Intelligent safety protection and fault prevention methods for low-voltage electrical appliances

Abstract

The present invention relates to the technical field of low-voltage electrical appliances, and discloses an intelligent safety protection and fault prevention method for low-voltage electrical appliances, including: obtaining the working state parameters of the target low-voltage electrical appliance, performing filtering processing, obtaining accurate working state parameters, inputting the accurate working state parameters into a fault analysis model, obtaining a fault analysis result, and if the fault analysis result is an operating fault, generating a safety protection scheme based on the fault type and the working state parameters. By implementing the present invention, a variety of protection strategies are adopted to comprehensively reduce the risk of functional failure caused by faults of intelligent low-voltage electrical appliances, extend the normal operation time of the equipment, and solve the problem in the related technology that the existing protection methods are difficult to effectively monitor and prevent these emerging faults due to the unique fault modes caused by the integration of multiple advanced modules in intelligent low-voltage electrical appliances.

Description

Technical Field

[0001] The invention relates to the field of low-voltage electrical appliances, and in particular to an intelligent safety protection and fault prevention method for low-voltage electrical appliances. Background Art

[0002] Low-voltage electrical appliances refer to electrical products that play the role of switching, protecting, controlling, regulating or converting in circuits with rated voltages of 1000V or less for AC and 1500V or less for DC. Traditional low-voltage electrical appliances include common circuit breakers, contactors, relays, fuses, knife switches, etc. Smart home appliances are a new form of low-voltage electrical appliances in the era of intelligence. They not only have the basic functions of traditional low-voltage electrical appliances, but also integrate advanced information technologies such as sensor technology, communication technology, artificial intelligence algorithms, etc., to achieve intelligent interaction, autonomous decision-making and remote monitoring.

[0003] Smart home appliances integrate complex electronic components and intelligent control modules. In addition to the overload, short circuit, leakage and poor contact failures of traditional low-voltage electrical appliances, there are also some special failure modes. For example, the computing power overload of the intelligent control chip. When processing a large number of complex instructions at the same time, the chip operation frequency rises sharply. If it is maintained above the normal frequency upper limit for a long time, it will cause the chip to overheat, affecting the accuracy of the operation and even causing chip damage. Another example is the communication module failure of the smart appliance. During the data transmission process, the packet error rate suddenly increases. If it exceeds the allowable error threshold, the appliance may lose connection with the external control system and be unable to receive or execute correct instructions. For some smart appliances with image recognition functions, the pixel grayscale deviation of the image sensor beyond the normal range will cause the recognition accuracy to decrease, affecting the normal functioning of the device's intelligent functions. Existing protection methods are difficult to effectively monitor and prevent these emerging faults, and innovative solutions are urgently needed. Summary of the invention

[0004] In view of this, the present invention provides an intelligent safety protection and fault prevention method for low-voltage electrical appliances to solve the problem in the related technology that the existing protection methods are difficult to effectively monitor and prevent the unique fault modes caused by the integration of multiple advanced modules in intelligent low-voltage electrical appliances.

[0005] The present invention provides an intelligent safety protection and fault prevention method for low-voltage electrical appliances, the method comprising: obtaining working state parameters of a target low-voltage electrical appliance, the working state parameters comprising image sensor parameters, communication module parameters, intelligent chip parameters and low-voltage electrical appliance parameters, the image sensor parameters comprising grayscale parameters, the communication module parameters comprising data packet error rate parameters, the intelligent chip parameters comprising chip temperature parameters and chip operation frequency parameters, the low-voltage electrical appliance parameters comprising electrical current parameters, electrical voltage parameters and electrical temperature parameters; filtering the working state parameters respectively to obtain accurate working state parameters, the accurate working state parameters comprising: accurate grayscale parameters, accurate data packet error rate parameters, accurate chip temperature parameters, and accurate chip operation frequency parameters. The precise working state parameters are input into the fault analysis model to obtain the fault analysis result. The fault analysis result is normal operation or operation failure. The operation failure includes at least one of the eight fault types, including chip computing power overload fault, communication module fault, image recognition fault, traditional overload fault, short circuit fault, leakage fault, poor electrical heat dissipation fault and re-analysis fault. If the fault analysis result is an operation failure, a safety protection plan is generated based on the fault type and working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection plan.

[0006] The intelligent safety protection and fault prevention method for low-voltage electrical appliances provided in this embodiment, firstly, through comprehensive multi-dimensional parameter collection covering image sensors, communication modules, smart chips and the low-voltage electrical appliances themselves, various key information of electrical appliance operation is captured in real time and accurately, and temperature and operation frequency monitoring are set for smart chips, communication modules monitor data packet error rate, and image sensors analyze grayscale parameters, etc. Compared with the traditional method of only focusing on basic electrical parameters, it greatly broadens the information acquisition surface, lays a solid foundation for subsequent accurate fault diagnosis and prevention, and effectively avoids the omission of potential faults due to missing information. Secondly, by tailoring filtering strategies for parameters with different characteristics, mean filtering is used for grayscale parameters, packet error rate parameters and chip operation frequency parameters to remove high-frequency random noise interference and ensure data stability; when the chip temperature and electrical temperature are close to the overheating warning, temperature-sensitive filtering is switched to accurately capture the trend of subtle temperature changes to avoid misjudgment due to temperature fluctuations. When the rate of change of electrical current is abnormal, noise suppression is enhanced to ensure that the current data truly reflects the load change. The accurate working state parameters after filtering input high-reliability data into the fault analysis model, significantly reducing the probability of misjudgment of faults caused by data noise and improving the reliability of the entire system. Then, by introducing a fault analysis model that integrates deep learning algorithms, the complex internal correlation between multivariate parameters and their change rates is fully explored. The model input layer reasonably allocates each precise parameter. The middle hidden layer uses the progressive operation of neurons to extract deep features from massive operating data. The output layer determines 8 types of faults with high accuracy, overcoming the limitations of traditional judgments based on simple thresholds. Whether it is a new type of fault such as the computing power overload of the smart chip, or a traditional overload or short circuit fault, it can be accurately identified, greatly improving the intelligence level and accuracy of fault diagnosis, and reducing the time and cost of manual troubleshooting. Finally, by customizing the generation of safety protection schemes based on different fault types, accurate protection is achieved. In the face of chip computing power overload, combined with chip temperature and task priority, non-critical tasks are reasonably suspended to release computing power, and the speed of the cooling fan is dynamically adjusted to ensure the stable operation of the chip from both software and hardware. When the communication module fails, the packet error rate is used to distinguish interference or hardware problems, and the channel is switched or restarted in a targeted manner and the communication strategy is optimized in combination with big data to ensure smooth data transmission. After the image recognition fault is repaired, the image sample self-learning optimization model during the fault period is used to improve the subsequent recognition accuracy. The intelligent safety protection and fault prevention method of low-voltage electrical appliances provided in this embodiment adopts a variety of protection strategies to comprehensively reduce the risk of functional failure caused by faults of intelligent low-voltage electrical appliances, extend the normal operation time of the equipment, and solve the problem that the existing protection methods are difficult to effectively monitor and prevent these emerging faults due to the unique fault mode caused by the integration of multiple advanced modules in the related technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0008] Figure 1 It is a flow chart of an intelligent safety protection and fault prevention method for a low-voltage electrical appliance according to an embodiment of the present invention. DETAILED DESCRIPTION

[0009] Low-voltage electrical appliances refer to electrical products that play the role of switching, protecting, controlling, regulating or converting in circuits with rated voltages of 1000V or less for AC and 1500V or less for DC. Traditional low-voltage electrical appliances include common circuit breakers, contactors, relays, fuses, knife switches, etc. Smart home appliances are a new form of low-voltage electrical appliances in the era of intelligence. They not only have the basic functions of traditional low-voltage electrical appliances, but also integrate advanced information technologies such as sensor technology, communication technology, artificial intelligence algorithms, etc., to achieve intelligent interaction, autonomous decision-making and remote monitoring.

[0010] Smart home appliances integrate complex electronic components and intelligent control modules. In addition to the overload, short circuit, leakage and poor contact failures of traditional low-voltage electrical appliances, there are also some special failure modes. For example, the computing power overload of the intelligent control chip. When processing a large number of complex instructions at the same time, the chip operation frequency rises sharply. If it is maintained above the normal frequency upper limit for a long time, it will cause the chip to overheat, affecting the accuracy of the operation and even causing chip damage. Another example is the communication module failure of the smart appliance. During the data transmission process, the packet error rate suddenly increases. If it exceeds the allowable error threshold, the appliance may lose connection with the external control system and be unable to receive or execute correct instructions. For some smart appliances with image recognition functions, the pixel grayscale deviation of the image sensor beyond the normal range will cause the recognition accuracy to decrease, affecting the normal functioning of the device's intelligent functions. Existing protection methods are difficult to effectively monitor and prevent these emerging faults, and innovative solutions are urgently needed.

[0011] The intelligent safety protection and fault prevention method for low-voltage electrical appliances provided in this embodiment, firstly, through comprehensive multi-dimensional parameter collection covering image sensors, communication modules, smart chips and the low-voltage electrical appliances themselves, various key information of electrical appliance operation is captured in real time and accurately, and temperature and operation frequency monitoring are set for smart chips, communication modules monitor data packet error rate, and image sensors analyze grayscale parameters, etc. Compared with the traditional method of only focusing on basic electrical parameters, it greatly broadens the information acquisition surface, lays a solid foundation for subsequent accurate fault diagnosis and prevention, and effectively avoids the omission of hidden fault risks due to missing information. Secondly, by tailoring filtering strategies for parameters with different characteristics, mean filtering is used for grayscale parameters, packet error rate parameters and chip operation frequency parameters to remove high-frequency random noise interference and ensure data stability; when the chip temperature and electrical temperature are close to the overheating warning, temperature-sensitive filtering is switched to accurately capture the trend of subtle temperature changes to avoid misjudgment due to temperature fluctuations. When the rate of change of electrical current is abnormal, noise suppression is enhanced to ensure that the current data truly reflects the load change. The accurate working state parameters after filtering input high-reliability data into the fault analysis model, significantly reducing the probability of misjudgment of faults caused by data noise and improving the reliability of the entire system. Then, by introducing a fault analysis model that integrates deep learning algorithms, the complex internal correlation between multivariate parameters and their change rates is fully explored. The model input layer reasonably allocates each precise parameter. The middle hidden layer uses the progressive operation of neurons to extract deep features from massive operating data. The output layer determines 8 types of faults with high accuracy, overcoming the limitations of traditional judgments based on simple thresholds. Whether it is a new type of fault such as the computing power overload of the smart chip, or a traditional overload or short circuit fault, it can be accurately identified, greatly improving the intelligence level and accuracy of fault diagnosis, and reducing the time and cost of manual troubleshooting. Finally, by customizing and generating safety protection solutions based on different fault types, accurate protection is achieved. In the face of chip computing power overload, combined with chip temperature and task priority, non-critical tasks are reasonably suspended to release computing power, and the speed of the cooling fan is dynamically adjusted to ensure stable operation of the chip from both software and hardware. When the communication module fails, the packet error rate is used to distinguish interference or hardware problems, and the channel is switched or restarted in a targeted manner and the communication strategy is optimized in combination with big data to ensure smooth data transmission. After the image recognition fault is repaired, the image sample self-learning optimization model during the fault period is used to improve the subsequent recognition accuracy. The intelligent safety protection and fault prevention method of low-voltage electrical appliances provided in this embodiment adopts a variety of protection strategies to comprehensively reduce the risk of functional failure caused by faults of intelligent low-voltage electrical appliances, extend the normal operation time of the equipment, and solve the problem that the existing protection methods are difficult to effectively monitor and prevent these emerging faults due to the unique fault mode caused by the integration of multiple advanced modules in the related technology.

