Circuit breaker control method and device for resident charging facilities, equipment and medium

Abstract

The invention discloses a circuit breaker control method and device for resident charging facilities, equipment and a medium, and the method comprises the steps: S1, obtaining the multi-mode safety perception data of the operation process of the resident charging facilities, and completing the synchronous sampling and real-time preprocessing to form a feature vector sequence; s2, performing alignment and joint modeling on the feature vector sequence to output a unified risk situation vector, and giving uncertainty; s3, according to the risk situation vector, fire behavior identification and thermal runaway prediction under the fire condition of the charging pile are executed, and a disconnection level and a confirmation window are determined; s4, determining a target phase and an optimal triggering moment of zero-crossover breaking by combining the phase of the power grid and response characteristics of an execution mechanism; and S5, implementing a control strategy of pre-excitation of the trip coil in advance and pile end cooperative current reduction in the confirmation window before contact separation. According to the invention, end-to-end rapid and zero-crossover accurate on-off is realized, arc energy and contact wear is substantially reduced, and intrinsic safety and fireproof capability of resident charging facilities are improved.

Description

Technical Field

[0001] This application relates to the field of electrical safety, and in particular to circuit breaker control methods, devices, equipment and media for residential charging facilities. Background Technology

[0002] A smart circuit breaker is a device that can detect circuit abnormalities and cut off the power supply. As a key component of an electrical protection system, it integrates intelligent technology and electrical protection functions, and has multiple advantages and functions, aiming to improve the safety, stability and reliability of electrical equipment. Intelligent circuit breakers, through sensors and intelligent processors, can achieve real-time monitoring and data analysis of circuit parameters, quickly detect abnormal conditions in the circuit, and implement rapid circuit breaker protection responses, effectively avoiding safety issues such as circuit overload and circuit breakage. However, traditional intelligent circuit breaker control methods have the following shortcomings: 1. Traditional intelligent circuit breaker control methods mostly target single sensor thresholds or study early-late stage fusion of "weighted average / feature splicing," failing to adequately handle multi-source asynchronous sampling and physical constraints, and lacking uncertainty output; 2. Traditional intelligent circuit breaker control methods typically set fixed thresholds for electrical anomalies and study triggering mechanisms for multi-cycle consistency confirmation, failing to fully consider the impact of fire development on disconnection timing and phase; 3. Traditional intelligent circuit breaker control methods typically treat the circuit breaker as an independent device, mainly studying fixed threshold triggering and actuator opening characteristics, lacking real-time coordination with charging piles, and failing to handle phase deviations caused by mechanical lag. Summary of the Invention

[0003] To address the aforementioned technical problems, this application provides a circuit breaker control method for residential charging facilities, which solves the technical problems in the prior art such as lack of multi-source asynchronous sampling and physical constraints, insufficient consideration of the impact of fire development on disconnection timing and phase, lack of real-time coordination with charging piles, and failure to handle phase deviation caused by mechanical lag.

[0004] This application is achieved through the following solution: A circuit breaker control method for residential charging facilities includes the following steps: S1. Acquire multimodal safety perception data of the operation of residential charging facilities, complete synchronous sampling and real-time preprocessing under a unified clock, construct multimodal time series feature data including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status, and form a feature vector sequence for subsequent fusion; S2. Based on deep fusion and risk assessment of multi-source asynchronous data, the feature vector sequence is aligned and jointly modeled, and a unified risk situation vector is formed and output with physical constraints, and the uncertainty is given. S3. Based on the risk situation vector, perform fire identification and thermal runaway prediction under the condition of charging pile fire, determine the current fire level and the predicted time limit, determine the disconnection level and confirmation window, and accelerate the decision response under the condition of charging pile fire. S4. Combining the power grid phase and the actuator response characteristics, determine the target phase and optimal triggering time for zero-crossover interruption by disconnection timing and phase planning; S5. Implement coordinated pre-power-off and control strategies. Within the confirmation window before contact separation, implement control strategies for pre-excitation of the trip coil and coordinated current reduction at the pile end to reduce arc energy during contact separation.

[0005] This application also provides a circuit breaker control device for residential charging facilities, including: The data acquisition and preprocessing module is used to acquire multimodal safety perception data during the operation of residential charging facilities. It completes synchronous sampling and real-time preprocessing under a unified clock, and constructs multimodal time series feature data including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status, forming a feature vector sequence for subsequent fusion. The risk assessment module is used to align and jointly model feature vector sequences based on deep fusion and risk assessment of multi-source asynchronous data, and to form and output a unified risk situation vector with physical constraints and give the uncertainty. The risk level identification module is used to identify the fire and predict thermal runaway under the condition of a charging pile catching fire based on the risk situation vector, determine the current fire level and the predicted time limit, determine the disconnection level and confirmation window, and accelerate the decision response under the condition of a charging pile catching fire. The disconnect timing and phase planning module is used to combine the power grid phase and actuator response characteristics to determine the target phase and optimal triggering time for zero crossover interruption through disconnect timing and phase planning. The collaborative execution module is used to execute collaborative pre-power-off and execution control strategies. Within the confirmation window before contact separation, it implements control strategies for pre-excitation of the trip coil and collaborative current reduction at the pile end to reduce the arc energy during contact separation.

[0006] This application also provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the circuit breaker control method for residential charging facilities.

[0007] This application also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the circuit breaker control method for residential charging facilities.

[0008] Compared with the prior art, this application has the following advantages: 1. This application proposes a multi-sensor deep fusion method based on hierarchical physical constraints and cross-modal attention mechanism, which fully handles multi-source asynchronous sampling and physical constraints, and realizes unified risk situation awareness and uncertainty output of multi-source signals such as electrical, thermal and environmental signals.

[0009] 2. This application addresses special working conditions such as charging pile fires, fully considers the impact of fire development on disconnection timing and phase, and designs a fast isolation control algorithm that minimizes fire awareness and zero-crossing energy to achieve an adaptive disconnection strategy associated with the fire level. 3. This application proposes a predictive current reduction and coil pre-excitation control mechanism that coordinates the circuit breaker and the charging pile. Before disconnection, it actively reduces arc energy and compensates for phase deviation caused by mechanical lag to ensure ultra-fast safe disconnection of ≤15ms.

[0010] 4. This application achieves end-to-end fast, zero-crossing precise interruption, significantly reducing arc energy and contact wear, and improving the inherent safety and fire resistance of residential charging facilities.

[0011] In addition to the purposes, features, and advantages described above, this application has other purposes, features, and advantages. A further detailed description of this application will be provided below with reference to the figures. Attached Figure Description

[0012] Figure 1 This is a schematic flowchart of a circuit breaker control method for residential charging facilities according to a preferred embodiment of this application; Figure 2 This is a schematic diagram of a circuit breaker control device for residential charging facilities according to a preferred embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device according to a preferred embodiment of this application; Figure 4 This is an internal structural diagram of a computer device according to a preferred embodiment of this application. Detailed Implementation

[0013] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0014] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a circuit breaker control device for residential charging facilities capable of performing the above functions. The following description uses a circuit breaker control device for residential charging facilities as an example to illustrate this embodiment and the subsequent embodiments.

[0015] like Figure 1 As shown, a preferred embodiment of this application provides a circuit breaker control method for residential charging facilities, including the following steps: S1. Acquire multimodal safety perception data of the operation of residential charging facilities, complete synchronous sampling and real-time preprocessing under a unified clock, construct multimodal time series feature data including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status, and form a feature vector sequence for subsequent fusion; S2. Based on deep fusion and risk assessment of multi-source asynchronous data, the feature vector sequence is aligned and jointly modeled, and a unified risk situation vector is formed and output with physical constraints, and the uncertainty is given. S3. Based on the risk situation vector, perform fire identification and thermal runaway prediction under the condition of charging pile fire, determine the current fire level and the predicted time limit, determine the disconnection level and confirmation window, and accelerate the decision response under the condition of charging pile fire. S4. Combining the power grid phase and the actuator response characteristics, determine the target phase and optimal triggering time for zero-crossover interruption by disconnection timing and phase planning; S5. Implement coordinated pre-power-off and control strategies. Within the confirmation window before contact separation, implement control strategies for pre-excitation of the trip coil and coordinated current reduction at the pile end to reduce arc energy during contact separation.

