A self-diagnosis system for intelligent power distribution cabinet

Abstract

The present invention relates to the field of self-diagnosis technology for power distribution cabinets, and in particular to a self-diagnosis system for an intelligent power distribution cabinet that acquires key parameters such as busbar temperature, cabinet humidity, path connectivity, current and voltage in real time through a parameter acquisition module in the cabinet, so as to realize intelligent monitoring of the power distribution cabinet. Compared with the traditional method that relies on manual inspection, it can automatically collect and analyze, significantly improve the real-time performance of monitoring and abnormal response speed, detect abnormalities and issue early warnings at an early stage, and avoid the spread of potential faults. Between the power distribution cabinets, a pre-association relationship is established through the power distribution cabinet communication module to achieve data synchronization between multiple cabinets. When an abnormal fluctuation is detected, the initial analysis module determines the abnormality type, and obtains the data of the associated distribution cabinet when necessary, and submits it to the Transformer network of the self-diagnosis module for in-depth analysis to clarify the source and diffusion path of the abnormality, solving the problem that it is difficult to locate the root cause of the fault in single-cabinet monitoring in the traditional system.

Description

Technical Field

[0001] The invention relates to the technical field of self-diagnosis of power distribution cabinets, and in particular to a self-diagnosis system of an intelligent power distribution cabinet. Background Art

[0002] In the actual operation of the power distribution system, there is usually a progressive power transmission relationship between distribution cabinets. The hierarchical structure of high-voltage, medium-voltage and low-voltage distribution cabinets in power distribution means that an abnormality in one distribution cabinet may cause operating problems in the next-level distribution cabinet. For example, when the main distribution cabinet experiences current or voltage fluctuations, it may cause abnormal fluctuations or unstable loads in the lower-level distribution cabinets directly connected to it. Therefore, the abnormality of a single distribution cabinet is often not a local problem, but may spread to other related distribution cabinets, especially when the load demand increases or the power supply pressure is high, the conductivity of this abnormality is more significant.

[0003] In the prior art, distribution cabinets usually rely on single-cabinet monitoring and passive maintenance. This method has the following significant disadvantages: Lack of real-time and intelligence: Traditional distribution cabinets usually rely on regular manual inspections or simple alarm systems to monitor anomalies. This method is difficult to provide real-time monitoring and response in the event of sudden abnormalities, which can easily lead to the accumulation of potential problems and affect the safety of the distribution system. Isolated single-cabinet monitoring: Existing distribution cabinets usually operate independently and can only monitor their own operating parameters, lacking linkage with other distribution cabinets. This isolated monitoring mode makes it impossible to analyze whether the source of an abnormality comes from the influence of the previous distribution cabinet when it occurs, especially the abnormal fluctuations of current, voltage, etc., which may be caused by the conduction of the upper distribution cabinet, but the single-cabinet monitoring cannot detect the source, resulting in inaccurate fault location. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a self-diagnosis system for an intelligent distribution cabinet, which analyzes the parameters in the cabinet through deep learning and combines the abnormal types to perform cross-distribution cabinet analysis and synchronization, effectively improving the accuracy of fault location and abnormal type identification.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A self-diagnosis system for an intelligent power distribution cabinet, comprising: a power distribution cabinet communication module, an in-cabinet parameter acquisition module, an initial analysis module and a self-diagnosis module connected in sequence, wherein the self-diagnosis module is connected to the power distribution cabinet communication module;

[0007] The cabinet parameter acquisition module is used to collect real-time operating parameters of the power distribution cabinet, and the operating parameters include bus temperature, cabinet humidity, connection status of each channel, and current and voltage;

[0008] The power distribution cabinet communication module is used to exchange data with the pre-associated power distribution cabinet, obtain the real-time operating parameters of the pre-associated power distribution cabinet, and is also used to synchronize the analysis results of the self-diagnosis module with the pre-associated power distribution cabinet;

[0009] The initial analysis module is used to determine the abnormality type through a feedforward neural network based on the real-time operating parameters of the current power distribution cabinet, and determine whether it is necessary to obtain the real-time operating parameters of the pre-associated power distribution cabinet according to the abnormality type. If not, a preset response strategy is matched according to the abnormality type; if necessary, the real-time operating parameters of the pre-associated power distribution cabinet are obtained through the power distribution cabinet communication module, and the real-time operating parameters of the pre-associated power distribution cabinet are integrated with the real-time operating parameters of the current power distribution cabinet and sent to the self-diagnosis module;

[0010] The self-diagnosis module is used to perform multi-dimensional feature analysis based on the real-time operating parameters of the pre-associated distribution cabinet and the real-time operating parameters of the current distribution cabinet through the Transformer network, determine the source of the abnormality and generate analysis results, and feed the analysis results back to the distribution cabinet communication module.

