Intelligent safety protection electric energy metering box

Abstract

This invention relates to the field of Internet of Things (IoT) technology and discloses an intelligent safety protection power metering box, comprising: a sensing module for collecting environmental and electrical status data within the box; a main control module for performing status prediction and fault diagnosis based on a spatiotemporal graph neural network model and generating protection commands; and an execution module for executing electrical interlocks or environmental adjustments according to the commands. This invention, by collecting multi-source data in real time, utilizing the main control module for intelligent prediction, and performing accurate fault diagnosis and risk simulation based on the residuals between prediction and actual measurements, ultimately drives the execution module to complete adaptive closed-loop protection operations. This achieves a shift in metering box safety protection from passive response to proactive early warning, helping to identify potential hazards and accurately locate root causes in the early stages of faults, effectively improving the safety and reliability of equipment operation, and providing technical support for predictive maintenance.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and in particular to an intelligent safety protection power metering box. Background Technology

[0002] As a key infrastructure of the power grid terminal, the electricity metering box undertakes the core functions of accurate electricity metering, billing, and basic power protection. It is a key node connecting the power supply system and end users. The stability and safety of its operation are not only directly related to the fairness and accuracy of electricity metering, but also of vital importance to ensuring the safety of users' property and the reliable operation of the regional power grid.

[0003] Existing electricity metering boxes are generally designed to provide a space that meets basic installation and physical protection requirements. The box typically integrates core components such as meters and circuit breakers, and its main function is limited to collecting and measuring electricity data and providing passive protection against high-voltage faults such as overloads and short circuits. Safety protection measures are also mostly limited to adding physical locks.

[0004] Current technologies typically treat electricity metering boxes as static, passive physical units rather than intelligent nodes capable of proactive sensing and response. This inherently passive approach to protection results in a lack of effective foresight and response capabilities when facing increasingly complex and ever-changing internal and external risks. Specifically, traditional metering boxes cannot continuously monitor key environmental parameters such as temperature, humidity, and vibration within the box, nor can they detect subtle signals that indicate electricity theft or early electrical faults, such as weak leakage currents, external magnetic field interference, or gradual changes in contact resistance. The consequence is severely delayed security protection, often only being discovered after a fault has occurred or even caused damage. This not only leads to metering disputes and equipment damage but can also cause serious safety accidents such as electrical fires, while significantly increasing the manpower costs and difficulty of subsequent maintenance and troubleshooting. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent safety protection power metering box, which solves the problems of passive safety protection and lack of intelligent diagnosis and early warning capabilities of traditional power metering boxes.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A smart safety protection power metering box, comprising:

[0008] Box and meter;

[0009] A sensing module is installed inside the enclosure. The sensing module includes an environmental sensor and an electrical sensor, which are used to continuously collect the state inside the enclosure and convert it into state data arranged in chronological order.

[0010] The main control module, located within the enclosure, is connected to the sensing module. The main control module is configured to process the time-ordered state data, the processing including:

[0011] Obtain the current moment data from the state data, and based on the current moment data, predict the state at the next moment using a spatiotemporal state model to obtain the predicted data;

[0012] Obtain the actual data for the next moment from the state data, compare the predicted data with the actual data for the next moment, and calculate the residual between the two;

[0013] Based on the residual, fault diagnosis and risk simulation are performed through pattern matching, and protection instructions are generated according to the simulation results; the execution module, connected to the main control module, is used to execute the protection instructions, and the execution module includes an interlocking mechanism and an environmental control mechanism.

[0014] Preferably, the main control module is configured to use a spatiotemporal graph neural network model for state prediction.

[0015] Preferably, the interlocking mechanism of the execution module includes mechanical interlocking components and electrical interlocking circuits.

[0016] Preferably, the environmental control mechanism of the execution module includes a heating element and a ventilation control element.

[0017] Preferred options also include:

[0018] An excitation module is installed inside the enclosure and is connected to the main control module. The excitation module is configured to generate a preset linear frequency modulation signal and inject it into the monitoring circuit in response to the instructions issued by the main control module.