[0016] Compared with the prior art, this embodiment has the following beneficial effects: 1. This embodiment proposes a multi-sensor deep fusion method based on hierarchical physical constraints and cross-modal attention mechanism, which fully handles multi-source asynchronous sampling and physical constraints, and realizes unified risk situation awareness and uncertainty output of multi-source signals such as electrical, thermal and environmental signals.

[0017] 2. This embodiment is designed for special working conditions such as charging pile fires. It fully considers the impact of fire development on disconnection timing and phase, and designs a fast isolation control algorithm that minimizes fire awareness and zero crossover energy to realize an adaptive disconnection strategy related to the fire level. 3. This embodiment proposes a predictive current reduction and coil pre-excitation control mechanism that coordinates the circuit breaker and the charging pile. Before disconnection, it actively reduces arc energy and compensates for phase deviation caused by mechanical lag to ensure ultra-fast safe disconnection of ≤15ms.

[0018] 4. This embodiment achieves end-to-end fast, zero-crossing precise disconnection, significantly reducing arc energy and contact wear, and improving the inherent safety and fire resistance of residential charging facilities.

[0019] In a preferred embodiment of this application, the circuit breaker control method for residential charging facilities further includes the step of: S6. Through action threshold calibration and event tracing, the protection action threshold is adjusted and verified in real time to control the false action rate and ensure the reliability of triggering decisions; at the same time, event root cause labels and key feature data are generated to achieve full-process recording and traceability.

[0020] This embodiment performs action threshold calibration and event tracing, and uses QCP / CTD (an online threshold calibration algorithm based on statistical coverage guarantee) to adjust and verify the protection action threshold in real time, so as to control the false action rate and ensure the reliability of triggering decisions. At the same time, it uses RCC (root cause classification recording module) to generate event root cause labels and key feature data to achieve full process recording and traceability.

[0021] In a preferred embodiment of this application, step S5 further includes the step: If the external collaboration fails to respond within the specified time, the hardware fast-track back-off module will automatically execute a fallback tripping action to complete the contact separation and ensure safety.

[0022] In this embodiment, if the external collaboration fails to respond within the specified time, such as if any indicator is found to exceed the safety envelope (15 ms total delay or...), (Arc energy limit), immediately abandon the current optimal plan, automatically switch to a conservative backoff strategy (select "nearest reachable zero-crossing point" or directly activate hardware fast channel HFT / FBS), thereby ensuring that a safe disconnection can still be completed in a deterministic manner under the worst operating conditions. Through this mechanism, this invention transforms "end-to-end ≤15 ms ultra-fast disconnection" and "controlled arc energy" from design goals into real-time monitored and enforceable safety hard constraints for the first time. It completely solves the reliability risks of traditional predictive zero-crossing technology in the event of coordinated disconnection or time delay fluctuations, achieving a true "zero-risk safety net" and significantly improving the intrinsic safety level of residential charging facilities in fire scenarios.

[0023] In a preferred embodiment of this application, step S1 specifically includes the following steps: S11. Synchronously collect multimodal safety perception data of the operation of residential charging facilities under a unified clock, including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status; S12. The collected multimodal safety perception data is preprocessed. The preprocessing process uses Goertzel array, TKEO energy operator, EWMA slope estimation, second-order difference threshold detection and Schmitt triggering mechanism to extract multimodal time series feature data of current anomaly, arc activity, temperature rise trend, smoke change and water immersion status. S13. Encapsulate the timestamped multimodal time series feature data into a feature vector sequence for subsequent fusion.

[0024] This embodiment acquires multimodal safety perception data of the operation of residential charging facilities, completes synchronous sampling and real-time preprocessing under a unified clock, and constructs multimodal time series feature data including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status, forming a feature vector sequence for subsequent fusion. It fully considers multi-source asynchronous sampling and physical constraints, and realizes unified risk situation perception of multiple source signals such as electrical, thermal and environmental signals.

[0025] In a preferred embodiment of this application, step S2 specifically includes the following steps: S21. Use MFGKS to complete asynchronous alignment and unified hidden state estimation of multi-rate feature vector sequences; In this embodiment, MFGKS (Multi-Factor Graph Kalman Smoother) is an innovative algorithm combining factor graph, Kalman smoothing, and fixed-lag smoothing techniques. It can handle multi-rate, asynchronously arriving multimodal data and estimates it by mapping these data to a unified risk latent state. MFGKS uses factor graphs to represent multi-source observation data, optimizes risk latent state estimation through MAP minimization, and performs extrapolation using Kalman smoothing within a fixed-lag window, ensuring real-time performance and numerical stability. Specifically, it includes: Unified hidden state of multimodal risk: Input data: Asynchronous observation data from different sensors (such as electrical, thermal, and environmental sensors), each mode may have different sampling rates and timestamps.

[0026] Output data: Unified hidden risk state Such as leakage intensity, arc activity, temperature rise rate, smoke intensity, etc.

[0027] The algorithm's function is to unify the hidden state. Multimodal data is mapped to the same physical quantity (such as security risk status). This avoids the errors caused by threshold voting in traditional methods for handling multi-source data, thereby improving the accuracy of fusion.

[0028] Factor Graph: Input data: Asynchronous observation data for each mode (e.g., current, temperature, smoke, etc.) and timestamps of the observations.

[0029] Output data: Optimization results for different factors in the factor graph, yielding the hidden states. The estimated value.

[0030] The algorithm's factor graph treats the data of each modality as an independent factor and a unified hidden state. The connection allows asynchronous data to flow in naturally without the need for forced synchronization or interpolation. Joint optimization using factor graphs enables the effective fusion of multimodal data.

[0031] Fixed-lag smoothing: Input data: Hidden state estimates from historical data and observation data at the current moment.

[0032] Output data: Smoothed hidden state estimate at the current time step And the corresponding uncertainty (covariance matrix).

[0033] This algorithm uses a fixed lag window to ensure real-time updates of the hidden state estimate and provides smoother estimation results by utilizing historical data. By smoothing the hidden state with a fixed lag, it avoids the delay problem in traditional Kalman filtering and ensures that the estimate at each time step achieves a balance between the system's real-time performance and stability.

[0034] MFGKS combines factor graph models, Kalman smoothing, and fixed-hysteresis smoothing techniques, specifically designed for efficient fusion and unified estimation of multimodal, asynchronously arriving sensor data. Traditional Kalman filtering methods require data synchronization, but they are often not directly applicable when processing multi-source asynchronous data. MFGKS, through its factor graph structure, allows data from different modalities to arrive asynchronously and be incorporated into the factor graph for joint optimization.

[0035] In this model, factor graph modeling is used: the observed data for each mode (such as current, temperature, smoke, etc.) are represented as an independent factor, along with the unified hidden state. Connect the factors. The factor graph optimizes the hidden state estimation using the MAP minimization method.

[0036] Kalman smoothing: The Kalman smoothing method is used to smoothly estimate hidden states. In multimodal data fusion, Kalman smoothing can utilize historical data to provide a more accurate and stable estimate of the hidden state at the current time step.

