Power grounding state monitoring method and system

Abstract

The invention discloses a power grounding state monitoring method and system, which is applied to the field of power equipment state monitoring, and comprises the following steps: synchronously obtaining a time sequence mechanical state signal and a time sequence electrical state signal of a grounding terminal; extracting and constructing a multi-dimensional feature vector containing static and dynamic feature components; performing differentiation evaluation and integration on the feature components by using a pre-constructed hybrid analysis model to generate an initial monitoring result; and introducing environmental parameters to dynamically calibrate the initial result to obtain a calibrated grounding state monitoring result. According to the electric power grounding state monitoring method provided by the invention, the mechanical tightness and the electrical connectivity information are fused, and the intelligent model is adopted to carry out collaborative analysis and environment adaptive calibration, so that the comprehensive and accurate sensing of the connection state of the grounding terminal is realized.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and in particular to a method and system for monitoring power grounding condition. Background Technology

[0002] As an important safety component of power equipment, the clamping force of the grounding terminal directly affects the grounding effect. Power equipment often operates outdoors or in harsh industrial environments. Factors such as temperature changes and vibrations may cause the grounding terminal to loosen. If the nut of the grounding terminal is loose, resulting in poor grounding, the equipment will not be able to effectively conduct current to the ground in the event of a fault, thereby threatening equipment safety and even causing serious consequences such as fire.

[0003] In existing technologies, monitoring methods for grounding terminals primarily rely on mechanical connection signals to infer the grounding status. This assessment model, which solely depends on mechanical signals, struggles to comprehensively and accurately reflect the complex deterioration process of grounding connections under the combined effects of multiple factors such as electrical corrosion and environmental changes. Particularly for electrical performance degradation caused by non-mechanical factors like contact surface corrosion and contamination, this method exhibits significant monitoring blind spots, resulting in diagnostic results lacking sufficient relevance and accuracy. This directly leads to insufficient accuracy in diagnosing grounding status under complex operating conditions, resulting in high false alarm and false negative rates, and hindering the provision of reliable data for preventative maintenance. Summary of the Invention

[0004] This invention provides a power grounding status monitoring method and system to solve the technical problem that it is difficult to comprehensively and accurately identify the complex degradation modes of grounding terminals due to the single monitoring dimension, so as to achieve the effect of coordinated perception and comprehensive evaluation of the mechanical connection and electrical performance status of grounding terminals.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for monitoring power grounding status, comprising: Acquire the timing mechanical status signal and timing electrical status signal of the target grounding terminal; The feature extraction results of the time-series mechanical state signal and the feature extraction results of the time-series electrical state signal are integrated, and a multi-dimensional feature vector is obtained based on the integrated results to characterize the comprehensive operating state of the target grounding terminal. The multi-dimensional feature vector includes at least static feature components and dynamic feature components. The multidimensional feature sequence is input into a pre-constructed grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature component and a second state evaluation vector corresponding to the dynamic feature component. Based on the first state evaluation vector and the second state evaluation vector, the initial state monitoring result is generated; At least one environmental parameter of the environment in which the target grounding terminal is located is obtained, and the initial state monitoring result is dynamically calibrated based on the environmental parameter to obtain the calibrated grounding state monitoring result.

[0006] As a preferred embodiment, the integration of the feature extraction results of the time-series mechanical state signal and the feature extraction results of the time-series electrical state signal, and the generation of a multi-dimensional feature vector based on the integration results to characterize the comprehensive operating state of the target grounding terminal, includes: The time-domain analysis of the time-series mechanical state signal is performed to extract the pressure change frequency as a dynamic feature component and the pressure amplitude as a static feature component. Time-domain analysis is performed on the time-series electrical state signal to calculate the fluctuation range of the grounding resistance in the power circuit as a dynamic characteristic component, and the standard deviation of the grounding resistance is calculated as a static characteristic component. The extracted pressure change frequency, pressure amplitude, fluctuation range, and standard deviation are synchronously aligned and combined according to their corresponding timestamps to construct the multidimensional feature vector.

[0007] As a preferred embodiment, the grounding state analysis model is a hybrid model. The step of inputting the multidimensional feature sequence into the pre-constructed grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature components and a second state evaluation vector corresponding to the dynamic feature components includes: The static feature components are input into the random forest model for static evaluation, and the first state evaluation vector is output. The dynamic feature components are input into the Long Short-Term Memory network model for dynamic evaluation, and the second state evaluation vector is output.

[0008] As one preferred embodiment, the step of generating initial state monitoring results based on the first state evaluation vector and the second state evaluation vector includes: The second state evaluation vector is standardized and converted into a probability vector with the same dimensions as the first state evaluation vector. Based on the attention mechanism, according to the multidimensional feature vector at the current time step, a fusion weight is dynamically generated between the first state evaluation vector and the probability vector. The first state evaluation vector and the probability vector are weighted and fused according to the fusion weight to obtain a comprehensive state probability distribution. Based on the comprehensive state probability distribution, the state category identifier and corresponding state probability of the target grounding terminal are output, and the initial state monitoring result is generated.

[0009] As one preferred embodiment, the step of dynamically calibrating the initial state monitoring results based on the environmental parameters to obtain calibrated grounding state monitoring results includes: Construct a fuzzy evaluation system, and map the initial state monitoring results to the corresponding membership degree representation based on the fuzzy evaluation system; Based on the environmental parameters, the weights of the fuzzy evaluation system are dynamically adjusted, and based on the adjusted fuzzy evaluation system, the membership identifier is processed to generate a calibration factor. The initial state monitoring results are corrected using the calibration factor to obtain the calibrated grounding state monitoring results.

[0010] Another embodiment of the present invention provides a power grounding status monitoring system, comprising: The acquisition module is used to acquire the timing mechanical state signal and timing electrical state signal of the target grounding terminal; An integration module is used to integrate the feature extraction results of the time-series mechanical state signal and the feature extraction results of the time-series electrical state signal, and obtain a multi-dimensional feature vector based on the integration results to characterize the comprehensive operating state of the target grounding terminal. The multi-dimensional feature vector includes at least static feature components and dynamic feature components. An evaluation module is used to input the multidimensional feature sequence into a pre-built grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature component and a second state evaluation vector corresponding to the dynamic feature component. The generation module is used to generate initial state monitoring results based on the first state evaluation vector and the second state evaluation vector; The calibration module is used to acquire at least one environmental parameter of the environment in which the target grounding terminal is located, and to dynamically calibrate the initial state monitoring results based on the environmental parameters to obtain calibrated grounding state monitoring results.

[0011] As one preferred embodiment, the integration module includes: The first time-domain analysis unit is used to perform time-domain analysis on the time-series mechanical state signal, extract the pressure change frequency as a dynamic feature component, and extract the pressure amplitude as a static feature component. The second time-domain analysis unit is used to perform time-domain analysis on the time-series electrical state signal, calculate the fluctuation range of the grounding resistance in the power circuit as a dynamic characteristic component, and calculate the standard deviation of the grounding resistance as a static characteristic component. The alignment unit is used to synchronously align and combine the extracted pressure change frequency, pressure amplitude, fluctuation range and standard deviation according to the corresponding timestamps to construct the multidimensional feature vector.

