracted to obtain the real-time residual voltage amplitude, including:

[0024] The instantaneous three-phase voltage data of the motor is low-pass filtered to obtain filtered three-phase voltage data;

[0025] The three-phase voltage data of the filter is transformed into the first and second axis components in the two-phase stationary coordinate system.

[0026] The first axis component is decomposed using a second-order generalized integrator to obtain the first in-phase component and the first quadrature component. The second axis component is also decomposed using a second-order generalized integrator to obtain the second in-phase component and the second quadrature component.

[0027] Based on the first in-phase component, the first quadrature component, the second in-phase component, and the second quadrature component, the first positive sequence axis component and the second positive sequence axis component are calculated. The expressions for the first positive sequence axis component and the second positive sequence axis component are as follows:

[0028] ;

[0029] ;

[0030] In the formula, It is any sampling timestamp. It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The first in-phase component, It is a sampling timestamp The second orthogonal component, It is a sampling timestamp The second positive sequence axis component, It is a sampling timestamp The first orthogonal component, It is a sampling timestamp The second in-phase component;

[0031] The real-time residual voltage amplitude is calculated based on the first and second positive sequence axis components. The expression for the real-time residual voltage amplitude is as follows:

[0032] ;

[0033] In the formula, It is any sampling timestamp. It is a sampling timestamp The real-time residual voltage amplitude, It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The second positive sequence axis component.

[0034] Furthermore, the instantaneous three-phase voltage data on the grid side includes sampling timestamps. Based on the instantaneous three-phase voltage data on the grid side, the zero-point time of the grid voltage is predicted, resulting in a zero-point timestamp set, including:

[0035] Using the time stamp when the conditions are met as the end time stamp, the instantaneous data of the three-phase voltage on the grid side corresponding to a preset fixed number of sampling time stamps are extracted to obtain the predicted reference voltage sequence.

[0036] The predicted reference voltage sequence is input into the zero-point prediction model to obtain the reference zero-point timestamp;

[0037] The reference zero-point timestamp is added sequentially to an integer multiple of a preset half-cycle parameter to obtain at least one adjusted zero-point timestamp, and a zero-point timestamp set is formed based on the reference zero-point timestamp and the adjusted zero-point timestamp.

[0038] The zero-point prediction model includes:

[0039] The input layer receives the predicted reference voltage sequence, normalizes the predicted reference voltage sequence to obtain the normalized voltage sequence, and transmits the normalized voltage sequence to the convolutional feature extraction layer.

[0040] The convolutional feature extraction layer is used to perform a one-dimensional convolution operation on the normalized voltage sequence, extract local waveform features, obtain a local feature map, and then transmit the local feature map to the pooling layer;

[0041] Pooling layers are used to downsample local feature maps to obtain dimensionality-reduced feature vectors, and then transmit the dimensionality-reduced feature vectors to fully connected layers;

[0042] The fully connected layer is used to perform linear transformations and nonlinear activations on the dimensionality-reduced feature vectors to obtain the baseline zero-point timestamp.

[0043] Furthermore, the zero-time prediction model is obtained through the following steps:

[0044] Acquire voltage waveform data under different power grid operating conditions; and perform sliding window sampling on the voltage waveform data to obtain a complete voltage sequence;

[0045] The complete voltage sequence is truncated to obtain a sample voltage sequence, and the time interval from the last sampling timestamp in the sample voltage sequence to the next zero-crossing point of the grid voltage is calculated. The model training dataset is constructed by using the sample voltage sequence as the input feature vector and the time interval corresponding to the sample voltage sequence as the prediction target value corresponding to the input feature vector.

[0046] The input layer, convolutional layer, pooling layer, fully connected layer and output layer are connected in sequence to obtain the initial prediction model architecture, and the initial prediction model architecture is initialized to obtain the initial zero-time model.

[0047] Input the preset number of input feature vectors into the model at the initial zero time point to obtain the prediction time interval value corresponding to the input feature vector;

[0048] Based on the input feature vector, the corresponding prediction target value, and the prediction time interval, the model loss value is calculated. The expression for the model loss value is:

[0049] ;

[0050] In the formula, It is a preset number of training sessions. It is the index of any input feature vector. It is the first The prediction time interval value corresponding to each input feature vector. It is the first The predicted target value corresponding to each input feature vector. It is the model loss value;

[0051] With the goal of minimizing the model loss, the initial zero-time model is iteratively updated until the preset iteration termination condition is met, thus obtaining the trained zero-time prediction model.

[0052] Furthermore, the instruction issuance timestamp is calculated based on the condition fulfillment timestamp and the zero-point timestamp set, including:

[0053] Calculate the difference between the baseline zero-point timestamp and the condition-meeting timestamp to obtain the baseline timestamp difference; calculate the difference between the adjustment zero-point timestamp and the condition-meeting timestamp to obtain the time difference of each adjustment zero-point timestamp.

[0054] When the difference between the reference timestamps is greater than the preset instruction start delay time parameter, the reference zero-point timestamp and the instruction start delay time parameter are subtracted to obtain the instruction issuance timestamp; or, when the difference between the reference timestamps is less than or equal to the instruction start delay time parameter, the time difference is compared with the instruction start delay time parameter to obtain the difference comparison result of each adjustment zero-point timestamp.

[0055] The earliest adjustment zero-point timestamp whose comparison result is greater than the instruction start delay time parameter is selected as the instruction reference zero-point timestamp. The instruction reference zero-point timestamp and the instruction start delay time parameter are subtracted to obtain the instruction issuance timestamp.

[0056] Secondly, this application also provides a control system for an automatic fault-switching soft starter for turbine air-cooled fan inverters, comprising:

[0057] The data acquisition module is used to acquire the fault codes and real-time operating data of the frequency converter within the current control cycle, and to perform fault classification judgment based on the fault codes and real-time operating data to obtain the fault level of the current control cycle.

[0058] The inverter stop control module is used to generate an inverter stop command when the fault level meets the preset switching requirements. The inverter stop command is used to instruct the inverter to stop output and to acquire instantaneous three-phase voltage data of the motor and instantaneous three-phase voltage data of the grid side in real time.

[0059] The motor residual voltage calculation module is used to extract the residual voltage change characteristics of the motor based on the instantaneous data of the motor's three-phase voltage, and obtain the real-time residual voltage amplitude.

[0060] The instruction issuance time calculation module is used to record the sampling timestamp corresponding to the real-time residual voltage amplitude as the condition compliance timestamp when the real-time residual voltage amplitude is less than the preset safety threshold. Based on the instantaneous data of the three-phase voltage on the grid side, the zero point time of the grid voltage is predicted to obtain the zero point timestamp set. The instruction issuance timestamp is calculated based on the condition compliance timestamp and the zero point timestamp set.

[0061] The soft-start command issuing module is used to generate a soft-starter startup command at the command issuance timestamp; the soft-starter startup command is used to instruct the soft-starter to start working.

[0062] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the control method of any of the turbine air-cooled fan inverter automatic fault switching soft starters of the first aspect of this application.

[0063] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements any of the control methods for automatic switching soft starters for turbine air-cooled fan inverters according to the first aspect of this application.

[0064] The aforementioned control method and system for an automatic switching soft starter for a turbine air-cooled fan inverter fault involves acquiring the inverter's fault code and real-time operating data within the current control cycle, and determining the fault level for the current control cycle based on the fault code and real-time operating data. When the fault level meets the preset switching requirements, an inverter stop command is generated. This inverter stop command instructs the inverter to stop outputting and acquires real-time instantaneous data of the motor's three-phase voltage and the grid's three-phase voltage. Based on the instantaneous data of the motor's three-phase voltage, the residual voltage change characteristics of the motor are extracted to obtain the real-time residual voltage amplitude. When the real-time residual voltage amplitude is less than a preset safety threshold, the sampling timestamp corresponding to the real-time residual voltage amplitude is recorded as the condition compliance timestamp. The zero-point time of the grid voltage is predicted based on the instantaneous data of the grid's three-phase voltage, resulting in a zero-point timestamp set. The command issuance timestamp is calculated based on the condition compliance timestamp and the zero-point timestamp set. At the command issuance timestamp, a soft starter start command is generated. This soft starter start command instructs the soft starter to start operating. This improved the electrical safety and success rate of the switching process, ensuring the continuous and stable operation of the turbine air-cooled fan under various fault conditions. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 A flowchart illustrating a control method for an automatic switching soft starter in case of a fault in a turbine air-cooled fan inverter, provided as an embodiment of this application;

[0067] Figure 2 A schematic diagram of the control system of an automatic switching soft starter for a turbine air-cooled fan inverter fault is provided in one embodiment of this application;

[0068] Figure 3 This is a schematic diagram of the computer device used in an embodiment of the present application to provide a control method for an automatic switching soft starter for a turbine air-cooled fan inverter fault. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0070] In one embodiment, such as Figure 1 As shown, a control method for an automatic switching soft starter for a turbine air-cooled fan inverter fault is provided. This embodiment illustrates the method by applying it to a switching terminal. It is understood that this method can also be applied to a server, and to a system including both a switching terminal and a server, and is implemented through the interaction between the switching terminal and the server. In this embodiment, the method includes the following steps S101-S105, wherein:

[0071] S101: Obtain the fault codes and real-time operating data of the frequency converter within the current control cycle, and determine the fault level based on the fault codes and real-time operating data to obtain the fault level of the current control cycle.