[0037] Fixed-lag smoothing: MFGKS uses a fixed-lag window for smoothing, ensuring real-time updates of the hidden state while handling multi-rate data. The fixed-lag smoothing method optimizes the current hidden state using historical data, enabling the system to respond quickly to new observations while maintaining stability.

[0038] The advantages and innovations of MFGKS technology include: (1) Efficient fusion of multimodal data: Traditional methods typically only handle single modalities or require synchronization between modalities, while MFGKS utilizes factor graphs to enable the effective fusion of data from different modalities without mandatory synchronization. Each modality is modeled using factor graphs, allowing for natural temporal alignment and joint optimization.

[0039] (2) Support for asynchronous data: MFGKS utilizes a flexible modeling structure based on factor graphs, allowing asynchronous data from different modalities to be directly incorporated into the model without the need for interpolation or alignment. This offers significant advantages when dealing with real-time data streams and packet loss.

[0040] (3) Uncertainty output: MFGKS not only outputs the estimated value of the hidden state x(t), but also the uncertainty (covariance matrix) of each estimate. This allows the system to provide more confidence information for subsequent decisions, helping to reduce the risk of misjudgment and misoperation.

[0041] (4) Improve the balance between real-time performance and stability: By using fixed-lag smoothing, MFGKS ensures that the hidden state update at each time step is both real-time and utilizes historical data to provide a more stable estimate. The use of fixed-lag smoothing ensures a good balance between system response speed and data stability.

[0042] (5) Improve the scalability and robustness of the system: Because MFGKS can effectively handle asynchronous, packet-loss, and multi-rate data, it can adapt to various sensor data inputs and environmental changes. This makes the algorithm robust in a variety of practical applications, and it is particularly suitable for real-time monitoring and early warning systems.

[0043] S22. Use CMTE to establish a cross-modal feature interaction model in the fusion layer to establish the interaction relationship between electrical, thermal and environmental information, rather than for classification. The principle of CMTE (Cross-Modal Temporal Embedding) is to embed the time slice features of different modalities within a window into the sequence { Construct uncertainty-weighted attention: ; in: This represents the uncertainty-weighted embedding vector of the i-th modal feature. Represents the original feature embedding vector of the i-th modality; Represents the query matrix, consisting of the weighted { The linear transformation is used to allow the current mode to actively "query" information from other modes. The key matrix is ​​represented by the weighted { Obtained through linear transformation, it can be used for querying other modalities; Represents a value matrix, consisting of the weighted { The linear transformation yields the actual feature content for weighted aggregation. This represents the query-key dot product similarity matrix, and calculates the raw attention score between each pair of modalities. This represents the attention scaling factor to prevent the dot product from becoming too large. Gradient vanishing; This represents the final attention-weighted output.

[0044] First, use the uncertainty given by MFGKS. Back-engineer a gating Amplify the weight of more reliable modal tokens. Then, multi-head attention is used to perform interactive weighting on the tokens of each modality. It automatically seeks the causal relationship of "abnormal current ↔ temperature rise ↔ smoke" to form a cross-modal coupled fusion representation.

[0045] After multi-head attention aggregation, a fused representation is obtained through feedforward and residual. And revert to continuous risk: ; in: The unified risk situation vector at time t is represented, and the final output is a multidimensional continuous risk value; The output layer weight matrix is ​​represented from the fused representation. A trainable linear transformation matrix to the risk dimension; This represents the output layer bias vector, which is the trainable bias term of the output layer. This represents the activation function for the S-shaped growth curve, which compresses values ​​to the (0,1) interval and is often used for probabilistic risk output. Mapping values ​​to (0, +∞) is more suitable for representing "risk intensity" than probability, avoiding... Gradient saturation problem.

[0046] The fused high-dimensional representation Regression to continuous risk intensity ∈ Instead of classification labels, this provides quantities that can be directly used in control calculations, such as threshold comparisons and phase optimization costs, thus achieving a closed-loop link where "sensor output is directly fed into control".

[0047] Then the output is: ; in: The dimensionless risk quantity representing the risk component related to leakage current, estimated by the multimodal fusion model as "leakage intensity / residual current imbalance", is used to characterize the severity of electric shock / leakage hazard in the current circuit. The risk component related to arc faults is the "arc activity" risk quantity estimated by the multimodal fusion model. It integrates current waveform distortion, arc characteristic signals, etc., and is used to characterize whether arc discharge exists and its degree of danger. The risk component related to overheating / thermal runaway is the "temperature rise rate and temperature over-limit trend" risk quantity estimated by the multimodal fusion model, which reflects the severity of the overheating / thermal runaway development of terminals, conductors and other parts. The risk component related to smoke / combustion is the "smoke rise intensity" risk quantity inferred from smoke sensing characteristics and their rate of change. It is used to characterize the presence of early fire signs such as smoldering or open flames.

[0048] Subsequent dosage (e.g., water immersion): It can be defined in the same way as: water immersion risk component, other extended risk components, etc., to represent specific risks such as water immersion short circuit, environmental anomalies, etc.

[0049] Finally, the coupling symptoms of electrical-thermal-environment are modeled in the fusion layer, and "continuous risk" rather than fault label is output for subsequent planning and control.

[0050] Within a fixed lag time window, the algorithm encapsulates preprocessed features from different modes such as electrical, thermal, and environmental into a token sequence in chronological order. Utilizing a multi-head attention mechanism based on uncertainty gating, it interactively weights the tokens of each mode, automatically mining cross-modal associations such as "abnormal current, rising temperature, and enhanced smoke." After multi-head attention aggregation, the resulting data is fused through a feedforward network and residual structure, and regressed into continuous risk intensity rather than discrete classification labels. This enables differentiable and quantifiable risk output for subsequent control stages.

[0051] Advantages and innovations of CMTE technology principles: Compared to traditional methods that simply concatenate features from different modalities or use fixed-weighted averaging, CMTE uses multi-head attention to perform learnable, context-sensitive weighting among tokens. It can automatically learn "which modality is more critical in the current scenario" (e.g., current + arc features are important during arc initiation, temperature rise + smoke are important during smoldering); and it has the ability to express complex nonlinear interactions (such as "current pumping first, then temperature rising"), rather than relying solely on linear superposition.

[0052] CMTE does not only look at a single moment, but builds a token sequence within a time window and uses self-attention to make correlations in both the time and modal dimensions simultaneously; it can identify the trend of "continuous slow temperature rise + slight leakage" instead of just looking at the absolute value at a certain instant; it can also capture chain signs of "electrical anomaly first, followed by temperature rise / smoke", thus giving a higher risk estimate in advance.

[0053] CMTE finally outputs continuous risk using a regression approach. Instead of classification probabilities, this allows subsequent modules (such as AEMZP phase planning, WVC window adjustment, and QCP / CTD threshold calibration) to... Treat it as a differentiable cost or constraint to achieve end-to-end linkage; S23. Embedding PGE in the fusion loss introduces differentiable physical constraint terms during model training to reduce false detections and ensure that the results comply with electrical and thermal constraints. PGE (Physics-Guided Embedding): Incorporates differentiable physical constraints into the fusion training objective. ; in: — Total loss (training objective): The total training loss function of the entire multi-source fusion model (MFGKS + CMTE, plus PGE). During training, this loss is minimized. It includes both the "fitted data" part and the penalty part for "meeting physical constraints".

[0054] This represents the data fitting loss (main task loss), which is the "main task" loss of the MFGKS+CMTE model itself when physical constraints are not considered.