[0019] Preferred options also include:

[0020] An excitation module is installed inside the enclosure and is connected to the main control module. The excitation module is configured to generate a preset linear frequency modulation signal and inject it into the monitoring loop in response to the instructions issued by the main control module.

[0021] The communication unit connected to the main control module interacts with the remote monitoring platform for data exchange.

[0022] The backup power supply module supplies power to the sensing module and the main control module when the main power supply is interrupted.

[0023] Preferably, before predicting the state at the next moment using the spatiotemporal state model, the main control module is further configured as follows:

[0024] The data at the current moment is input into the spatiotemporal state model to construct and update a spatiotemporal graph model representing the current state of the power metering box.

[0025] Preferably, when calculating the residual between the two, the main control module is further configured to:

[0026] Both the predicted data and the actual data at the next moment are treated as matrices, and a residual matrix is ​​obtained by matrix subtraction. The residual is the residual matrix.

[0027] Preferably, the main control module is further configured to perform the risk simulation as follows:

[0028] The residual is input into a preset digital twin model, and the digital twin model is driven to perform a health assessment on the power metering box to generate the simulation result.

[0029] A safety protection method for an intelligent safety protection electricity metering box includes the following steps:

[0030] S1. The sensing module collects environmental and electrical parameters inside the power metering box in real time to obtain multi-source heterogeneous status data.

[0031] S2. Input the status data into the main control module to construct and update the spatiotemporal graph model representing the current status of the power metering box;

[0032] S3. Predict the future state of the power metering box based on a spatiotemporal graph neural network to obtain predicted state data;

[0033] S4. Compare the predicted state data with the actual state data collected at the next moment, and calculate the residual matrix;

[0034] S5. Analyze the residual matrix, drive the digital twin model to perform fault root cause diagnosis and health assessment, and generate control commands;

[0035] S6. The control command is sent to the execution module to perform electrical safety interlocking or environmental adjustment operations, and the execution result is fed back to form an adaptive closed-loop control.

[0036] In summary, the present invention has at least one of the following beneficial technical effects:

[0037] 1. This invention predicts the future state of the metering box by constructing a spatiotemporal graph neural network model and detects anomalies based on the residual between the prediction and the actual measurement. Compared with the traditional technology that relies on fixed thresholds for post-event alarms, this invention helps to realize the transformation from passive response to active response, and identifies abnormal precursors in the early stage of equipment failure, thereby effectively improving the reliability and safety of power supply.

[0038] 2. This invention integrates data from two types of heterogeneous sensors—environmental and electrical—and uses a pattern matching algorithm to compare real-time residuals with a pre-defined fault mode library. Compared to the single and vague alarm information of traditional technologies, this invention improves the accuracy of fault root cause diagnosis, helps to more clearly distinguish different events, provides clearer decision support for operation and maintenance response, and thus shortens fault troubleshooting time.

[0039] 3. By setting up an execution module that includes interlocking and environmental control mechanisms, and converting diagnostic decision results into automated physical control commands, this invention constructs an adaptive closed-loop control link compared to the traditional technology that relies entirely on manual on-site handling. This link can automatically perform operations such as environmental adjustment or safety interlocking, helping the equipment maintain a more ideal and safe working state without human intervention, and providing the possibility of reducing operation and maintenance costs.

[0040] 4. By adding an excitation module that can actively inject linear frequency modulation signals into the monitoring circuit, this invention provides an active detection method for the health status of the line, compared with the traditional technology that can only perform passive parameter monitoring. It can more sensitively detect hidden faults such as increased contact resistance caused by loose screws, contact oxidation, etc., and further move the safety protection checkpoint forward, providing effective technical support for predictive maintenance. Attached Figure Description

[0041] Figure 1 This is a block diagram of the overall structure of the present invention;

[0042] Figure 2 This is a schematic diagram of the main control module's workflow in this invention;

[0043] Figure 3 This is a schematic diagram illustrating the construction of the spatiotemporal graph model of the present invention;

[0044] Figure 4 This is a schematic diagram illustrating the working principle of the interlocking mechanism of the execution module of the present invention;

[0045] Figure 5 This is a schematic diagram illustrating the active detection working principle of the excitation module of the present invention;

[0046] Figure 6 This is a flowchart of the intelligent security protection method of the present invention.