[0011] Furthermore, the cabinet parameter acquisition module includes a temperature sensor, a humidity sensor, a current sensor, a voltage sensor and a microprocessor, and the temperature sensor, the humidity sensor, the current sensor and the voltage sensor are respectively connected to the microprocessor.

[0012] Furthermore, the pre-associated distribution cabinets are pre-associated in the following manner: based on the progressive hierarchy of high-voltage, medium-voltage and low-voltage distribution cabinets in power distribution, the upper-level main distribution cabinet is pre-associated with the distribution cabinets it directly supplies power to.

[0013] Furthermore, the abnormality types include temperature abnormality, humidity abnormality, current abnormality and voltage abnormality.

[0014] Further, the method of determining the abnormality type based on the real-time operating parameters of the power distribution cabinet through a feedforward neural network and determining whether it is necessary to obtain the real-time operating parameters of the pre-associated power distribution cabinet according to the abnormality type includes the following steps:

[0015] After receiving the real-time operating parameters of the power distribution cabinet, denoising and normalization are performed to generate an input feature vector;

[0016] The preprocessed feature vector is input into the feedforward neural network, the feature vector is processed by multiple layers of neurons, and the outlier value is output;

[0017] If the abnormal value is greater than the preset abnormal threshold, the abnormal type is determined through the output classification node of the neural network; if the abnormal value is less than or equal to the preset abnormal threshold, no operation is performed;

[0018] Based on the determined abnormality type, determine whether it is necessary to obtain the operating parameters of the pre-associated distribution cabinet. If the abnormality type is current abnormality or voltage fluctuation, request the real-time operating parameters of the pre-associated distribution cabinet through the distribution cabinet communication module; if the abnormality type is temperature abnormality or humidity abnormality, match the preset response strategy of this distribution cabinet.

[0019] Furthermore, the multi-dimensional feature analysis based on the real-time operating parameters of the pre-associated power distribution cabinet and the real-time operating parameters of the power distribution cabinet through the Transformer network includes the following steps:

[0020] Standardize the real-time operating parameters of the current distribution cabinet and the pre-associated distribution cabinet, and encode these features into input tensors, adding time series and position encoding;

[0021] The encoded feature tensor is input into the self-attention layer of the Transformer network, and the weighted feature vector is output by calculating the attention weights between each time step;

[0022] The weighted feature vector output by the self-attention mechanism is input into the fully connected layer of the Transformer network for feature aggregation and anomaly score calculation. By comparing it with the preset anomaly threshold, it is determined whether there is an anomaly, and the location and impact range of the anomaly are determined.

[0023] Furthermore, the formula of the Transformer network is as follows:

[0024] ;

[0025] Among them, Output is the calculation result of the abnormal source; softmax is the normalization function; , and The query vector, key vector, and value vector generated for temperature, humidity, current, and voltage parameters, respectively; , and They are , and The corresponding weighting matrix; is the dimension of the key vector.

[0026] Furthermore, the analysis results include the abnormal source distribution cabinet, the abnormality type and the severity of the abnormality.

[0027] Furthermore, the data synchronization of the analysis result of the self-diagnosis module to the pre-associated power distribution cabinet includes: the pre-associated power distribution cabinet integrates the synchronized analysis result into its own abnormality monitoring record, updates its own abnormality type and performs safety adjustment.