[0012] As one preferred embodiment, the grounding state analysis model is a hybrid model, and the evaluation module includes: The static evaluation unit is used to input the static feature components into the random forest model for static evaluation and output the first state evaluation vector. The dynamic evaluation unit is used to input the dynamic feature components into the long short-term memory network model for dynamic evaluation and output the second state evaluation vector.

[0013] As one preferred embodiment, the generation module includes: The standardization processing unit is used to standardize the second state evaluation vector and convert it into a probability vector with the same dimensions as the first state evaluation vector. The weighted fusion unit is used to dynamically generate fusion weights between the first state evaluation vector and the probability vector based on the multidimensional feature vector at the current time step, according to the attention mechanism, and to perform weighted fusion of the first state evaluation vector and the probability vector according to the fusion weights to obtain a comprehensive state probability distribution. The probability generation unit is used to output the state category identifier and corresponding state probability of the target grounding terminal according to the comprehensive state probability distribution, and generate the initial state monitoring result.

[0014] As one preferred embodiment, the calibration module includes: The system construction unit is used to construct a fuzzy evaluation system and map the initial state monitoring results into corresponding membership degree representations based on the fuzzy evaluation system. The dynamic adjustment unit is used to dynamically adjust the weights of the fuzzy evaluation system based on the environmental parameters, and to process the membership identifier based on the adjusted fuzzy evaluation system to generate a calibration factor. The result calibration unit is used to correct the initial state monitoring result using the calibration factor to obtain the calibrated grounding state monitoring result.

[0015] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: (1) This invention constructs a comprehensive state feature vector that simultaneously includes static and dynamic feature components by synchronously acquiring and fusing the time-series mechanical and electrical state signals of the grounding terminal. A pre-built hybrid analysis model is then used to differentiate and integrate this vector, achieving a comprehensive and in-depth collaborative perception of the connection status from both mechanical tightness and electrical connectivity dimensions. This method overcomes the limitations of single-parameter monitoring and can effectively distinguish grounding status anomalies caused by different reasons such as loose nuts, contact surface corrosion, and environmental interference, significantly improving the accuracy of status diagnosis and the specificity of fault mode identification.

[0016] (2) This invention introduces a dynamic calibration mechanism based on environmental parameters to perform secondary correction on the initial monitoring results of the model, effectively eliminating the interference of external variables such as environmental temperature and humidity on the diagnostic conclusions, and significantly enhancing the environmental robustness and result stability of the monitoring system. This invention can adaptively adjust the evaluation weights according to real-time operating conditions, so that the final output grounding status monitoring results not only reflect the current connection performance, but also have the ability to predict the evolution of potential risks, thereby providing operation and maintenance personnel with more accurate and reliable status warnings and maintenance guidance, and strongly supporting the intelligent transformation of power system safe operation from "post-event handling" to "pre-event prevention". Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a power grounding status monitoring method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the working process of the power grounding status monitoring method in one embodiment of the present invention; Figure 3 This is a schematic diagram of a power grounding status monitoring system in one embodiment of the present invention.

[0018] Figure label: Among them, 11 is the acquisition module, 12 is the integration module, 13 is the evaluation module, 14 is the generation module, and 15 is the calibration module. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0021] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0022] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0023] One embodiment of the present invention provides a method for monitoring the power grounding status. For details, please refer to [link / reference]. Figure 1 , Figure 1 The diagram shown illustrates a power grounding status monitoring method according to one embodiment of the present invention, which includes steps S1 to S5: S1: Acquire the timing mechanical status signal and timing electrical status signal of the target grounding terminal; During power system operation, the safety status of grounding terminals directly depends on the stability of their mechanical connections and the effectiveness of their electrical conductivity. Monitoring only one dimension of the grounding system's status cannot fully reflect its true safety level. Monitoring only the mechanical clamping condition may miss electrical grounding failures caused by contact corrosion; monitoring only the electrical resistance makes it difficult to detect mechanical connection failures caused by loose nuts in advance. Therefore, step S1 simultaneously acquires the timing-series mechanical and electrical status signals of the target grounding terminal, laying a data foundation for subsequent accurate analysis of the grounding status.

[0024] In this invention, the time-series mechanical state signal is a dynamic signal used to characterize the stability of the mechanical connection of the target grounding terminal, reflecting the change in the clamping force between the grounding terminal nut and the screw. Since the reliability of the mechanical connection of the grounding terminal directly depends on the nut clamping force—when the nut loosens due to vibration, temperature changes, or other factors, the clamping force will decrease accordingly. If this is not detected in time, it may lead to poor contact of the grounding terminal; conversely, excessive clamping force may cause deformation of the terminal or screw, also affecting connection stability. Therefore, the system uses a ring pressure sensor to monitor the clamping force between the nut and the washer in real time. The sensor converts the clamping force (physical quantity) at each instant into a continuous analog electrical signal, which is then converted into a digital quantity by a high-precision analog-to-digital converter, while simultaneously marking the corresponding time information, forming a signal sequence that dynamically changes over time, namely the time-series mechanical state signal.

[0025] The timing electrical state signal is a dynamic signal used to characterize the electrical conductivity of a target grounding terminal, reflecting the changes in the resistance value of the grounding loop. Since the core function of the grounding terminal is to conduct fault current to the ground during power equipment failures, and this function relies on the low-resistance conductivity of the grounding loop—if the grounding loop resistance is too high, the fault current cannot be effectively discharged, potentially leading to safety hazards such as live equipment casings and malfunctioning protection devices; abnormal fluctuations in resistance may indicate problems such as contact corrosion or loose wiring in the grounding loop. Therefore, the system uses a grounding resistance testing circuit to measure the grounding resistance between the grounding terminal and the ground in real time. The measuring circuit converts the resistance value (physical quantity) at each instant into an analog electrical signal, which is then converted from analog to digital and timestamped synchronously with the timing mechanical state signal, forming a dynamic sequence of resistance values ​​over time, i.e., the timing electrical state signal.

[0026] Specifically, a comprehensive self-test is first performed on the ring pressure sensor that collects mechanical status signals and the resistance measurement circuit that collects electrical status signals. During the self-test, three verifications need to be completed: Sensor circuit connectivity test: Apply a standard test signal to the signal transmission link between the ring pressure sensor and the resistance measurement circuit to check whether the feedback signal is normal, so as to eliminate problems such as open circuit or poor contact. Signal output stability verification: Control the ring pressure sensor to be in a standard pressure environment, connect the resistance measurement circuit to a standard resistance load, continuously collect 10 sets of output signals, calculate the signal fluctuation amplitude, and ensure that the fluctuation value does not exceed the preset threshold (±2%). Accuracy calibration of resistance measurement circuit: Connect the circuit to a standard resistor with a known resistance value (such as 1Ω or 10Ω), compare the deviation between the measured value and the standard value, and correct the deviation to the allowable range (±0.5%) through an algorithm.