[0047] The components include: 1. Housing; 2. Metering instrument; 3. Sensing module; 31. Environmental sensor; 32. Electrical sensor; 4. Main control module; 5. Execution module; 51. Interlocking mechanism; 511. Mechanical interlocking component; 512. Electrical interlocking circuit; 52. Environmental control mechanism; 521. Heating component; 522. Ventilation control component; 6. Excitation module; 7. Communication unit; and 8. Backup power supply module. Detailed Implementation

[0048] The technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Please see the appendix Figure 1 -Appendix Figure 6 This invention provides an intelligent safety protection power metering box, comprising:

[0050] Box 1 and meter 2;

[0051] The sensing module 3 is installed inside the enclosure 1. The sensing module 3 includes an environmental sensor 31 and an electrical sensor 32, which are used to continuously collect the state inside the enclosure 1 and convert it into state data arranged in chronological order.

[0052] The main control module 4 is located inside the housing 1. The main control module 4 is connected to the sensing module 3. The main control module 4 is configured to process the state data arranged in chronological order.

[0053] The execution module 5 is connected to the main control module 4 and is used to execute the protection commands generated by the main control module 4. The execution module 5 includes an interlocking mechanism 51 and an environmental control mechanism 52.

[0054] Specifically, the intelligent safety protection power metering box of this embodiment includes a box 1 as the basic load-bearing structure and a meter 2 for completing power metering. The core innovation lies in a highly integrated intelligent safety system, which consists of a sensing module 3, a main control module 4 and an execution module 5.

[0055] The sensing module 3 is the core of the entire intelligent system, responsible for constructing a panoramic digital image of the internal state of the metering box. In order to achieve comprehensive and blind-spot-free monitoring, the sensing module 3 in this embodiment integrates the following two types of sensors: environmental sensor 31: This type of sensor focuses on monitoring the physical environment and structural integrity of the box 1, including high-precision digital temperature and humidity sensors such as the SHT series, door magnetic sensors for monitoring the opening and closing status of the box door, and triaxial accelerometers for monitoring violent damage or abnormal tilting.

[0056] Electrical Sensors 32: These sensors are designed to monitor minute changes in core electrical circuits. They include high-precision micro-current / zero-sequence current transformers for measuring leakage currents in the mA range, high-sensitivity Hall-effect magnetic field sensors for monitoring external strong magnetic interference, and voltage and current sampling circuits for high-frequency sampling of the main circuit.

[0057] Data from all sensors is aggregated to the main control module 4 via shielded cables or internal buses, forming a multi-dimensional, heterogeneous, time-sequential state data stream that lays the foundation for subsequent intelligent analysis.

[0058] The main control module 4, as the core control unit of this invention, can be a microcontroller (MCU), digital signal processor (DSP), or system-on-a-chip (SoC) with computing capabilities in an embedded system. The main control module 4 is connected to the sensing module 3, and its core task is to process the state data provided by the sensing module 3 in chronological order to achieve predictive maintenance and proactive safety protection. The processing includes the following core steps:

[0059] The main control module 4, as the core control unit of the present invention, can be physically implemented as a microcontroller (MCU), digital signal processor (DSP), or system-on-a-chip (SoC) with computing capabilities in an embedded system.

[0060] The main control module 4 is electrically connected to the sensing module 3. Its core task is to process the state data provided by the sensing module 3, which is arranged in chronological order, to achieve predictive maintenance and proactive safety protection. The processing flow is a data-driven, closed-loop intelligent workflow, which specifically includes the following three logically interconnected stages:

[0061] Phase 1: State prediction and dynamic model construction;

[0062] This stage forms the basis for all subsequent analyses. The main control module 4 first acquires the current moment's data from the chronologically ordered state data. To achieve accurate predictions, the main control module 4 performs a crucial preprocessing step:

[0063] The current data is input into a spatiotemporal state model to construct and update a spatiotemporal graph model that represents the current complete state of the power metering box. Specifically, the main control module 4 abstracts each sensor in the sensing module 3 as a node in a graph structure, and defines the edges between nodes according to the physical proximity or electrical coupling relationship between the sensors, thereby constructing a spatiotemporal graph that describes the system topology.