[0055]

[0056] : Error in the evolution of hidden states over time (dynamic model error), where: Represents the state transition matrix (how risk naturally increases / decreases over time); The precision matrix representing process noise (the smaller the Q value, the more stringent the precision); This represents the actual observation value of the m-th sensor at time k; This represents the theoretical observation value predicted by the observation equation from the hidden state x; This represents the observation residuals (weighted squared error). : Measurement accuracy matrix of the m-th sensor (the lower the noise, the greater the weight). ; in, This represents the squared error of the fusion risk in the j-th dimension; This represents the average risk across each dimension (leakage, arc flash, temperature, smoke, etc.). Represents the logarithm of the variance of the uncertainty in the model prediction; This indicates that the uncertainty reported by the model itself is used to standardize the error; the second term represents the Gaussian negative log-likelihood. Represents the RCD constraint penalty term, a physical / regulatory constraint penalty function related to RCD, which characterizes the degree to which "the model output violates the residual current limit"; Once the risk assessment model implies the leakage intensity Tends to exceed regulatory limits The losses increased dramatically, forcing the model to give a more conservative risk estimate near the boundary to avoid the dangerous output of "violations that were not reported".

[0057] This indicates a consistency penalty term for the hot model. This forces the physical balance of "rate of temperature rise = heat generation power - heat dissipation power" to hold; if the model overestimates the "risk of temperature rise" and the "power / heat dissipation" does not support it, the penalty will pull it back to the physically interpretable range. Specifically: Indicates heat capacity; : Rate of temperature rise; Power loss (heating power); Current temperature; Ambient temperature; Thermal resistance; This represents the upper limit penalty term for arc energy, a penalty function for the condition that "the arc energy in the short time window after prediction separation should not exceed the safe upper limit". For "expected current" "and "arc voltage" "Integration, estimating arc energy" As long as the predicted arc energy exceeds the allowable upper limit. Losses will penalize the model, prompting it to reduce the amount of dangerous risk or drive earlier disconnection; in: Indicates the moment when the contacts begin to separate; Indicates the arc duration window; Indicates instantaneous current; Indicates instantaneous arc pressure; Indicates the expected arc energy; This indicates the maximum allowable arc energy limit.

[0058] Make the physical laws (residual current limit, heat-power conservation, arc energy limit) into differentiable penalties. This is optimized together with the data fitting loss. The model trained in this way naturally adheres to the electrical / thermal boundaries, resulting in more stable performance and fewer false alarms when deployed online.

[0059] Where: RCD constraint (residual current limit constraint): the leakage-related output of the fusion model at the current sample / time, for example, the input: 1. Leakage risk components (t); 2. By The equivalent residual current obtained by mapping; 3. The corresponding RCD operating current limit for the line / equipment; 4. If necessary, operating condition correction parameters may be included, such as equivalent limits under different rated currents and environmental conditions.

[0060] Output: 1. The scalar of the RCD constraint loss calculated for the current sample / batch is used as part of the total training loss; 2. By backpropagation, the model parameters are indirectly corrected. The distribution of (t) makes the leakage risk output of the finally trained model during the inference stage naturally satisfy the behavior of "being more conservative when approaching the limit".

[0061] This constraint is used to constrain the output of leakage-related risks, ensuring that the leakage risk component (or equivalent residual current) given by the fusion model does not remain in the "exceeding regulatory limits but not triggering" range for an extended period. Furthermore, by imposing a significant penalty on "exceeding limits" during training, the model automatically provides a more conservative risk assessment when approaching the limits stipulated by the RCD law, reducing the dangerous output of "violations but no reports," and providing reliable input for subsequent protection actions and threshold calibration.

[0062] Thermal model consistency (RC model): This is used to constrain the consistency between the temperature rise-related risk output and the thermophysical model, preventing the model from giving unreasonable results such as "high temperature rise risk, but the corresponding heat generation power / heat dissipation conditions do not support it" under data noise. By forcing the first-order RC thermal equilibrium relationship of "temperature rise rate ≈ heat generation power − heat dissipation power" to hold, the physical interpretability and robustness of temperature rise risk are improved, and false detections caused by physical inconsistencies are reduced.

[0063] enter: 1. The fusion model outputs the temperature rise-related data for the current sample; 2. Heat / heat dissipation information obtained from electrical and structural parameters: 3. Preset first-order RC thermal model parameters (thermal resistance) Heat capacity wait).

[0064] Arc energy cap: This constraint ensures that the model's estimate of the "short-window arc energy after separation" does not exceed a preset safety cap when predicting disconnection timing and risk levels. If the predicted arc energy exceeds this cap, this constraint applies a stronger penalty to the corresponding output during training, prompting the model to either assign a higher risk level, driving earlier disconnection in subsequent control, or to favor disconnection timing with lower arc energy in the planning process. This provides a physical safety boundary for the zero-crossing energy minimization planning (AEMZP) in subsequent S4.

[0065] This embodiment is based on the electrical regulations limit and the heat-power balance relationship. It constructs a set of differentiable penalty functions by combining the residual current limit, the first-order RC thermal model, and the upper limit of arc energy. During the training phase of the multi-source fusion model, these functions are embedded as physical constraints into the total loss for joint optimization. This ensures that the leakage risk, temperature rise risk, and arc energy prediction of the model output naturally meet the electrical and thermal safety boundaries, thereby reducing false detections and missed detections caused by physical inconsistencies and improving the reliability of subsequent fire identification and disconnection timing planning.

[0066] S24. Output a unified risk situation vector r(t) and uncertainty for subsequent real-time control.

[0067] In a preferred embodiment of this application, step S3 specifically includes the following steps: S31. Based on the risk situation vector, use FSC to determine the fire level under the condition of a charging pile catching fire; The Fire Situation Classifier (FSC) is typically used as a general tabular data classifier in traditional Gradient Boosting Decision Trees. It directly takes raw sensor values ​​or simple statistical features as input and outputs binary (normal / fault) or multi-class fault type labels, mainly for offline diagnosis or non-real-time alarm scenarios. This application innovatively transforms it into a Fire Situation Classifier (FSC) specifically for residential charging facilities. It uses the unified risk situation vector r(t) and its dynamic increment ∆r(t) output in step S2 as core input features. The output is defined as a three-stage fire level with clear fire safety significance {smoke, open flame, and progression}. Through model compression and inference optimization, it ensures that the single prediction latency is <0.2 ms, so that the classification results can be directly used as control variables to seamlessly drive subsequent execution strategies such as adaptive confirmation window (WVC), zero-crossing phase planning (AEMZP), and hardware fast backoff. This achieves millisecond-level closed-loop coordination of the entire perception-identification-control chain, significantly improving the early identification and accurate isolation capabilities of charging fires. The specific implementation is as follows: An additive model using gradient boosting trees is employed. ; "Current risk intensity" "and the rate of risk growth" "By overlaying and fitting the tree model, we can find the nonlinear discrimination boundary to give the working condition level of 'smoke / open flame / progression,' thus transforming continuous risk into control level and driving subsequent time series / window optimization, where:" This represents the model's final predicted value or score for the i-th sample; This represents the input feature vector of the model, which is the unified risk situation vector output in step S2 (including dimensions such as leakage risk, arc risk, temperature rise risk, and smoke risk). This represents the dynamic increment of the risk situation vector, and is usually used together with r(t) as an input feature; This represents the index of the k-th decision tree in the gradient boosting model; This represents the total number of decision trees in the model (a hyperparameter, typically ranging from 50 to 300). Let represent the learning function of the k-th decision tree, which maps the input features to a numerical offset or class score; Represents the function space formed by the CART (Classification and Regression Tree) decision tree; This represents the target loss function, which in FSC uses multi-class cross-entropy loss to measure the deviation between the predicted value and the true label; This represents the structure regularization term for the k-th tree, typically including a penalty for the number of leaf nodes and L2 regularization of leaf weights to prevent overfitting; it outputs the fire severity level. Finally, enter... Its increment The algorithm obtains c and the confidence score. It uses the "level" as a control variable in steps S4 and S5, rather than solely for alarm purposes.