[0028] The beneficial effect of the present invention is that the present invention acquires key parameters such as busbar temperature, humidity in the cabinet, connectivity status of each channel, current and voltage in real time through the parameter acquisition module in the cabinet, so as to realize real-time monitoring of the distribution cabinet. Compared with the traditional method of relying on manual inspection or simple alarm, the present solution improves the real-time and intelligent level of monitoring through automated real-time acquisition and analysis, can detect abnormalities in time and issue early warnings, and effectively prevent the accumulation of potential problems and the further spread of faults. In addition, the distribution cabinet communication module in the present solution allows the establishment of pre-association relationships between different distribution cabinets, and can exchange and synchronize the operation data of the pre-associated upper or lower distribution cabinets. When abnormal conditions such as current and voltage fluctuations that may be transmitted by the upper level are detected, the initial analysis module can automatically obtain the operation data of the associated distribution cabinet, and pass it to the self-diagnosis module for in-depth analysis after integration, so as to effectively determine the source and diffusion path of the abnormality. This linkage mechanism greatly enhances the accuracy of fault location and solves the problem that the existing technology cannot trace the source due to the isolation of monitoring. It combines the deep learning technology of feedforward neural network and Transformer network. The initial analysis module uses a feedforward neural network to identify the type of anomaly and automatically determines whether it is necessary to call the data of the associated distribution cabinet, ensuring that data is exchanged only when necessary, optimizing the use of system resources. Subsequently, the self-diagnosis module uses the Transformer network to perform multi-dimensional feature analysis, identifies key abnormal factors and conduction paths through the self-attention mechanism, and generates accurate diagnostic results. This technology enables in-depth analysis of different types of anomalies (such as temperature, current or voltage anomalies), solving the problem that traditional distribution cabinets cannot achieve multi-dimensional anomaly analysis. Through the linkage synchronization function of the diagnostic results, it ensures that abnormal situations are coordinated across distribution cabinets, effectively improving the overall safety and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a structural schematic diagram of a self-diagnosis system of an intelligent power distribution cabinet in the present invention.

[0030] Figure 2 It is a flowchart of the steps of performing multi-dimensional feature analysis based on the real-time operating parameters of the pre-associated distribution cabinet and the real-time operating parameters of the distribution cabinet through the Transformer network in the present invention. DETAILED DESCRIPTION

[0031] See also Figure 1-Figure 2 As shown, the present invention relates to a self-diagnosis system of an intelligent power distribution cabinet, comprising: a power distribution cabinet communication module, an in-cabinet parameter acquisition module, an initial analysis module and a self-diagnosis module connected in sequence, wherein the self-diagnosis module is connected to the power distribution cabinet communication module;

[0032] The cabinet parameter acquisition module is used to collect real-time operating parameters of the power distribution cabinet, and the operating parameters include bus temperature, cabinet humidity, connection status of each channel, and current and voltage;

[0033] The power distribution cabinet communication module is used to exchange data with the pre-associated power distribution cabinet, obtain the real-time operating parameters of the pre-associated power distribution cabinet, and is also used to synchronize the analysis results of the self-diagnosis module with the pre-associated power distribution cabinet;

[0034] The initial analysis module is used to determine the abnormality type through a feedforward neural network based on the real-time operating parameters of the current power distribution cabinet, and determine whether it is necessary to obtain the real-time operating parameters of the pre-associated power distribution cabinet according to the abnormality type. If not, a preset response strategy is matched according to the abnormality type; if necessary, the real-time operating parameters of the pre-associated power distribution cabinet are obtained through the power distribution cabinet communication module, and the real-time operating parameters of the pre-associated power distribution cabinet are integrated with the real-time operating parameters of the current power distribution cabinet and sent to the self-diagnosis module;

[0035] The self-diagnosis module is used to perform multi-dimensional feature analysis based on the real-time operating parameters of the pre-associated distribution cabinet and the real-time operating parameters of the current distribution cabinet through the Transformer network, determine the source of the abnormality and generate analysis results, and feed the analysis results back to the distribution cabinet communication module.

[0036] Furthermore, the cabinet parameter acquisition module includes a temperature sensor, a humidity sensor, a current sensor, a voltage sensor and a microprocessor, and the temperature sensor, the humidity sensor, the current sensor and the voltage sensor are respectively connected to the microprocessor.