[0064] After the spatiotemporal graph model is constructed and updated, the main control module 4 uses the spatiotemporal state model to predict the state at the next moment based on the data at the current moment, and obtains the predicted data.

[0065] The spatiotemporal state model is a spatiotemporal graph neural network model. The specific calculation process for state prediction performed by the main control module 4 is as follows:

[0066] Input data preparation: The main control module 4 for constructing the historical state sequence does not use only a single data point of "current moment". Instead, it extracts a historical time window of preset length N from the state data arranged in chronological order, including the current moment, such as the data of the past 60 seconds.

[0067] This will form a historical state data sequence, in which the data at each time step contains the reading values ​​of all sensor nodes at that moment, i.e., the feature matrix, as well as the association structure between the sensors, i.e., the adjacency matrix.

[0068] Spatial Feature Extraction: Graph Convolutional Network Aggregates Neighborhood Information. For each time step in the above historical data sequence, the spatiotemporal state model will use its built-in Graph Convolutional Network (GCN) layer for processing.

[0069] The role of GCN is to learn spatial dependencies. Specifically, at time t, the GCN layer, based on the graph's topology and defined by the adjacency matrix, weights and aggregates the information of each sensor node itself with the information of its directly connected neighbor nodes, as shown in the following formula:

[0070] H t =GCN(A,X) t );

[0071] In the formula, H t X is the node hidden feature matrix obtained after GCN processing at time t; GCN is Graph Convolutional Network; A is the nearest neighbor matrix; X t Let be the original feature matrix of the nodes at time t.

[0072] Output H t It is a new feature matrix, H tEach node in the system not only contains its own original information, but also incorporates the spatial context information of its neighborhood. For example, the temperature characteristics of a node now also implicitly contain the temperature conditions of its surrounding nodes.

[0073] By performing this operation one by one at each of the N time steps in the historical sequence, the model obtains a new set of time series rich in spatial information.

[0074] Temporal Feature Extraction: Recurrent Neural Networks Capture Evolution Patterns. Subsequently, the model takes this set of spatially rich time series as a whole and inputs it into its built-in recurrent neural network layer, preferably a variant such as GRU or LSTM.

[0075] Subsequently, the sequence is processed in chronological order to capture how the features of each node, which incorporates spatial information, evolve over time. For example, the model can learn dynamic patterns such as "the temperature in a certain area has been rising continuously over the past 60 seconds, and the rate of increase is accelerating."

[0076] Generate prediction data: Output the final prediction result. After the recurrent neural network processes the entire time series, the hidden state of its last time step contains the final summary of the spatiotemporal information of the past N times.

[0077] The main control module 4 achieves high-precision prediction of the future state of the power metering box through the above method, laying a solid foundation for subsequent residual calculation and anomaly diagnosis.

[0078] Phase Two: Residual Calculation and Anomaly Quantification

[0079] This stage is the core of anomaly detection, and its purpose is to quantify potential problems by comparing the differences between model predictions and physical reality.

[0080] First, after obtaining the predicted data, the main control module 4 then acquires the actual data for the next moment from the state data arranged in chronological order.

[0081] Subsequently, the main control module 4 compares the predicted data with the actual data at the next time step and calculates the residual between the two. To achieve refined analysis, the main control module 4 is further configured to treat both the predicted data and the actual data at the previous time step as matrices, and obtain a residual matrix through matrix subtraction. The residual is this residual matrix, and the calculation formula is as follows: RR(t+1)=Y actual (t+1)-Y predicted (t+1);

[0082] In the formula, RR(t+1) is the residual matrix calculated at time t+1; Y actual (t+1) is a matrix composed of the actual state data acquired from sensing module 3 at time t+1; Ypredicted (t+1) is a matrix consisting of predicted data for the state at time t+1, generated based on the data at time t.