[0068] S32. Use TRE to predict thermal runaway time to limit; TRE (Thermal Runaway Extrapolation) is based on a first-order RC equivalent thermal model to predict the temperature evolution of key thermal nodes (such as terminals, wires, and contacts) in charging facilities in order to calculate the remaining time to reach a preset danger threshold. The specific principle is as follows: Steady-state temperature rise: = ; in, The ambient temperature (obtained in real time via an environmental sensor); Equivalent thermal resistance (determined by material properties); The current heat generation power (derived from the multi-source fusion risk vector in step S2) (Estimation of power-related dimensions in the data). Then, the exponential decay of temperature over time is described using the first-order thermal inertia equation: ; in, Current temperature (output by a temperature sensor or fusion layer); Equivalent heat capacity (as determined by the system); (·) is the natural exponential function; To predict time offset; Given a preset thermal runaway risk threshold (For example, based on the material's temperature resistance limit being set at 150°C), the temperature equation can be solved analytically to deduce the maximum temperature. Remaining time τ: ; Among them, if ≥ ,but = 0 (trigger immediately); if < and < ,but = ∞ (no risk); otherwise, ensure the denominator ( - To avoid logarithmic invalidity; In practical applications, multiple key hot nodes are calculated independently. Values, and take the minimum value. = min( , ,..., This is the most conservative estimate of the thermal runaway time limit. Directly input the confirmation window adaptive control (WVC) for subsequent step S33 and the disconnection timing plan (AEMZP) for step S4 to ensure that the disconnection action is completed before thermal runaway.

[0069] This model has low computational complexity, involving only basic floating-point operations, with a single execution latency of less than 0.1 ms, supporting millisecond-level real-time response. Furthermore, through collaboration with the front-end fusion module, it ensures parameter accuracy. and The dynamic updates enable accurate prediction of fire progress.

[0070] Traditional thermal runaway prediction methods typically employ fixed temperature / temperature rise rate threshold triggering, polynomial trend extrapolation, empirical statistical models, or complex multi-level thermal network simulations. These methods can only provide a fuzzy judgment on whether runaway is imminent, or require calculation time of more than a second, making it difficult to provide accurate hard constraints on remaining time that can be used for millisecond-level control. This embodiment innovatively transforms the classical first-order RC thermal model into a real-time analytical thermal runaway extrapolation model (TRE). By solving the temperature exponential decay equation in parallel for multiple key thermal nodes, it directly outputs the physically meaningful "shortest remaining time to runout". This is used as a hard deadline for subsequent zero-crossing phase planning (AEMZP), safety envelope supervision (SES), and adaptive confirmation window (WVC). Meanwhile, the required heat output P and the current temperature T0 are provided in real time by the front-end multi-source deep fusion module, requiring only a few floating-point operations (delay <0.1 ms). This achieves full-link closed-loop coordination of perception-prediction-control, thereby upgrading thermal runaway prediction from the traditional "passive alarm" to "active, precisely timed safety constraints", significantly improving the timeliness and certainty of disconnection during the fire progression stage.

[0071] S33. Use WVC to adaptively adjust the width of the confirmation window according to the fire severity level to improve response speed.

[0072] WVC (Window Validation Control) is used to dynamically adjust the width of the confirmation window for protective actions based on the current urgency of the fire. This is to achieve rapid response in fire situations while suppressing false alarms. The specific principle is as follows: WVC uses the fire severity level c output in step S31 and the shortest thermal runaway time limit output in step S32. And the required confirmation window width is calculated in real time based on the fusion uncertainty output in step S2. The calculation formula is as follows: ; in, The ordinal number representing the severity level of the fire (smoke level 1, open flame level 2, and progress level 3). Indicates the shortest remaining time to limit predicted for all critical hot nodes (unit: ms); This represents the mean or maximum value of the diagonal elements of the state covariance matrix output by the multi-source fusion module, used to characterize the uncertainty of the current risk estimate; , , This indicates a calibrable weighting coefficient, used to adjust the intensity of the influence of the three factors on the window width; These represent the upper and lower limits of the confirmation window (typical values ​​are 10–15 ms and 1–3 ms, respectively). (·) denotes a truncation function, ensuring that the calculation result always falls within [ , Within the interval; When the fire severity level is higher, the remaining thermal runaway time is shorter, and the fusion uncertainty is lower (i.e., the more certain the model is), the subtraction term in the formula becomes larger, leading to a larger confirmation window. W Automatic shortening accelerates the confirmation and execution of protection decisions; conversely, during normal operation or when uncertainty is high, W Maintaining a longer value effectively filters out transient interference; Calculated confirmation window width The special condition identification process directly used in step S3 serves as the duration requirement for the continuous fulfillment of triggering conditions in the risk situation, and further constrains the disconnection timing planning in step S4 and the collaborative execution strategy in step S5. This algorithm involves only basic arithmetic operations and one truncation operation, with a computational latency of less than 0.05ms. It can update in real time within a millisecond-level control cycle, ensuring the system adaptively balances speed and reliability under different fire situations.

[0073] Traditional protection devices typically use a fixed duration (e.g., 20ms, 50ms, 100ms) or a simple multi-cycle voting mechanism for the confirmation window (consistency confirmation period for preventing false triggering). This cannot be flexibly adjusted according to the actual urgency of the fire, easily leading to contradictions such as "too slow response when the fire is severe" or "too many false triggers during normal fluctuations." This embodiment innovatively proposes a verification window adaptive control algorithm (WVC) based on fire level, remaining time to limit, and fusion uncertainty. Through the aforementioned monotonically decreasing mapping function, the confirmation window width is adjusted... W It can dynamically shrink according to the following three real-time risk indicators: The more severe the fire severity level (c), the higher the rank(c) → W The shorter; Remaining time of thermal runaway The shorter → W The shorter; The smaller the fusion uncertainty σ, the more confident the model. W The shorter; This achieves an adaptive acceleration mechanism that follows the principle of "the more dangerous, the closer, the more certain → the shorter the confirmation window → the faster the decision response," while simultaneously... Reasonable calibration of the function and coefficients α1~α3 ensures a longer confirmation window under normal operating conditions to suppress false triggering. WVC upgrades the originally rigid fixed delay to a dynamic soft constraint deeply bound to the fire situation, becoming a key accelerator in the patented "fire awareness and zero-crossing energy minimization rapid isolation control" link. While ensuring a controllable false triggering rate, it compresses the end-to-end response time under the worst fire conditions to within 10 ms, significantly improving the real-time performance and reliability of fire prevention and control for residential charging facilities.

[0074] In a preferred embodiment of this application, step S4 specifically includes the following steps: S41. By combining the grid phase and actuator response characteristics with AEMZP, plan the target phase and optimal triggering time τ for zero-crossover interruption. * ; AEMZP (Arc Energy Minimized Zero-crossing Planning) proactively selects an optimal triggering time that minimizes the expected arc energy, while satisfying an end-to-end delay budget of ≤15 ms. The core formula and process are as follows: Expected arc energy (approximate): ; in, Indicates the trigger time The expected arc energy (in J) is a variable. Indicates a direct proportion (proportional relationship); Indicates the instant the actual contacts separate. The instantaneous current value (unit: A); Indicates the actual contact separation time (= ); This represents the energy metric exponent (a hyperparameter, typically p=2). Indicates the angular frequency of the power grid. =2πf (f is 50 Hz or 60 Hz); Indicates the triggering time of the trip coil (a variable the algorithm needs to decide); This represents the random deviation in the mechanism's response (the random delay from triggering to actual separation, which follows a certain distribution). This indicates the real-time phase of the current grid current (measured in real-time by a phase detection circuit); first, the current grid frequency is acquired in real-time. With phase In the future [ , Enumerate all reachable zero-crossing moments within the time range. ( = 1, 2, …); calculate the candidate trigger time for each zero crossing point. ,in For the pre-calibrated average mechanical hysteresis, the expected arc energy cost for each candidate is then calculated. Finally, the triggering time with the minimum arc energy is selected from all candidates that satisfy the following constraints: , + + ≤ (No later than the thermal runaway time limit output in step S32), if there are no candidates that fully satisfy the constraints, the selection degenerates into choosing the "nearest reachable zero crossing point".