[0012] According to an embodiment of the present invention, an embodiment of an intelligent safety protection and fault prevention method for low-voltage electrical appliances is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0013] In this embodiment, an intelligent safety protection and fault prevention method for low-voltage electrical appliances is provided, which is applied to an intelligent low-voltage electrical appliance system. The intelligent low-voltage electrical appliance system includes an image sensor module, a communication module, an intelligent chip module, and a low-voltage electrical appliance module. Figure 1 is a flow chart of an intelligent safety protection and fault prevention method for a low-voltage electrical appliance according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0014] Step S101, obtaining the working state parameters of the target low-voltage electrical appliance, the working state parameters include image sensor parameters, communication module parameters, smart chip parameters and low-voltage electrical appliance parameters, the image sensor parameters include grayscale parameters, the communication module parameters include data packet error rate parameters, the smart chip parameters include chip temperature parameters and chip operation frequency parameters, the low-voltage electrical appliance parameters include electrical current parameters, electrical voltage parameters and electrical temperature parameters.

[0015] Specifically, obtaining the working state parameters of the target low-voltage electrical appliance includes:

[0016] The image sensor is used to collect images. Each time a frame of image is collected, the sliding window algorithm is used to calculate the pixel grayscale deviation of the frame image, and the average pixel grayscale deviation of consecutive A frame images is taken as the grayscale parameter, where 9≤A≤10, and A is an integer.

[0017] Specifically, smart low-voltage electrical appliances rely on image recognition functions to achieve intelligent operation in many scenarios, such as smart camera monitoring in smart homes and gesture recognition of smart appliances. By accurately monitoring the grayscale deviation of pixels, changes in image quality can be captured sensitively, because grayscale deviation may be caused by factors such as fluctuations in lighting conditions, lens contamination, and image sensor failure. Taking the average value of multiple consecutive frames of images as the grayscale parameter can effectively smooth grayscale anomalies caused by accidental factors such as momentary light flickering and slight jitter, avoid misjudgment, and provide a reliable basis for subsequent judgment of image recognition-type failures. Limiting the value range of A is to strike a balance between data stability and real-time performance. If the A value is too small, the grayscale parameter is easily affected by instantaneous interference; if the A value is too large, although the stability is enhanced, it will delay the response to sudden image problems. 9-10 frames can not only reflect the overall trend within a certain period of time, but also detect abnormal changes in time.

[0018] The error rate of the data packet is collected by using the error monitor of the communication module. After each B data packets are received, the error rate of the data packet is calculated to obtain a data packet error rate parameter, where 50≤B≤150, and B is an integer.

[0019] Specifically, during the data transmission process of the communication module, the error monitor monitors the transmission status of each data packet in real time. After receiving each B data packet, the data packet error rate calculation program is started. The error monitor checks the B data packets one by one according to the CRC cyclic redundancy check algorithm and counts the number of error data packets. The number of error data packets is divided by B to obtain the data packet error rate at the current stage, which is output as the data packet error rate parameter. For example, if 100 data packets are received and 5 data packets are found to be erroneous after verification, then the data packet error rate is 5÷100=5%. As the data continues to be transmitted, this calculation process is repeated periodically to update the data packet error rate parameter in real time. In the intelligent low-voltage electrical system, the communication module is responsible for the data interaction between the device and the external control system, cloud server or other intelligent devices, and its stability is crucial. In different operating scenarios, the frequency and amount of data transmission vary. Setting the value range of 50≤B≤150 can not only adapt to the rapid monitoring of the error rate when a small number of key instructions are transmitted in a short period of time, but also provide a statistically significant data packet error rate in scenarios such as system updates and remote monitoring data transmission. Through periodic calculations, communication anomalies caused by signal interference, degradation of communication channel quality, aging of communication module hardware, etc. can be discovered in a timely manner, providing key indicators for judging communication module type failures, ensuring the reliability of data transmission, and avoiding electrical appliance misoperation or fault diagnosis deviations caused by erroneous data.

[0020] The temperature sensor is used to collect the chip temperature value based on the first sampling frequency, and the collected chip temperature value is differentially processed to obtain the chip temperature parameter. The first sampling frequency is once every 0.5 seconds.

[0021] Specifically, the temperature sensor for the smart chip quickly and frequently collects the temperature value of the chip surface or key parts at a first sampling frequency of once every 0.5 seconds. Since the chip temperature may change rapidly, especially when the high-load operation and heat dissipation conditions fluctuate, high-frequency sampling can capture subtle temperature changes. After each temperature value is collected, differential processing is immediately performed, that is, the difference between the current temperature value and the temperature value at the previous sampling moment is calculated to obtain the chip temperature change rate. The current temperature value and the corresponding temperature change rate are output as chip temperature parameters. For example, the chip temperature collected at time t1 is 60°C, and the temperature collected at time t2 after 0.5 seconds is 62°C. Then the temperature change rate is (62-60) ÷ 0.5 = 4°C / s, and the chip temperature parameter includes two key information of 62°C and 4°C / s. The smart chip is the core control unit of the intelligent low-voltage electrical appliance, and its performance is closely related to temperature. Excessive temperature will cause the chip computing performance to decline, shorten its life, or even directly damage it. The sampling frequency of 0.5 seconds can timely track the temperature dynamics of the chip under different workloads and heat dissipation environments, and provide real-time data support for early warning of chip computing power overload, poor heat dissipation and other faults. By simultaneously obtaining the temperature value and the rate of change, not only can the current thermal state of the chip be known, but also the rising or falling trend of the temperature can be judged, and the cause of abnormal chip temperature can be analyzed more accurately. For example, continuous temperature rise may indicate computing power overload, and sudden temperature changes may be related to sudden failures of the cooling system, so that targeted protection measures can be taken.

[0022] The chip operation frequency is collected based on the second sampling frequency by using a frequency counter, and the chip operation frequency is first-order differentiated to obtain the chip operation frequency parameter, wherein the second sampling frequency is once per second.

[0023] Specifically, the frequency counter monitors the chip operation frequency at a second sampling frequency of once per second. During the operation of the chip, the operation frequency will change dynamically according to the complexity and number of the tasks performed. The frequency counter records the current operation frequency value of the chip every 1 second, and then performs a first-order difference operation on the frequency value, that is, calculates the difference between the current sampling frequency and the previous sampling frequency to obtain the chip operation frequency change rate. The chip operation frequency value and its change rate together constitute the chip operation frequency parameter. For example, at two consecutive 1-second sampling moments, the chip operation frequency is measured to be 2GHz and 2.2GHz respectively, then the operation frequency change rate is (2.2-2) ÷ 1 = 0.2GHz / s, and the chip operation frequency parameters are 2.2GHz and 0.2GHz / s. Monitoring the chip operation frequency and its change rate is crucial to determine the chip computing power usage and whether it is overloaded. When intelligent low-voltage electrical appliances perform complex intelligent tasks (such as multi-task parallel processing, high-definition image real-time analysis, etc.), the chip operation frequency will increase significantly. Through a sampling frequency of once per second, the dynamic changes of the chip operation frequency can be effectively captured without excessively increasing the system burden. The frequency change rate can reflect the increase or decrease trend of the chip's computing power demand. When combined with parameters such as chip temperature and system load rate, it can more accurately determine whether the chip is at risk of computing power overload, and take protective measures such as suspending non-critical tasks in advance to ensure stable operation of the chip.

[0024] The current sensor is used to collect the working current of the electrical appliance based on the third sampling frequency to obtain the current parameters of the electrical appliance, and the third sampling frequency is once every 0.1 seconds.

[0025] Specifically, the current sensor monitors the working current of the low-voltage electrical appliance in real time at a third sampling frequency of once every 0.1 seconds. Since the change in the working current of the electrical appliance can directly reflect its load state and operating conditions, high-frequency sampling can capture instantaneous current fluctuations. The current sensor directly outputs each current value collected as an electrical appliance current parameter. At the same time, the system can quickly calculate the current change rate based on the current values ​​of adjacent sampling points for subsequent analysis. For example, at two consecutive 0.1 second moments, the measured current values ​​are 5A and 5.3A respectively, then the current change rate is (5.3-5) ÷ 0.1 = 3A / s, and these two values ​​will be involved in the subsequent fault judgment process. For low-voltage electrical appliances, current is one of the key operating parameters. Traditional faults such as overload and short circuit, as well as some new faults caused by abnormal load fluctuations, will be clearly reflected in the current. The sampling frequency of 0.1 seconds can keenly capture abnormal conditions such as large current shocks at the moment of motor startup and sudden increase in current of electrical appliances due to faults, providing the most direct data basis for timely judgment of traditional overload faults and short circuit faults. The calculation of the current change rate helps to distinguish the normal gradual change of current from the sudden change caused by a fault, thus improving the accuracy of fault diagnosis.

[0026] The voltage sensor is used to collect the working voltage of the electrical appliance based on a fourth sampling frequency to obtain the voltage parameter of the electrical appliance, and the fourth sampling frequency is once every 0.1 seconds.

[0027] Specifically, the voltage sensor also collects the working voltage of the electrical appliance at the fourth sampling frequency of once every 0.1 seconds. Similar to current monitoring, the voltage sensor obtains the voltage value of the electrical appliance access terminal in real time and outputs it as the electrical appliance voltage parameter. At the same time, the system calculates the voltage change rate based on the continuously sampled voltage data to reflect the voltage fluctuation. For example, if the voltage values ​​at two adjacent 0.1 second sampling moments are 220V and 218V, respectively, then the voltage change rate is (218-220) ÷ 0.1 = -2V / s. These voltage parameters and current parameters cooperate with each other to comprehensively judge the operating status of the electrical appliance. Stable voltage supply is a prerequisite for the normal operation of low-voltage electrical appliances. Abnormal voltage fluctuations may be caused by factors such as power grid fluctuations and failures in the internal voltage stabilization circuit of the electrical appliance. High-frequency sampling of 0.1 seconds ensures that small changes in voltage can be detected in time, whether it is the risk of electrical appliance shutdown caused by a sudden drop in voltage or potential power supply problems implied by a slow rise in voltage. They can be accurately captured. The voltage change rate combined with the current parameters can more comprehensively analyze the operating conditions of electrical appliances. For example, when judging a leakage fault, there is a correlation between voltage fluctuations and leakage current changes. Collaborative analysis can improve the reliability of fault diagnosis.

[0028] The temperature of a preset part of the electrical appliance is collected by using a temperature sensor based on a fifth sampling frequency to obtain a temperature parameter of the electrical appliance, and the third sampling frequency is once per second.

[0029] Specifically, the temperature sensor collects the temperature near key heating components, heat sinks and other parts at the fifth sampling frequency of once per second. The collected temperature value is directly output as the temperature parameter of the electrical appliance, and the system will also calculate the temperature change rate for subsequent fault analysis. For example, the temperature of the electrical heat sink is collected at a certain moment as 50°C, and 1 second later it is collected as 52°C, then the temperature change rate is (52-50) ÷ 1 = 2°C / s, and the electrical temperature parameters include 52°C and 2°C / s, which provide a basis for judging the poor heat dissipation type fault of the electrical appliance. Monitoring the temperature of key parts of electrical appliances is very important for evaluating their heat dissipation performance and overall operating status. When the electrical appliance is running for a long time, working at high load or the heat dissipation system fails, the temperature of specific parts will rise. The sampling frequency of once per second can continuously track the temperature changes of these parts without excessively occupying system resources, and the temperature change rate can intuitively reflect the speed of temperature rise or fall. When combined with parameters such as chip temperature and operating current, it can accurately determine whether the electrical appliance has a problem of poor heat dissipation, and take measures such as strengthening heat dissipation and reducing load in time to avoid damage to the electrical appliance due to overheating.

[0030] Step S102, filtering the working state parameters respectively to obtain accurate working state parameters, which include: accurate grayscale parameters, accurate data packet error rate parameters, accurate chip temperature parameters, accurate chip operation frequency parameters, accurate electrical current parameters, accurate electrical voltage parameters and accurate electrical temperature parameters.