[0083] The residual matrix is ​​a multidimensional map of abnormal symptoms, where the value and location of each element precisely indicate which sensor, at what time, and to what extent it deviated from the expected normality.

[0084] Phase Three: Diagnosis, Deduction, and Command Generation

[0085] This stage marks the end of intelligent decision-making. Based on residuals, the main control module 4 performs parallel fault diagnosis and risk simulation through pattern matching, and finally generates protection commands by comprehensively analyzing the results.

[0086] Fault Diagnosis: The main control module 4 is further configured to match the residual matrix with a preset fault mode library. This fault mode library pre-stores characteristic residual modes corresponding to various typical fault states, such as unpacking and electricity theft, magnetic field interference, and internal overheating. If the residual mode successfully matches any mode in the state fault mode library, a diagnostic conclusion containing the specific fault type is generated.

[0087] Risk simulation: Unlike pattern matching, which is used to quickly diagnose known faults, risk simulation aims to assess the long-term health trends and early potential risks of a system.

[0088] Therefore, when performing risk simulation, the main control module 4 is further configured to input the residual matrix into a preset digital twin model and drive the state digital twin model to perform a health assessment of the power metering box in order to generate simulation results.

[0089] At the heart of this digital twin model is a recursively updated health index assessment algorithm. This algorithm quantifies the health of the electricity meter box as an index that varies within the range [0, 1].

[0090] Even if the residuals do not trigger a clear failure threshold, the above model can dynamically assess the evolution of system health based on the small, continuous trend of residual changes, thereby achieving early warning of potential risks. The update formula for this health index is as follows:

[0091] H(t) = H(t-1) - λ·W⊙R(t);

[0092] In the formula, H(t) is the health index at the current time t, which is the inference result generated in this evaluation; H(t-1) is the historical health index at the previous time t-1, which reflects the continuity of the health status; R(t) is the residual matrix calculated from the second stage; W is a preset weight matrix with the same dimension as the residual matrix, the elements of which reflect the influence weight of the residuals of different sensor nodes on the overall health of the system; λ is an adjustable attenuation coefficient used to control the rate at which the health decreases with the residual amplitude.

[0093] Finally, the main control module 4 intelligently integrates the outputs of the two paths mentioned above, taking into account the urgency of the fault diagnosis and the predictability of the risk simulation. Based on the final simulation results, it generates and issues the optimal protection command to the execution module 5.

[0094] For example, when H(t) first falls below a preset "early warning" threshold, such as 0.8, a "recommend maintenance" instruction is generated and reported to the remote platform; when H(t) falls below a "alarm" threshold, such as 0.6, a higher-level alarm instruction may be generated.

[0095] The execution module 5 is responsible for translating the decisions of the main control module 4 into actual actions in the physical world.

[0096] Interlocking mechanism 51: Provides both electrical and mechanical safety protection, including a mechanical interlocking component 511 driven by an electromagnet for physically locking the circuit breaker handle; and an electrical interlocking circuit 512 integrated with the main circuit breaker, such as in the form of an intelligent trip unit, for quickly performing the tripping operation.

[0097] Environmental control mechanism 52: Enables intelligent steady-state regulation of the environment inside the chamber. Includes a surface-insulated PTC heating element 521 for precise temperature rise when the risk of condensation is high; and a ventilation control element 522 with a dust filter and a miniature fan for active ventilation when the temperature or humidity is too high.

[0098] In a preferred embodiment of the present invention, in order to achieve the most comprehensive protection capability, the following modules are also integrated: Excitation module 6: an active detection unit that, under the instruction of the main control module 4, can generate a preset linear frequency modulation signal and weakly inject it into the monitoring circuit. By analyzing the amplitude attenuation and phase shift of the response signal, the main control module 4 can calculate the impedance spectrum of the circuit, thereby more sensitively detecting early hidden dangers caused by loose screws, contact oxidation, etc. than passive monitoring.