[0072] Specifically, the current control cycle refers to a fixed time window during which the switching terminal cyclically executes fault detection tasks. Its duration can be set according to the data acquisition frequency and equipment protection requirements in actual operation, for example, 10ms. The fault code is an integer value read from the frequency converter, which can be denoted as... ,in Each integer value corresponds to a fault type within the frequency converter, such as overcurrent, overvoltage, or overheating. The mapping relationship between fault code values ​​and fault types can be set according to the frequency converter's specifications and model. Real-time operating data can be used to characterize the frequency converter's operating status; real-time operating data can include DC bus voltage values, IGBT junction temperature values, and output frequency values, etc. Parameters included in the real-time operating data can have sampling timestamps, which can be used to characterize the acquisition time of the corresponding data. The switching terminal obtains the frequency converter's fault codes and real-time operating data within the current control cycle, and comprehensively determines the severity level of the fault within the current cycle based on the fault codes and real-time operating data to obtain the fault level for the current control cycle. The fault level can be an enumerated value, where a fault level value of 0 represents a minor fault in the current cycle, and a value of 1 represents a severe fault in the current cycle.

[0073] S102, when the fault level meets the preset switching requirements, a frequency converter stop operation command is generated; the frequency converter stop operation command is used to instruct the frequency converter to stop output and acquire the instantaneous data of the motor three-phase voltage and the instantaneous data of the grid three-phase voltage in real time.

[0074] Specifically, the preset switching requirement refers to the current cycle's fault level being a severe fault. When the fault level is an enumerated value, the preset switching requirement can be set to fault level 1. When the fault level of the current fault cycle meets the preset switching requirement, the switching terminal can send an inverter stop command to the inverter. The inverter stop command can be used to instruct the inverter's internal control logic to trigger pulse blocking, causing the inverter to immediately stop outputting PWM modulated voltage to the motor, thereby cutting off the drive to the motor. Simultaneously, the inverter stop command can also be used to instruct the switching terminal to acquire real-time instantaneous data of the motor's three-phase voltage and the grid's three-phase voltage. The instantaneous data of the motor's three-phase voltage refers to the instantaneous value of the three-phase voltage at the inverter output terminal, i.e., the motor cable connection terminal, at a discrete sampling timestamp. Its mathematical form can be... The instantaneous three-phase voltage data on the grid side refers to the instantaneous three-phase voltage values ​​at the soft starter's input terminal, i.e., the power frequency power supply side. The mathematical form can be... , For the first Each sampling timestamp This is the first sampling timestamp of the current control cycle, and it can be recorded. The latest sampling timestamp, , and These are the motor's sampling timestamps. The instantaneous values ​​of phase A voltage, phase B voltage, and phase C voltage represent the values ​​at the sampling timestamps, respectively. The instantaneous voltage values ​​of phases A, B, and C to the neutral point, in volts; similarly, the instantaneous three-phase voltage data on the grid side... , , These are the sampling timestamps of the power grid. The instantaneous voltage values ​​of phases A, B, and C on the grid side are recorded. Phases A, B, and C are spatially separated by 120° electrical degrees, forming the original three-phase stationary coordinate system, i.e., the abc coordinate system. The sampling period for both the instantaneous three-phase voltage data from the motor and the grid side can be the standard sampling interval. For example, 0.5ms.

[0075] S103 extracts the residual voltage change characteristics of the motor based on the instantaneous data of the three-phase voltage of the motor, and obtains the real-time residual voltage amplitude.

[0076] Specifically, the switching terminal can extract the residual voltage change of the motor based on the instantaneous data of the motor's three-phase voltage to obtain the real-time residual voltage amplitude. Residual voltage refers to the back electromotive force generated in the stator windings due to residual magnetism induction while the motor rotor continues to rotate due to inertia after the inverter stops outputting. This residual voltage still exists at the motor cable end after the inverter stops outputting, and its amplitude decays exponentially with time. (Any sampling timestamp) Real-time residual voltage amplitude Indicates the current sampling timestamp The measured instantaneous effective value of the residual voltage is expressed in volts (V). The extraction process may include low-pass filtering, Clarke transform, positive sequence component extraction, and amplitude and phase calculation of the instantaneous three-phase voltage data of the motor.

[0077] S104. When the real-time residual voltage amplitude is less than the preset safety threshold, the sampling timestamp corresponding to the real-time residual voltage amplitude is recorded as the condition compliance timestamp. The zero point time of the grid voltage is predicted based on the instantaneous data of the three-phase voltage on the grid side, and the zero point timestamp set is obtained. The instruction issuance timestamp is calculated based on the condition compliance timestamp and the zero point timestamp set.

[0078] Specifically, the preset safety threshold can be used to characterize whether the voltage value corresponding to the real-time residual voltage amplitude is safe. It can be set according to the standard of motor residual voltage during actual operation switching, and can be set to 20% of the motor's rated voltage. For example, for a motor with a rated voltage of 380V, the safety threshold can be set to 76V. (Latest timestamp) Real-time residual voltage amplitude When the current sampling timestamp is first lower than the preset safety threshold, the switching terminal will change the current sampling timestamp. Record as a timestamp of the conditions being met This indicates that the residual voltage has dropped to a level that allows for safe switching. The switching terminal then predicts the timestamp when the conditions are met based on the instantaneous three-phase voltage data from the grid side. The sampling timestamps corresponding to several grid voltage zero-crossing moments, i.e., the moments when the voltage changes from negative to positive or from positive to negative, form a zero-point timestamp set. Finally, based on the condition-meeting timestamps and the zero-point timestamp set, a specific command issuance timestamp is calculated, ensuring that after the command is issued at that moment, the contactor's main contacts close precisely near the grid voltage zero-crossing point, thereby suppressing inrush current.

[0079] S105, at the timestamp of the instruction issuance, generates a soft starter startup instruction; the soft starter startup instruction is used to instruct the soft starter to start working.

[0080] Specifically, at the command issuance timestamp, the switching terminal generates and sends a soft starter start command. The soft starter start command can be a digital output signal, typically a relay contact closure or a 24V DC level signal. The soft starter start command is output at the absolute moment corresponding to the calculated command issuance timestamp to control the closing of the AC contactor in the soft starter branch, connecting the soft starter's input terminal to the mains frequency power supply. It also sends a start command to the soft starter's control terminals, causing the soft starter to gradually conduct the thyristor according to a preset start curve (e.g., initial voltage 30%, ramp time 10s), smoothly accelerating the motor from its current residual speed to the mains frequency speed. After the start-up process is complete, the bypass contactor inside the soft starter automatically closes, short-circuiting the thyristor, and the motor continues to run at a constant speed at the mains frequency, thus completing a seamless switch from inverter failure to soft starter-driven operation.

[0081] This embodiment provides a control method for an automatic switching soft starter for turbine air-cooled fans in the event of inverter failure. First, it classifies and judges inverter fault codes and real-time operating data, triggering switching only in cases of severe faults to avoid unnecessary downtime caused by minor faults. Then, after the inverter stops outputting, it acquires real-time instantaneous three-phase voltage data of the motor and the grid, and monitors the residual voltage amplitude of the motor. Switching is only allowed when the real-time residual voltage amplitude is less than a preset safety threshold. Finally, it calculates the command issuance timestamp by predicting the grid voltage zero point and combining it with the condition compliance timestamp, ensuring the soft starter engages at the optimal time. This improves the electrical safety and success rate of the switching process, ensuring the continuous and stable operation of the turbine air-cooled fan under different fault conditions.

[0082] In one embodiment, the real-time operating data includes DC bus voltage, IGBT junction temperature, and output frequency. All three values ​​have a sampling timestamp. Fault classification is determined based on the fault code and the real-time operating data to obtain the fault level for the current control cycle, including:

[0083] S201, calculate the average value of the DC bus voltage corresponding to each sampling timestamp to obtain the average value of the DC bus voltage, and calculate the DC bus voltage deviation rate based on the normal DC bus voltage value and the average value of the DC bus voltage.

[0084] Specifically, within the current control cycle, real-time operating data can include DC bus voltage, IGBT junction temperature, and output frequency, all of which include a sampling timestamp. Each sampling timestamp... The DC bus voltage value can be denoted as: The normal DC bus voltage value is a preset reference value, which can be denoted as... This is the rated DC bus voltage of the frequency converter, which can be provided in the frequency converter manufacturer's manual or obtained through on-site calibration. The switching terminal first calculates the arithmetic mean of the DC bus voltage values ​​within the current control cycle: ,in It is the sampling timestamp at the start of the current control cycle. It is the total number of sampling timestamps included within a control cycle. It is the average DC bus voltage of the current control cycle. The unit is volts. Then, the switching terminal calculates the DC bus voltage deviation rate based on the average DC bus voltage and the normal DC bus voltage value: The DC bus voltage deviation rate reflects the relative degree to which the DC bus voltage deviates from the normal value within the current control cycle. The value range is generally between 0 and 1. The larger the value, the more severe the voltage fluctuation. It is one of the characteristics for judging whether a major fault has occurred.

[0085] S202, extract the maximum value of the IGBT junction temperature corresponding to the starting sampling timestamp within the current control cycle, and use it as the maximum IGBT junction temperature. Calculate the difference between the maximum IGBT junction temperature and the rated upper limit of the IGBT junction temperature to obtain the IGBT junction temperature difference.