[0031] Specifically, the above step S102 includes:

[0032] The grayscale parameters, data packet error rate parameters and chip operation frequency parameters are subjected to mean filtering to obtain accurate grayscale parameters, accurate data packet error rate parameters and accurate chip operation frequency parameters respectively.

[0033] Specifically, the grayscale parameters are easily affected by high-frequency random noises such as instantaneous changes in illumination and slight jitters of image acquisition devices. The use of mean filtering can effectively smooth these noises, so that the obtained accurate grayscale parameters can better reflect the real grayscale change trend of the image, and provide a reliable basis for the subsequent accurate judgment of image recognition faults. Because the grayscale deviations caused by these instantaneous interferences are often isolated and high-frequency, they can be "averaged out" by taking the average value over a period of time, highlighting the grayscale changes caused by factors such as the image's own content and the gradual change of ambient light. During the data transmission process, the packet error rate occasionally has individual abnormal values ​​due to factors such as instantaneous electromagnetic interference and short-term channel congestion. Mean filtering uses error rate samples of multiple consecutive time periods to average, avoiding misjudgment of communication module failures due to single abnormal interference. Only when the error rate is generally high over a period of time, it indicates that there may be a stable communication problem, ensuring the accuracy of communication fault judgment. Although the chip operation frequency is relatively stable, it may also have instantaneous peaks or troughs during task switching and short-term system resource competition. The mean filter processes the frequency value and its rate of change separately, which can not only remove the frequency measurement fluctuations caused by these short-term factors, but also retain its overall change trend, so that the precise chip operation frequency parameters can truly reflect the dynamic changes in the chip computing power requirements, and assist in judging faults such as chip computing power overload.

[0034] The chip temperature parameters are filtered. When it is detected that the chip temperature in the chip temperature parameters exceeds the first preset overheat warning temperature range or the chip temperature change rate exceeds the first preset overheat warning temperature change rate range, it is switched to the temperature-sensitive filtering mode to obtain accurate chip temperature parameters.

[0035] Specifically, the temperature monitoring module built into the system compares the chip temperature parameters with the preset first overheat warning temperature in real time, which is set as T_warn. This value is pre-set according to the chip model, heat dissipation design and reliability requirements. When the chip temperature meets T_warn-5℃≤T_chip≤T_warn+5℃, or by continuously monitoring the chip temperature, the temperature difference between two adjacent sampling moments is calculated. If the cumulative temperature change is greater than 3℃ within 2 seconds, that is, 4 consecutive sampling points, the filter mode switching instruction is immediately triggered. At this time, the temperature-sensitive filtering algorithm is enabled. Compared with conventional filtering, this algorithm will greatly reduce the filtering strength of high-frequency noise and increase the sensitivity to small temperature changes. For example, when using low-pass filtering, the cutoff frequency is reduced to allow more low-frequency signals that reflect the subtle rise or fall trend of temperature to pass, accurately capture the subtle changes of chip temperature when it is close to overheating, and output accurate chip temperature parameters. Chip temperature is a key indicator for judging the working state of the chip and whether it is facing the risk of overheating. In the normal working temperature range, a certain degree of filtering can remove measurement noise and ensure data stability. However, when the chip temperature approaches the overheat warning temperature, even a slight temperature rise may indicate that the heat dissipation problem is aggravated and the risk of computing power overload is increased. At this time, if conventional filtering is continued, key temperature change details may be filtered out, resulting in failure to detect the deterioration of the chip's thermal state in time. The temperature-sensitive filtering mode adjusts the filtering parameters to focus on temperature trend changes, ensuring accurate monitoring of temperature dynamics under the critical thermal state of the chip, providing an accurate basis for taking heat dissipation enhancement measures in advance, adjusting chip loads, and other operations, to prevent chip damage due to overheating.

[0036] The electrical current parameters are filtered. When it is detected that the electrical current change rate in the electrical current parameters exceeds the preset current warning change rate range, the noise suppression intensity value is switched to the preset value to obtain accurate electrical current parameters.

[0037] Specifically, the current sensor collects the working current of the appliance in real time at a sampling frequency of once every 0.1 seconds and calculates the current change rate. The system continuously monitors the current change rate. Once it detects that the working current of the appliance is greater than the preset current warning change rate range, it immediately sends instructions to the filter module to adjust the high-frequency noise suppression parameters. The filter originally used for current parameter filtering will dynamically adjust its parameters such as high-frequency cutoff frequency and attenuation coefficient according to the preset value. For example, the high-frequency cutoff frequency is reduced to enhance the attenuation ability of high-frequency noise, so that the current signal can be filtered to highlight the stable current change trend, remove the current spike noise caused by factors such as instantaneous switching of electromagnetic components inside the appliance and high-frequency interference of the power grid, and output accurate appliance current parameters. The current of the appliance has a relatively stable change law during normal operation, such as the current rises slowly when the motor starts smoothly and the current fluctuation is small when the load of the appliance is stable. However, when a fault occurs, such as a short circuit, severe overload or abnormal load changes caused by some internal component failures, the current change rate will increase sharply. At this time, the ordinary filtering setting may not be able to effectively eliminate the strong high-frequency noise interference caused by the fault, resulting in distortion of the current parameters and affecting the accuracy of fault judgment. By dynamically enhancing the high-frequency noise suppression strength, it is possible to accurately restore the true current change trend when the current changes suddenly, and quickly and accurately determine whether there are traditional overload faults, short-circuit faults, etc., to ensure the safe operation of electrical appliances.

[0038] The electrical appliance voltage parameters are filtered. When it is detected that the fluctuation amplitude of the electrical appliance voltage in the electrical appliance voltage parameters exceeds the preset voltage warning range for C consecutive sampling cycles or the voltage change rate exceeds the preset voltage change rate warning range, the adaptive filtering mode of the voltage signal is started to obtain accurate electrical appliance voltage parameters.

[0039] Specifically, the voltage sensor collects the working voltage of the electrical appliance every 0.1 seconds, and the system calculates the voltage fluctuation amplitude and voltage change rate in real time. When it is detected that the fluctuation amplitude of the electrical appliance voltage in the electrical appliance voltage parameter exceeds the preset voltage warning range for C consecutive sampling cycles or the voltage change rate exceeds the preset voltage change rate warning range, such as when it is monitored that the voltage fluctuation amplitude is greater than 0.5V for C consecutive sampling cycles, C is set between 4-6, according to the voltage stability requirements and the common cycle setting of power grid fluctuations, or the voltage change rate is greater than 2V / s, the adaptive filtering program is started, firstly, the spectrum analysis of the current voltage data is performed to identify the main interference frequency components, and then the filter parameters are dynamically adjusted according to the spectrum characteristics. For example, if it is found that the power frequency interference of 50Hz or 60Hz is strong, a notch filter is set near this frequency to enhance the filtering effect of power frequency interference and its harmonics; if there is high-frequency pulse interference, the high-pass filter parameters are optimized to accurately remove various interferences and output accurate electrical appliance voltage parameters. The working voltage of the electrical appliance is affected by many factors such as the power supply quality of the power grid and the performance of the internal voltage stabilization circuit of the electrical appliance, and the fluctuation situation is relatively complex. Under normal circumstances, voltage fluctuations are relatively small and regular. However, when the power grid fluctuates, the voltage stabilization circuit of the appliance fails, or there is external electromagnetic interference, the voltage will fluctuate with a large amplitude or change rapidly. By setting the continuous sampling cycle judgment conditions and adaptive filtering mode, on the one hand, it is possible to avoid the false start of complex filtering due to momentary accidental fluctuations and ensure the real-time performance of the system; on the other hand, when continuous abnormal fluctuations do occur, the adaptive filtering can be accurately adjusted according to the actual interference characteristics to ensure that the precise appliance voltage parameters truly reflect the power supply status of the appliance, assist in judging leakage faults, power supply abnormalities and other problems, and improve the system's ability to diagnose voltage-related faults.

[0040] The appliance temperature parameters are filtered. When it is detected that the appliance temperature in the appliance temperature parameters exceeds the second preset overheat warning temperature range or the appliance temperature change rate exceeds the second preset overheat warning temperature change rate range, it is switched to the temperature-sensitive filtering mode to obtain accurate appliance temperature parameters.

[0041] Specifically, similar to the processing of chip temperature, for the temperature of electrical appliances, the system collects the temperature of the preset parts of the electrical appliances at a sampling frequency of once per second, and compares the temperature parameters of the electrical appliances with the second overheat warning temperature in real time. The temperature parameter of the electrical appliances is set as T_warn_app, which is set according to the type of electrical appliances, heat dissipation design and safe operating temperature range. For example, when T_warn_app-5℃≤T_app≤T_warn_app+5℃ is satisfied, or the second preset overheat warning temperature change rate range is within 2 seconds, and the cumulative temperature change of 2 consecutive sampling points is greater than 3℃, it switches to the temperature-sensitive filtering mode. The filter reduces the attenuation of high-frequency noise and focuses on capturing subtle changes in temperature. For example, the parameters of the low-pass filter are adjusted to allow more low-frequency signals that reflect the trend of slow temperature rise or fall to pass through, and accurate electrical appliance temperature parameters are output to accurately reflect the thermal state changes of the electrical appliances. When the electrical appliances run for a long time, work under high load or have problems with the heat dissipation system, the temperature of key parts will rise. Similar to the chip temperature, moderate filtering in the normal temperature range ensures data stability, and when approaching the overheat warning temperature, subtle temperature changes contain key fault information. For example, in the early stage of heat sink blockage, the temperature of the appliance may rise slowly. If this change is ignored due to filtering, the discovery and treatment of poor heat dissipation faults may be delayed. The temperature-sensitive filtering mode ensures that when the appliance faces the risk of overheating, the temperature trend is accurately monitored, providing accurate temperature basis for timely taking heat dissipation improvement measures and adjusting the appliance load, avoiding damage to the appliance due to overheating and extending its service life.

[0042] Step S103, input the precise working state parameters into the fault analysis model to obtain the fault analysis result. The fault analysis result is normal operation or operation fault. The operation fault includes at least one of 8 fault types, including: chip computing power overload fault, communication module fault, image recognition fault, traditional overload fault, short circuit fault, leakage fault, poor electrical heat dissipation fault and re-analysis fault.

[0043] Specifically, the fault analysis model includes 1 input layer, 3 intermediate hidden layers and 1 output layer. The input layer has 14 neurons, the number of neurons in the intermediate hidden layer is 128, 64 and 32 respectively, and the output layer has 7 neurons. Step S103 includes:

[0044] Step S1031, distribute the precise grayscale parameters, precise data packet error rate parameters, precise chip temperature parameters, precise chip operation frequency parameters, precise electrical current parameters, precise electrical voltage parameters and precise electrical temperature parameters to the 14 neurons of the input layer according to a predetermined corresponding relationship.

[0045] Specifically, according to the pre-set correspondence rules between precise parameters and neurons, the system corresponds precise grayscale parameters, precise packet error rate parameters, precise chip temperature parameters, precise chip operation frequency parameters and their first-order differences, precise electrical current parameters and their change rates, precise electrical voltage parameters and their change rates, and precise electrical temperature parameters and their change rates to 14 neurons in the input layer, and sets 2 neurons to receive precise grayscale parameters, one to receive the current grayscale value, and the other to receive the grayscale change rate, so as to fully capture the image sensor data characteristics; for the precise chip temperature parameters and their change rates, 2 neurons are also assigned to correspond to them, and so on, to ensure that each parameter and its key change information have a dedicated information transmission channel to enter the neural network. These neurons are like data entrances, strictly following the predetermined layout, and orderly accepting various precise working status parameters, building an initial architecture for subsequent deep processing. Different working status parameters reflect the operating status of different modules and different levels of low-voltage electrical appliances, each of which contains unique fault clues. Through the carefully designed correspondence, the parameters are accurately distributed to the input layer neurons, which can ensure that the data is reasonably classified and orderly integrated from the source to avoid information confusion. This allows the neural network to perform targeted initial processing based on the characteristics of the data received by each neuron, allowing each key parameter to find the most suitable analysis path in the neural network, laying the foundation for accurately mining fault characteristics and improving the accuracy and efficiency of fault diagnosis.