[0037] In some embodiments, the parameter acquisition module in the cabinet serves as a core monitoring unit and includes high-precision acquisition devices such as temperature sensors, humidity sensors, current sensors, and voltage sensors. Each sensor is installed at key nodes such as busbars and access connection parts, and obtains parameter information such as busbar temperature, cabinet humidity, connectivity status of each access, and current and voltage in real time through the data acquisition bus. The microprocessor configured in the module performs denoising, calibration, and standardization on the data to ensure data consistency in the subsequent analysis process. Data acquisition and preprocessing are performed at a millisecond rate to ensure that the system can obtain accurate and real-time data in the early stage of parameter abnormalities. The distribution cabinet communication module realizes data exchange and linkage control between systems. The module uses a high-speed industrial Ethernet interface or a wireless communication chip, and is embedded with a multi-layer anti-interference filtering circuit and an encryption chip to maintain stable transmission performance in an electromagnetic interference environment while ensuring data transmission security. The communication module is used to synchronize data and share status with the pre-associated distribution cabinet. When an abnormality is detected, data can be exchanged smoothly between multiple cabinets, providing a hardware basis for cross-cabinet linkage response. The feedforward neural network hardware architecture of the initial analysis module quickly analyzes the real-time operating parameters provided by the cabinet parameter acquisition module. The feedforward neural network is deployed in a local large model in the power distribution cabinet, and a parallel data processing structure is used to support the rapid classification of multiple parameters. The module determines whether there is an abnormality through GPU calculation, analyzes the abnormality type based on the preliminary detection results, and determines whether it is necessary to obtain real-time data from the pre-associated power distribution cabinet. If there is no need to obtain the data of the associated power distribution cabinet, the preset response strategy is directly matched and executed. The response strategy can be to control the start of the heat dissipation device or notify the staff to check and handle it; if the current or voltage is detected. The associated abnormality, the initial analysis module obtains the real-time data of the associated power distribution cabinet through the power distribution cabinet communication module, and integrates the data and transmits it to the self-diagnosis module. As a high-performance computing unit, the self-diagnosis module realizes multi-dimensional data analysis based on the hardware-accelerated Transformer network. The self-attention mechanism of the Transformer network processes multi-dimensional data through parallel acceleration, and analyzes the correlation between the features of the current power distribution cabinet and the pre-associated power distribution cabinet. Through the deep feature analysis of the network, the self-diagnosis module can identify the conduction source and impact range of the abnormality and form an accurate diagnosis report.

[0038] Furthermore, the pre-associated distribution cabinets are pre-associated in the following manner: based on the progressive hierarchy of high-voltage, medium-voltage and low-voltage distribution cabinets in power distribution, the upper-level main distribution cabinet is pre-associated with the distribution cabinets it directly supplies power to.

[0039] Specifically, in terms of hardware implementation, the communication module in the main distribution cabinet integrates the device identification unit and the power transmission node identification module, which are used to detect and record the device number, location information and operating status of each distribution cabinet that it directly supplies power to. The device identification unit of the main distribution cabinet automatically obtains the identity information of all directly powered distribution cabinets by reading the unique identification code (UID) of the communication module of the distribution cabinet connected downstream to complete the preliminary identification. Subsequently, the main distribution cabinet registers these directly connected distribution cabinets as pre-associated devices to build a hierarchical progressive relationship between power transmission nodes. The data exchange between the main distribution cabinet and the distribution cabinet is carried out using a 5G network to ensure stable and low-latency associated data transmission in a complex electromagnetic environment. The communication module of each distribution cabinet has built-in multi-stage anti-interference filtering chips and automatic error correction circuits to ensure the integrity and accuracy of data synchronization. When the main distribution cabinet detects fluctuations in its own operating status (such as abnormal current or voltage), it can automatically send a diagnostic signal to all pre-associated distribution cabinets, requesting to obtain the operating parameters of the downstream distribution cabinets for integrated analysis, to ensure the stable operation of the hierarchical transmission and associated coordination functions in the power system.

[0040] Furthermore, the abnormality types include temperature abnormality, humidity abnormality, current abnormality and voltage abnormality.

[0041] Further, the method of determining the abnormality type based on the real-time operating parameters of the power distribution cabinet through a feedforward neural network and determining whether it is necessary to obtain the real-time operating parameters of the pre-associated power distribution cabinet according to the abnormality type includes the following steps:

[0042] After receiving the real-time operating parameters of the power distribution cabinet, denoising and normalization are performed to generate an input feature vector;

[0043] The preprocessed feature vector is input into the feedforward neural network, the feature vector is processed by multiple layers of neurons, and the outlier value is output;

[0044] If the abnormal value is greater than the preset abnormal threshold, the abnormal type is determined through the output classification node of the neural network; if the abnormal value is less than or equal to the preset abnormal threshold, no operation is performed;

[0045] Based on the determined abnormality type, determine whether it is necessary to obtain the operating parameters of the pre-associated distribution cabinet. If the abnormality type is current abnormality or voltage fluctuation, request the real-time operating parameters of the pre-associated distribution cabinet through the distribution cabinet communication module; if the abnormality type is temperature abnormality or humidity abnormality, match the preset response strategy of this distribution cabinet.