[0086] Specifically, within the current control period, any sampling timestamp The junction temperature of an IGBT can be denoted as: The IGBT junction temperature can be obtained by measuring the IGBT module using a thermistor or temperature sensor inside the inverter, and is used to characterize the temperature status of the IGBT module. The rated upper limit of the IGBT junction temperature is a preset device safety threshold, which can be denoted as... This can be set according to the IGBT device datasheet, for example, 150°C. The switching terminal extracts the maximum value within the current control cycle. And calculate the difference between the two: The IGBT junction temperature difference value was obtained. For example, when This indicates that the junction temperature has exceeded the rated upper limit, constituting severe overheating; even if A negative or near-zero value also indicates insufficient heat dissipation. The IGBT junction temperature difference can be used to determine whether a major failure is caused by overheating.

[0087] S203, extract the output frequency value corresponding to the starting sampling timestamp in the current control cycle as the starting output frequency value, and extract the output frequency value corresponding to the last sampling timestamp in the current control cycle as the ending output frequency value.

[0088] Specifically, within the current control period, any sampling timestamp The output frequency value can be denoted as The output frequency value is the value recorded by the inverter at the sampling timestamp. The fundamental frequency output to the motor, in Hertz, can be obtained from the inverter's internal controller. The switching terminal extracts the sampling timestamp at the start of the current control cycle. Corresponding output frequency value The starting output frequency value for the current control cycle. It also extracts the output frequency value corresponding to the sampling timestamp at the end of the current control cycle. As the termination output frequency value .

[0089] S204 calculates the output frequency change rate based on the starting output frequency value, the ending output frequency value, and the control cycle duration parameter, and constructs a periodic change feature vector based on the DC bus voltage deviation rate, the IGBT junction temperature difference value, and the output frequency change rate.

[0090] Specifically, the switching terminal is based on the initial output frequency value. Termination output frequency value and control cycle duration parameters The output frequency change rate is calculated; the expression for the output frequency change rate is: , This is the control cycle duration parameter, i.e., the actual time length of one control cycle. The output frequency change rate is calculated from this, with the unit being Hz / s. Then, the switching terminal will adjust the DC bus voltage deviation rate. IGBT junction temperature difference Output frequency change rate By arranging them into a three-dimensional vector in a fixed order, a periodically changing characteristic vector can be obtained. .

[0091] S205 identifies the intersection of fault codes and minor fault code sets to obtain the minor fault intersection result, and identifies the intersection of fault codes and major fault code sets to obtain the major fault intersection result.

[0092] Specifically, the minor fault code set and the major fault code set can be two integer sets pre-constructed based on the inverter manufacturer's manual and field experience. The minor fault code set can be denoted as... The minor fault code set is used to characterize the integer values ​​of fault codes corresponding to self-recoverable minor faults, for example... Severe fault code sets can be denoted as The severe fault code set is used to characterize the integer values ​​of fault codes corresponding to severe faults that require downtime, for example... The intersection of the terminal identification fault code and the minor fault code set is used to obtain the minor fault intersection result. It also identifies the intersection of fault codes and severe fault code sets to obtain the severe fault intersection results. .

[0093] S206, based on the fault code mapping rule set, performs fault level analysis on the intersection results of minor faults, the intersection results of severe faults, and the periodic change feature vector to obtain the fault level.

[0094] Specifically, the fault code mapping rule set can be a predefined set of judgment logic rules used to indicate the intersection results of minor faults. Intersection results with severe faults If none of the results are empty sets, then the intersection results of minor faults are used. Intersection results of severe faults The fault level is obtained by mapping the values ​​of various indicators in the periodic variation feature vector. The switching terminal can then perform fault level analysis on the intersection results of minor and major faults and the periodic variation feature vector based on the fault code mapping rule set to obtain the fault level. The obtained fault level can be an enumerated value, where 0 represents a minor fault and 1 represents a major fault.

[0095] This embodiment provides a control method for an automatic fault switching soft starter for a turbine air-cooled fan inverter. First, in each control cycle, multiple points are sampled for the DC bus voltage, IGBT junction temperature, and output frequency, and the deviation rate, temperature difference, and frequency change rate are calculated to form a periodic feature vector. Simultaneously, real-time fault codes are matched with preset sets of light and heavy fault codes. Finally, based on a fault code mapping rule set, considering both code attribution and feature vector exceeding limits, the fault is classified as either a light or heavy fault. This method not only utilizes the classification information of the codes themselves but also incorporates dynamic monitoring of multi-dimensional electrical characteristics, effectively avoiding misjudgments based on a single threshold. It provides a reliable input basis for whether to trigger a switching operation, thereby reducing unnecessary downtime and improving system continuity.

[0096] In one embodiment, based on the fault code mapping rule set, fault level analysis is performed on the intersection results of minor faults, the intersection results of severe faults, and the periodic change feature vector to obtain the fault level, including:

[0097] S301, when the intersection result of a minor fault is a fault code or the intersection result of a major fault is a fault code, the intersection result of a minor fault, the intersection result of a major fault, and the periodic change feature vector are mapped based on the fault code mapping rule set to obtain the fault level.

[0098] Specifically, when the intersection result of a minor fault and a major fault is a fault code, it indicates that the current fault code belongs to the minor fault code set or the major fault code set, respectively. The switching terminal queries the fault code mapping rule set based on the minor fault intersection result, the major fault intersection result, and the periodic change feature vector to obtain the current fault level. The fault code mapping rule set is a predefined lookup table that can be set according to the DC bus voltage deviation rate, IGBT junction temperature difference, and output frequency change rate under minor and major fault levels in actual operation, as well as the form of the fault level. Any rule can take the form of… ; ,in It is the first one Rule 1 It is the fault level. It is the first one Rule 1 This is the result of a minor fault intersection. This is the result of the intersection of severe faults. It is the DC bus voltage deviation rate. It is the IGBT junction temperature difference value. It is the rate of change of the output frequency.

[0099] S302, or when the intersection results of minor faults and major faults are both empty sets, calculate the difference between the periodic change feature vector and the normal operating condition mean vector to obtain the feature deviation degree.

[0100] Specifically, if the intersection results of minor and major faults are both empty sets, meaning the current fault code belongs to neither the minor nor major fault code set, then the fault level cannot be directly determined by the code. In this case, the switching terminal calculates the current periodic change feature vector. Vector of mean values ​​under normal operating conditions The difference between them yields the feature deviation. Here, the normal operating condition mean vector is a reference vector obtained beforehand through offline statistics, which can be denoted as... The method for constructing the mean vector under normal operating conditions can be as follows: During the normal operation of the frequency converter without any fault alarms, continuously collect the periodic change feature vectors of multiple control cycles, and calculate the arithmetic mean of the periodic change feature vectors. For example, the mean of the normal DC bus voltage deviation rate. Typically close to 0, the average IGBT junction temperature difference The value is negative (because the normal junction temperature is below the rated upper limit), and the mean frequency change rate is... Approaching 0. Specifically, the difference can be quantified using Mahalanobis distance: ,in It is a preset covariance matrix under normal operating conditions, which can also be obtained through offline statistics. This is the characteristic deviation degree. The characteristic deviation degree is a non-negative real number; the larger the value, the more serious the deviation of the current operating state from the normal operating condition.

[0101] S303, calculate the difference between the characteristic deviation degree and the preset deviation threshold, and map the fault level by the characteristic deviation degree threshold difference based on the deviation level mapping rule set.

[0102] Specifically, the preset deviation threshold is a pre-defined constant, which can be denoted as: The preset deviation threshold can be obtained based on the 3σ principle in the statistical process of the mean vector under normal operating conditions, for example, a value of 3. The switching terminal first calculates the difference between the characteristic deviation degree and the deviation threshold: Feature deviation threshold difference The value can be positive, zero, or negative. Then, based on the deviation level mapping rule set, the fault level is obtained by mapping the difference in feature deviation thresholds. The deviation level mapping rule set is a function that maps a range of differences to fault levels, and its form can be: This indicates that the difference between the current operating status and the normal operating condition is within the allowable range, i.e. If, then it is mapped to a minor fault; if This indicates that if the deviation exceeds the normal range, it is mapped as a major fault.

[0103] This embodiment provides a control method for an automatic fault-switching soft starter for a turbine air-cooled fan inverter. First, it determines whether the fault code matches a preset set of minor or major faults. If a match is found, the fault level is directly output based on the code mapping rules and the exceeding of characteristic limits. If the code does not match any set, the Mahalanobis distance between the periodically changing characteristic vector and the mean vector of normal operating conditions is calculated to obtain the characteristic deviation. The difference between this deviation and the deviation threshold is then used to map the fault level. This method utilizes explicit fault code information and covers unknown code scenarios through statistical deviation analysis of multi-dimensional characteristic quantities, achieving accurate classification of all fault conditions.

[0104] In one embodiment, the residual voltage change characteristics of the motor are extracted based on the instantaneous data of the motor's three-phase voltage to obtain the real-time residual voltage amplitude, including:

[0105] S401 performs low-pass filtering on the instantaneous three-phase voltage data of the motor to obtain filtered three-phase voltage data.