[0046] Step S1032, control the first hidden layer to receive input layer data, use 128 neurons to perform weighted sum operation on the input layer data based on the initial weight value range, and then perform nonlinear transformation through the activation function to extract preliminary deep features between precise grayscale parameters, precise data packet error rate parameters, precise chip temperature parameters, precise chip operation frequency parameters, precise electrical current parameters, precise electrical voltage parameters and precise electrical temperature parameters. The preliminary deep features include change trend features and correlation relationship features.

[0047] Specifically, after the 128 neurons in the first hidden layer receive data from the input layer, they first perform a weighted summation operation on the input data according to the initial weight value range determined by the initialization method. Each neuron has a set of weight vectors, which are randomly initialized according to the appropriate value range generated by the Xavier algorithm based on the number of neurons in the input layer and the number of neurons in the current hidden layer during the model initialization phase. Taking the neurons that receive precise chip temperature parameters and their change rates as an example, their weight vectors will perform a dot product operation with the corresponding input data to achieve a weighted combination of the data and highlight the impact of certain key temperature change trends. After completing the weighted summation, the neurons are then nonlinearly transformed through the activation function. For the part where the weighted summation result is less than 0, the ReLU function sets its output to 0, retaining only the positive part, introducing nonlinear factors, so that the neurons can learn more complex relationships between the input data, convert the original linear combination data into a feature representation with a higher level of abstraction, and extract preliminary deep features covering the trend and correlation of each parameter change. These features are no longer simple parameter values, but high-dimensional vectors containing potential fault information. Xavier weight initialization can ensure the stability of gradients in the back propagation process in the early stage of neural network training, avoid the problem of gradient vanishing or gradient exploding due to improper weight initialization, and enable each neuron to smoothly learn effective information from the input data. The introduction of the ReLU activation function, on the one hand, accelerates the calculation process of neurons. Compared with traditional functions, ReLU is more efficient in calculation; on the other hand, its nonlinear characteristics break the limitations of linear models, allowing neurons to capture complex patterns in the data, such as the correlation between the rising trend of precise chip temperature parameters and the simultaneous increase in precise chip operation frequency. This correlation may indicate the risk of chip computing power overload. Through the conversion of the activation function, these potential fault clues can be initially excavated, providing strong support for subsequent deeper fault diagnosis. Among them, the Xavier algorithm is a weight initialization method widely used in the initialization process of neural networks. The core purpose is to ensure that in the early stage of model training, the gradient can be stably propagated between the layers of the neural network, avoiding the problem of gradient vanishing or gradient exploding, so that neurons can effectively learn features from the input data. The ReLU function, or the rectified linear unit function, outputs the input value itself when the input value is greater than 0; and when the input value is less than or equal to 0, the output is 0.

[0048] Step S1033, control the second hidden layer to randomly discard D% connections using 64 neurons when receiving the output of the first hidden layer, and use the optimization algorithm to allocate an adaptive learning rate to the connection weights based on the historical gradient square sum of the precise grayscale parameters, precise data packet error rate parameters, precise chip temperature parameters, precise chip operating frequency parameters, precise electrical current parameters, precise electrical voltage parameters and precise electrical temperature parameters, and converge to the optimal solution.

[0049] Specifically, the 64 neurons in the second hidden layer first perform the Dropout operation when receiving the output of the first hidden layer. According to the set ratio D%, some connections between neurons are randomly disconnected. For example, if D takes the value of 20, it means that about 20% of the connections will be randomly discarded, so that each neuron can only rely on the remaining connections to receive input data in this training iteration. D is generally taken as 18≤D≤22, and is fine-tuned during the model training process according to the need to prevent overfitting. The Dropout process occurs dynamically during the forward propagation of each training batch, and the connections discarded between different batches are different. Then, the historical gradient square sum of the precise grayscale parameters, precise packet error rate parameters, etc. of the Adagrad optimization algorithm is used to assign an adaptive learning rate to the connection weight. During the model training process, the Adagrad algorithm will continuously record the gradient update history of each parameter. For parameters that are frequently updated and have large gradient changes, its learning rate will be automatically reduced to make its steps smaller in subsequent updates to avoid skipping the optimal solution; while for parameters that are slowly updated and have stable gradients, a relatively large learning rate is maintained to accelerate convergence. Through such adaptive adjustments, neurons can adjust weights more accurately according to the dynamic characteristics of input data, further explore more discriminative fault features, and improve the model's ability to identify complex fault patterns. The application of Dropout technology effectively prevents excessive dependence between neurons and avoids overfitting. In neural network training, as the number of layers increases, neurons may over-adapt to the specific patterns of training data, resulting in a decrease in generalization ability. Dropout forces neurons to learn different feature combinations by randomly discarding connections, enhancing the robustness of the model so that it can better cope with fault diagnosis needs in various actual operating scenarios. The adaptive learning rate mechanism of the Adagrad optimization algorithm fully considers the changing characteristics of different parameters during the training process, so that when facing complex and changeable low-voltage electrical operation data, the model can dynamically adjust the learning strategy according to the importance and frequency of changes of parameters, quickly converge to the optimal fault diagnosis model parameter configuration, efficiently extract fault features, and accurately identify fault types.

[0050] Step S1034, controlling 32 neurons of the third hidden layer to normalize the E bytes of the second hidden layer data to obtain a hidden layer output feature vector.

[0051] Specifically, the 32 neurons in the third hidden layer use batch normalization technology to normalize the data output by the second hidden layer. The data volume is E bytes, where E is generally 32. According to the data processing requirements of the model's intermediate layer and the optimization setting of computing resources, in the forward propagation process of each training batch with 32 bytes, the mean and variance of the batch data are first calculated, and then each data point is standardized to make its mean 0 and variance 1. Taking the processing of feature data related to electrical voltage as an example, the electrical voltage parameters and their change rates collected in different batches may have large numerical differences due to factors such as power grid fluctuations and electrical load switching. Batch Normalization can eliminate these internal covariate offsets and unify the data into a relatively stable distribution range. After normalization, the distribution of data becomes more regular, so that neurons can learn the deep features in the data more stably and efficiently during subsequent operations, avoid the instability of the learning process caused by fluctuations between data batches, and further transform the deep features into more abstract high-level feature vectors that can better distinguish different fault types, providing a solid input foundation for accurate fault judgment in the output layer. BatchNormalization plays a crucial stabilizing role in neural network training. As the model depth increases, the distribution of the middle layer data will gradually change. This phenomenon is called internal covariate shift, which will make model training more difficult and convergence slower. By normalizing each batch of data, the data always maintains relatively consistent distribution characteristics before entering the neuron, which not only speeds up the model training process, but also improves the generalization ability of the model, allowing the model to focus on learning the essential characteristics of the fault from the data rather than being affected by random fluctuations between data batches, thereby improving the diagnostic accuracy of low-voltage electrical faults under various working conditions and ensuring that the output high-level feature vector has stronger fault discrimination.

[0052] Step S1035, controlling the seven neurons of the output layer to receive the hidden layer output feature vector output by the third hidden layer, and using an activation function to convert the neuron input value into a normalized vector representing the probability of each fault type.

[0053] Specifically, the 8 neurons in the output layer receive the high-level feature vector output by the third hidden layer and process it using the Softmax activation function. The Softmax function converts the input value of each neuron into a probability value between 0 and 1, and the sum of the probability values ​​output by all 8 neurons is 1. For example, assuming that the model receives a set of high-level feature vectors containing features such as rapid rise in chip temperature, chip operation frequency approaching the upper limit, and increased image grayscale deviation, after calculation by the Softmax function, a probability vector similar to [0.05, 0.03, 0.1, 0.02, 0.01, 0.01, 0.78, 0] may be obtained, among which the neuron corresponding to the chip computing power overload fault has the highest probability (0.78), and the last neuron represents a fault-free situation, with a probability of 0 indicating that a fault is currently detected. If a probability vector of [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.2] is obtained, it means that the current operating state is likely to be fault-free, and the probability of each fault type is relatively low. Theoretically, for the classification task including 8 fault types and no-fault state, using 9 neurons is a more intuitive design method. Each neuron corresponds to a specific state, and the probability distribution of each state is output through the Softmax function. However, in the current design, 8 neurons are used, and the no-fault state is regarded as a special "background" category. When the probability vector output by the output layer neuron does not have a probability corresponding to any fault type higher than the set threshold, it is assumed to be in a no-fault state. This design simplifies the model structure to a certain extent, reduces the amount of calculation and number of parameters associated with a neuron, and helps to improve the training efficiency and running speed of the model. This streamlined design is more advantageous in low-voltage electrical fault diagnosis scenarios with limited resources or high real-time requirements. The Softmax activation function can convert the output of the neural network into a reasonable probability distribution, clearly presenting the relative probability of various fault types and no-fault states. During the model training process, by comparing with the real fault label, the cross entropy loss function is used to measure the difference between the model prediction and the actual situation, prompting the model to continuously adjust the weight so that the output fault probability vector is as close to the real fault type and no-fault state as possible, achieving accurate fault classification and improving the accuracy and reliability of fault diagnosis.

[0054] Step S1036, obtaining a fault analysis result according to the normalized vector of each fault type probability and a preset fault judgment threshold.

[0055] The system pre-sets a preset fault judgment threshold, which is usually 0.5. The preset fault judgment threshold can be adjusted within a certain range according to the requirements of the actual application scenario for fault sensitivity. For example, for scenarios that require high reliability and do not allow potential faults to be missed, the threshold can be appropriately lowered. When the probability corresponding to a certain fault type in the output layer is greater than the threshold, it is preliminarily determined that this type of fault has occurred. For example, if the probability of a chip computing power overload fault is 0.78, which is greater than 0.5, it will be included in the list of suspected faults. If the probability of the last neuron is greater than 0.5, it is directly judged as no fault. If there are multiple fault types with probabilities greater than the threshold, further integrated analysis is required. Considering the correlation and exclusivity between faults, the fault association rules mined from expert knowledge and historical fault data are used for judgment. For example, short-circuit faults and traditional overload faults may occur at the same time in some cases, but if a short circuit occurs, its current change characteristics will mask the overload characteristics. At this time, the short-circuit fault is mainly judged; for image recognition faults and electrical heat dissipation faults, they are usually independent of each other. If the probability of both is higher than the threshold, they are marked as different faults, and finally a complete and accurate fault analysis result is formed, including possible single faults or multiple concurrent faults. Setting the fault judgment threshold provides a clear judgment standard, converting the probability vector output by Softmax into an actual fault judgment decision, so that the model can quickly determine whether there is a fault and the type of fault in a complex fault probability distribution. For the case where the probability of multiple faults is higher than the threshold, considering the fault correlation and exclusivity for integrated analysis is based on the physical characteristics and fault occurrence mechanism of the actual operation of low-voltage electrical appliances. There may be causal relationships, mutual influences or independent occurrences between different faults. Judging based solely on the probability threshold may lead to misjudgment or missed judgment. By introducing expert knowledge and historical data mining rules, complex fault scenarios can be analyzed more accurately to ensure that the fault analysis results are both comprehensive and accurate, providing a reliable basis for the subsequent targeted safety protection measures.

[0056] Step S104: if the fault analysis result is an operating fault, a safety protection plan is generated based on the fault type and the working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection plan.

[0057] In some optional embodiments, when the fault type includes a chip computing power overload fault, the data packet task priority data is obtained, the pause task data is determined based on the data packet task priority data, the dynamic speed data is calculated based on the precise chip temperature parameters and the preset fan speed formula, a safety protection plan is generated based on the pause task data and the dynamic speed data, and the intelligent low-voltage electrical system is controlled to execute the safety protection plan.