[0027] After passing the self-test, the ring pressure sensor and resistance measurement circuit automatically switch to a low-power standby state. In this state, the device maintains its real-time response capability to external signals while reducing unnecessary energy consumption. During the data acquisition phase, the ring pressure sensor needs to be deployed between the sensor nut and washer of the target grounding terminal. Utilizing its ring structure to adapt to the terminal screw installation requirements, the sensor's force-bearing surface adheres to the nut and washer, converting the nut clamping force into a 4-20mA analog electrical signal in real time. This signal is the time-series mechanical status signal. The resistance measurement circuit is connected to the grounding loop through a dedicated terminal block. A four-terminal measurement method is used to eliminate the influence of wiring resistance, and the resistance value of the grounding loop is detected in real time, forming a time-series electrical status signal. To avoid analysis deviations caused by the time difference between the two types of signals, a high-precision analog-to-digital converter (16-bit sampling accuracy, sampling frequency dynamically adjustable within the range of 1Hz-10Hz according to system requirements) is used to synchronously acquire the two types of analog signals. After converting them into digital signals, the system clock module marks each group of digital signals with a unified timestamp, ensuring that each group of mechanical status data and electrical status data corresponds one-to-one in the time dimension. Finally, the time-stamped mechanical status signals and time-stamped electrical status signals are transmitted to the integration module via the CAN bus communication interface to complete the data acquisition process.

[0028] The method in step S1 is used to acquire signals. On the one hand, the self-test process ensures the performance reliability of the acquisition equipment and avoids invalid data acquisition due to equipment failure. On the other hand, the design of synchronous acquisition and timestamp marking eliminates the time deviation between mechanical and electrical status signals, ensuring that the two types of data can be analyzed together. At the same time, the low-power standby mode and dynamically adjustable sampling frequency take into account both data real-time performance and system energy consumption control, providing high-quality and highly consistent raw data support for the subsequent accurate analysis of grounding status.

[0029] S2: Integrate the feature extraction results of the time-series mechanical state signal and the feature extraction results of the time-series electrical state signal, and obtain a multi-dimensional feature vector based on the integration results to characterize the comprehensive operating state of the target grounding terminal. The multi-dimensional feature vector includes at least static feature components and dynamic feature components. After acquiring the time-series mechanical and electrical state signals, although these two types of raw signals can reflect the basic state of the grounding terminal, directly using them for grounding state analysis presents problems such as signal noise interference, single feature dimensions, and insufficient correlation between mechanical and electrical states. Relying solely on the raw signals makes it difficult to accurately capture subtle features of faults such as loosening or corrosion of the grounding terminal, and it also fails to establish a collaborative analysis relationship between mechanical connection status and electrical conduction status. Therefore, in step S2, through feature extraction and integration, key features are mined from the two types of raw signals, and a multi-dimensional feature vector is constructed. This transforms the scattered signal data into structured features that can directly support subsequent intelligent analysis, achieving an accurate characterization of the comprehensive operating state of the grounding terminal.

[0030] Among them, static characteristic components are feature parameters used to characterize the "instantaneous attributes" of the target grounding terminal's state at a specific point in time or within a preset time window. Their core value lies in quantifying the current mechanical connection strength and electrical conductivity of the grounding terminal, serving as a direct basis for judging whether the grounding state meets safety standards. Static characteristic components include the pressure amplitude extracted from the time-series mechanical state signal (nut pressure data) and the standard deviation of grounding resistance extracted from the time-series electrical state signal (grounding resistance data). Dynamic characteristic components are feature parameters used to characterize the trend of the target grounding terminal's state changing over time. Their core value lies in capturing the dynamic evolution of the grounding terminal's mechanical connection or electrical conductivity state, and identifying potential progressive faults in advance (such as slow nut loosening or gradual corrosion of the contact surface). Dynamic characteristic components include the pressure change frequency extracted from the time-series mechanical state signal and the grounding resistance fluctuation range extracted from the time-series electrical state signal.

[0031] Specifically, the first step is to preprocess the two types of raw signals. For high-frequency noise and power frequency interference, low-pass filtering or moving average filtering algorithms are used for smoothing. The selection of the filtering algorithm requires determining the noise frequency band through signal spectrum analysis. If the noise is concentrated above 20Hz, a low-pass filter with a cutoff frequency of 20Hz is selected; if the noise exhibits random pulse characteristics, a moving average filter with a window size of 5 is used. This smooths the fluctuations in the raw signal, retains the effective low-frequency components reflecting changes in grounding status, and avoids data distortion caused by transient interference. After filtering, outlier detection and removal are performed on the two types of signals using the 3σ criterion. The mean and standard deviation of each signal dataset are calculated first. Data exceeding the range of "mean ± 3 times standard deviation" are identified as outliers. The occurrence time and type of numerical change of the outlier data are automatically marked and removed from the dataset, ensuring that subsequent feature extraction is based on a reliable and consistent data foundation. The processed signals are then normalized using the Z-Score normalization method, which scales the data corresponding to the time-series mechanical state signals and the data corresponding to the time-series electrical state signals to the same numerical range, typically [-1, 1]. This eliminates the imbalance of feature weights caused by the difference in dimensions between the two types of signals, ensuring that each feature has equal analytical priority during subsequent feature fusion.

[0032] After preprocessing the signals, time-domain analysis is performed on the two types of signals to extract key features.

[0033] Preferably, in one embodiment of the present invention, the feature extraction results of the time-series mechanical state signal and the feature extraction results of the time-series electrical state signal are integrated, and a multi-dimensional feature vector for characterizing the comprehensive operating state of the target grounding terminal is obtained based on the integrated results, including: Time-domain analysis is performed on the time-series mechanical state signal to extract the pressure change frequency as a dynamic feature component and the pressure amplitude as a static feature component. Time-domain analysis is performed on the time-series electrical state signal to calculate the fluctuation range of the grounding resistance in the power circuit as a dynamic characteristic component, and the standard deviation of the grounding resistance is calculated as a static characteristic component. The extracted pressure change frequency, pressure amplitude, fluctuation range, and standard deviation are synchronously aligned and combined according to the corresponding timestamps to construct a multidimensional feature vector.

[0034] Specifically, for time-series mechanical state signals, the time-domain signal is converted to the frequency domain using a spectrum analysis tool to identify the frequency components of periodic fluctuations in the signal. This frequency is the pressure change frequency, used to reflect the periodic pressure changes caused by mechanical vibration or nut loosening, and serves as a dynamic characteristic component. At the same time, the peak-to-peak value of the signal within a preset time window is calculated. The preset time window can be set to 1 minute. The peak-to-peak value is the difference between the maximum and minimum values ​​within the window, which serves as the pressure amplitude, used to characterize the dynamic range of nut tightening force, and serves as a static characteristic component.

[0035] For time-series electrical state signals, the difference between the maximum and minimum resistance values ​​within the same time window is calculated to obtain the fluctuation range of the grounding resistance, which is used to reflect the instantaneous changes in the electrical conduction state and serves as a dynamic characteristic component. At the same time, the standard deviation of the resistance data within this time window is calculated, and the dispersion of the resistance value is quantified by the size of the standard deviation. The smaller the standard deviation, the more stable the grounding resistance is, and vice versa, it indicates that there is an abnormal risk in the electrical conduction state, which is used as a static characteristic component.