[0106] Specifically, for any sampling timestamp Switching terminal to monitor instantaneous data of three-phase voltage of its motor Perform low-pass filtering to obtain filtered three-phase voltage data. Specifically, the switching terminal can use a digital low-pass filter to filter the instantaneous voltage values ​​of each phase in the instantaneous three-phase voltage data of the motor, such as a second-order Butterworth low-pass filter. The second-order Butterworth low-pass filter is a recursive filter with the flattest amplitude response; its amplitude-frequency characteristics are the flattest and ripple-free in the passband, and monotonically attenuated in the stopband. It is often used to extract the fundamental frequency or low-frequency components from noisy signals. The switching terminal sets its cutoff frequency according to the normal operating frequency of the instantaneous three-phase voltage data of the motor. , such as setting The filter operates at 200Hz, meaning it filters out high-frequency noise above 200Hz while retaining the 50Hz fundamental residual signal and a few low-order harmonic components. The coefficients of the filter's recursive difference equation are pre-calculated using a bilinear transform. The filter applies this to each sampling timestamp. The formula for the instantaneous data of the three-phase voltage of the motor is: ,in These are the filter coefficients. The obtained filtered three-phase voltage data can be expressed as: ,in , , These are the motor's sampling timestamps. The instantaneous values ​​of the A-phase filter voltage, B-phase filter voltage, and C-phase filter voltage. For example, using a sampling rate... and cutoff frequency Taking the sampling rate as an example, the process of determining the filter coefficients can be as follows: First, based on the sampling rate... and the set cutoff frequency Calculate the normalized cutoff angular frequency Then, the simulated prototype transfer function based on the Butterworth low-pass filter. The coefficients of the difference equation are obtained by mapping the result to the digital domain through a bilinear transformation. The specific calculation formula is as follows: Let... ,but , , , , Substitute Calculation yields , , , , .

[0107] S402 performs coordinate system transformation on the filtered three-phase voltage data, resulting in the first and second axis components in a two-phase stationary coordinate system.

[0108] Specifically, for any sampling timestamp The switching terminal will transfer the filtered three-phase voltage data in the three-phase stationary coordinate system (abc coordinate system). The system is transformed to a two-phase stationary coordinate system (αβ coordinate system), where the α-axis is aligned with the A-phase axis and the β-axis leads the α-axis by 90° electrical degrees. This coordinate system transformation can be called the Clarke transformation, and its transformation matrix is: . Any sampling timestamp Substituting the filtered three-phase voltage data, the first axis component is calculated. Second axis component : , The first axis component obtained is... That is, the α-axis component and the second axis component. This is the β-axis component. These two components constitute the orthogonal representation of the voltage vector in the stationary coordinate system.

[0109] S403, the first axis component is decomposed based on the second-order generalized integrator to obtain the first in-phase component and the first quadrature component, and the second axis component is decomposed based on the second-order generalized integrator to obtain the second in-phase component and the second quadrature component.

[0110] Specifically, a second-order generalized integrator is an adaptive filter based on a second-order transfer function, capable of extracting in-phase and quadrature components with the same frequency as the input fundamental wave from the input signal. For any sampling timestamp... First axis component The terminal switches to input the data into a second-order generalized integrator, which contains two outputs: in-phase components. and orthogonal components Its transfer function is: (For in-phase components) (For orthogonal components), where The resonant angular frequency can be adjusted in real time by the frequency-locked loop inside the second-order generalized integrator to track the fundamental angular frequency of the input signal; the fundamental angular frequency refers to the angular frequency of the fundamental component in the current motor residual voltage signal. This is the gain coefficient, usually taken as... After discretization, at each sampling timestamp The first in-phase component is calculated recursively. and the first orthogonal component Similarly, the second axis component... The second in-phase component is obtained by inputting it into the second second-order generalized integrator. Second orthogonal component .

[0111] S404, based on the first in-phase component, the first quadrature component, the second in-phase component, and the second quadrature component, the first positive sequence axis component and the second positive sequence axis component are calculated, wherein the expressions for the first positive sequence axis component and the second positive sequence axis component are:

[0112] ;

[0113] ;

[0114] In the formula, It is any sampling timestamp. It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The first in-phase component, It is a sampling timestamp The second orthogonal component, It is a sampling timestamp The second positive sequence axis component, It is a sampling timestamp The first orthogonal component, It is a sampling timestamp The second in-phase component.

[0115] Specifically, in a power system, three-phase voltages may contain positive-sequence, negative-sequence, and zero-sequence components. The positive-sequence component is the effective component for generating the rotating magnetic field, while the negative-sequence and zero-sequence components lead to calculation errors. To obtain a clean residual voltage fundamental positive-sequence component, for any sampling timestamp... The switching terminal utilizes the principle of the symmetric component method to switch the first in-phase component in the αβ coordinate system. First orthogonal component Second in-phase component Second orthogonal component The conversion is performed to calculate the first positive sequence axis component of the sampling timestamp. Second positive sequence axis components The physical meaning of this formula is: utilizing the symmetry property of the positive-sequence components in the αβ coordinate system, by changing the orthogonal components of the β-axis... In-phase component with α axis Combination, and the orthogonal components of the α-axis In-phase component with β axis By combining and canceling out the negative sequence components and some harmonics, the first positive sequence axis component, which represents only the positive sequence component of the fundamental frequency, is extracted. (Corresponding to the positive sequence component of the α-axis) and the second positive sequence component (Corresponding to the positive sequence component of the β axis).

[0116] S405, based on the first positive sequence axis component and the second positive sequence axis component, the real-time residual voltage amplitude is calculated. The expression for the real-time residual voltage amplitude is:

[0117] ;

[0118] In the formula, It is any sampling timestamp. It is a sampling timestamp The real-time residual voltage amplitude, It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The second positive sequence axis component.

[0119] Specifically, for any sampling timestamp The switching terminal is composed of the first positive sequence axis component. Second positive sequence axis components Reconstruct the real-time residual voltage amplitude of the residual voltage fundamental wave. Real-time residual voltage amplitude Defined as the magnitude of the positive-sequence voltage vector, with units of volts (V). The switching terminal can also calculate the real-time residual voltage phase based on the first and second positive-sequence axis components. Real-time residual phase Defined as the electrical angle of the positive-sequence voltage vector relative to the α-axis, the calculation formula is: The unit is radians (rad), and the value range is usually [range missing]. arrive It should be noted that in actual calculations, it is necessary to consider... and The sign is used to determine the quadrant, for example, by using the atan2 function to obtain the correct phase angle.

[0120] This embodiment provides a control method for an automatic switching soft starter for a turbine air-cooled fan inverter fault. First, the instantaneous three-phase voltage data of the motor is low-pass filtered to eliminate high-frequency noise. Then, the orthogonal components in the two-phase stationary coordinate system are obtained through Clarke transform. Next, the in-phase and orthogonal components of each component are extracted using a second-order generalized integrator. Then, the positive-sequence axis component is calculated based on the principle of symmetrical components. Finally, the real-time residual voltage amplitude and phase are accurately calculated from the positive-sequence component. This method can extract the pure fundamental positive-sequence component from the residual voltage after the inverter stops, accurately reflecting the residual magnetic field state of the motor, without relying on a speed sensor. This provides high-precision input data for subsequent residual voltage safety threshold judgment and zero-crossing synchronous switching, avoiding misjudgments caused by noise and harmonics.

[0121] In one embodiment, the instantaneous three-phase voltage data on the grid side includes sampling timestamps. Based on the instantaneous three-phase voltage data on the grid side, the zero-point time of the grid voltage is predicted, resulting in a zero-point timestamp set, including:

[0122] S501 uses the condition compliance time stamp as the end time stamp, and extracts the instantaneous data of the three-phase voltage on the grid side corresponding to a preset fixed number of sampling time stamps to obtain the predicted reference voltage sequence.

[0123] Specifically, the timestamp of meeting the conditions This is the sampling timestamp when the previously recorded real-time residual voltage amplitude first falls below the safety threshold. The preset fixed duration refers to a predetermined integer, which can be denoted as... For example This indicates the number of sampling timestamps to be captured. The fixed duration can be set based on the sampling period of the instantaneous three-phase voltage data from the grid side. The switching terminal can... As the endpoint timestamp, backtrack and extract... Each sampling timestamp is used to obtain a sampling timestamp sequence. The instantaneous three-phase voltage data of the grid side corresponding to these sampling timestamps in the sampling timestamp sequence are extracted sequentially. Arranged in chronological order, the predicted reference voltage sequence is obtained: , Its length is The predicted reference voltage sequence can characterize the waveform variation trend of the grid voltage over a period of time before the condition is met.

[0124] S502, input the predicted reference voltage sequence into the zero-point prediction model to obtain the reference zero-point timestamp.

[0125] Specifically, the zero-time prediction model is a pre-trained one-dimensional convolutional neural network whose input is the prediction reference voltage sequence. The output is a baseline zero-point timestamp, which is a scalar representing the time since the condition was met. The time interval until the next positive zero-crossing of the grid voltage can be denoted as... The zero-point prediction model can be trained using three-phase voltage waveform data under various power grid operating conditions, and can adapt to complex situations such as harmonics, noise, and frequency fluctuations.

[0126] S503, the reference zero-point timestamp is added sequentially to an integer multiple of a preset half-cycle parameter to obtain at least one adjusted zero-point timestamp, and a zero-point timestamp set is formed based on the reference zero-point timestamp and the adjusted zero-point timestamp.