[0075] Calculated optimal trigger time The output is directly fed to step S5 to execute the coordinated pre-power-off and control strategy for precise timing alignment of the coil pre-excitation (SPFC) and current reduction command (SS-CD). This algorithm requires only dozens of floating-point operations and a few loops, with a single execution delay of less than 0.3ms. It supports millisecond-level real-time planning, ensuring safe isolation with minimal arc energy under fire conditions, while also taking into account mechanical hysteresis compensation and hard constraints on thermal runaway time.

[0076] Traditional circuit breaker phase-selective disconnection technology typically employs passive waiting for the nearest zero-crossing point, fixed-delay triggering, or simple prediction of the current cycle's zero-crossing point. This results in problems such as large response delays, inability to account for thermal runaway time constraints, and insufficient mechanical hysteresis compensation leading to deviations from the actual disconnection time from the zero-crossing point. This embodiment innovatively proposes a zero-crossing energy minimization planning algorithm (AEMZP), which for the first time uses arc energy minimization as an explicit optimization objective. This is achieved by enumerating multiple candidate zero-crossing points and predicting arc energy costs, combined with the mechanism's average hysteresis. Execution link budget Latest moment of thermal runaway Equal hard constraints to achieve optimal triggering time Active planning; at the same time, a statistical form of the expected arc energy is introduced (considering the randomness of Δ), and it automatically degenerates into "nearest reachable zero crossing" when the safety margin is insufficient, ensuring that separation can still be completed within 15 ms with the minimum arc energy even under the worst fire conditions.

[0077] Through this algorithm, this embodiment upgrades the traditional passive "waiting for zero point" to active "selecting the optimal zero point". Under the premise of ensuring that the circuit breaks before thermal runaway, the instantaneous current and arc energy are reduced to the theoretical minimum, significantly reducing the risk of contact erosion and arc reignition, and realizing true predictive zero-crossing precise interruption.

[0078] S42. SES verifies that the end-to-end action delay is completed within 15ms. SES (Safety Envelope Supervision) imposes inequality constraints on end-to-end budget and arc energy: ; in: This represents the latency of the signal acquisition and preprocessing stage, the total time taken from the initial sensor sampling to the completion of feature extraction in step S1, typically 2–4 ms. This represents the latency of the multi-source deep fusion stage. The time taken for a complete forward inference cycle of MFGKS + CMTE is typically 3-5 ms. This indicates the time delay during the disconnection of timing and phase planning stages, and the total calculation time for the entire process of step S3 (FSC+TRE+WVC) + step S41 (AEMZP), typically 2-4 ms. This represents the execution link time budget, the longest deterministic time from coil ignition to actual complete contact separation, typically 4–7 ms; Indicates the optimal triggering time The energy of the separated arc predicted by the model below; This indicates the maximum allowable arc energy limit; continuously monitor whether the sum of the delays for the four segments of "acquisition / fusion / planning / execution" does not exceed 15 ms, and estimate the time required for each segment. The algorithm checks whether the arc energy at the point of impact is below the safety limit; if any condition is not met, a fallback strategy is triggered to ensure safety even in the worst-case scenario. The fallback strategy involves selecting "nearest reachable zero crossover" or directly switching to a hardware fast path. This algorithm guarantees that both safety and speed objectives are met even in the worst-case scenario.

[0079] Traditional circuit breaker protection systems typically only record the event after it has been activated, or simply reset the watchdog for a single timeout. They lack real-time prediction and active backoff mechanisms for the entire link's time delay and arc energy, making them prone to disconnection failure or severe contact erosion in extreme fire conditions due to planning errors or coordination delays. This embodiment innovatively proposes a Security Envelope Supervision (SES) mechanism, which monitors the actual latency of the four stages of perception, fusion, planning, and execution during the planning phase. ) and the best trigger time Corresponding expected arc energy Perform real-time inequality verification; if any metric is found to exceed the safety envelope (15ms total latency or...) (Arc energy limit), immediately abandon the current optimal plan, automatically switch to a conservative backoff strategy (select "nearest reachable zero crossover point" or directly start the hardware fast channel HFT / FBS), thereby ensuring that a safe disconnection can still be completed in a deterministic manner under the worst operating conditions.

[0080] Through this mechanism, this embodiment transforms "end-to-end ≤15 ms ultra-fast disconnection" and "controlled arc energy" from design goals into hard safety constraints that can be monitored in real time and enforced. This completely solves the reliability risks of traditional predictive zero-crossover technology when there is a loss of coordination or time delay fluctuations, achieving a true "zero-risk safety net" and significantly improving the intrinsic safety level of residential charging facilities in fire scenarios.

[0081] S43. When the safety margin is insufficient, the phase that can achieve zero crossover will be automatically selected.

[0082] In a preferred embodiment of this application, step S5 specifically includes the following steps: S51. Send a current reduction command to the charging pile via SS-CD to quickly reduce the charging current to below the safety threshold in order to reduce the arc energy at the moment of separation. SS-CD (Stepwise Sharp Current Drop) actively and rapidly reduces the charging current to below a safe threshold before contact separation by sending a precisely timed PWM duty cycle step command to the charging pile, thereby significantly reducing the arc energy at the moment of separation. The specific principle is as follows: SS-CD is based on a first-order exponential response model of charging piles and electric vehicles to CP signal duty cycle commands: ; in, This is the actual charging current; The target current; This is the current. This is the moment the flow reduction command is issued; This is the current response time constant between the charging pile and the vehicle (typical value 3-6 ms, calibrated by the protocol).

[0083] To ensure the actual separation time = + The current does not exceed the preset safety threshold. (Typically ≤50A), the algorithm solves in reverse order using the following steps: 1. Based on the target separation time Using the response time constant, we can deduce the theoretical target current required: ; 2. Linearly map the theoretical target current to the CP duty cycle command: ; Wherein, α and β are the charging pile duty cycle-available current calibration coefficients, determined according to the GB / T 27930 protocol. : Target control guidance signal duty cycle command value; 3. Calculate the latest time when the current reduction command is issued: ( The optimal trigger time output by AEMZP in step S4-1; The average mechanical hysteresis of the mechanism; This is the magnetization margin of the coil, typically 0.5–1 ms. 4. In The duty cycle step command is sent to the charging pile in a single operation via CAN or PLC communication. (Typically, the duty cycle is reduced directly from the normal value to 5%–10%), causing the current to decay rapidly within 5–8 ms before separation. the following.

[0084] If the communication delay exceeds the preset threshold or no confirmation response is received from the charging pile, the subsequent hardware fast track (HFT / FBS) will automatically take over to ensure that safety is not affected by the failure of coordination.

[0085] The algorithm involves only basic exponential and arithmetic operations, with a single execution delay of less than 0.1 ms. It can accurately achieve current soft landing within the planned zero-crossover separation window. Together with coil pre-excitation control (SPFC), it forms a predictive low-arc energy execution link, which significantly reduces the arc energy and ablation risk during contact separation.