[0036] After acquiring each feature component, feature integration is required. Based on the timestamp, the four types of features extracted—pressure change frequency, pressure amplitude, grounding resistance fluctuation range, and grounding resistance standard deviation—are matched one-to-one according to the acquisition time. This ensures that the same timestamp contains two types of static features and two types of dynamic features with mechanical and electrical dimensions. This combination forms a multi-dimensional feature vector containing four components. Each vector can fully characterize the mechanical connection stability and electrical conductivity effectiveness of the grounding terminal at the corresponding time point.

[0037] In one embodiment of the present invention, in order to further improve the targeting and analysis efficiency of the multidimensional feature vectors, the above feature extraction process is optimized, and the specific operation is as follows: First, determine the size of the time window for feature extraction. Based on the typical cycle of grounding terminal fault changes—loose nuts usually show slow changes on the order of hours, while contact corrosion shows gradual changes on the order of days—set the time window to 5 minutes. This window length can avoid excessive feature fluctuations caused by an excessively short window, and also avoid delayed capture of fault features caused by an excessively long window.

[0038] When performing time-domain analysis on the time-series mechanical state signal, the pressure change frequency is extracted using a short-time Fourier transform. The 5-minute time window is divided into 10 short time segments, each 30 seconds long. After performing a Fourier transform on each segment, the average frequency is taken. This average value is used as the pressure change frequency within the 5-minute window to improve the stability of the frequency characteristics. For the extraction of pressure amplitude, instantaneous pulse interference is first removed within the 5-minute window. The difference threshold between adjacent data is used for judgment. If the difference exceeds 20% of the average value, it is judged as a pulse. Then, the peak-to-peak value of the remaining data is calculated to ensure that the pressure amplitude can truly reflect the actual range of change in the nut tightening force.

[0039] When performing time-domain analysis on time-series electrical state signals, the calculation of the grounding resistance fluctuation range requires first performing linear interpolation on the resistance data within a 5-minute window to fill in any possible data gaps (such as sampling gaps caused by brief communication interruptions). Then, the difference between the maximum and minimum values ​​of the interpolated data is calculated to avoid the data gaps affecting the accuracy of the fluctuation range. The calculation of the grounding resistance standard deviation uses a weighted standard deviation algorithm, assigning higher weights to the data collected later within the window. For example, the weighting coefficient increases linearly with time, and the weight of later data is set to 1.2 times that of earlier data, so that the standard deviation can better reflect the recent stability of the grounding resistance, thereby capturing potential electrical fault trends in advance.

[0040] S3: Input the multidimensional feature sequence into the pre-built grounding state analysis model to obtain the first state evaluation vector corresponding to the static feature component and the second state evaluation vector corresponding to the dynamic feature component; After obtaining the multidimensional feature sequence representing the comprehensive operating status of the grounding terminal, it is necessary to conduct targeted analysis of the static and dynamic features in the feature sequence using a professional model. The static feature component reflects the current state attributes of the grounding terminal, while the dynamic feature component reflects the trend of state change. The attributes of the two types of features are fundamentally different from the analysis requirements. If a single model is used for unified analysis, the detailed information of the static features may be obscured by the dynamic trend, or the change pattern of the dynamic features may not be fully explored, thus affecting the accuracy of the grounding status assessment.

[0041] In this invention, the hybrid architecture design of the grounding state analysis model stems from the essential difference in attributes between static and dynamic feature components: static feature components focus on the state attributes of the grounding terminal at a certain moment, and their core value lies in quantifying "whether the current state meets safety standards," requiring the model to capture the nonlinear mapping relationship between feature values ​​and state labels; dynamic feature components focus on the evolution trend of the grounding terminal state over time, and their core value lies in identifying "whether the state is developing towards a fault," requiring the model to mine the feature change patterns and long-term dependencies in the time-series dimension. If a single model is used for unified analysis, the detailed information of static features is easily obscured by time-series fluctuations, or the trend information of dynamic features cannot be fully learned. Therefore, the model adopts a hybrid architecture of "random forest + long short-term memory network," with each type of module specifically processing one type of feature, achieving complementary advantages.

[0042] In the pre-construction phase of the model, the training of two types of models is first completed based on the historical dataset. The historical dataset comes from the grounding terminal operation data collected in the early stage, which includes multi-dimensional feature sequences under various fault states such as normal state, loose nuts, contact corrosion, and transient resistance. Each feature sequence is labeled with the corresponding actual operating state label.

[0043] For the random forest model module, all static feature components are extracted from the historical multidimensional feature sequence and divided into a static training set and a static test set in a 7:3 ratio. The random forest model parameters are initialized, with 100 decision trees, a maximum tree depth of 10, and a minimum number of sample splits of 5. The model is trained using the static training set. Through ensemble learning of multiple decision trees, the nonlinear mapping relationship between static feature components and ground state labels is captured. After training, the model performance is verified using the static test set. The model's classification accuracy, recall, and F1 score are calculated. When the accuracy exceeds 92% and the F1 score exceeds 0.9, the random forest model module is considered to have completed training. This module ultimately has the ability to convert the input static feature components into state evaluation results, and the output vector is the first state evaluation vector.

[0044] For the Long Short-Term Memory (LSTM) network model module, all dynamic feature components are extracted from historical multidimensional feature sequences and divided into dynamic training and dynamic test sets in a 7:3 ratio. Considering the temporal characteristics of dynamic features, the dynamic training set is divided into temporal windows of length 10, forming temporal samples suitable for LTM network input. The LTM network model parameters are initialized with two hidden layers, 64 hidden units per layer, ReLU activation function, Adam optimizer, and a learning rate of 0.001. The model is trained using the dynamic training set, leveraging the LTM network's gating mechanism to capture the long-term dependencies of dynamic feature components over time and learn the evolution patterns of dynamic features under different fault states. An early stopping mechanism is used to prevent overfitting during training; training stops when the validation set loss does not decrease for five consecutive epochs. After training, the model performance is validated using the dynamic test set, with classification accuracy, recall, and F1 score as evaluation metrics. If the accuracy exceeds 90% and the F1 score exceeds 0.88, the LTM network model module is considered to have completed training. This module has the ability to convert the input dynamic feature components into state evaluation results, and the output vector is the second state evaluation vector.

[0045] Preferably, in one embodiment of the present invention, the grounding state analysis model is a hybrid model. A multi-dimensional feature sequence is input into the pre-constructed grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature components and a second state evaluation vector corresponding to the dynamic feature components, including: The static feature components are input into the random forest model for static evaluation, and the first state evaluation vector is output. The dynamic feature components are input into the Long Short-Term Memory network model for dynamic evaluation, and the second state evaluation vector is output.