[0127] Specifically, the preset half-cycle parameter is the fundamental half-cycle value of the grid voltage, which can be denoted as: The preset half-cycle parameters can be set according to the fundamental frequency period of the grid voltage. For example, for a standard 50Hz grid, the fundamental frequency period is... Then the preset half-cycle parameter Switch terminals using a base zero-timestamp. Starting from, add sequentially Generate at least one adjusted zero-point timestamp by multiplying the integer multiple of the timestamp: , And so on. Then, the baseline zero-point timestamp and all adjusted zero-point timestamps are combined to form a zero-point timestamp sequence. This sequence provides multiple candidate zero-crossing times, allowing subsequent steps to select the most suitable command issuance time based on the residual pressure reaching the target time and the contactor delay time, thus ensuring the timeliness and safety of the switching.

[0128] The zero-point prediction model includes:

[0129] The input layer receives the predicted reference voltage sequence, normalizes the predicted reference voltage sequence to obtain the normalized voltage sequence, and then transmits the normalized voltage sequence to the convolutional feature extraction layer.

[0130] Specifically, the input layer is used to receive the predicted reference voltage sequence. First, normalization is performed on each channel (phases A, B, and C) to eliminate the influence of amplitude differences between different samples. Normalization can use the mean and standard deviation of the training dataset: for the... Mutually( (corresponding to A, B, and C respectively), calculate the mean of all training samples for that phase. and standard deviation Then, the instantaneous voltage values ​​of the corresponding phases in the predicted reference voltage sequence are standardized to obtain the corresponding normalized values: The normalized voltage sequence is obtained. The data is then transmitted to the convolutional feature extraction layer.

[0131] The convolutional feature extraction layer is used to perform one-dimensional convolution operations on the normalized voltage sequence, extract local waveform features, obtain local feature maps, and then transmit the local feature maps to the pooling layer.

[0132] Specifically, the convolutional feature extraction layer consists of multiple stacked one-dimensional convolutional layers, used to extract abstract features step by step from the normalized voltage sequence. The first convolutional layer directly receives the normalized voltage sequence output from the input layer. Suppose that this layer contains There are learnable convolutional kernels, each with a size of . , The length of the time window, for example The convolutional kernel slides along the time dimension, and each sampling timestamp corresponds to a window within the input sequence. The sampled timestamps and the normalized values ​​of the three channels are weighted and summed, then a bias is added, and the result is passed through a non-linear activation function (such as ReLU) to output a feature value. Specifically, the first convolutional layer is performed at the sampled timestamps. The The output of each convolution kernel is , It is the first in the normalized voltage sequence Mutually( These can correspond to A, B, and C respectively at the sampling timestamps. The normalized value, It is the first Each convolutional kernel at the offset ,aisle Weight on, For the corresponding bias term, It is the ReLU function ( The outputs of all convolutional kernels are stacked along the channel dimension to obtain the output feature map of the first layer. ,in The time length after convolution. The number of convolution kernels. The second convolution layer uses... For input, containing There are n convolutional kernels, each with a size of 1. The second layer Each convolutional kernel at the sampling timestamp The output is , It is the output feature map of the first layer. The Middle Each channel at the sampling timestamp The normalized value, and These are the weights and biases for the second layer. Similarly, more convolutional layers can be added as needed. Each layer takes the feature map output from the previous layer as input, progressively extracting higher-order local waveform patterns such as slope changes, peak shapes, and three-phase phase differences. The weights and biases of all convolutional layers are obtained through offline training and remain fixed during inference. The final convolutional layer output is a local feature map. And transmit it to the pooling layer.

[0133] Pooling layers are used to downsample local feature maps to obtain dimensionality-reduced feature vectors, which are then passed to fully connected layers.

[0134] Specifically, pooling layers target local feature maps. Downsampling is performed to reduce the resolution of the time dimension while preserving salient features, resulting in a dimensionality-reduced feature vector. Specifically, max pooling can be used, with a pooling window size of [value missing]. ,For example The step size is also This means taking the maximum value within a non-overlapping window. Let the time length of the feature map after pooling become... The number of channels remains at If unchanged, then pooling output .Will Flattening it into a one-dimensional vector involves sequentially concatenating the eigenvalues ​​from all time steps and all channels to obtain a dimensionality-reduced eigenvector. ,in Each element in this vector corresponds to the activation intensity of a specific feature channel at a given sampling timestamp after pooling.

[0135] The fully connected layer is used to perform linear transformations and nonlinear activations on the dimensionality-reduced feature vectors to obtain the baseline zero-point timestamp.

[0136] Specifically, fully connected layers extract dimensionality-reduced feature vectors. The higher-order features are used to calculate the linear regression value: and the linear regression values and the latest sampling timestamp in the predicted reference voltage sequence (i.e., the condition met timestamp) Add them together to get the baseline zero-point timestamp. For example, multiple fully connected layers (linear transformation plus nonlinear activation) can be set to extract higher-order features. For instance, the first fully connected layer computes... ,in This is the weight matrix of the first fully connected layer. It is the bias term of the first fully connected layer. It uses the ReLU activation function. The last fully connected layer does not contain an activation function and directly outputs the linear regression value. ,in It represents the total number of fully connected layers.

[0137] This embodiment provides a control method for an automatic fault-switching soft starter for a turbine air-cooled fan inverter. First, a fixed-length three-phase grid voltage sequence is extracted, starting from the condition-meeting timestamp, to construct a multi-channel predictive reference sequence. This sequence is then input into a one-dimensional convolutional neural network model. Through progressive processing including input layer normalization, multi-channel convolution to extract spatiotemporal features, pooling dimensionality reduction, and fully connected regression, the time interval from the current moment to the next zero-crossing point is accurately predicted, yielding a reference zero-point timestamp. Finally, half-cycle parameters are superimposed to generate multiple candidate zero-point timestamps to form a sequence. This method fully exploits the phase relationship between phases, enhances the model's robustness to harmonics and noise, and provides reliable input data for optimizing subsequent command issuance times.

[0138] In one embodiment, the zero-point prediction model is obtained through the following steps:

[0139] S601 acquires voltage waveform data under different power grid operating conditions; and performs sliding window sampling on the voltage waveform data to obtain a complete voltage sequence.

[0140] Specifically, different power grid operating conditions can include standard sinusoidal wave operating conditions (frequency 50Hz, no harmonics), harmonic-containing operating conditions (e.g., 5%–15% of the 3rd, 5th, and 7th harmonics), noise-containing operating conditions (signal-to-noise ratio 30dB–60dB), frequency fluctuation operating conditions (changing at a rate of 0.1–0.5Hz / s within the range of 49.5Hz–50.5Hz), and phase jump operating conditions (±10°–±30°). Under each operating condition, the switching terminal continuously collects instantaneous three-phase voltage data from the power grid side, records the sampling timestamp of each sample value, and obtains the complete voltage sequence under different operating conditions.

[0141] S602, the complete voltage sequence is truncated to obtain the sample voltage sequence, and the time interval from the last sampling timestamp in the sample voltage sequence to the next zero-crossing point of the grid voltage is calculated; the sample voltage sequence is used as the input feature vector, and the time interval corresponding to the sample voltage sequence is used as the prediction target value corresponding to the input feature vector to construct the model training dataset.

[0142] Specifically, the switching terminal truncates each complete voltage sequence to obtain a sample voltage sequence. Specifically, for a random sampling timestamp in the complete voltage sequence, the switching terminal uses that sampling timestamp as the endpoint and truncates the instantaneous three-phase voltage data from the grid side corresponding to a number of sampling timestamps with a preset fixed duration, forming a sample voltage sequence. And calculate the sample voltage sequence. The time interval between the last sampling timestamp and the next zero-crossing point of the grid voltage. Switching terminals to sample voltage sequences For an input feature vector With sample voltage sequence Corresponding time interval The model training dataset is constructed by using the predicted target values ​​corresponding to the input feature vectors. For example, after construction, the mean value of each channel (phases A, B, and C) is calculated for all input feature vectors in the training set. and standard deviation ( ), and save it for subsequent normalization.

[0143] S603 connects the input layer, convolutional layer, pooling layer, fully connected layer and output layer in sequence to obtain the initial prediction model architecture, and initializes the initial prediction model architecture to obtain the initial zero-time model.

[0144] Specifically, the switching terminal connects the input layer, convolutional layer, pooling layer, fully connected layer, and output layer in a concatenated manner to obtain the initial prediction model architecture. Then, all trainable parameters (convolutional kernel weights and biases, fully connected layer weights and biases) in the initial prediction model architecture are initialized to obtain the initial zero-time model. Specifically, the initialization method can employ Xavier initialization or He initialization; for example, starting with a mean of 0 and a variance of... Random sampling is performed from a normal distribution to ensure gradient stability during the initial training phase. This is the number of input neurons in the current layer. For convolutional layers, Equal to the time window length of the convolution kernel Multiply by the number of input channels; for example, if the first convolutional layer has 3 input channels, then... For fully connected layers, It is equal to the dimension of the reduced-dimensional feature vector.

[0145] S604 inputs a preset number of input feature vectors into the initial zero-time model to obtain the prediction time interval value corresponding to the input feature vectors.

[0146] Specifically, the preset training quantity can be set according to the number of sample vectors used in actual training, and can be denoted as: The terminal selects a preset number of input feature vectors from the model training dataset and inputs them into the model at the initial zero point time, obtaining the prediction time interval value corresponding to the input feature vector.

[0147] S605, based on the input feature vector, the corresponding prediction target value, and the prediction time interval, the model loss value is calculated. The expression for the model loss value is:

[0148] ;

[0149] In the formula, It is a preset number of training sessions. It is the index of any input feature vector. It is the first The prediction time interval value corresponding to each input feature vector. It is the first The predicted target value corresponding to each input feature vector. It is the model loss value.