[0086] Traditional circuit breaker protection systems typically only record the event after it has been activated, or simply reset the watchdog for a single timeout. They lack real-time prediction and active backoff mechanisms for the entire link's time delay and arc energy, making them prone to disconnection failure or severe contact erosion in extreme fire conditions due to planning errors or coordination delays. This embodiment innovatively proposes a Security Envelope Supervision (SES) mechanism, which monitors the actual latency of the four stages of perception, fusion, planning, and execution during the planning phase. The expected arc energy corresponding to the optimal trigger time τ* ( Perform real-time inequality verification; if any metric exceeds the safety envelope (15ms total delay or ... (Arc energy limit), immediately abandon the current optimal plan, automatically switch to a conservative backoff strategy (select "nearest reachable zero crossover point" or directly start the hardware fast channel HFT / FBS), thereby ensuring that a safe disconnection can still be completed in a deterministic manner under the worst operating conditions.

[0087] Through this mechanism, this embodiment transforms "end-to-end ≤15 ms ultra-fast disconnection" and "controlled arc energy" from design goals into hard safety constraints that can be monitored in real time and enforced. This completely solves the reliability risks of traditional predictive zero-crossover technology when there is a loss of coordination or time delay fluctuations, achieving a true "zero-risk safety net" and significantly improving the intrinsic safety level of residential charging facilities in fire scenarios.

[0088] S52, through SPFC and optimal triggering time τ * Predictive pre-excitation is applied to the trip coil to compensate for the randomness of mechanical hysteresis and minimize the arc energy at the moment of separation.

[0089] SPFC (Single Pulse Feedforward Compensation) uses the optimal triggering time planned... Beforehand, a precisely timed single-pulse pre-excitation current is applied to the trip coil to ensure accurate timing of the zero-crossing phase at the moment of contact separation. This compensates for the randomness of mechanical hysteresis and minimizes the arc energy at the moment of separation. The specific principle is as follows: SPFC models the trip coil as a first-order RL circuit, and its current response equation is: ; in, To apply voltage; For coil resistance (compensated in real time according to temperature) ); It represents the coil inductance.

[0090] To ensure that the coil current reaches the mechanism's starting threshold at the target time. The algorithm performs the analysis and calculation according to the following steps: 1. The inverse solution reaches the starting threshold. Minimum required pulse width: ; 2. Calculate the single-pulse ignition timing: ; in The optimal trigger time output by AEMZP in step S4-1; The average mechanical hysteresis of the mechanism is pre-calibrated, with a typical value of 2–4 ms; The calculated pulse width 3. In The amplitude V and duration are applied to the trip coil drive circuit at all times. A single pulse causes the coil current to... - The moment just arrived This triggers the mechanism to move and causes the actual contacts to separate. = + Precisely align the zero-crossing phase of the plan.

[0091] 4. Real-time temperature compensation: Dynamically corrects the resistance based on the coil temperature sensor readings. This ensures the accuracy of pulse width calculation under different ambient temperatures and aging conditions.

[0092] If external temperature or mechanical characteristics drift, the algorithm supports online adaptive adjustment. and Threshold. This algorithm involves only basic logarithms and four arithmetic operations, with a single execution latency of less than 0.05 ms, enabling precise pre-excitation within a millisecond-level control cycle.

[0093] By combining with the step current reduction algorithm (SS-CD), SPFC upgrades the traditional passive tripping with continuous power supply or fixed pulse width to a feedforward predictive excitation based on a physical model. This ensures that the jitter at the moment of contact separation is compressed to within ±0.5 ms, achieving a zero crossover hit rate of over 95%. It significantly reduces the instantaneous current and arc energy during separation, achieving truly ultra-fast, precise, and low-wear safe disconnection.

[0094] Traditional circuit breakers often use continuous energization or fixed pulse width triggering for coil excitation, which has problems such as large mechanical lag randomness, difficulty in accurately controlling the separation time, and easy to miss the optimal zero crossover phase, resulting in the actual arc energy being significantly higher than the theoretical value. This embodiment innovatively proposes a single-pulse feedforward pre-excitation algorithm (SPFC), which for the first time treats the trip coil as a first-order RL system and performs analytical inverse solution, combined with the optimal triggering time planned by AEMZP. With the calibrated average mechanical hysteresis Accurately calculate the ignition timing of a single pulse. With pulse width This achieves "feedforward precision hit" for zero crossover phase at the moment of contact separation; at the same time, temperature compensation and pulse width adaptation are introduced to ensure that the jitter at the moment of separation is compressed to within ±0.5 ms under different ambient temperatures and aging conditions.

[0095] Through this algorithm, this embodiment upgrades the traditional passive and discrete tripping execution to an active, predictable, and compensable predictive pre-excitation control. Together with step current reduction (SS-CD), it forms a millisecond-level low-arc energy collaborative execution link, which completely solves the phase deviation problem caused by mechanical lag. This increases the zero-crossing hit rate from 60-70% in the traditional method to over 95%, achieving true "ultra-fast, zero-crossing precise interruption" and significantly reducing the risk of contact wear and arc reignition.

[0096] In a preferred embodiment of this application, step S6 specifically includes the following steps: S61. QCP / CTD is used to achieve online statistical calibration and coverage guarantee of protection action threshold; The specific operation of QCP / CTD (Quantile-based Calibration with Probability Coverage / Coverage Threshold Determination) is as follows: Define the degree of inconsistency under the "normal state": ; in, Indicates the degree of inconsistency at the current moment; This represents the maximum risk component at time t; This represents the value of the j-th dimension risk component at time t; This represents the basic protection action threshold; it compares the current maximum risk component with a basic threshold. By subtracting from the standard value, we obtain the "inconsistency degree". It measures how close / exceeds the trigger boundary of the current sample and serves as the raw material data for subsequent statistical calibration.

[0097] In rolling calibration set Take the upper quantile: ; in: This represents the 1−α quantile calibration value; Indicates the upper limit of the target error rate; Representing historical moments Non-uniformity samples; Indicates the historical sample index; Indicates a rolling calibration buffer; a rolling "calibration buffer" Above, estimated quantiles By statistically analyzing the "approach level" over a past period, the trigger threshold can be dynamically raised or lowered to ensure that coverage / false alarm rate meets the target. controlled.

[0098] Trigger threshold: ; Set the base threshold Add statistical calibration To obtain the final action threshold It will trigger immediately as soon as the current maximum risk exceeds it, and the long-term false trigger rate is within [a certain range]. Internally verifiable.

[0099] The algorithm transforms the "action threshold" into a data-adaptive and coverage-guaranteed quantitative indicator, satisfying both compliance and reliability requirements.

[0100] Traditional circuit breaker protection threshold calibration often relies on offline empirical settings, fixed thresholds, or simple averaging filtering, which cannot adapt to environmental noise, sensor drift, or changes in operating conditions in real time. This results in large fluctuations in the false trip rate or the need for frequent manual parameter adjustments. In contrast, this embodiment innovatively proposes an online threshold calibration algorithm based on statistical coverage guarantee (QCP / CTD), which for the first time incorporates the "non-consistency of the highest risk component". As calibration material, the high quantile of the rolling buffer C is used. Dynamic compensation base threshold To achieve the threshold The algorithm features adaptive statistical calibration and error rate coverage guarantee (long-term confidence level ≥1−α). It requires no manual intervention, only basic quantile calculation (delay <0.1 ms), and directly feeds the calibration results back to the fusion layer (S2) and the identification layer (S3), forming a closed loop throughout the process. This ensures that the error rate remains constant within 1% to 5% under noise interference or changes in operating conditions, while improving the reliability and compliance of the overall decision-making.

[0101] Through this algorithm, this embodiment upgrades the traditional static threshold to a data-driven dynamic confidence threshold mechanism, which completely solves the problem of threshold failure in complex environments for residential charging facilities, realizes "online learning" protection parameter optimization, and significantly reduces the risk of false triggering and missed detection.