[0046] In some embodiments, first, the real-time operating parameters of the power distribution cabinet are received and obtained, including busbar temperature, humidity in the cabinet, current and voltage of each channel. Data denoising is performed on these raw data to eliminate random noise caused by environmental interference or equipment fluctuations to ensure the accuracy of data input. Subsequently, the system normalizes the denoised data, converts parameters of different dimensions into dimensionless features, and generates a unified feature vector input to adapt to the input standard of the feedforward neural network. After the data preprocessing is completed, the feature vector is input into the feedforward neural network for anomaly detection analysis. The feedforward neural network consists of multiple layers of neurons, and each layer of neurons transforms and maps the features of the input data step by step through weighting and biasing to extract potential patterns of different anomaly types. The outlier value (score) generated at the output of the network is used to evaluate whether the current data state exceeds the preset threshold. If the outlier value is greater than the set threshold, it means that there may be an abnormality in the system operation state, and the next step of abnormal type determination is entered; if the outlier value is less than or equal to the threshold, it is determined to be a normal state, and no subsequent operations are performed, and the system maintains a normal monitoring state. In the case where the outlier value exceeds the threshold, the feedforward neural network further determines the type of anomaly by outputting a classification node. Specifically, the classification nodes in the network structure set corresponding activation functions for each type of anomaly (temperature anomaly, humidity anomaly, current anomaly, and voltage anomaly) so that the final output value points to a specific anomaly category. If the activation value of the classification node indicates that the anomaly type is current anomaly or voltage anomaly, the system determines that the anomaly may be affected by the associated distribution cabinet, so it requests the real-time operating parameters of the pre-associated distribution cabinet through the distribution cabinet communication module. This operation allows the system to compare and analyze the abnormal state of the current distribution cabinet with the data of the pre-associated distribution cabinet to accurately determine whether there is abnormal conduction across cabinets. If it is determined to be temperature or humidity anomaly, the system believes that the anomaly is limited to the inside of this distribution cabinet and does not need to be associated with the data of other distribution cabinets. The system will directly match the preset response strategy of this distribution cabinet and perform corresponding processing for the anomaly.

[0047] Furthermore, the multi-dimensional feature analysis based on the real-time operating parameters of the pre-associated power distribution cabinet and the real-time operating parameters of the power distribution cabinet through the Transformer network includes the following steps:

[0048] Standardize the real-time operating parameters of the current distribution cabinet and the pre-associated distribution cabinet, and encode these features into input tensors, adding time series and position encoding;

[0049] The encoded feature tensor is input into the self-attention layer of the Transformer network, and the weighted feature vector is output by calculating the attention weights between each time step;

[0050] The weighted feature vector output by the self-attention mechanism is input into the fully connected layer of the Transformer network for feature aggregation and anomaly score calculation. By comparing it with the preset anomaly threshold, it is determined whether there is an anomaly, and the location and impact range of the anomaly are determined.

[0051] In some embodiments, first, the real-time operating parameters of the power distribution cabinet and the pre-associated power distribution cabinet (such as busbar temperature, humidity in the cabinet, current, voltage, etc.) are input into the system, and the data enters the standardized processing flow. The standardization step aims to eliminate dimensional differences and scale the parameters of different physical quantities to the same numerical range to ensure that the influence of each parameter on the model is consistent. After standardization, the data is encoded in the form of an input tensor. This input tensor contains information of multiple dimensions, and by adding time series encoding and position encoding to retain the time dependency and spatial position relationship of the parameters, it ensures that the model can capture the time-varying pattern of features and the physical location of features in the system. The processed input tensor enters the self-attention layer of the Transformer network. In the self-attention mechanism, the feature vector of each time step is associated with the feature vector of other time steps. The self-attention mechanism calculates the attention weight matrix of each time step through the generated query vector, key vector and value vector. Specifically, the self-attention layer calculates the degree of attention of each time step to other time steps, and reflects the relative importance of data at each moment through these weights. These weight matrices are applied to the value vector to generate a weighted feature vector. The weighted feature vector reflects the relationship between different parameters over time, providing a basis for capturing key time steps under abnormal conditions. The weighted feature vector output by the self-attention layer is further input to the fully connected layer of the Transformer network, which integrates the feature information of different time steps through feature aggregation operations to generate anomaly scores. In feature aggregation, the model calculates the contribution of each feature through the fully connected layer, maps the weighted feature vector to anomaly score values, and thus summarizes the overall data pattern in a high dimension. This anomaly score reflects the synergistic state between multi-dimensional parameters and can capture the comprehensive performance of system anomalies in multi-time series and multi-channel features. Finally, the anomaly score is compared and analyzed with the system's preset anomaly threshold. If the score exceeds the threshold, the system is judged to be in an abnormal state, and the specific location and impact range of the anomaly are determined based on the attention distribution in feature aggregation. This process accurately locates the source of the anomaly by identifying important time steps and key features, providing a decision-making basis for further diagnosis and countermeasures. This method not only improves the accuracy of anomaly detection, but also enhances the system's processing capability in identifying anomaly propagation paths, thereby ensuring the overall stability of the distribution cabinet system.