[0150] Specifically, the switching terminal calculates the model loss value based on the input feature vector, the corresponding prediction target value, and the prediction time interval. Model loss value Used to quantify the difference between the model's predicted values ​​and the actual values. For each input feature vector Switching terminals to calculate the predicted time interval value Compared with the predicted target value The differences are squared, then the sum of the squared differences of all samples is divided by . To obtain the average loss value of the current batch. The smaller the loss value, the more accurate the model's prediction.

[0151] S606 aims to minimize the model loss value by iteratively updating the initial zero-time model until the preset iteration termination condition is met, thus obtaining the trained zero-time prediction model.

[0152] Specifically, switching terminals aims to minimize the model loss value. The initial zero-time model is iteratively updated until a preset termination condition is met, resulting in a trained zero-time prediction model. Specifically, the backpropagation algorithm can be used to calculate the gradient of the model loss value with respect to each trainable parameter. Then, an optimizer, such as the Adam optimizer, with an initial learning rate of 0.001, updates the parameters according to the gradient direction. ,in This represents all parameters of the model. The learning rate is used. Parameters are updated after each batch is processed until a preset iteration termination condition is met. For example, the preset iteration termination condition could be reaching the maximum number of iterations, such as 200 rounds, or it could be selecting a certain proportion of input feature vectors and their corresponding prediction target values ​​from the model training dataset to form a validation set. The preset iteration termination condition could be that the model loss value on the validation set no longer decreases for 10 consecutive rounds.

[0153] This embodiment provides a control method for an automatic switching soft starter for a turbine air-cooled fan inverter during faults. First, three-phase grid voltage waveform data under multiple operating conditions is collected, and a training dataset is constructed through sliding window sampling and labeling. Then, an initial neural network model consisting of an input layer, convolutional layer, pooling layer, fully connected layer, and output layer is built. The model is iteratively optimized using the training data and mean squared error as the loss function, ultimately obtaining a trained zero-crossing time prediction model. This model can learn the nonlinear mapping between voltage waveforms and zero-crossing intervals, exhibiting stronger resistance to harmonics, noise, and frequency fluctuations. This provides a reliable computational foundation for subsequent high-precision zero-crossing prediction, thereby further improving the accuracy of switching command timing.

[0154] In one embodiment, the instruction issuance timestamp is calculated based on the condition fulfillment timestamp and the zero-point timestamp set, including:

[0155] S701, calculate the difference between the baseline zero-point timestamp and the condition compliance timestamp to obtain the baseline timestamp difference; calculate the difference between the adjustment zero-point timestamp and the condition compliance timestamp to obtain the time difference of each adjustment zero-point timestamp.

[0156] Specifically, the terminal switches to calculate the difference between the baseline zero-point timestamp and the condition-meeting timestamp in the zero-point timestamp set to obtain the baseline timestamp difference. The unit is seconds, and this difference represents the time interval from the moment the residual pressure reaches the target to the zero-crossing point of the benchmark. Simultaneously, for each adjusted zero-point timestamp in the zero-point timestamp sequence... ( ), calculate the difference between it and the timestamp of meeting the conditions. This yields the time difference corresponding to each adjusted zero-point timestamp.

[0157] S702, when the difference between the reference timestamps is greater than the preset instruction start delay time parameter, the reference zero-point timestamp and the instruction start delay time parameter are subtracted to obtain the instruction issuance timestamp; or, when the difference between the reference timestamps is less than or equal to the instruction start delay time parameter, the time difference is compared with the instruction start delay time parameter to obtain the difference comparison result of each adjusted zero-point timestamp.

[0158] Specifically, the preset instruction start delay time parameter This is the inherent operating delay time of the contactor from coil energization to main contact closure, measured in seconds. The command start delay time parameter can be obtained from the contactor's datasheet. (The difference between the reference timestamp and the actual time delay is not specified in the original text.) This indicates that there is sufficient time margin between the moment the condition is met and the reference zero-crossing point to compensate for the contactor delay. At this time, the command is issued with a timestamp. Calculate using the following formula: At that moment, a closing command is issued for the soft starter branch contactor, after which... After the mechanical delay, the contactor's main contacts will be exactly at the reference zero-crossing point. The circuit is closed at all times, so that the soft starter is engaged at the zero-crossing point of the grid voltage, effectively suppressing the inrush current.

[0159] S703, select the earliest adjustment zero-point timestamp whose comparison result is greater than the instruction start delay time parameter, and use it as the instruction reference zero-point timestamp. Then, subtract the instruction reference zero-point timestamp from the instruction start delay time parameter to obtain the instruction issuance timestamp.

[0160] Specifically, if In other words, the time interval between the reference zero-crossing point and the moment when the condition is met is insufficient to fully compensate for the contactor delay. If the command is issued according to the reference zero point, the contact closing time will lag behind the zero-crossing point, failing to achieve the effect of zero-voltage connection. At this time, the switching terminal checks the time difference corresponding to each zero-point adjustment timestamp in the order of the zero-point timestamp adjustment. For each adjustment of the zero-point timestamp ( ), switch terminals and compare their time differences and The difference comparison results are obtained: If This indicates that the zero-point adjustment provides sufficient lead time, and the zero-point adjustment timestamp is... The zero-point timestamp is used as the reference for the instruction, and the instruction issuance timestamp is calculated according to the following formula: ;like Then continue checking the next adjustment of the zero-point timestamp. .

[0161] This embodiment provides a control method for an automatic switching soft starter for a turbine air-cooled fan inverter. First, it calculates the time difference between the reference zero point and each adjusted zero point and the condition fulfillment timestamp. Then, it compares the reference timestamp difference with the command start delay time parameter. If the reference difference is greater than the delay, the command issuance timestamp is directly calculated based on the reference zero point. If the reference difference is not greater than the delay, the time differences of each adjusted zero point are further compared to prepare for selecting the earliest zero point that satisfies delay compensation. The method fully considers the impact of the inherent contactor operating delay on the switching timing. Through numerical comparison and candidate zero point screening, it ensures that the command issuance time accurately compensates for the delay, ensuring that the actual activation time of the soft starter falls near the grid voltage zero-crossing point, thereby maximally suppressing inrush current and ensuring the stable operation of the equipment and the power grid.

[0162] In the aforementioned control method for an automatic switching soft starter for a turbine air-cooled fan inverter fault, the method involves acquiring the inverter's fault code and real-time operating data within the current control cycle, and determining the fault level for the current control cycle based on the fault code and real-time operating data. When the fault level meets the preset switching requirements, an inverter stop command is generated. This inverter stop command instructs the inverter to stop outputting and acquires real-time instantaneous data of the motor's three-phase voltage and the grid's three-phase voltage. Based on the instantaneous data of the motor's three-phase voltage, the residual voltage change characteristics of the motor are extracted to obtain the real-time residual voltage amplitude. When the real-time residual voltage amplitude is less than a preset safety threshold, the sampling timestamp corresponding to the real-time residual voltage amplitude is recorded as the condition compliance timestamp. Based on the instantaneous data of the grid's three-phase voltage, the zero-point time of the grid voltage is predicted to obtain a zero-point timestamp set. The command issuance timestamp is calculated based on the condition compliance timestamp and the zero-point timestamp set. At the command issuance timestamp, a soft starter start command is generated. This soft starter start command instructs the soft starter to start operating. Improving the electrical safety and success rate of the switching process ensures the continuous and stable operation of the turbine air-cooled fan under various fault conditions.

[0163] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0164] Based on the same inventive concept, this application also provides a control system for an automatic switching soft starter for a turbine air-cooled fan inverter fault, used to implement the control method for the aforementioned automatic switching soft starter for a turbine air-cooled fan inverter fault. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the control system for an automatic switching soft starter for a turbine air-cooled fan inverter fault provided below can be found in the limitations of the control method for an automatic switching soft starter for a turbine air-cooled fan inverter fault described above, and will not be repeated here.

[0165] In one exemplary embodiment, such as Figure 2As shown, a control system 200 for an automatic switching soft starter for a turbine air-cooled fan inverter fault is provided, comprising:

[0166] The data acquisition module 201 is used to acquire the fault codes and real-time operating data of the frequency converter in the current control cycle, and to perform fault classification judgment based on the fault codes and real-time operating data to obtain the fault level of the current control cycle.

[0167] The inverter stop control module 202 is used to generate an inverter stop command when the fault level meets the preset switching requirements; the inverter stop command is used to instruct the inverter to stop output and to acquire instantaneous three-phase voltage data of the motor and instantaneous three-phase voltage data of the grid side in real time.

[0168] The motor residual voltage calculation module 203 is used to extract the residual voltage change characteristics of the motor based on the instantaneous data of the three-phase voltage of the motor, and obtain the real-time residual voltage amplitude.

[0169] The instruction issuance time calculation module 204 is used to record the sampling timestamp corresponding to the real-time residual voltage amplitude as the condition compliance timestamp when the real-time residual voltage amplitude is less than the preset safety threshold, predict the zero point time of the grid voltage based on the instantaneous data of the three-phase voltage on the grid side, obtain the zero point timestamp set, and calculate the instruction issuance timestamp based on the condition compliance timestamp and the zero point timestamp set.

[0170] The soft-start command issuing module 205 is used to generate a soft-starter startup command at the command issuance timestamp; the soft-starter startup command is used to instruct the soft-starter to start working.