[0099] Communication Unit 7: An industrial-grade wireless communication module, with statuses such as NB-IoT / 4G and LTE, is responsible for reporting its status data, diagnostic logs, and alarm information to the remote monitoring platform in real time via an encrypted protocol, and for receiving remote commands from the platform.

[0100] Backup power supply module 8: The last line of defense to ensure the system can still operate under extreme conditions. It consists of a passive inductive energy harvesting circuit and a supercapacitor. When the main power is interrupted, the electrical energy stored in the supercapacitor powers the sensing module 3 and the main control module 4 through a DC-DC conversion circuit, ensuring that the system has enough energy to complete the recording and reporting of the last fault information.

[0101] In summary, this embodiment constructs a complete closed-loop control system from multi-dimensional perception, intelligent prediction, accurate diagnosis to proactive execution through the precise coordination of the above functional modules, thereby achieving intelligent safety protection for the electricity metering box.

[0102] Please see the appendix Figure 1 -Appendix Figure 6 This invention provides a safety protection method for an intelligent safety protection electricity metering box, comprising the following steps:

[0103] S1. Real-time acquisition of environmental and electrical parameters inside the power metering box through sensing module 3 to obtain multi-source heterogeneous status data;

[0104] Specifically, the main control module 4 embedded in the metering box will continuously send data acquisition commands to the sensing module 3 at a preset frequency.

[0105] Parallelism and differentiated frequency of data acquisition: The acquisition process is parallel and multi-rate. For rapidly changing electrical parameters, such as zero-sequence current acquired by a high-precision micro-current transformer and magnetic field strength acquired by a Hall sensor, the main control module 4 will sample at a higher frequency, such as 1kHz, to capture transient signals that may indicate electricity theft or short-circuit faults.

[0106] For environmental parameters that change relatively slowly, such as temperature and humidity collected by temperature and humidity sensors, and vibration and tilt angle collected by triaxial accelerometers, a lower frequency, such as 1Hz, can be used for acquisition to save system resources.

[0107] The formation of multi-source heterogeneous data: The collected data are of various types, such as current, magnetic field, temperature, humidity, vibration, etc., with different units and dimensions, and come from multiple sensors distributed in different locations in the box. Therefore, they constitute typical multi-source heterogeneous state data.

[0108] Data preprocessing: Before sending the data to the next step, the main control module 4 will also perform necessary preprocessing, including digital filtering of high-frequency signals to remove noise interference, normalization of all acquired data to eliminate the influence of different units, and adding a precise timestamp to each frame of data, ultimately forming a structured, time-aligned multidimensional data stream, laying the foundation for accurate analysis of the subsequent model.

[0109] S2. Input the status data into the main control module 4 to construct and update the spatiotemporal diagram model representing the current status of the power metering box;

[0110] Specifically, the main control module 4 does not view the data from each sensor in isolation, but incorporates the inherent relationships between them into the model.

[0111] The construction logic of the spatiotemporal graph: The main control module 4 first logically constructs a "graph". Each node in the graph corresponds to a physical sensor in the perception module 3, and the edges between nodes represent the relationships between sensors. These relationships can be:

[0112] Physical proximity: For example, a temperature sensor node installed on the top of the enclosure and a door magnetic sensor node installed on the door are connected by an edge because they are physically close.

[0113] Electrical coupling relationship: For example, the current sensor node on the incoming line side and the magnetic field sensor node near meter 2 are connected by an edge due to the strong correlation in the electricity theft scenario. The weight of the edge can be preset or learned to represent the strength of the correlation.

[0114] Dynamic updates of the model: This spatiotemporal graph model is not static. As new state data continuously flows in, the main control module 4 will use online learning or incremental learning strategies to continuously update the parameters of the graph model.

[0115] This allows the model to adapt to the gradual aging process of the device itself, such as the slow increase in resistance of a certain contact or seasonal changes in the environment, such as the different temperature and humidity baselines in summer and winter, ensuring the effectiveness of the model throughout its entire life cycle.

[0116] S3. Based on the spatiotemporal graph neural network, the future state of the power metering box is predicted to obtain the predicted state data. Specifically, the main control module 4 uses its powerful computing capabilities to run a pre-trained spatiotemporal graph neural network and spatiotemporal state model to make short-term predictions of the future state of the system.