[0052] In the training of the Transformer network, it is necessary to collect multiple real-time operating parameters such as temperature, humidity, current and voltage to construct a multi-dimensional time series data set. Abnormal events (such as current surge and temperature exceeding the limit) are pre-identified and manually annotated to ensure that each data has an accurate abnormal type label and severity level. The annotated data set contains rich samples of normal and abnormal states so that the model can learn to identify abnormal feature patterns and different types of faults. In the data preprocessing stage, all operating parameters are standardized to convert numerical features of different dimensions into the same range. Subsequently, parameters such as temperature, humidity, current and voltage generate query vectors, key vectors and value vectors respectively for self-attention calculation in the Transformer network. In addition, each input data is added with time series encoding and position encoding according to the timestamp and physical location. The time encoding helps the model identify the temporal relationship of the anomaly, while the position encoding is used to capture the correlation of the anomaly in the physical location of the distribution cabinet, ensuring that the model can analyze the characteristics of the time and space dimensions at the same time. After feature encoding, the annotated data is input into the Transformer network for training. The self-attention layer of the model calculates the attention weights between each time step and identifies the key temporal features. In each iteration, through forward propagation, the model gradually adjusts the weight matrices of the query, key, and value vectors to maximize the accuracy and precision of anomaly detection. During the training process, the cross entropy loss function is used to calculate the deviation between the predicted output and the true label. The model parameters are continuously adjusted through back propagation to reduce the error value, ensuring that the model can accurately distinguish between normal and abnormal states and make reasonable judgments on the type and severity of anomalies. In order to improve the performance of the Transformer network under complex data, the learning rate decay strategy and early stopping mechanism are used in the model training process to prevent the model from overfitting or underfitting during the training process. In addition, by using the multi-head attention mechanism in the self-attention mechanism, the model can analyze abnormal features from multiple angles at the same time, further improving the model's recognition ability and classification accuracy for multiple anomalies. After the training is completed, the final model is tested on the validation set and the test set to ensure the generalization ability of the model and its high accuracy in anomaly detection and source determination in the actual distribution cabinet monitoring scenario.

[0053] Furthermore, the formula of the Transformer network is as follows:

[0054] ;

[0055] Among them, Output is the calculation result of the abnormal source; softmax is the normalization function; , and The query vector, key vector, and value vector generated for temperature, humidity, current, and voltage parameters, respectively; , and They are , and The corresponding weighting matrix; is the dimension of the key vector.

[0056] Specifically, the softmax function is used for normalization to convert the attention weights into probability distributions. This operation ensures that each feature can properly reflect its relative importance in multidimensional analysis, allowing the system to give higher weights to key features and help identify important abnormal features. Key, key vector Sum value vector Perform linear transformation to form the overall feature matrix; weighted matrix , and They are used to linearly map the query, key, and value vectors, respectively, to encode the importance of each feature. The introduction of these weighted matrices enables the network to accurately calculate the associated weights of different features, providing a basis for multi-dimensional feature analysis. The dimension of the key vector The square root of is used to scale the score of the entire feature matrix to prevent the eigenvalue from being too large and causing unstable calculations. This scaling step normalizes the feature matrix numerically to improve the stability and accuracy of the network in complex data analysis. After the calculation is completed, the final result is the same as the value vector Multiplication makes the calculation results of anomaly detection more interpretable and can reflect the specific location and degree of anomalies. The product output of this part is matched with the original feature value to ensure that the final result retains the association with the actual parameter.