[0171] Furthermore, the real-time operating data includes DC bus voltage, IGBT junction temperature, and output frequency. All three values ​​are accompanied by a sampling timestamp. Based on the fault codes and real-time operating data, a fault classification is determined to obtain the fault level for the current control cycle, including:

[0172] The average value of the DC bus voltage corresponding to each sampling timestamp is calculated to obtain the average DC bus voltage. Based on the normal DC bus voltage value and the average DC bus voltage, the DC bus voltage deviation rate is calculated.

[0173] Extract the maximum value of the IGBT junction temperature corresponding to the starting sampling timestamp within the current control cycle, and use it as the maximum IGBT junction temperature. Calculate the difference between the maximum IGBT junction temperature and the rated upper limit of the IGBT junction temperature to obtain the IGBT junction temperature difference.

[0174] Extract the output frequency value corresponding to the first sampling timestamp in the current control cycle as the starting output frequency value, and extract the output frequency value corresponding to the last sampling timestamp in the current control cycle as the ending output frequency value.

[0175] Based on the starting output frequency value, the ending output frequency value, and the control cycle duration parameter, the output frequency change rate is calculated, and a periodic change feature vector is constructed based on the DC bus voltage deviation rate, IGBT junction temperature difference value, and output frequency change rate.

[0176] Identify the intersection of fault codes and minor fault code sets to obtain the minor fault intersection result; identify the intersection of fault codes and major fault code sets to obtain the major fault intersection result.

[0177] Based on the fault code mapping rule set, fault level analysis is performed on the intersection results of minor faults, the intersection results of severe faults, and the periodic change feature vector to obtain the fault level.

[0178] Furthermore, based on the fault code mapping rule set, fault level analysis is performed on the intersection results of minor faults, the intersection results of severe faults, and the periodic change feature vectors to obtain the fault level, including:

[0179] When the intersection result of a minor fault is a fault code or the intersection result of a major fault is a fault code, the intersection result of a minor fault, the intersection result of a major fault, and the periodic change feature vector are mapped based on the fault code mapping rule set to obtain the fault level;

[0180] Alternatively, when the intersection results of minor faults and major faults are both empty sets, the difference between the periodic change feature vector and the normal operating condition mean vector is calculated to obtain the feature deviation degree.

[0181] The characteristic deviation degree is calculated and the characteristic deviation degree threshold difference is set to a preset deviation threshold. Based on the deviation level mapping rule set, the fault level is obtained by mapping the characteristic deviation degree threshold difference.

[0182] Furthermore, based on the instantaneous data of the motor's three-phase voltage, the residual voltage change characteristics of the motor are extracted to obtain the real-time residual voltage amplitude, including:

[0183] The instantaneous three-phase voltage data of the motor is low-pass filtered to obtain filtered three-phase voltage data;

[0184] The three-phase voltage data of the filter is transformed into the first and second axis components in the two-phase stationary coordinate system.

[0185] The first axis component is decomposed using a second-order generalized integrator to obtain the first in-phase component and the first quadrature component. The second axis component is also decomposed using a second-order generalized integrator to obtain the second in-phase component and the second quadrature component.

[0186] Based on the first in-phase component, the first quadrature component, the second in-phase component, and the second quadrature component, the first positive sequence axis component and the second positive sequence axis component are calculated. The expressions for the first positive sequence axis component and the second positive sequence axis component are as follows:

[0187] ;

[0188] ;

[0189] In the formula, It is any sampling timestamp. It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The first in-phase component, It is a sampling timestamp The second orthogonal component, It is a sampling timestamp The second positive sequence axis component, It is a sampling timestamp The first orthogonal component, It is a sampling timestamp The second in-phase component;

[0190] The real-time residual voltage amplitude is calculated based on the first and second positive sequence axis components. The expression for the real-time residual voltage amplitude is as follows:

[0191] ;

[0192] In the formula, It is any sampling timestamp. It is a sampling timestamp The real-time residual voltage amplitude, It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The second positive sequence axis component.

[0193] Furthermore, the instantaneous three-phase voltage data on the grid side includes sampling timestamps. Based on the instantaneous three-phase voltage data on the grid side, the zero-point time of the grid voltage is predicted, resulting in a zero-point timestamp set, including:

[0194] Using the time stamp when the conditions are met as the end time stamp, the instantaneous data of the three-phase voltage on the grid side corresponding to a preset fixed number of sampling time stamps are extracted to obtain the predicted reference voltage sequence.

[0195] The predicted reference voltage sequence is input into the zero-point prediction model to obtain the reference zero-point timestamp;

[0196] The reference zero-point timestamp is added sequentially to an integer multiple of a preset half-cycle parameter to obtain at least one adjusted zero-point timestamp, and a zero-point timestamp set is formed based on the reference zero-point timestamp and the adjusted zero-point timestamp.

[0197] The zero-point prediction model includes:

[0198] The input layer receives the predicted reference voltage sequence, normalizes the predicted reference voltage sequence to obtain the normalized voltage sequence, and transmits the normalized voltage sequence to the convolutional feature extraction layer.

[0199] The convolutional feature extraction layer is used to perform a one-dimensional convolution operation on the normalized voltage sequence, extract local waveform features, obtain a local feature map, and then transmit the local feature map to the pooling layer;

[0200] Pooling layers are used to downsample local feature maps to obtain dimensionality-reduced feature vectors, and then transmit the dimensionality-reduced feature vectors to fully connected layers;

[0201] The fully connected layer is used to perform linear transformations and nonlinear activations on the dimensionality-reduced feature vectors to obtain the baseline zero-point timestamp.

[0202] Furthermore, the zero-time prediction model is obtained through the following steps:

[0203] Acquire voltage waveform data under different power grid operating conditions; and perform sliding window sampling on the voltage waveform data to obtain a complete voltage sequence;

[0204] The complete voltage sequence is truncated to obtain a sample voltage sequence, and the time interval from the last sampling timestamp in the sample voltage sequence to the next zero-crossing point of the grid voltage is calculated. The model training dataset is constructed by using the sample voltage sequence as the input feature vector and the time interval corresponding to the sample voltage sequence as the prediction target value corresponding to the input feature vector.

[0205] The input layer, convolutional layer, pooling layer, fully connected layer and output layer are connected in sequence to obtain the initial prediction model architecture, and the initial prediction model architecture is initialized to obtain the initial zero-time model.

[0206] Input the preset number of input feature vectors into the model at the initial zero time point to obtain the prediction time interval value corresponding to the input feature vector;

[0207] Based on the input feature vector, the corresponding prediction target value, and the prediction time interval, the model loss value is calculated. The expression for the model loss value is:

[0208] ;

[0209] In the formula, It is a preset number of training sessions. It is the index of any input feature vector. It is the first The prediction time interval value corresponding to each input feature vector. It is the first The predicted target value corresponding to each input feature vector. It is the model loss value;

[0210] With the goal of minimizing the model loss, the initial zero-time model is iteratively updated until the preset iteration termination condition is met, thus obtaining the trained zero-time prediction model.

[0211] Furthermore, the instruction issuance timestamp is calculated based on the condition fulfillment timestamp and the zero-point timestamp set, including:

[0212] Calculate the difference between the baseline zero-point timestamp and the condition-meeting timestamp to obtain the baseline timestamp difference; calculate the difference between the adjustment zero-point timestamp and the condition-meeting timestamp to obtain the time difference of each adjustment zero-point timestamp.

[0213] When the difference between the reference timestamps is greater than the preset instruction start delay time parameter, the reference zero-point timestamp and the instruction start delay time parameter are subtracted to obtain the instruction issuance timestamp; or, when the difference between the reference timestamps is less than or equal to the instruction start delay time parameter, the time difference is compared with the instruction start delay time parameter to obtain the difference comparison result of each adjustment zero-point timestamp.

[0214] The earliest adjustment zero-point timestamp whose comparison result is greater than the instruction start delay time parameter is selected as the instruction reference zero-point timestamp. The instruction reference zero-point timestamp and the instruction start delay time parameter are subtracted to obtain the instruction issuance timestamp.

[0215] In one embodiment, such as Figure 3 A computer device is provided, comprising:

[0216] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301: the memory stores application code executable by at least one of the processors, the application code being executed by at least one of the processors to enable at least one of the processors to perform a control method for an automatic switching soft starter for a turbine air-cooled fan inverter fault as described above.

[0217] Computer equipment may also include: sensor 303.

[0218] The processor 301, memory 302 and sensor 303 can be connected via a bus or other means, with the bus being an example in the figure.

[0219] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0220] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0221] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A control method for automatically switching a soft starter in case of inverter failure in a turbine air-cooled fan, characterized in that, The method includes: Obtain the fault codes and real-time operating data of the frequency converter within the current control cycle, and perform fault classification judgment based on the fault codes and real-time operating data to obtain the fault level of the current control cycle; When the fault level meets the preset switching requirements, a frequency converter stop operation command is generated; the frequency converter stop operation command is used to instruct the frequency converter to stop output and acquire instantaneous three-phase voltage data of the motor and instantaneous three-phase voltage data of the grid side in real time; Based on the instantaneous data of the three-phase voltage of the motor, the residual voltage change characteristics of the motor are extracted to obtain the real-time residual voltage amplitude; When the real-time residual voltage amplitude is less than the preset safety threshold, the sampling timestamp corresponding to the real-time residual voltage amplitude is recorded as the condition compliance timestamp. Based on the instantaneous data of the three-phase voltage on the grid side, the zero point time of the grid voltage is predicted to obtain the zero point timestamp set. The instruction issuance timestamp is calculated based on the condition compliance timestamp and the zero point timestamp set. At the timestamp of the instruction issuance, a soft launcher startup instruction is generated; the soft launcher startup instruction is used to instruct the soft launcher to start working.