[0058] Specifically, during the operation of the intelligent low-voltage electrical system, when the fault analysis model determines that a chip computing power overload fault occurs, the system immediately starts the task management module to deeply monitor all data packet tasks currently being executed and queued for execution. When each data packet task enters the system, it is assigned different priority labels according to pre-set rules. These rules comprehensively consider factors such as the importance of the task source, the requirements for real-time performance, and the urgency of task execution. By querying the task priority database in the task management module, the priority data of all data packet tasks currently in operation and in the queue can be quickly and accurately obtained. Based on the acquired data packet task priority data, the system uses an intelligent decision-making algorithm to determine the suspended task. First, a computing power threshold interval is set, and the lower and upper limits of the threshold are dynamically adjusted according to the chip's performance specifications, current temperature conditions, and historical computing power overload recovery experience. When the chip computing power is determined to be overloaded, the total computing power required by the running tasks is compared with the currently set computing power threshold upper limit. If the threshold is exceeded, the suspended objects are screened starting from the lowest priority task. In the screening process, the relevance of the tasks is considered to avoid errors or interruptions in other tasks closely related to it due to the suspension of a task. For example, for some subtasks belonging to the same data processing flow, if one of them is suspended, the data of the entire process may be missing or confused. At this time, after comprehensive evaluation, the tasks that can be safely suspended and have the least impact on the overall system are determined as suspended task data, and output in the form of a list for subsequent steps to call. The system's built-in temperature sensor continuously and accurately monitors the chip temperature and outputs accurate chip temperature parameters. At the same time, a preset fan speed formula is stored: fan speed = f (chip temperature, ambient temperature, heat sink temperature gradient, etc.), where chip temperature is the key variable. When the chip computing power is detected to be overloaded and the temperature has an upward trend, the precise chip temperature parameters are substituted into the formula, combined with the current ambient temperature and auxiliary parameters such as the heat sink temperature gradient, and the dynamic speed data that can effectively control the chip temperature is quickly calculated. The heat sink temperature gradient is calculated by arranging temperature sensors at different positions on the heat sink. For example, if the chip temperature rises to 70°C, the ambient temperature is 30°C, and the heat sink temperature gradient is 5°C / cm, it is calculated after substituting it into the formula that the fan should be increased to 3000RPM to enhance the heat dissipation effect and avoid chip damage due to overheating. The system generates a detailed safety protection plan based on the determined pause task data and the calculated dynamic speed data. The plan first sends instructions to the task scheduling unit of the smart chip, and suspends the corresponding data packet tasks in order according to the pause task data list to release the chip computing resources. At the same time, it sends instructions containing dynamic speed data to the cooling fan control module to drive the fan to run at the specified speed. During the execution process, the chip temperature, computing power load, and the recovery conditions of the paused tasks are continuously monitored.For example, when the chip temperature drops to a safe range and the computing load is lower than the set threshold, the suspended tasks are gradually resumed according to the task priority and pause order, ensuring that the system smoothly transitions from the computing overload state to the normal operating state, ensuring the normal function of electrical appliances and stable data transmission throughout the process.

[0059] In some optional embodiments, when the fault type includes a communication module type fault, the communication module fault type is determined based on the precise data packet error rate parameters and a preset error rate formula, and the communication module fault types include interference type faults and hardware type faults; a safety protection plan is generated based on the communication module fault type and the precise data packet error rate parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection plan.

[0060] Specifically, when the fault analysis model determines that a communication module fault occurs, the system quickly focuses on the precise packet error rate parameter. This parameter is continuously collected, processed and updated by the error monitor of the communication module according to specific sampling rules. At the same time, the system has built-in preset error rate formulas, which are based on the analysis and induction of a large amount of experimental data of the communication module under different working conditions and the derivation of the communication theory model. For interference type fault determination, the formula is: interference index = g (accurate packet error rate change frequency, error rate peak time distribution, background electromagnetic noise intensity, etc.), where the precise packet error rate parameter is one of the key inputs. The system monitors the dynamic changes of the packet error rate in real time, combines the background electromagnetic noise intensity in the environment, and analyzes the time period law of the error rate peak. The background electromagnetic noise intensity is collected by a special electromagnetic sensor. For example, if the packet error rate fluctuates frequently within a certain period of time, and the fluctuation amplitude is highly coincident with the period of strong electromagnetic interference in the environment, and the error rate peak is mostly concentrated in the high-frequency start-stop period of the external device, a higher interference index is calculated by substituting it into the preset formula, and it is initially determined to be an interference type fault. For hardware fault determination: hardware fault probability = h (duration of continuous high error rate, error rate variation with temperature, port signal attenuation, etc.). When the precise data packet error rate remains high for a long time and has no obvious correlation with the temperature change of the communication module, and the port signal attenuation is far beyond the normal range, a higher hardware fault probability is calculated by substituting it into the formula, which tends to be determined as a hardware fault. The port signal is monitored in real time using the signal strength detection module. By comprehensively using these formulas and multi-source data, the fault type of the communication module can be accurately distinguished. The causes of communication module failure are complex and diverse. Accurately distinguishing between interference-type and hardware-type failures is a prerequisite for taking targeted protection measures. It is difficult to determine the root cause of the fault based solely on the data packet error rate value. The preset error rate formula integrates many key factors, comprehensively considers the dynamic changes of the error rate, environmental impact, and the characteristics of the hardware itself, avoids misjudgment, and provides guidance for subsequent effective fault repair. The communication module is susceptible to interference. After accurately judging it as an interference-type fault, shielding, channel adjustment and other measures can be taken as a priority, rather than blindly replacing hardware, saving costs and time.

[0061] If it is determined to be an interference-type fault, the system first starts the channel optimization strategy. According to the real-time monitoring of the electromagnetic environment spectrum information (collected by the spectrum analyzer module), combined with the distribution of the packet error rate in different frequency bands, the system intelligently selects a backup channel with less interference for data transmission switching. For example, if it is found that the current channel is seriously interfered by a fixed frequency, and the packet error rate in another frequency band is extremely low, the communication module is quickly instructed to switch to the backup channel to ensure the stability of data transmission. At the same time, the communication protocol parameters are dynamically adjusted to reduce the data transmission rate and increase the check redundancy bit. By reducing the rate, the risk of data loss caused by interference is reduced, and the error correction capability of the data is improved by increasing the check redundancy bit, such as upgrading the original CRC-16 check to CRC-32 to enhance the reliability of data in an interference environment. Continuously monitor the precise packet error rate parameters. When the error rate drops to the normal range, gradually restore the original channel and communication protocol parameters to ensure the stable operation of the communication module.

[0062] If it is determined to be a hardware fault, the hardware self-test program is immediately triggered, and the built-in self-test circuit of the communication module detects the key hardware components one by one, and accurately locates the faulty component according to the preset hardware fault threshold. For example, if the signal output power of a certain transceiver chip is detected to be far below the normal threshold, combined with other related parameter abnormalities, the chip is determined to be the source of the fault. Once the faulty component is located, the system sends an alarm message to the operation and maintenance management module of the intelligent low-voltage electrical system, including the fault component number, fault type and detailed fault parameters, so that the operation and maintenance personnel can respond quickly and prepare to replace spare parts. At the same time, start the backup communication module or temporarily reduce the priority of the electrical function to ensure the limited transmission of key data and avoid major losses due to communication interruptions. After the operation and maintenance personnel complete the hardware replacement, the system automatically calibrates and tests the new hardware, and determines whether the hardware has returned to normal working state based on parameters such as the packet error rate. If normal, the system will resume full function operation.

[0063] In some optional embodiments, when the fault type includes an image recognition type fault, the calibration position is determined based on the precise grayscale parameters; a safety protection scheme is generated based on the calibration position and the precise grayscale parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme.

[0064] Specifically, when the fault analysis model determines that an image recognition fault has occurred, the system immediately retrieves and deeply analyzes the precise grayscale parameters. This parameter is calculated by the image sensor for each frame of image captured, and the sliding window algorithm is used to calculate the pixel grayscale deviation of the frame of image. The average pixel grayscale deviation of multiple consecutive frames of images is taken as the grayscale parameter, which is continuously generated to accurately reflect the changes in the grayscale characteristics of the image. The system has a built-in calibration position recognition model based on grayscale features, which is trained based on a large number of image grayscale data under known fault scenarios and standard grayscale templates during normal operation. The model compares the current precise grayscale parameters with various pre-stored standard grayscale patterns, and uses the pattern matching algorithm to find the area with the most significant difference. For example, if in a smart home security scenario, the grayscale value of a key area of ​​the monitoring screen is stable in a specific range under normal conditions, when an image recognition failure occurs, the grayscale value of the area changes suddenly due to abnormal lighting, lens contamination, etc. The model analyzes the precise grayscale parameters and combines the pixel coordinate information of the image to accurately locate the area with the largest grayscale abnormal change and determine it as the calibration position, which is usually marked in the form of a pixel coordinate range, such as a rectangular area from (x1, y1) to (x2, y2). Accurately locating the calibration position is the key starting point for repairing image recognition failures. Image recognition failures caused by different reasons often present local grayscale anomalies on the image. Through in-depth mining of precise grayscale parameters, the calibration position recognition model can quickly focus on the area where the root cause of the problem is located, avoid blind processing of the entire image, save computing resources, and improve fault repair efficiency. For example, in the smart access control system, the calibration position where the grayscale of the face recognition area is abnormal due to lens contamination can be quickly located, and subsequent targeted cleaning or correction operations can be accurately implemented to ensure normal recognition of the access control system. According to the determined calibration position, the system first adjusts the parameters of the image acquisition device in a targeted manner. If the calibration position is located in the edge area of ​​the image and the grayscale deviation shows uneven illumination, the system automatically controls the aperture size, exposure time and other parameters of the image sensor to optimize the illumination conditions in the area. For example, if the upper right corner of the image is found to be dark and the grayscale value is low, the image sensor is instructed to increase the aperture and extend the exposure time to increase the image brightness in the area and bring the grayscale value closer to the normal range. If the calibration position is concentrated in the key recognition area in the center of the image and the grayscale fluctuation indicates that there may be lens contamination, the system triggers the automatic cleaning device to clean the lens. After cleaning, the image is collected again to compare the precise grayscale parameters before and after cleaning to determine whether the fault has been repaired. At the same time, the threshold of the image recognition algorithm is dynamically adjusted in combination with the precise grayscale parameters of the overall image. For example, in the intelligent warehousing and logistics system, the grayscale changes in the cargo recognition area are caused by dust. After cleaning the lens, the grayscale threshold of the cargo contour recognition is appropriately relaxed or tightened according to the grayscale of the newly collected image to improve the recognition accuracy and ensure a smooth logistics process. During the entire process, the system monitors the image acquisition, processing and recognition effects in real time to ensure that the fault is effectively resolved and the image recognition function returns to normal.The safety protection scheme customized based on calibration position and precise grayscale parameters can directly hit the key points of image recognition failure. For grayscale anomalies in different locations and for different reasons, hardware parameter adjustment, automatic cleaning, algorithm optimization and other measures are taken to coordinate and repair the failure from multiple aspects to ensure the quality of image acquisition and the accuracy of subsequent recognition. Real-time monitoring ensures timely detection of problems and continuous optimization, so that the intelligent low-voltage electrical system can operate stably in image recognition-related application scenarios and avoid functional failure or misjudgment caused by image recognition failure.

[0065] In an optional manner of some embodiments, the fault period is determined based on precise grayscale parameters; the image samples during the fault period are marked with erroneous recognition areas to obtain marked image sample data; and the marked image sample data is used to train an image recognition model.