[0117] How spatiotemporal graph neural networks work: This neural network model typically consists of two core parts:

[0118] One part is graph convolutional networks, which aggregate information from neighboring nodes in a graph to learn the spatial dependencies of states; the other part is recurrent neural networks, such as temporal convolutional networks, which learn the evolution of each node's own data over time, i.e., temporal dependencies.

[0119] Prediction process: Specifically, the main control module 4 will input the spatiotemporal graph data containing all node information from the past period, for example, the first 60 seconds, as a sequence into the spatiotemporal state model.

[0120] The model uses complex nonlinear calculations to output a prediction of the state values ​​of all nodes at the next future moment, for example, one second. This result is the predicted state data. For example, the model can predict the temperature at the incoming terminal in the next second based on the current, voltage, and internal temperature of the enclosure.

[0121] S4. Compare the predicted state data with the actual state data collected at the next moment and calculate the residual matrix; specifically, the system detects anomalies through comparison.

[0122] Residual calculation: When the real state data of the next moment is collected by step S1, the main control module 4 will immediately subtract it from the predicted state data generated by step S3 element by element. The result is a vector or matrix with the same dimension as the number of sensors, i.e., the residual matrix.

[0123] The significance of residuals: When the system is running normally, because the model prediction is accurate enough, all the element values ​​in the residual matrix should fluctuate within a very small range near zero. Once the actual value measured by one or more sensors deviates significantly from the model's predicted value, the corresponding residual value will increase sharply, indicating the occurrence of an abnormal event.

[0124] S5. Analyze the residual matrix to drive the digital twin model to perform root cause diagnosis and health assessment, and generate control commands;

[0125] Specifically, this step will be initiated when a significant excess of residuals is detected, to conduct in-depth fault diagnosis and decision-making.

[0126] Driving the digital twin model: The main control module 4 internally constructs a digital twin model corresponding to the height of the physical metering box. This model not only includes the geometric structure of the metering box, but also embeds simulation algorithms for its electrical, thermodynamic and other physical processes. The residual matrix generated by S4 is input to this digital twin model.

[0127] Fault root cause diagnosis: After receiving the residuals, the digital twin model will perform simulation and deduction. For example, if the residual matrix shows "sudden increase in zero-sequence current" and "slight increase in internal temperature", the digital twin model will simulate a variety of possible faults and calculate what residuals these faults will produce in its virtual model.

[0128] By comparing the simulated residuals with the actual observed residuals, the model can pinpoint the most likely root cause of the failure with extremely high confidence.

[0129] Health Assessment: In addition to real-time fault diagnosis, the main control module 4 also performs long-term trend analysis on residuals to assess the health of the system. For example, if the residual of a certain contact is not exceeded, but its variance increases slowly over time, it may indicate that poor contact is worsening. The system will then lower the health score of that component and recommend preventative maintenance.

[0130] Generate control commands: Based on the determined root causes of the fault and the health assessment results, the decision subsystem of the main control module 4 will automatically generate the optimal control commands according to the preset expert rule base.

[0131] S6. Send the control command to the execution module 5 to perform electrical safety interlocking or environmental adjustment operations, and feed back the execution results to form an adaptive closed-loop control.

[0132] Specifically, the results of intelligent analysis will be translated into actual physical protective actions.

[0133] Command execution: The main control module 4 accurately sends the generated control commands to the execution module 5 through the internal bus. The drive circuit of the execution module 5 parses the commands and controls the corresponding actuators to perform actions, such as driving the electromagnet to perform electrical safety interlocks or starting the PTC heating element to perform environmental adjustment operations.

[0134] Execution result feedback and closed loop: After the execution action is completed, the execution module 5 will send a status signal back to the main control module 4 to confirm that the instruction has been executed.

[0135] More importantly, the effects of this protective action, such as the decrease in humidity inside the chamber after heating is started, will be collected again by the sensing module 3 in step S1 and input as new state data into the spatiotemporal graph model in S2.