[0046] Specifically, the grounding status analysis model was first optimized and enhanced through parameter optimization and training. For the random forest model, during the historical static feature data preprocessing stage, the feature importance score was calculated using a temporarily trained random forest model. Redundant features with scores below 0.3 were removed, and only high-value static features were retained to construct the training set. The model parameters were adjusted to 150 decision trees, a maximum tree depth of 12, and a minimum number of sample splits of 8. The model was trained using 5-fold cross-validation until the classification accuracy on the static test set exceeded 95% and the F1 score exceeded 0.93, thus completing the optimization of the random forest model. For the Long Short-Term Memory (LSTM) network model, historical dynamic feature data was normalized and scaled to [0,1] using Min-Max, and divided into time-series samples of length 10. The model structure was adjusted to two hidden layers (128 hidden units per layer), and a dropout layer with a dropout probability of 0.2 was added after the hidden layers. The Adam optimizer was selected (initial learning rate of 0.001, which was halved if the validation set loss did not decrease for three consecutive epochs). The model was trained using an early stopping mechanism until the classification accuracy on the dynamic test set exceeded 93% and the F1 score exceeded 0.91, thus completing the optimization of the LSM network model.

[0047] During model deployment and runtime, the real-time multidimensional feature sequence is first preprocessed: after obtaining static and dynamic feature components, outlier checks are performed on the static feature components. If the static feature value exceeds a reasonable range at a certain moment (e.g., pressure amplitude exceeds 90% of the sensor range, grounding resistance standard deviation exceeds 0.6Ω), the static feature value from the previous moment is used instead. The dynamic feature components are smoothed using a moving average with a window size of 3, and then divided into time-series samples with a length of 10. The preprocessed static feature components are input into the optimized random forest model time-by-time. The model outputs a two-dimensional first state evaluation vector (the two elements are the probability of normal state and the probability of fault state, respectively, with a sum of 1) through ensemble voting of 150 decision trees. The smoothed dynamic feature time-series samples are input into the optimized long short-term memory network model. The model learns the time-series pattern through a gating mechanism and outputs a two-dimensional second state evaluation vector through a softmax activation function, finally obtaining two-class state evaluation vectors corresponding to each moment.

[0048] S4: Generate initial state monitoring results based on the first state evaluation vector and the second state evaluation vector; After obtaining the first state evaluation vector corresponding to the static feature component and the second state evaluation vector corresponding to the dynamic feature component, although the two types of vectors reflect the grounding terminal status from the dimensions of static attributes and dynamic trends respectively, there are problems of inconsistent dimensions and mismatched weights. The first state evaluation vector is a directly output probability vector (the elements are normal / fault probabilities), and the initial output of the second state evaluation vector is a time-series risk score (the numerical range is different from the probability vector). If the results of the two types of vectors are directly merged, it is easy to cause an imbalance in the evaluation weights of dynamic features, and it is impossible to accurately reflect the synergistic relationship between static attributes and dynamic trends.

[0049] Preferably, in one embodiment of the present invention, generating an initial state monitoring result based on a first state evaluation vector and a second state evaluation vector includes: The second state evaluation vector is standardized and converted into a probability vector with the same dimensions as the first state evaluation vector. Based on the attention mechanism, the fusion weight between the first state evaluation vector and the probability vector is dynamically generated according to the multi-dimensional feature vector of the current time step. The first state evaluation vector and the probability vector are weighted and fused according to the fusion weight to obtain the comprehensive state probability distribution. Based on the comprehensive state probability distribution, the state category identifier and corresponding state probability of the target grounding terminal are output to generate the initial state monitoring results.

[0050] Specifically, the second-state evaluation vector is first standardized. The historical time window length is determined to be the previous 10 time steps, and the mean and standard deviation of the initial output temporal risk scores of the second-state evaluation vector within this window are calculated. The temporal risk score of the current time step is preprocessed by subtracting the historical mean from the current score and then dividing by the historical standard deviation to obtain the standardized temporal risk score. This eliminates the influence of differences in the benchmark scores across different scenarios. For example, in scenarios with significant environmental interference, the historical temporal risk score mean is high; preprocessing can prevent scores within the normal range from being misjudged as high-risk. The standardized temporal risk score is then substituted into the Sigmoid function, mapped to the [0,1] interval, and converted into a probability vector. This vector has the same dimensions as the first-state evaluation vector and can more accurately reflect the relative abnormality of the current dynamic characteristics.

[0051] Its dimension conversion formula is: in, The first element of the transformed probability vector output for the Long Short-Term Memory network model is... Represents normal probability, the second element This represents the failure probability, thus converting the outputs of both models into probability vectors in the same semantic space; After standardization, a fusion weight is dynamically generated between the first-state evaluation vector and the transformed probability vector based on an attention mechanism. The core of the attention mechanism is to determine the importance of static and dynamic features to the current state evaluation based on the multi-dimensional feature vector at the current time step, and then assign different weights accordingly. First, a context feature vector for the current time step needs to be constructed. This vector is formed by concatenating the hidden states of the Long Short-Term Memory network, the first-state evaluation vector, and the multi-dimensional feature vector of the current time step. After concatenation, a fully connected layer is used for linear transformation and dimensionality reduction to obtain a dimensionally adapted context feature vector. This context feature vector is then input into the attention mechanism, and the fusion weight is calculated using the Sigmoid activation function. Here, is a learnable parameter that is optimized through backpropagation along with the Random Forest model and the Long Short-Term Memory network model during model training. The training objective is to minimize the cross-entropy loss of grounding state classification. The fusion weight ranges from [0,1]. A larger value indicates higher importance of the transformed probability vector (dynamic feature evaluation result), while a smaller value indicates higher importance of the first-state evaluation vector (static feature evaluation result).

[0052] The formula for calculating attention weights is as follows: in, For the Sigmoid function, and For learnable parameters, , As part of the RF-LSTM hybrid model, it is optimized together with random forest and long short-term memory network through backpropagation during model training. The training objective is to minimize the cross-entropy loss of grounded state classification. Attention weights; This indicates the importance of the LSTM output.

[0053] The first state evaluation vector and the transformed probability vector are weighted and fused based on the generated fusion weights to obtain the comprehensive state probability distribution. Specifically, the transformed probability vector is multiplied by the fusion weight, the first state evaluation vector is multiplied by one and subtracted from the fusion weight, and the two results are added together to obtain the comprehensive state probability distribution. For example, if the first state evaluation vector is [value missing], the transformed probability vector is [value missing], and the fusion weight is 0.6, then the comprehensive state probability distribution is [value missing]. This distribution is also a two-dimensional vector, with elements corresponding to the probabilities of normal and fault states, respectively, and the sum of the two elements is 1. Finally, the initial state monitoring results are generated based on the comprehensive state probability distribution. A state determination threshold is set, which is adjustable, with a recommended range of 0.5 to 0.7. The threshold is set to balance recall and precision in binary classification problems. A lower threshold is chosen if the tolerance for false negatives is low, and a higher threshold is chosen if the tolerance for false positives is low. The specific value needs to be verified and determined based on historical data.