[0066] Specifically, the system continuously records the timestamp information of image acquisition and stores it synchronously with the precise grayscale parameters. When it is determined that an image recognition fault occurs, the fault period is precisely defined based on the mutation moment of the precise grayscale parameters and the moment of recovery. For example, if the precise grayscale parameters continue to fluctuate abnormally from time t1 and deviate from the normal grayscale range until the grayscale parameters return to stability at time t2, then the [t1, t2] time period is the fault period. For all image samples collected during the fault period, the image analysis algorithm is used for processing. The algorithm is based on deep learning image segmentation technology and is pre-trained to identify the misidentified areas in the image. It uses image samples during normal operation as positive samples and image samples during the fault period as negative samples to learn the differences in image features under different states. When processing images during the fault period, through feature extraction and comparison, the areas with large differences from normal image features that cause recognition errors are marked, such as marking the image with frames of different colors to generate labeled image sample data. These data contain the original image information, the coordinates of the misidentified areas, and related feature descriptions, providing rich materials for subsequent model training. By clearly identifying the fault period and marking the misidentification area, the key information when the image recognition fault occurs can be accurately captured. This information reflects the specific manifestations of the image problem caused by the fault, and provides an intuitive and detailed basis for in-depth analysis of the fault cause and optimization of the image recognition model. The generated labeled image sample data is divided into a training set, a validation set, and a test set according to a certain ratio, usually 70%, 15%, and 15% to ensure the scientificity and effectiveness of the model training. The training set is input into the existing image recognition model and iteratively trained using the back propagation algorithm. The model adjusts the internal neuron weights according to the features of the misidentification area in the labeled image sample data, and learns how to avoid misidentification in similar situations. For example, in the pet recognition model in the smart home scenario, if a large number of misjudgments occur during the fault period due to changes in the pet's posture, the model optimizes the neuron connection weights by learning the pet posture, light and shadow changes and other features in the labeled image samples, so that it can accurately recognize the next time it encounters similar postures and light and shadow conditions. After each round of training, the model is verified using the validation set, and the improvement of the model performance is observed by calculating evaluation indicators such as accuracy and recall. If the performance of the model on the validation set does not reach the expected improvement target, such as the accuracy rate is less than 2%, the model's hyperparameters are adjusted and the next round of training continues. After multiple rounds of iterative training, until the model reaches satisfactory performance indicators on the test set, such as an accuracy rate of more than 95%, it is considered that the model has been effectively optimized for image recognition failures and can be put into subsequent actual operation to significantly improve the accuracy and robustness of image recognition. Using labeled image sample data during the failure period to train the image recognition model is an effective way to learn optimization from practical failures.By allowing the model to face complex scenarios that lead to misidentification and adjusting its own parameters in a targeted manner, it is possible to continuously improve the model's ability to adapt to various potential fault factors, overcome image recognition problems caused by environmental changes, equipment aging, etc., ensure that the intelligent low-voltage electrical system always maintains a high image recognition accuracy during long-term operation, and expand its reliable application in different scenarios.

[0067] In an optional manner of some embodiments, when the fault type includes a leakage type fault, leakage current data, voltage phase change data, and quantum state collapse data collected by quantum sensors at at least three different positions on the target low-voltage electrical appliance are obtained, and the leakage point coordinate data is calculated through the leakage current data, voltage phase change data, quantum state collapse data and a pre-trained quantum positioning model; a safety protection plan is generated based on the leakage point coordinate data, and the intelligent low-voltage electrical appliance system is controlled to execute the safety protection plan.

[0068] Specifically, when the fault analysis model determines that a leakage fault has occurred, the system quickly activates the quantum sensor network deployed in the key parts of the target low-voltage electrical appliance. These quantum sensors are designed based on cutting-edge principles such as quantum tunneling and quantum entanglement, have ultra-high sensitivity, and can accurately capture changes in electrical signals at the microscopic level. At least three sensors are placed in areas prone to leakage risks such as the incoming line end of the appliance, near the main heating components, and at the grounding connection of the shell to ensure all-round monitoring. The sensor synchronously collects leakage current data at an extremely high sampling frequency, captures subtle current fluctuations in real time, and reflects the size and change trend of the leakage current; at the same time, using the built-in high-precision phase measurement module, accurately measures the voltage phase change data, and records the dynamic information of the voltage and current phase difference at different positions, providing key clues for analyzing the leakage path; in addition, with the quantum characteristics, the sensor can keenly perceive the quantum state collapse phenomenon. Once leakage occurs, the slight disturbance of the surrounding electric field will cause the instability of the quantum state. The sensor captures these quantum state collapse data, packages them together with the leakage current and voltage phase change data, and transmits them to the system core processor in real time through a high-speed data transmission channel. Traditional leakage detection methods often rely on conventional electrical sensors with limited accuracy, making it difficult to accurately detect faults in the early stages of leakage and when the leakage current is small. Quantum sensors use quantum effects to break through the sensitivity bottleneck of traditional detection, capture leakage signs at the microscopic level, and achieve comprehensive coverage with multi-location deployment to ensure that no potential leakage areas are missed. For example, in complex industrial electrical equipment, leakage may occur in internal hidden lines. Quantum sensors, with their high sensitivity and multi-location layout, can detect extremely weak leakage signals in advance, buy precious time for timely protection measures, and effectively prevent safety hazards caused by leakage accidents. The system pre-uses a large amount of simulated leakage experiments and actual leakage fault case data to build and train a quantum positioning model. The model is based on a deep learning architecture, integrating the principles of quantum mechanics and traditional electromagnetic theory, and can deeply explore the complex internal correlations between leakage current, voltage phase change, and quantum state collapse data. After receiving the above three types of data collected in real time, the data is first pre-processed, including normalization, denoising, and other operations to ensure data quality. The processed data is then input into the quantum positioning model, and the neural network in the model conducts a comprehensive analysis of multi-source data based on the training and learning patterns. For example, through the gradient change of the leakage current between sensors at different positions, combined with the distribution characteristics of the voltage phase difference, and the probability density distribution of quantum state collapse, complex mathematical algorithms (such as positioning algorithms based on multivariate regression and neural network back propagation optimization) are used to calculate the coordinate data of the leakage point in three-dimensional space, and accurately output in the form of (x, y, z). The positioning accuracy can reach the millimeter level, far exceeding traditional positioning methods. It is difficult to accurately locate the leakage point relying solely on a certain type of leakage data, because the leakage phenomenon is interfered by multiple factors in a complex electrical environment.The quantum positioning model integrates multiple quantum data, gives full play to the advantages of each data, and outlines the leakage characteristics from different angles. The leakage current gradient reflects the direction of the leakage path, the voltage phase difference assists in determining the location of the leakage branch, and the quantum state collapse data provides the details of the microscopic electric field disturbance. The three work together to break through the limitations of traditional positioning and achieve high-precision leakage point positioning. In the smart home scenario, when an electrical appliance leaks electricity, the leakage source can be found quickly and accurately, whether it is an internal circuit board or an external power cord. Accurate positioning facilitates timely repair, avoids the risk of electric shock to users, and ensures the safety of home electricity use. After obtaining accurate leakage point coordinate data, the system immediately generates a targeted safety protection plan. First, send a command to the power control module of the intelligent low-voltage electrical appliance to instantly cut off the power supply of the faulty electrical appliance to prevent the leakage accident from further expanding and ensure the safety of personnel and equipment. If the leakage point is located in a repairable part inside the appliance, such as a certain area of ​​the circuit board, the system automatically retrieves the pre-stored maintenance knowledge base, and generates detailed maintenance instructions based on the specific component location corresponding to the coordinates of the leakage point, including the component model that needs to be replaced, the list of maintenance tools, and the schematic diagram of the operation steps, which are pushed to the mobile phone of the operation and maintenance personnel through the intelligent terminal to assist them in quickly carrying out maintenance work; at the same time, the backup power supply is started to ensure the temporary power supply of other key parts of the system, maintain the operation of basic functions, and avoid the paralysis of the entire intelligent low-voltage electrical system due to leakage of a single appliance. For some leakage faults that are difficult to repair quickly on site, such as those involving large-scale damage to the core components of the appliance or requiring professional disassembly of the equipment, the system not only cuts off the power supply, but also sends early warning signals to the surrounding related appliances to notify them to adjust their operating status to avoid being affected by the leakage fault; at the same time, the detailed information of the leakage fault is uploaded to the cloud server, which is convenient for the remote expert team to conduct fault analysis and subsequent maintenance guidance. After the professional personnel arrive with appropriate tools, they will be thoroughly repaired to ensure that the whole process is orderly and the risk and impact of leakage are minimized. Customized safety protection solutions based on the coordinates of the leakage point achieve precise control from fault discovery to repair. Rapid power off is the primary measure to ensure safety and avoid the risk of electric shock; personalized maintenance instructions and auxiliary means are provided for leakage points with different repair difficulties, which not only facilitates on-site operation and maintenance, but also ensures that the system can still maintain some functions under complex faults and reduce losses. Uploading information to related electrical warnings and the cloud further expands the breadth and depth of fault response, and uses the power of multiple parties to coordinate and solve leakage problems, ensuring the stable and safe operation of intelligent low-voltage electrical systems in various scenarios.

[0069] In some optional methods of the embodiments, when the fault type includes a re-analysis type fault, the working state parameters, precise working state parameters, preliminary depth features and normalized vectors of the probability of each fault type within a preset time interval are obtained to generate expert library input data; the expert library input data is sent to an expert diagnosis center to obtain expert library output data; a safety protection plan is generated based on the expert library output data, and the intelligent low-voltage electrical system is controlled to execute the safety protection plan.

[0070] Specifically, when the fault analysis model determines that a reanalysis fault occurs, it means that the current fault characteristics are unclear or have a low match with the known fault mode, and it is difficult to quickly determine the root cause of the fault by relying on the conventional diagnostic process. At this time, the system automatically traces back to the preset time interval before the fault occurs, which is generally set according to the type of electrical appliance and the complexity of the operating environment. For example, industrial-grade complex equipment is set to 60 seconds before the fault to 30 seconds after the fault occurs, and relatively simple smart home appliances are set to 30 seconds before the fault to 15 seconds after the fault occurs. The working state parameters during this period are fully collected. These parameters cover the grayscale parameters derived from the image data collected by the image sensor, the data packet error rate parameters of the communication module, the temperature and operation frequency parameters of the smart chip, and the current, voltage, temperature parameters of the low-voltage electrical appliance itself, and other all-round operating information. At the same time, the system extracts the precise working state parameters obtained after filtering. These precise parameters remove the noise interference in the original data, and can more truly reflect the subtle changes in the operating state of the electrical appliance, providing a reliable basis for in-depth analysis. In addition, it also includes the preliminary deep features extracted by the first hidden layer of the fault analysis model through a series of operations such as Xavier initialization and ReLU function. These features contain the changing trends and correlations between the parameters, which are the key intermediate information for further mining fault clues. Finally, the normalized vector of the probability of each fault type output by the fault analysis model is included. This vector intuitively presents the model's possible judgment of different fault types. Although it is currently judged as a reanalysis type fault, these probability information can still provide a reference for experts to judge the tendency and uncertainty of the model. The system integrates and packages the above multi-source and multi-level data in a predetermined structured format to generate expert library input data. The system transmits the generated expert library input data to the expert diagnosis center in real time through a pre-configured high-speed and secure communication link. The expert diagnosis center starts the intelligent task allocation system, and accurately distributes tasks to the expert group or intelligent diagnosis module that is best at handling such problems based on key information such as fault type, electrical appliance category, and data characteristics, and comprehensively obtains the expert library output data. The data includes detailed fault cause analysis, such as rare failure modes of a new type of electronic component under specific working conditions, or complex electrical performance degradation caused by the interweaving of multiple factors; it also includes targeted repair suggestions, such as replacing specific model components, adjusting certain circuit parameters, optimizing software algorithms and other specific operation steps, providing authoritative guidance for the subsequent generation of safety protection solutions. After receiving the output data from the expert library, the system immediately starts the safety protection solution generation module. This module first parses the fault cause analysis and repair suggestions in the output data and converts them into executable operation instruction sequences.If the cause of the failure is that a key chip overheats and causes abnormal performance, and experts recommend optimizing heat dissipation and adjusting chip loads, the system generates the following instructions: Send instructions to the temperature control module of the smart chip to increase the speed of the cooling fan, dynamically adjust according to the preset temperature-speed curve, and ensure that the chip temperature quickly falls back to a safe range; at the same time, according to the task priority algorithm, suspend some non-critical tasks to reduce the chip computing burden and avoid overheating again. In the case of component replacement, the system automatically searches the internal component inventory database to confirm whether there is stock. If so, it generates a detailed operation process for replacing components, including power-off sequence, component removal steps, and precautions for installing new components, and sends these instructions to the smart terminal of the operation and maintenance personnel to assist them in on-site operations; if there is no stock, immediately send an emergency purchase request to the procurement department, and start the backup equipment or temporarily adjust the system operation mode to reduce the functional dependence of the faulty electrical appliances, maintain basic operation, and wait for spare parts replacement. During the entire execution process, the system continuously monitors the progress of fault repair, verifies the repair effect by collecting relevant parameters, ensures that the fault is completely resolved, and the intelligent low-voltage electrical system resumes normal and stable operation.