2. The method according to claim 1, characterized in that, The real-time operating data includes DC bus voltage, IGBT junction temperature, and output frequency. Each of these values ​​carries a sampling timestamp. The fault classification judgment based on the fault code and the real-time operating data to obtain the fault level for the current control cycle includes: The average value of the DC bus voltage corresponding to each sampling timestamp is calculated to obtain the average DC bus voltage. Based on the normal DC bus voltage value and the average DC bus voltage, the DC bus voltage deviation rate is calculated. Extract the maximum value of the IGBT junction temperature corresponding to the sampling timestamp at the beginning of the current control cycle, and take it as the maximum IGBT junction temperature. Calculate the difference between the maximum IGBT junction temperature and the rated upper limit of the IGBT junction temperature to obtain the IGBT junction temperature difference. Extract the output frequency value corresponding to the first sampling timestamp in the current control cycle as the starting output frequency value, and extract the output frequency value corresponding to the last sampling timestamp in the current control cycle as the ending output frequency value. Based on the starting output frequency value, the ending output frequency value, and the control cycle duration parameter, the output frequency change rate is calculated, and a periodic change feature vector is constructed based on the DC bus voltage deviation rate, the IGBT junction temperature difference value, and the output frequency change rate. Identify the intersection of the fault codes and the set of minor fault codes to obtain the minor fault intersection result; identify the intersection of the fault codes and the set of major fault codes to obtain the major fault intersection result. Based on the fault code mapping rule set, the fault level is analyzed by the intersection results of the mild faults, the intersection results of the severe faults, and the periodic change feature vector to obtain the fault level.

3. The method according to claim 2, characterized in that, The fault level is obtained by performing fault code mapping rule set-based analysis on the intersection results of the minor faults, the intersection results of the severe faults, and the periodic variation feature vectors to obtain the fault level, including: When the intersection result of the minor fault is the fault code or the intersection result of the severe fault is the fault code, the intersection result of the minor fault, the intersection result of the severe fault, and the periodic change feature vector are mapped based on the fault code mapping rule set to obtain the fault level; Alternatively, if the intersection results of the minor fault and the intersection results of the severe fault are both empty sets, calculate the difference between the periodic change feature vector and the normal operating condition mean vector to obtain the feature deviation degree; Calculate the difference between the feature deviation degree and the preset deviation threshold, and based on the deviation level mapping rule set, map the fault level from the feature deviation degree threshold difference.

4. The method according to claim 2, characterized in that, The step of extracting the residual voltage change characteristics of the motor based on the instantaneous data of the motor's three-phase voltage to obtain the real-time residual voltage amplitude includes: The instantaneous three-phase voltage data of the motor is low-pass filtered to obtain filtered three-phase voltage data; The filtered three-phase voltage data is transformed to obtain the first axis component and the second axis component in a two-phase stationary coordinate system. The first axis component is decomposed based on the second-order generalized integrator to obtain the first in-phase component and the first quadrature component. The second axis component is decomposed based on the second-order generalized integrator to obtain the second in-phase component and the second quadrature component. Based on the first in-phase component, the first quadrature component, the second in-phase component, and the second quadrature component, the first positive sequence axis component and the second positive sequence axis component are calculated, wherein the expressions for the first positive sequence axis component and the second positive sequence axis component are: ； ； In the formula, It is any sampling timestamp. It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The first in-phase component, It is a sampling timestamp The second orthogonal component, It is a sampling timestamp The second positive sequence axis component, It is a sampling timestamp The first orthogonal component, It is a sampling timestamp The second in-phase component; The real-time residual voltage amplitude is calculated based on the first positive sequence axis component and the second positive sequence axis component. The expression for the real-time residual voltage amplitude is as follows: ； In the formula, It is any sampling timestamp. It is a sampling timestamp The real-time residual voltage amplitude, It is a sampling timestamp The first positive sequence axis component, It is a sampling timestamp The second positive sequence axis component.

5. The method according to claim 4, characterized in that, The instantaneous three-phase voltage data on the grid side includes the sampling timestamps. The zero-point timestamp set obtained by predicting the grid voltage zero-point time based on the instantaneous three-phase voltage data on the grid side includes: Using the time stamp when the conditions are met as the end time stamp, a preset fixed number of sampling time stamps corresponding to the instantaneous data of the three-phase voltage on the grid side are extracted to obtain the predicted reference voltage sequence. The predicted reference voltage sequence is input into the zero-point prediction model to obtain the reference zero-point timestamp; The reference zero-point timestamp is added sequentially to an integer multiple of a preset half-cycle parameter to obtain at least one adjusted zero-point timestamp, and the zero-point timestamp set is formed based on the reference zero-point timestamp and the adjusted zero-point timestamp. The zero-point prediction model includes: The input layer is used to receive the predicted reference voltage sequence, normalize the predicted reference voltage sequence to obtain a normalized voltage sequence, and transmit the normalized voltage sequence to the convolutional feature extraction layer. The convolutional feature extraction layer is used to perform a one-dimensional convolution operation on the normalized voltage sequence, extract local waveform features, obtain a local feature map, and transmit the local feature map to the pooling layer; The pooling layer is used to downsample the local feature map to obtain a dimensionality-reduced feature vector, and then transmit the dimensionality-reduced feature vector to the fully connected layer. The fully connected layer is used to perform linear transformation and nonlinear activation on the dimensionality-reduced feature vector to obtain the reference zero-point timestamp.

6. The method according to claim 5, characterized in that, The zero-point prediction model is obtained through the following steps: Acquire voltage waveform data under different power grid operating conditions; and perform sliding window sampling on the voltage waveform data to obtain a complete voltage sequence; The complete voltage sequence is truncated to obtain a sample voltage sequence, and the time interval from the last sampling timestamp in the sample voltage sequence to the next zero-crossing point of the grid voltage is calculated; the sample voltage sequence is used as the input feature vector, and the time interval corresponding to the sample voltage sequence is used as the prediction target value corresponding to the input feature vector to construct a model training dataset; The input layer, the convolutional layer, the pooling layer, the fully connected layer, and the output layer are connected sequentially to obtain the initial prediction model architecture. The initial prediction model architecture is then initialized to obtain the initial zero-time model. The input feature vectors of a preset number of training iterations are input into the initial zero-point time model to obtain the predicted time interval value corresponding to the input feature vectors; Based on the input feature vector, the prediction target value corresponding to the input feature vector, and the prediction time interval value, the model loss value is calculated, and the expression for the model loss value is: ； In the formula, It is a preset number of training sessions. It is the index of any input feature vector. It is the first The prediction time interval value corresponding to each input feature vector. It is the first The predicted target value corresponding to each input feature vector. It is the model loss value; With the goal of minimizing the model loss value, the initial zero-time model is iteratively updated until a preset iteration termination condition is met, thereby obtaining the trained zero-time prediction model.

7. The method according to claim 5, characterized in that, The calculation of the instruction issuance timestamp based on the condition fulfillment timestamp and the zero-point timestamp set includes: Calculate the difference between the baseline zero-point timestamp and the condition-meeting timestamp to obtain the baseline timestamp difference; calculate the difference between the adjustment zero-point timestamp and the condition-meeting timestamp to obtain the time difference of each adjustment zero-point timestamp. When the difference between the reference timestamps is greater than the preset instruction start delay time parameter, the reference zero-point timestamp and the instruction start delay time parameter are subtracted to obtain the instruction issuance timestamp; or, when the difference between the reference timestamps is less than or equal to the instruction start delay time parameter, the time difference and the instruction start delay time parameter are compared to obtain the difference comparison result of each of the adjusted zero-point timestamps. The earliest adjustment zero-point timestamp in the comparison result where the time difference is greater than the instruction start delay time parameter is selected as the instruction reference zero-point timestamp. The instruction reference zero-point timestamp and the instruction start delay time parameter are subtracted to obtain the instruction issuance timestamp.

8. A control system for an automatic switching soft starter for a turbine air-cooled fan inverter fault, characterized in that, The system includes: The data acquisition module is used to acquire the fault codes and real-time operating data of the frequency converter within the current control cycle, and to perform fault classification judgment based on the fault codes and real-time operating data to obtain the fault level of the current control cycle. The inverter stop control module is used to generate an inverter stop command when the fault level meets the preset switching requirements; the inverter stop command is used to instruct the inverter to stop output and acquire instantaneous three-phase voltage data of the motor and instantaneous three-phase voltage data of the grid side in real time. The motor residual voltage calculation module is used to extract the residual voltage change characteristics of the motor based on the instantaneous data of the three-phase voltage of the motor, and obtain the real-time residual voltage amplitude. The instruction issuance time calculation module is used to record the sampling timestamp corresponding to the real-time residual voltage amplitude as the condition compliance timestamp when the real-time residual voltage amplitude is less than the preset safety threshold, predict the zero point time of the grid voltage based on the instantaneous data of the three-phase voltage on the grid side, obtain the zero point timestamp set, and calculate the instruction issuance timestamp based on the condition compliance timestamp and the zero point timestamp set. The soft-start command issuing module is used to generate a soft-starter startup command at the command issuance timestamp; the soft-starter startup command is used to instruct the soft-starter to start working.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images