[0054] The fault state probability in the comprehensive state probability distribution is compared with a threshold. If the fault state probability is greater than the threshold, the state category of the target grounding terminal is determined to be fault, and the state category identifier is output as fault and the corresponding fault state probability. If the fault state probability is less than or equal to the threshold, the state category is determined to be normal, and the state category identifier is output as normal and the corresponding normal state probability, thereby generating a complete initial state monitoring result.

[0055] S5: Obtain at least one environmental parameter of the environment in which the target grounding terminal is located, and dynamically calibrate the initial state monitoring results based on the environmental parameters to obtain the calibrated grounding state monitoring results.

[0056] While the initial state monitoring results integrate evaluation information from both static and dynamic characteristics, they do not consider the impact of environmental parameters on the operating status of the grounding terminal. Since the mechanical connection stability and electrical conductivity of the grounding terminal change significantly with environmental variations—for example, high temperature and humidity environments accelerate metal corrosion leading to increased grounding resistance, while low temperature environments may cause the nut to appear loose due to thermal expansion and contraction—ignoring these environmental factors can easily lead to misjudgments in the initial monitoring results, failing to accurately reflect the actual operating status. Therefore, step S5 needs to introduce environmental parameters and use a fuzzy evaluation system to dynamically calibrate the initial state monitoring results, correcting deviations caused by environmental interference and obtaining grounding state monitoring results that better reflect the actual scenario.

[0057] Preferably, in one embodiment of the present invention, the initial state monitoring results are dynamically calibrated based on environmental parameters to obtain calibrated grounding state monitoring results, including: Construct a fuzzy evaluation system, and map the initial state monitoring results to the corresponding membership degree representation based on the fuzzy evaluation system; Based on environmental parameters, the weights of the fuzzy evaluation system are dynamically adjusted, and based on the adjusted fuzzy evaluation system, the membership identifiers are processed to generate calibration factors. The initial state monitoring results are corrected using a calibration factor to obtain the calibrated grounding state monitoring results.

[0058] Specifically, the environmental parameters of the target grounding terminal are first obtained. Ambient temperature and humidity, which significantly affect the grounding state, are selected as core parameters. These are collected in real-time by temperature and humidity sensors deployed near the grounding terminal, with the collection frequency consistent with the generation frequency of the grounding feature sequence (e.g., once per minute), ensuring a one-to-one correspondence between the environmental parameters and the initial state monitoring results in the time dimension. After collection, the environmental parameters are preprocessed. Min-max normalization is used to scale the temperature values ​​(in degrees Celsius) and humidity values ​​(in relative humidity) to the [0,1] range, eliminating the impact of dimensional differences on subsequent weight adjustments.

[0059] Next, a fuzzy evaluation system was constructed, and the initial state monitoring results were mapped to membership degree representations. The fuzzy evaluation system was constructed with grounding state categories as the evaluation objects, and "normal," "suspected fault," and "fault" as three evaluation levels, each corresponding to a specific characteristic range. Specifically, the normal level corresponds to a fault probability ≤ 0.3 in the comprehensive state probability distribution, the suspected fault level corresponds to a fault probability of 0.3~0.7, and the fault level corresponds to a fault probability ≥ 0.7.

[0060] Based on expert experience, membership functions for each evaluation level are defined, and a triangular membership function is used to map the initial state monitoring results: Let the fault probability in the initial state monitoring results be P. If P ≤ 0.3, the membership degree of the normal level is 1, and the membership degree of the suspected fault and the fault level is 0; if 0.3 < P < 0.7, the membership degree of the normal level is (0.7-P) / 0.4, the membership degree of the suspected fault level is (P-0.3) / 0.4, and the membership degree of the fault level is 0; if P ≥ 0.7, the membership degree of the normal level and the suspected fault level is 0, and the membership degree of the fault level is 1. Through this mapping process, the probability values ​​of the initial state monitoring results are converted into membership degree representations in fuzzy mathematics, forming a membership degree vector. The vector elements correspond to the membership degrees of the normal, suspected fault, and fault levels, respectively.

[0061] The weights of the fuzzy evaluation system are dynamically adjusted based on environmental parameters, specifically including: The initial weights for each evaluation level are set as follows: the initial weight for the normal level is 0.3, the initial weight for the suspected fault level is 0.5, and the initial weight for the fault level is 0.2. The initial weights are set based on the fact that the suspected fault status in the initial state monitoring results needs to be given special attention, so they are assigned higher weights.

[0062] The weights are adjusted based on the pre-treated environmental parameters: an environmental impact coefficient K is defined, which is calculated from the standardized temperature value T and humidity value H, using the formula K=0.6T+0.4H. Temperature has a greater impact on corrosion and mechanical properties, so it is assigned a higher coefficient.

[0063] When K < 0.3, the initial weights remain unchanged; when 0.3 ≤ K < 0.7, the weight of the fault level is increased to 0.3, the weight of the suspected fault level is adjusted to 0.4, and the weight of the normal level remains at 0.3; when K ≥ 0.7, the weight of the fault level is further increased to 0.4, the weight of the suspected fault level is adjusted to 0.3, and the weight of the normal level is reduced to 0.3. After the weight adjustment, normalization processing is required to ensure that the sum of the weights of the three evaluation levels is 1, resulting in the adjusted weight vector.

[0064] The membership description is processed based on the adjusted fuzzy evaluation system to generate calibration factors. Specifically, the membership vector and the adjusted weight vector are multiplied by a dot product to obtain the fuzzy comprehensive evaluation result S, where the value of S is in the range of [0,1].

[0065] The calibration factor C is determined based on the numerical range of S: if S≤0.3 (biased to normal), the calibration factor C=0.9, which is used to reduce the initial failure probability and correct oversensitivity under low environmental influence; if 0.3<S<0.7 (biased to suspected failure), the calibration factor C=1.0, keeping the initial failure probability unchanged; if S≥0.7 (biased to failure), the calibration factor C=1.1, which is used to increase the initial failure probability and strengthen risk identification under high environmental influence.

[0066] Finally, the initial state monitoring results are corrected using a calibration factor to obtain the calibrated grounding state monitoring results. Specifically, the correction object is the fault probability P in the initial state monitoring results, and the correction formula is P'=P×C, where P' is the calibrated fault probability.

[0067] After correction, boundary constraints need to be applied to P': if P' > 1, then set it to 1; if P' < 0, then set it to 0, to ensure the reasonableness of the probability value. The grounding status category is re-determined based on P': P' ≤ 0.3 is considered normal, 0.3 < P' < 0.7 is considered a suspected fault, and P' ≥ 0.7 is considered a fault. Simultaneously, the calibrated fault probability P' is output, forming a complete calibrated grounding status monitoring result.

[0068] In another embodiment of the present invention, after receiving the grounding status monitoring result of the target grounding terminal, the system will ensure that the user obtains fault information and responds in a timely manner through sound and light coordinated alarm, real-time status monitoring and dynamic upgrading of alarm intensity, and realize data traceability of the entire alarm process.