[0071] In some optional embodiments, when the fault type is a traditional overload fault, the system first performs an in-depth analysis of the current parameters of the precise electrical appliances, and determines the degree of overload in combination with the preset current overload threshold. If the overload is relatively light, such as the current exceeds the rated value by 10%-30%, the system generates instructions to adjust the task scheduling strategy of the smart chip. By querying the list of currently running tasks and their power consumption models, some low-priority, non-critical tasks are suspended to reduce the overall power demand and alleviate the pressure of current overload. For example, in a smart home system, ongoing non-emergency data backup tasks or scheduled environmental monitoring data upload tasks are suspended. At the same time, pay close attention to the temperature parameters of precise electrical appliances. Since overload will cause the heating of electrical appliances to intensify, the heat dissipation measures are dynamically adjusted based on the comparison of the temperature rise rate with the preset overheating protection temperature curve. If the temperature rises rapidly, send instructions to the cooling fan control module, and gradually increase the fan speed according to the built-in temperature-fan speed intelligent adjustment algorithm to enhance the heat dissipation capacity and ensure that the temperature of the electrical appliances is maintained in a safe range. If the overload is serious and the current exceeds the rated value by more than 30%, in addition to the above measures, a power-off command is immediately sent to the power control module to cut off the power supply of the appliance to prevent damage to the appliance, fire and other dangerous situations caused by long-term severe overload. After the power is off, the overload alarm information is pushed to the user through the smart terminal to prompt the user to check the load situation and restart the power after eliminating the overload source. The system sends the generated instructions to the corresponding modules in order of priority, such as sending the task scheduling instruction to the smart chip first, and then sending the fan speed adjustment instruction or power-off instruction after confirming that the task is suspended. During the execution process, key parameters such as electrical current and temperature are continuously collected to verify the effectiveness of the protection measures. If it is found that the current has not dropped to a reasonable range or the temperature continues to rise, further strengthening measures are taken, such as adjusting the fan speed again, expanding the task suspension range, etc., until the fault is eliminated and the appliance resumes normal operation.

[0072] When the fault type is a short-circuit fault, the system quickly triggers the short-circuit protection mechanism based on the instantaneous mutation characteristics of the precise electrical current parameters. First, an emergency power-off command is sent to the power control module, requiring the electrical power to be cut off within milliseconds. This process is achieved through high-speed relays or intelligent solid-state switches to ensure that the short-circuit current can be quickly blocked to protect the internal circuits and components of the electrical appliances from damage caused by large current shocks. At the same time, the system starts the fault troubleshooting program, and uses the short-circuit fault location algorithm combined with the changes in the precise electrical current and voltage parameters at the moment of the short circuit to roughly determine the short-circuit area. For example, if the voltage drops to zero and the current is extremely large on a local circuit branch, it is preliminarily determined that the short-circuit point is near the branch. Combined with the built-in electrical circuit topology diagram and component layout information, a detailed troubleshooting guide is generated, including a list of components that may be short-circuited, the troubleshooting sequence, and a simple detection method, and this information is pushed to the intelligent terminal of the operation and maintenance personnel to assist them in quickly locating and repairing the short-circuit fault. The power-off command is executed first to ensure safety. While the operation and maintenance personnel are repairing the fault according to the troubleshooting instructions, the system continuously monitors the progress of the repair, and can judge by detecting changes in parameters such as the circuit resistance returning to the normal range and the insulation resistance reaching the standard value. Once it is confirmed that the short circuit fault has been eliminated, the system gradually restores the power supply to the appliance according to the preset power-on sequence, while closely monitoring the current, voltage and other parameters after the initial power-on to ensure that the appliance restarts stably and resumes normal operation.

[0073] When the fault type is a poor heat dissipation fault of an electrical appliance, the degree of poor heat dissipation is comprehensively judged based on the precise electrical temperature parameters, especially the temperature of the key heating component parts continues to rise and exceeds the normal operating temperature range, combined with the heat sink temperature, ambient temperature and precise electrical current parameters. If the degree of poor heat dissipation is relatively mild, the system first adjusts the speed of the cooling fan, and by querying the built-in temperature-fan speed association table (based on heat dissipation design theory and a large amount of experimental data), the fan speed is appropriately increased according to the current temperature increase to enhance air convection heat dissipation. At the same time, check whether the heat dissipation duct is unobstructed, and use the duct pressure sensor to monitor the pressure difference between the inlet and outlet of the duct. If the pressure difference is abnormal, it indicates that the duct may be blocked, and generate a duct cleaning prompt message, which is pushed to the intelligent terminal of the operation and maintenance personnel to guide them to clean up dust, debris and other blockages and restore the duct to smooth. If the degree of poor heat dissipation is serious and the temperature rises sharply and approaches the heat resistance limit of the component, in addition to the above measures, the system immediately takes load reduction measures, similar to the task scheduling adjustment in the traditional overload fault handling, suspends some non-critical tasks, reduces the overall power consumption of the electrical appliance, reduces heat generation, relieves heat dissipation pressure from the source, and avoids damage to components due to overheating. The system executes instructions such as fan speed adjustment, air duct cleaning reminder, and load reduction in sequence, and continuously monitors the temperature changes of the appliance during the execution process. If the temperature begins to fall back to the normal range, the suspended tasks are gradually resumed, and the fan speed is optimized to a reasonable level to maintain a balance between heat dissipation and operating power consumption. If the temperature is still not effectively controlled, further increase the load reduction or check whether there are other faults in the cooling system hardware until the heat dissipation problem is resolved to ensure that the appliance can operate stably at an appropriate temperature.

[0074] In some embodiments, the processing strategy for the simultaneous occurrence of multiple faults of low-voltage electrical appliances (updated)

[0075] When the target low-voltage electrical appliance has several faults at the same time, the system needs to prioritize the faults that occur at the same time according to the severity of the faults and potential risks. Short-circuit faults are given the highest priority because of the huge instantaneous current, which can easily cause catastrophic consequences such as burning of electrical components and fire. The second is the traditional overload fault. If it is in an overload state for a long time, although it does not explode instantly like a short circuit, it will gradually damage the electrical appliance and affect its service life. There is also a risk of overheating and fire when the overload is serious. The fault of poor heat dissipation of electrical appliances ranks third. It will increase the internal temperature of the electrical appliance, accelerate the aging of components, and reduce performance. If not handled in time, it may eventually cause other faults. Re-analysis faults, communication module faults, image recognition faults, etc. are relatively less threatening to the immediate operation safety of the equipment, and have a lower priority, but they still need to be handled in time to ensure the integrity and reliability of the overall function of the electrical appliance. When judging the priority, the system will comprehensively consider the change trend of key parameters corresponding to each fault type, historical fault data, and the preset fault hazard level table to ensure accurate and reasonable sorting. At the same time, if there are multiple faults of the same priority, such as interference faults and image recognition faults in communication module faults, the system will further sort them according to the degree of impact of the faults on the currently executing task. If the current appliance is performing a real-time data transmission task, then the interference fault has a more direct impact on the task and has a slightly higher priority than the image recognition fault; if the image monitoring-based operation is in progress, the image recognition fault will be processed first. This process is achieved by analyzing the task execution process, data flow, and the interference of each fault on the key links, ensuring that resource allocation and processing order are more accurate in complex multi-fault scenarios.

[0076] If a short-circuit fault occurs concurrently with other faults, the system immediately focuses all its attention on short-circuit protection. As mentioned above, an emergency power-off command is sent to the power control module to cut off the power supply of the appliance within milliseconds to prevent the short-circuit current from causing further damage to the equipment. At the same time, other concurrent fault types and related parameter states are recorded for subsequent processing. If a short-circuit fault occurs at the same time as a poor heat dissipation fault of the appliance, after the power is cut off, in addition to troubleshooting and repairing the short-circuit point, a comprehensive inspection of the heat dissipation system is also required. Check whether the cooling fan is damaged by the short-circuit impact, whether the heat dissipation duct is deformed due to high temperature or blocked by foreign objects, etc., and make preparations for heat dissipation repair in advance, so that the heat dissipation function can be quickly restored after the short-circuit fault is repaired, and ensure the stable operation of the equipment after restarting. Prioritize ensuring that the power-off command is executed quickly and accurately. After the short-circuit fault is eliminated, the corresponding protection plan is generated and executed in sequence according to the priority order of other faults. For example, if there is also a fault of poor electrical heat dissipation, before re-powering on, first clean the heat dissipation duct and check the fan and other heat dissipation components to ensure that the cooling system has initially returned to normal, and then gradually restore power supply to avoid the risk of other faults being aggravated by heat dissipation problems after power-on.

[0077] If the traditional overload fault and the poor heat dissipation fault occur simultaneously, the system adopts a collaborative processing strategy in view of the mutual correlation and mutual influence of these two faults. For the case where the overload is relatively light and the poor heat dissipation is also relatively light, on the one hand, some low-priority tasks are suspended according to the power consumption model to reduce power consumption and alleviate the overload; on the other hand, the temperature-fan speed association table is queried to appropriately increase the speed of the cooling fan to enhance heat dissipation. At the same time, the current and temperature parameters are continuously monitored to determine the development trend of the fault. If the overload is heavy or the poor heat dissipation is serious and close to the heat resistance limit of the component, the power supply is immediately cut off on the basis of taking the above-mentioned load reduction and heat dissipation enhancement measures to prevent equipment damage. Instructions are executed in sequence according to the order of task scheduling, fan speed adjustment, and power off, and the operating parameters of the electrical appliances are closely monitored during the execution process. If it is found that the current or temperature is not effectively controlled, the strategy is further adjusted, such as expanding the task suspension range, checking whether there are other hidden faults in the heat dissipation hardware, etc., until the risks of both faults are effectively controlled and the equipment resumes stable operation.

[0078] When there are multiple other combinations of faults, for combinations involving re-analysis faults, communication module faults, image recognition faults, and other faults, follow the principle of first handling faults that endanger the safe operation of the equipment, and then handling functional faults. For example, if a communication module fault and an electrical heat dissipation fault occur at the same time, solve the heat dissipation problem first to ensure the normal operation of the equipment, and then take corresponding measures such as channel switching and hardware self-test to repair the communication fault based on the type of communication module fault. During the entire processing process, the system maintains real-time monitoring of all key parameters and dynamically adjusts protection measures according to the progress of fault repair. After each fault is handled or a phased repair result is achieved, re-evaluate the remaining fault conditions and optimize the subsequent processing process to ensure that in multiple concurrent fault scenarios, the equipment can be restored to normal operation in an orderly manner, and each functional module gradually resumes normal collaboration.