[0136] The main control module 4 evaluates the effectiveness of the intervention by observing whether the residual returns to normal after the control command is executed, and adjusts the subsequent control strategy accordingly. This constitutes an adaptive closed-loop control from perception to analysis to decision-making to execution and then to re-perception, ensuring the intelligence, efficiency and reliability of the entire protection process.

Claims

1. An intelligent safety protection power metering box, characterized in that, include: Box (1) and meter (2); The sensing module (3) is located inside the enclosure (1). The sensing module (3) includes an environmental sensor (31) and an electrical sensor (32) for continuously collecting the state inside the enclosure (1) and converting it into state data arranged in chronological order. A main control module (4) is located inside the housing (1). The main control module (4) is connected to the sensing module (3). The main control module (4) is configured to process the state data arranged in chronological order. The processing includes: Obtain the current moment data from the state data, and based on the current moment data, predict the state at the next moment using a spatiotemporal state model to obtain the predicted data; Obtain the actual data for the next moment from the state data, compare the predicted data with the actual data for the next moment, and calculate the residual between the two; Based on the residuals, fault diagnosis and risk simulation are performed through pattern matching, and protection instructions are generated based on the simulation results. When calculating the residual between the two, the main control module (4) is configured as follows: Both the predicted data and the actual data at the next moment are treated as matrices, and a residual matrix is ​​obtained by matrix subtraction. The residual is the residual matrix. When performing the risk simulation, the main control module (4) is configured as follows: The residual is input into a preset digital twin model, and the digital twin model is driven to perform a health assessment on the power metering box to generate the inference result; The main control module (4) is configured as follows: The residual is matched with a preset fault mode library. If the match is successful, a diagnostic conclusion containing the specific fault type is generated. The execution module (5) is connected to the main control module (4) and is used to execute the protection command. The execution module (5) includes an interlocking mechanism (51) and an environmental control mechanism (52).

2. The intelligent safety protection power metering box according to claim 1, characterized in that, The main control module (4) is configured to use a spatiotemporal graph neural network model for state prediction.

3. The intelligent safety protection power metering box according to claim 1, characterized in that, The interlocking mechanism (51) of the execution module (5) includes a mechanical interlocking component (511) and an electrical interlocking circuit (512).

4. The intelligent safety protection power metering box according to claim 1, characterized in that, The environmental control mechanism (52) of the execution module (5) includes a heating element (521) and a ventilation control element (522).

5. The intelligent safety protection power metering box according to claim 1, characterized in that, Also includes: An excitation module (6) is installed inside the housing (1). The excitation module (6) is connected to the main control module (4). The excitation module (6) is configured to generate a preset linear frequency modulation signal and inject it into the monitoring circuit in response to the instruction issued by the main control module (4). The communication unit (7) connected to the main control module (4) interacts with the remote monitoring platform for data exchange; The backup power supply module (8) supplies power to the sensing module (3) and the main control module (4) when the main power supply is interrupted.

6. The intelligent safety protection power metering box according to claim 1, characterized in that, Before predicting the state at the next moment through the spatiotemporal state model, the main control module (4) is further configured as follows: The data at the current moment is input into the spatiotemporal state model to construct and update a spatiotemporal graph model representing the current state of the power metering box.

7. A safety protection method for an intelligent safety protection energy metering box, wherein the method is applied to an intelligent safety protection energy metering box as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. The environmental and electrical parameters inside the power metering box are collected in real time through the sensing module (3) to obtain multi-source heterogeneous status data; S2. Input the status data into the main control module (4) to construct and update the spatiotemporal diagram model representing the current status of the power metering box; S3. Predict the future state of the power metering box based on a spatiotemporal graph neural network to obtain predicted state data; S4. Compare the predicted state data with the actual state data collected at the next moment, and calculate the residual matrix; S5. Analyze the residual matrix, drive the digital twin model to perform fault root cause diagnosis and health assessment, and generate control commands; S6. The control command is sent to the execution module (5) to perform electrical safety interlocking or environmental adjustment operations, and the execution result is fed back to form an adaptive closed-loop control.

Images