[0069] The multimodal alarm module mainly consists of a monitoring alarm unit and supporting signal transmission and control components. The monitoring alarm unit includes an alarm housing, buzzer, alarm light, operation buttons, internal circuit board, and power supply battery. The alarm housing provides physical protection and a mounting platform for the module. The internal circuit board is used to parse alarm commands and generate control signals for the buzzer and alarm light. The buzzer outputs an audible alarm signal and supports volume gradation using pulse width modulation technology. The alarm light is controlled by an RGB LED driver chip, outputting only a bright red signal to indicate the alarm status. The power supply battery is an 18650 lithium battery pack, with a 3-cell series design to achieve a 12.6V charging cutoff voltage, a 9V discharging cutoff voltage, and a 3000mAh capacity. A full charge allows the module to operate continuously for approximately 48 hours. For extended monitoring, an external charger can be used. The operation buttons allow users to manually confirm or deactivate the alarm, enabling human-machine interaction.

[0070] In addition, the module also includes a hardware watchdog component and non-volatile memory: the hardware watchdog component is used to monitor the execution status of alarm commands and prevent command loss due to hardware failure; the non-volatile memory is used to store alarm process data, including command generation time, hardware response time, user feedback time, etc., to provide data support for post-event auditing and root cause analysis.

[0071] When the target grounding terminal is determined to be in a fault state, the multi-modal alarm module immediately initiates its workflow, which includes: Command generation and parsing: After the intelligent analysis and prediction module determines the fault, the module triggers initialization, generates standardized alarm commands in binary code according to the preset protocol, calls the monitoring alarm device to parse them into buzzer and alarm light control signals, and at the same time starts the hardware watchdog to monitor command execution.

[0072] Sound and light synchronized alarm: The control signal is transmitted synchronously to the buzzer and alarm light through the parallel interface to avoid timing deviation; the signal cycle is aligned by the global clock to ensure that the sound and light are activated at the same time, the buzzer sounds at the basic volume, and the alarm light lights up red.

[0073] Status monitoring and response: Start a thread to monitor hardware feedback (signal strength, device health status) and open a user response window; if the user presses a button within a preset time, the alarm will stop; if there is no feedback, the buzzer volume will be turned up to the maximum and a secondary alarm command will be sent to the upper-level monitoring platform.

[0074] Data logging: Non-volatile memory records instruction generation time, hardware response time, user feedback time, and device status data in real time. Data is not lost when power is off and is available for subsequent analysis.

[0075] The power grounding status monitoring method in this embodiment of the invention revolves around the accurate monitoring, analysis, and early warning of the operating status of grounding terminals. It constructs a complete technical solution through multi-module collaboration, solving problems in traditional grounding monitoring such as reliance on manual inspections, delayed fault identification, and susceptibility to environmental interference. This achieves intelligent closed-loop management of grounding status from data acquisition to risk early warning. For details, please refer to... Figure 2 , Figure 2 The diagram shows a working process of a power grounding status monitoring method according to one embodiment of the present invention.

[0076] Another embodiment of the present invention provides a power grounding status monitoring system, for details, see [link to documentation]. Figure 3 , Figure 3 The diagram shown illustrates a power grounding status monitoring system according to one embodiment of the present invention, which includes: Acquisition module 11 is used to acquire the timing mechanical state signal and timing electrical state signal of the target grounding terminal; Integration module 12 is used to integrate the feature extraction results of time-series mechanical state signals and the feature extraction results of time-series electrical state signals, and obtain a multi-dimensional feature vector based on the integration results to characterize the comprehensive operating state of the target grounding terminal. The multi-dimensional feature vector includes at least static feature components and dynamic feature components. Evaluation module 13 is used to input the multidimensional feature sequence into the pre-built grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature component and a second state evaluation vector corresponding to the dynamic feature component. The generation module 14 is used to generate initial state monitoring results based on the first state evaluation vector and the second state evaluation vector; The calibration module 15 is used to acquire at least one environmental parameter of the environment in which the target grounding terminal is located, and to dynamically calibrate the initial state monitoring results based on the environmental parameters to obtain the calibrated grounding state monitoring results.

[0077] Preferably, in one embodiment of the present invention, the integration module includes: The first time-domain analysis unit is used to perform time-domain analysis on the time-series mechanical state signal, extract the pressure change frequency as a dynamic feature component, and extract the pressure amplitude as a static feature component. The second time-domain analysis unit is used to perform time-domain analysis on the time-series electrical state signal, calculate the fluctuation range of the grounding resistance in the power circuit as a dynamic characteristic component, and calculate the standard deviation of the grounding resistance as a static characteristic component. The alignment unit is used to synchronously align and combine the extracted pressure change frequency, pressure amplitude, fluctuation range and standard deviation according to the corresponding timestamps to construct a multi-dimensional feature vector.

[0078] Preferably, in one embodiment of the present invention, the grounding state analysis model is a hybrid model, and the evaluation module includes: The static evaluation unit is used to input static feature components into the random forest model for static evaluation and output the first state evaluation vector. The dynamic evaluation unit is used to input dynamic feature components into the long short-term memory network model for dynamic evaluation and output a second-state evaluation vector.

[0079] Preferably, in one embodiment of the present invention, the generation module includes: The standardization processing unit is used to standardize the second state evaluation vector and convert it into a probability vector with the same dimensions as the first state evaluation vector. The weighted fusion unit is used to dynamically generate fusion weights between the first state evaluation vector and the probability vector based on the multi-dimensional feature vector of the current time step, and to perform weighted fusion of the first state evaluation vector and the probability vector according to the fusion weights to obtain the comprehensive state probability distribution. The probability generation unit is used to output the state category identifier and corresponding state probability of the target grounding terminal according to the comprehensive state probability distribution, and generate the initial state monitoring results.

[0080] Preferably, in one embodiment of the present invention, the calibration module includes: The system construction unit is used to construct a fuzzy evaluation system and map the initial state monitoring results to the corresponding membership degree representation based on the fuzzy evaluation system. The dynamic adjustment unit is used to dynamically adjust the weights of the fuzzy evaluation system based on environmental parameters, and to process the membership identifiers based on the adjusted fuzzy evaluation system to generate calibration factors. The result calibration unit is used to correct the initial state monitoring results using calibration factors to obtain calibrated grounding state monitoring results.

[0081] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: (1) This invention constructs a comprehensive state feature vector that simultaneously includes static and dynamic feature components by synchronously acquiring and fusing the time-series mechanical and electrical state signals of the grounding terminal. A pre-built hybrid analysis model is then used to differentiate and integrate this vector, achieving a comprehensive and in-depth collaborative perception of the connection status from both mechanical tightness and electrical connectivity dimensions. This method overcomes the limitations of single-parameter monitoring and can effectively distinguish grounding status anomalies caused by different reasons such as loose nuts, contact surface corrosion, and environmental interference, significantly improving the accuracy of status diagnosis and the specificity of fault mode identification.