[0057] Furthermore, the analysis results include the abnormal source distribution cabinet, the abnormality type and the severity of the abnormality.

[0058] First, by correlating the operating parameters of the distribution cabinet and the pre-associated distribution cabinet, the system determines the specific source of the abnormal signal and marks the distribution cabinet at the source of the abnormality. Secondly, the system identifies the type of abnormality through the joint analysis of the feedforward neural network and the Transformer network. Based on the classification results output by the network, the system can accurately distinguish between temperature abnormalities, humidity abnormalities, current abnormalities or voltage abnormalities to ensure that the determination of the abnormal type is consistent with the actual situation. This classification result provides a reliable basis for formulating corresponding response strategies. For example, when it is determined that the current is abnormal, the system will check the current state of the pre-associated distribution cabinet to avoid the spread of systemic faults caused by current fluctuations. The analysis results also include the severity of the abnormality, that is, the severity of the abnormality is quantified by the abnormality score output by the model. This severity is based on the degree of deviation of the abnormal value from the normal threshold. The higher the value, the greater the potential impact of the abnormality on the system. The severity information enables the system to prioritize high-risk abnormalities in subsequent responses and hierarchically manage low-risk abnormalities, optimize the allocation of maintenance resources, and improve the overall response efficiency.

[0059] Furthermore, the data synchronization of the analysis result of the self-diagnosis module to the pre-associated power distribution cabinet includes: the pre-associated power distribution cabinet integrates the synchronized analysis result into its own abnormality monitoring record, updates its own abnormality type and performs safety adjustment.

[0060] Specifically, first, the analysis results generated by the self-diagnosis module are synchronized to the pre-associated distribution cabinet through the distribution cabinet communication module. This process adopts a multi-channel communication protocol to ensure the rapidity and data integrity of the distribution system in the linkage response. The analysis results include information such as the distribution cabinet where the abnormality is sourced, the type of abnormality, and the severity of the abnormality. These data provide a comprehensive status reference for the pre-associated distribution cabinet. After receiving the data, the abnormality monitoring system of the pre-associated distribution cabinet integrates the synchronized analysis results into its abnormality monitoring record. The record update distinguishes the abnormal data of different distribution cabinets through timestamps and source tags, so that each distribution cabinet has a global understanding of the systemic abnormality, thereby providing a data basis for further abnormality tracing and collaborative management of multiple distribution cabinets. Through this record update, the pre-associated distribution cabinet can accumulate and analyze historical abnormal data, which is helpful for subsequent abnormal pattern recognition and trend analysis. Based on the updated record, the pre-associated distribution cabinet starts the corresponding safety adjustment mechanism according to the abnormality type and severity in the synchronization result. Safety adjustment includes measures such as automatic correction of abnormal parameters, load redistribution, or protection status improvement. For example, when the synchronization results indicate that the voltage of the upper-level distribution cabinet is abnormal and the severity is high, the pre-associated lower-level distribution cabinet will automatically reduce its load demand to relieve system pressure and avoid abnormal conduction. These safety adjustment measures are implemented through preset response strategies, effectively reducing the impact of abnormalities on the entire distribution network.

[0061] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A self-diagnosis system for an intelligent power distribution cabinet, characterized in that: include: A power distribution cabinet communication module, an in-cabinet parameter acquisition module, an initial analysis module and a self-diagnosis module connected in sequence, wherein the self-diagnosis module is connected to the power distribution cabinet communication module; The cabinet parameter acquisition module is used to collect real-time operating parameters of the power distribution cabinet, and the operating parameters include bus temperature, cabinet humidity, connection status of each channel, and current and voltage; The power distribution cabinet communication module is used to exchange data with the pre-associated power distribution cabinet, obtain the real-time operating parameters of the pre-associated power distribution cabinet, and is also used to synchronize the analysis results of the self-diagnosis module with the pre-associated power distribution cabinet; The initial analysis module is used to determine the abnormality type through a feedforward neural network based on the real-time operating parameters of the current power distribution cabinet, and determine whether it is necessary to obtain the real-time operating parameters of the pre-associated power distribution cabinet according to the abnormality type. If not, a preset response strategy is matched according to the abnormality type; if necessary, the real-time operating parameters of the pre-associated power distribution cabinet are obtained through the power distribution cabinet communication module, and the real-time operating parameters of the pre-associated power distribution cabinet are integrated with the real-time operating parameters of the current power distribution cabinet and sent to the self-diagnosis module; The self-diagnosis module is used to perform multi-dimensional feature analysis based on the real-time operating parameters of the pre-associated distribution cabinet and the real-time operating parameters of the current distribution cabinet through the Transformer network, determine the source of the abnormality and generate analysis results, and feed the analysis results back to the distribution cabinet communication module.