[0079] The intelligent safety protection and fault prevention method for low-voltage electrical appliances provided in this embodiment, firstly, through comprehensive multi-dimensional parameter collection covering image sensors, communication modules, smart chips and the low-voltage electrical appliances themselves, various key information of electrical appliance operation is captured in real time and accurately, and temperature and operation frequency monitoring are set for smart chips, communication modules monitor data packet error rate, and image sensors analyze grayscale parameters, etc. Compared with the traditional method of only focusing on basic electrical parameters, it greatly broadens the information acquisition surface, lays a solid foundation for subsequent accurate fault diagnosis and prevention, and effectively avoids the omission of potential faults due to missing information. Secondly, by tailoring filtering strategies for parameters with different characteristics, mean filtering is used for grayscale parameters, packet error rate parameters and chip operation frequency parameters to remove high-frequency random noise interference and ensure data stability; when the chip temperature and electrical temperature are close to the overheating warning, temperature-sensitive filtering is switched to accurately capture the trend of subtle temperature changes to avoid misjudgment due to temperature fluctuations. When the rate of change of electrical current is abnormal, noise suppression is enhanced to ensure that the current data truly reflects the load change. The accurate working state parameters after filtering input high-reliability data into the fault analysis model, significantly reducing the probability of misjudgment of faults caused by data noise and improving the reliability of the entire system. Then, by introducing a fault analysis model that integrates deep learning algorithms, the complex internal correlation between multivariate parameters and their change rates is fully explored. The model input layer reasonably allocates each precise parameter. The middle hidden layer uses the progressive operation of neurons to extract deep features from massive operating data. The output layer determines 8 types of faults with high accuracy, overcoming the limitations of traditional judgments based on simple thresholds. Whether it is a new type of fault such as the computing power overload of the smart chip, or a traditional overload or short circuit fault, it can be accurately identified, greatly improving the intelligence level and accuracy of fault diagnosis, and reducing the time and cost of manual troubleshooting. Finally, by customizing the generation of safety protection schemes based on different fault types, accurate protection is achieved. In the face of chip computing power overload, combined with chip temperature and task priority, non-critical tasks are reasonably suspended to release computing power, and the speed of the cooling fan is dynamically adjusted to ensure the stable operation of the chip from both software and hardware. When the communication module fails, the packet error rate is used to distinguish interference or hardware problems, and the channel is switched or restarted in a targeted manner and the communication strategy is optimized in combination with big data to ensure smooth data transmission. After the image recognition fault is repaired, the image sample self-learning optimization model during the fault period is used to improve the subsequent recognition accuracy. The intelligent safety protection and fault prevention method of low-voltage electrical appliances provided in this embodiment adopts a variety of protection strategies to comprehensively reduce the risk of functional failure caused by faults of intelligent low-voltage electrical appliances, extend the normal operation time of the equipment, and solve the problem that the existing protection methods are difficult to effectively monitor and prevent these emerging faults due to the unique fault mode caused by the integration of multiple advanced modules in the related technology.

[0080] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. An intelligent safety protection and fault prevention method for low-voltage electrical appliances, applied to an intelligent low-voltage electrical appliance system, wherein the intelligent low-voltage electrical appliance system comprises an image sensor module, a communication module, an intelligent chip module and a low-voltage electrical appliance module, characterized in that: The method comprises: Acquire working state parameters of the target low-voltage electrical appliance, the working state parameters including image sensor parameters, communication module parameters, smart chip parameters and low-voltage electrical appliance parameters, the image sensor parameters including grayscale parameters, the communication module parameters including data packet error rate parameters, the smart chip parameters including chip temperature parameters and chip operation frequency parameters, the low-voltage electrical appliance parameters including electrical current parameters, electrical voltage parameters and electrical temperature parameters; The working state parameters are filtered respectively to obtain accurate working state parameters, the accurate working state parameters include: accurate grayscale parameters, accurate data packet error rate parameters, accurate chip temperature parameters, accurate chip operation frequency parameters, accurate electrical current parameters, accurate electrical voltage parameters and accurate electrical temperature parameters; The precise working state parameters are input into the fault analysis model to obtain a fault analysis result, wherein the fault analysis result is normal operation or operation failure, and the operation failure includes at least one of eight fault types, including: chip computing power overload fault, communication module fault, image recognition fault, traditional overload fault, short circuit fault, leakage fault, poor electrical heat dissipation fault and re-analysis fault; If the fault analysis result is an operating fault, a safety protection scheme is generated based on the fault type and working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme.

2. The method according to claim 1, characterized in that The step of obtaining the working state parameters of the target low-voltage electrical appliance includes: The image sensor is used to collect images. For each frame of image collected, the sliding window algorithm is used to calculate the pixel grayscale deviation of the frame image, and the average pixel grayscale deviation of consecutive A frame images is taken as the grayscale parameter. The error rate of the data packet is collected by using the error monitor of the communication module. After receiving B data packets, the error rate of the data packet is calculated once to obtain the error rate parameter of the data packet. Using a temperature sensor to collect chip temperature values ​​based on a first sampling frequency, performing differential processing on the collected chip temperature values ​​to obtain chip temperature parameters; Using a frequency counter to collect the chip operation frequency based on the second sampling frequency, performing a first-order difference on the chip operation frequency, and obtaining a chip operation frequency parameter; Using a current sensor to collect the working current of the electrical appliance based on a third sampling frequency to obtain a current parameter of the electrical appliance; Using a voltage sensor to collect the working voltage of the electrical appliance based on a fourth sampling frequency to obtain a voltage parameter of the electrical appliance; The temperature of a preset part of the electrical appliance is collected by using a temperature sensor based on a fifth sampling frequency to obtain a temperature parameter of the electrical appliance.

3. The method according to claim 2, characterized in that The filtering process is performed on the working state parameters to obtain accurate working state parameters, including: A mean filtering method is used for the grayscale parameter, the data packet error rate parameter and the chip operation frequency parameter to obtain an accurate grayscale parameter, an accurate data packet error rate parameter and an accurate chip operation frequency parameter respectively; Filter the chip temperature parameters, and when it is detected that the chip temperature in the chip temperature parameters exceeds the first preset overheat warning temperature range or the chip temperature change rate exceeds the first preset overheat warning temperature change rate range, switch to the temperature sensitive filtering mode to obtain accurate chip temperature parameters; Filter the electrical current parameters. When it is detected that the electrical current change rate in the electrical current parameters exceeds the preset current warning change rate range, switch the noise suppression strength value to the preset value to obtain accurate electrical current parameters. Filter the electrical appliance voltage parameters. When it is detected that the fluctuation amplitude of the electrical appliance voltage in the electrical appliance voltage parameters exceeds the preset voltage warning range for C consecutive sampling cycles or the voltage change rate exceeds the preset voltage change rate warning range, start the adaptive filtering mode of the voltage signal to obtain accurate electrical appliance voltage parameters; The appliance temperature parameters are filtered. When it is detected that the appliance temperature in the appliance temperature parameters exceeds the second preset overheat warning temperature range or the appliance temperature change rate exceeds the second preset overheat warning temperature change rate range, it is switched to the temperature-sensitive filtering mode to obtain accurate appliance temperature parameters.

4. The method according to claim 3, characterized in that The fault analysis model includes 1 input layer, 3 intermediate hidden layers and 1 output layer. The input layer has 14 neurons, the number of neurons in the intermediate hidden layer is 128, 64 and 32 respectively, and the output layer has 8 neurons. The step of inputting the precise working state parameters into a fault analysis model to obtain a fault analysis result includes: Distribute the precise grayscale parameter, precise data packet error rate parameter, precise chip temperature parameter, precise chip operation frequency parameter, precise electrical current parameter, precise electrical voltage parameter and precise electrical temperature parameter to 14 neurons of the input layer according to a predetermined corresponding relationship; Controlling the first hidden layer to receive input layer data, using 128 neurons to perform weighted summation operation on the input layer data based on the initial weight value range, and then performing nonlinear transformation through the activation function to extract the preliminary depth features between the precise grayscale parameter, the precise data packet error rate parameter, the precise chip temperature parameter, the precise chip operation frequency parameter, the precise electrical current parameter, the precise electrical voltage parameter and the precise electrical temperature parameter, wherein the preliminary depth features include change trend features and correlation relationship features; Control the second hidden layer to randomly discard D% connections using 64 neurons when receiving the output of the first hidden layer, and use an optimization algorithm to allocate an adaptive learning rate to the connection weights according to the historical gradient square sum of the precise grayscale parameter, the precise data packet error rate parameter, the precise chip temperature parameter, the precise chip operation frequency parameter, the precise electrical current parameter, the precise electrical voltage parameter, and the precise electrical temperature parameter, and converge to the optimal solution; Control 32 neurons in the third hidden layer to normalize E bytes of the second hidden layer data to obtain a hidden layer output feature vector; The 8 neurons in the control output layer receive the hidden layer output feature vector output by the third hidden layer, and use the activation function to convert the neuron input value into a normalized vector representing the probability of each fault type; The fault analysis result is obtained according to the normalized vector of the probability of each fault type and the preset fault judgment threshold.

5. The method according to claim 4, characterized in that If the fault analysis result is an operating fault, a safety protection scheme is generated based on the fault type and the working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme, including: When the fault type includes a chip computing power overload fault, obtaining data packet task priority data; Determining pause task data based on the data packet task priority data; Dynamic speed data is calculated based on precise chip temperature parameters and preset fan speed formula; A safety protection scheme is generated based on the pause task data and the dynamic speed data, and the intelligent low-voltage electrical appliance system is controlled to execute the safety protection scheme.

6. The method according to claim 5, characterized in that If the fault analysis result is an operating fault, a safety protection scheme is generated based on the fault type and the working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme, including: When the fault type includes a communication module type fault, determining the communication module fault type based on the precise data packet error rate parameter and the preset error rate formula, the communication module fault type includes an interference type fault and a hardware type fault; A safety protection scheme is generated based on the communication module fault type and the precise data packet error rate parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme.

7. The method according to claim 5, characterized in that If the fault analysis result is an operating fault, a safety protection scheme is generated based on the fault type and the working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme, including: When the fault type includes an image recognition type fault, determining a calibration position based on the precise grayscale parameter; A safety protection scheme is generated based on the calibration position and the precise grayscale parameters, and the intelligent low-voltage electrical appliance system is controlled to execute the safety protection scheme.

8. The method according to claim 7, characterized in that After generating a safety protection scheme based on the calibration position and the precise grayscale parameter, and controlling the intelligent low-voltage electrical appliance system to execute the safety protection scheme, the method further includes: Determining a fault period based on the precise grayscale parameter; Marking the error recognition area of ​​the image samples during the fault period to obtain marked image sample data; The image recognition model is trained using the labeled image sample data.

9. The method according to claim 8, characterized in that If the fault analysis result is an operating fault, a safety protection scheme is generated based on the fault type and the working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme, including: When the fault type includes a leakage fault, leakage current data, voltage phase change data, and quantum state collapse data collected by quantum sensors at at least three different positions on the target low-voltage electrical appliance are obtained, The leakage point coordinate data is calculated through leakage current data, voltage phase change data, quantum state collapse data and pre-trained quantum positioning model; A safety protection scheme is generated based on the leakage point coordinate data, and the intelligent low-voltage electrical appliance system is controlled to execute the safety protection scheme.

10. The method according to claim 9, characterized in that If the fault analysis result is an operating fault, a safety protection scheme is generated based on the fault type and the working state parameters, and the intelligent low-voltage electrical system is controlled to execute the safety protection scheme, including: When the fault type includes a reanalysis fault, the working state parameters, precise working state parameters, preliminary depth features and normalized vectors of the probability of each fault type within a preset time interval are obtained to generate expert database input data; Send the expert database input data to the expert diagnosis center and obtain the expert database output data; A safety protection scheme is generated based on the output data of the expert database, and the intelligent low-voltage electrical appliance system is controlled to execute the safety protection scheme.

Images