[0082] (2) This invention introduces a dynamic calibration mechanism based on environmental parameters to perform secondary correction on the initial monitoring results of the model, effectively eliminating the interference of external variables such as environmental temperature and humidity on the diagnostic conclusions, and significantly enhancing the environmental robustness and result stability of the monitoring system. This invention can adaptively adjust the evaluation weights according to real-time operating conditions, so that the final output grounding status monitoring results not only reflect the current connection performance, but also have the ability to predict the evolution of potential risks, thereby providing operation and maintenance personnel with more accurate and reliable status warnings and maintenance guidance, and strongly supporting the intelligent transformation of power system safe operation from "post-event handling" to "pre-event prevention".

[0083] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for monitoring the status of power grounding, characterized in that, include: Acquire the timing mechanical status signal and timing electrical status signal of the target grounding terminal; The feature extraction results of the time-series mechanical state signal and the feature extraction results of the time-series electrical state signal are integrated, and a multi-dimensional feature vector is obtained based on the integrated results to characterize the comprehensive operating state of the target grounding terminal. The multi-dimensional feature vector includes at least static feature components and dynamic feature components. The multidimensional feature sequence is input into a pre-constructed grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature component and a second state evaluation vector corresponding to the dynamic feature component. Based on the first state evaluation vector and the second state evaluation vector, the initial state monitoring result is generated; At least one environmental parameter of the environment in which the target grounding terminal is located is obtained, and the initial state monitoring result is dynamically calibrated based on the environmental parameter to obtain the calibrated grounding state monitoring result.

2. The power grounding status monitoring method as described in claim 1, characterized in that, The feature extraction results of the integrated time-series mechanical state signal and the feature extraction results of the integrated time-series electrical state signal are used to obtain a multi-dimensional feature vector characterizing the comprehensive operating state of the target grounding terminal, including: The time-domain analysis of the time-series mechanical state signal is performed to extract the pressure change frequency as a dynamic feature component and the pressure amplitude as a static feature component. Time-domain analysis is performed on the time-series electrical state signal to calculate the fluctuation range of the grounding resistance in the power circuit as a dynamic characteristic component, and the standard deviation of the grounding resistance is calculated as a static characteristic component. The extracted pressure change frequency, pressure amplitude, fluctuation range, and standard deviation are synchronously aligned and combined according to their corresponding timestamps to construct the multidimensional feature vector.

3. The power grounding status monitoring method as described in claim 1, characterized in that, The grounding state analysis model is a hybrid model. The step of inputting the multidimensional feature sequence into the pre-constructed grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature components and a second state evaluation vector corresponding to the dynamic feature components includes: The static feature components are input into the random forest model for static evaluation, and the first state evaluation vector is output. The dynamic feature components are input into the Long Short-Term Memory network model for dynamic evaluation, and the second state evaluation vector is output.

4. The power grounding status monitoring method as described in claim 1, characterized in that, The step of generating initial state monitoring results based on the first state evaluation vector and the second state evaluation vector includes: The second state evaluation vector is standardized and converted into a probability vector with the same dimensions as the first state evaluation vector. Based on the attention mechanism, according to the multidimensional feature vector at the current time step, a fusion weight is dynamically generated between the first state evaluation vector and the probability vector. The first state evaluation vector and the probability vector are weighted and fused according to the fusion weight to obtain a comprehensive state probability distribution. Based on the comprehensive state probability distribution, the state category identifier and corresponding state probability of the target grounding terminal are output, and the initial state monitoring result is generated.

5. The power grounding status monitoring method as described in claim 1, characterized in that, The step of dynamically calibrating the initial state monitoring results based on the environmental parameters to obtain calibrated grounding state monitoring results includes: Construct a fuzzy evaluation system, and map the initial state monitoring results to the corresponding membership degree representation based on the fuzzy evaluation system; Based on the environmental parameters, the weights of the fuzzy evaluation system are dynamically adjusted, and based on the adjusted fuzzy evaluation system, the membership identifier is processed to generate a calibration factor. The initial state monitoring results are corrected using the calibration factor to obtain the calibrated grounding state monitoring results.

6. A power grounding status monitoring system, characterized in that, include: The acquisition module is used to acquire the timing mechanical state signal and timing electrical state signal of the target grounding terminal; An integration module is used to integrate the feature extraction results of the time-series mechanical state signal and the feature extraction results of the time-series electrical state signal, and obtain a multi-dimensional feature vector based on the integration results to characterize the comprehensive operating state of the target grounding terminal. The multi-dimensional feature vector includes at least static feature components and dynamic feature components. An evaluation module is used to input the multidimensional feature sequence into a pre-built grounding state analysis model to obtain a first state evaluation vector corresponding to the static feature component and a second state evaluation vector corresponding to the dynamic feature component. The generation module is used to generate initial state monitoring results based on the first state evaluation vector and the second state evaluation vector; The calibration module is used to acquire at least one environmental parameter of the environment in which the target grounding terminal is located, and to dynamically calibrate the initial state monitoring results based on the environmental parameters to obtain calibrated grounding state monitoring results.

7. A power grounding status monitoring system as described in claim 6, characterized in that, The integration module includes: The first time-domain analysis unit is used to perform time-domain analysis on the time-series mechanical state signal, extract the pressure change frequency as a dynamic feature component, and extract the pressure amplitude as a static feature component. The second time-domain analysis unit is used to perform time-domain analysis on the time-series electrical state signal, calculate the fluctuation range of the grounding resistance in the power circuit as a dynamic characteristic component, and calculate the standard deviation of the grounding resistance as a static characteristic component. The alignment unit is used to synchronously align and combine the extracted pressure change frequency, pressure amplitude, fluctuation range and standard deviation according to the corresponding timestamps to construct the multidimensional feature vector.

8. A power grounding status monitoring system as described in claim 6, characterized in that, The grounding state analysis model is a hybrid model, and the evaluation module includes: The static evaluation unit is used to input the static feature components into the random forest model for static evaluation and output the first state evaluation vector. The dynamic evaluation unit is used to input the dynamic feature components into the long short-term memory network model for dynamic evaluation and output the second state evaluation vector.

9. A power grounding status monitoring system as described in claim 6, characterized in that, The generation module includes: The standardization processing unit is used to standardize the second state evaluation vector and convert it into a probability vector with the same dimensions as the first state evaluation vector. The weighted fusion unit is used to dynamically generate fusion weights between the first state evaluation vector and the probability vector based on the multidimensional feature vector at the current time step, according to the attention mechanism, and to perform weighted fusion of the first state evaluation vector and the probability vector according to the fusion weights to obtain a comprehensive state probability distribution. The probability generation unit is used to output the state category identifier and corresponding state probability of the target grounding terminal according to the comprehensive state probability distribution, and generate the initial state monitoring result.

10. A power grounding status monitoring system as described in claim 6, characterized in that, The calibration module includes: The system construction unit is used to construct a fuzzy evaluation system and map the initial state monitoring results into corresponding membership degree representations based on the fuzzy evaluation system. The dynamic adjustment unit is used to dynamically adjust the weights of the fuzzy evaluation system based on the environmental parameters, and to process the membership identifier based on the adjusted fuzzy evaluation system to generate a calibration factor. The result calibration unit is used to correct the initial state monitoring result using the calibration factor to obtain the calibrated grounding state monitoring result.

Images