2. The self-diagnosis system of an intelligent power distribution cabinet according to claim 1 is characterized in that: The cabinet parameter acquisition module comprises a temperature sensor, a humidity sensor, a current sensor, a voltage sensor and a microprocessor, and the temperature sensor, the humidity sensor, the current sensor and the voltage sensor are connected to the microprocessor respectively.

3. The self-diagnosis system of an intelligent power distribution cabinet according to claim 1 is characterized in that: The pre-associated distribution cabinets are pre-associated in the following manner: based on the progressive hierarchy of high-voltage, medium-voltage and low-voltage distribution cabinets in power distribution, the upper-level main distribution cabinet is pre-associated with the distribution cabinets it directly supplies power to.

4. The self-diagnosis system of an intelligent power distribution cabinet according to claim 1, characterized in that: The abnormality types include temperature abnormality, humidity abnormality, current abnormality and voltage abnormality.

5. The self-diagnosis system of an intelligent power distribution cabinet according to claim 4, characterized in that: The method of determining the abnormality type based on the real-time operating parameters of the power distribution cabinet through a feedforward neural network and determining whether it is necessary to obtain the real-time operating parameters of the pre-associated power distribution cabinet according to the abnormality type includes the following steps: After receiving the real-time operating parameters of the power distribution cabinet, denoising and normalization are performed to generate an input feature vector; The preprocessed feature vector is input into the feedforward neural network, the feature vector is processed by multiple layers of neurons, and the outlier value is output; If the abnormal value is greater than the preset abnormal threshold, the abnormal type is determined through the output classification node of the neural network; if the abnormal value is less than or equal to the preset abnormal threshold, no operation is performed; Based on the determined abnormality type, determine whether it is necessary to obtain the operating parameters of the pre-associated distribution cabinet. If the abnormality type is current abnormality or voltage fluctuation, request the real-time operating parameters of the pre-associated distribution cabinet through the distribution cabinet communication module; if the abnormality type is temperature abnormality or humidity abnormality, match the preset response strategy of this distribution cabinet.

6. The self-diagnosis system of an intelligent power distribution cabinet according to claim 1, characterized in that: The multi-dimensional feature analysis based on the real-time operating parameters of the pre-associated power distribution cabinet and the real-time operating parameters of the power distribution cabinet through the Transformer network includes the following steps: Standardize the real-time operating parameters of the current distribution cabinet and the pre-associated distribution cabinet, and encode these features into input tensors, adding time series and position encoding; The encoded feature tensor is input into the self-attention layer of the Transformer network, and the weighted feature vector is output by calculating the attention weights between each time step; The weighted feature vector output by the self-attention mechanism is input into the fully connected layer of the Transformer network for feature aggregation and anomaly score calculation. By comparing it with the preset anomaly threshold, it is determined whether there is an anomaly, and the location and impact range of the anomaly are determined.

7. The self-diagnosis system of an intelligent power distribution cabinet according to claim 6, characterized in that: The formula of the Transformer network is as follows: ； Among them, Output is the calculation result of the abnormal source; softmax is the normalization function; , and The query vector, key vector, and value vector generated for temperature, humidity, current, and voltage parameters, respectively; , and They are , and The corresponding weighting matrix; is the dimension of the key vector.

8. The self-diagnosis system of an intelligent power distribution cabinet according to claim 1, characterized in that: The analysis results include the abnormal source distribution cabinet, the abnormal type and the severity of the abnormality.

9. The self-diagnosis system of an intelligent power distribution cabinet according to claim 8, characterized in that: The data synchronization of the analysis result of the self-diagnosis module to the pre-associated power distribution cabinet includes: the pre-associated power distribution cabinet integrates the synchronized analysis result into its own abnormal monitoring record, updates its own abnormal type and performs safety adjustment.

Images