[0102] S62. After the protection action is triggered, the RCC is used to automatically classify and identify the root cause of the event and record the event, complete the full-element traceable archiving, and realize the accurate location of the fault and compliant evidence collection. After a protection action is triggered, the RCC (Root Cause Classification Recorder) automatically classifies and identifies the root cause of the event and completes full-element traceable archiving, achieving accurate fault location and compliant evidence collection. The specific principle is as follows: RCC uses the high-dimensional fusion representation output by the multi-source deep fusion module in step S2. Using the unified input feature, multiple parallel independent logistic regression classifiers are used to calculate the probability of occurrence for each of the predefined root cause category sets k ∈ {leakage, arc flash, overheating, smoke, water immersion, external short circuit, other}: ; Where σ(·) is the sigmoid activation function; , The trainable weights and biases corresponding to the k-th root cause (obtained offline through supervised training using labeled event data); Then, based on the preset threshold value Generate binary root cause labels: = 1, if > ; = 0, otherwise; Typical values ​​range from 0.7 to 0.95, and can be set according to the importance of the categories. Simultaneously, RCC automatically extracts the contribution of key features for each root cause classification for interpretability analysis. Typical methods include: 1. Absolute gradient sensitivity | / |; 2. Or the absolute value of the logistic regression coefficient | ,i|; In generating the root cause label set { Afterwards, RCC will form a structured event log with the following information and store it in non-volatile memory (EEPROM or Flash): 1. Root cause label vector { } and its confidence level { }; 2. Ranking of key feature contributions; 3. Complete timeline of events: (Optimal trigger time) (Coil ignition time) (Contact separation moment) (At the moment of the current reduction instruction) The fusion risk vector r(t) and the uncertainty sequence action results and hardware state snapshot 50-100 ms before the trigger; The above records are stored in encrypted timestamp format and can be exported through a dedicated diagnostic interface or remote operation and maintenance platform for use in fire safety compliance evidence collection, insurance claims, and product iteration analysis.

[0103] This module has extremely low computational overhead (single inference latency <0.1 ms), does not affect the real-time protection link, and is executed serially only after the action is triggered. By directly using the deep fusion intermediate representation for multi-task root cause classification, RCC achieves true end-to-end automatic root cause diagnosis and full-process traceability recording, completely solving the pain point of traditional circuit breakers that "only know the action but not the cause," enabling residential charging facilities to have complete event black box functionality, and significantly improving the system's auditability, maintainability, and fire safety compliance.

[0104] Traditional circuit breaker event logging systems typically only log threshold crossing times, action types, and raw sensor values, lacking automatic root cause analysis and interpretable contribution. This leads to fault tracing relying on manual analysis, which is inefficient and prone to errors. In contrast, this embodiment innovatively proposes a Root Cause Classification (RCC) logging module, which for the first time integrates multi-source deep fusion representation. As input, the probability of occurrence of each root cause (leakage, arcing, overheating, smoke, water immersion, etc.) is calculated in real time by a multi-head logistic regression unit. With binary labels It automatically extracts key feature contributions (gradients or tree paths) and integrates them with the event timeline for unified archiving, realizing a closed-loop mechanism of "end-to-end automatic root cause diagnosis + quantifiable traceability". This module has low computational complexity (latency <0.1 ms), requires no additional hardware, relies only on the output of the fusion layer, and supports online auditing and AI-assisted review.

[0105] Through this module, this embodiment upgrades the traditional passive log recording into an active and intelligent root cause classification and tracing system, completely solving the "black box" problem of fault diagnosis of residential charging facilities, shifting event evidence collection from manual reliance to data-driven, and significantly improving the system's maintainability, compliance and fire prevention capabilities.

[0106] S63. Archive key feature data and control parameters to achieve traceability and verification.

[0107] like Figure 2 As shown, another preferred embodiment of this application also provides a circuit breaker control device for residential charging facilities, including: The data acquisition and preprocessing module is used to acquire multimodal safety perception data during the operation of residential charging facilities. It completes synchronous sampling and real-time preprocessing under a unified clock, and constructs multimodal time series feature data including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status, forming a feature vector sequence for subsequent fusion. The risk assessment module is used to align and jointly model feature vector sequences based on deep fusion and risk assessment of multi-source asynchronous data, and to form and output a unified risk situation vector with physical constraints and give the uncertainty. The risk level identification module is used to identify the fire and predict thermal runaway under the condition of a charging pile catching fire based on the risk situation vector, determine the current fire level and the predicted time limit, determine the disconnection level and confirmation window, and accelerate the decision response under the condition of a charging pile catching fire. The disconnect timing and phase planning module is used to combine the power grid phase and actuator response characteristics to determine the target phase and optimal triggering time for zero crossover interruption through disconnect timing and phase planning. The collaborative execution module is used to execute collaborative pre-power-off and execution control strategies. Within the confirmation window before contact separation, it implements control strategies for pre-excitation of the trip coil and collaborative current reduction at the pile end to reduce the arc energy during contact separation.

[0108] The circuit breaker control device for residential charging facilities provided in this embodiment adopts the circuit breaker control method for residential charging facilities in the above embodiments. It solves the technical problems in the prior art where bandwidth allocation and loss compensation are lagging due to the discrete control mechanism, making it difficult to meet the power system's requirements for real-time and reliable communication, and unable to synchronously respond to changes in service demand and line loss fluctuations, resulting in insufficient communication link stability and low resource utilization. Compared with the prior art, the beneficial effects of the circuit breaker control device for residential charging facilities provided in this embodiment are the same as those of the circuit breaker control method for residential charging facilities provided in the above embodiments. Moreover, other technical features in the circuit breaker control device for residential charging facilities are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0109] In a preferred embodiment of this application, the circuit breaker control device for residential charging facilities further includes: The calibration and traceability module is used to adjust and verify the protection action threshold in real time through action threshold calibration and event traceability, so as to control the false action rate and ensure the reliability of triggering decisions; at the same time, it generates event root cause labels and key feature data to achieve full process recording and traceability.

[0110] like Figure 3 As shown, a preferred embodiment of this embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the circuit breaker control method for residential charging facilities described in the above embodiment.

[0111] This embodiment provides an electronic device that employs the circuit breaker control method for residential charging facilities described in the above embodiments. This addresses the technical problems in the prior art where bandwidth allocation and loss compensation, due to the lag of the discrete control mechanism, are difficult to meet the power system's requirements for real-time communication and reliability, and cannot synchronously respond to changes in service demand and fluctuations in line loss, resulting in insufficient communication link stability and low resource utilization. Compared with the prior art, the beneficial effects of the electronic device provided in this embodiment are the same as those of the circuit breaker control method for residential charging facilities provided in the above embodiments. Furthermore, other technical features of the electronic device are the same as those disclosed in the methods of the above embodiments, and will not be elaborated upon here.

[0112] like Figure 4 As shown in the preferred embodiment, this embodiment also provides a computer device, which may be a terminal or a liveness detection server, and its internal structure diagram may be as follows. Figure 4 As shown. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with other external computer devices via a network connection. When the computer program is executed by the processor, it implements the steps of the circuit breaker control method for residential charging facilities described above.

[0113] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the solution of this embodiment, and does not constitute a limitation on the computer device to which the solution of this embodiment is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0114] The computer equipment provided in this application adopts the circuit breaker control method for residential charging facilities in the above embodiments, which solves the technical problems in the prior art where bandwidth allocation and loss compensation are lagging due to the discrete control mechanism, making it difficult to meet the power system's requirements for real-time and reliable communication, and unable to synchronously respond to changes in business demand and fluctuations in line loss, resulting in insufficient communication link stability and low resource utilization. Compared with the prior art, the beneficial effects of the computer equipment provided in this embodiment are the same as those of the circuit breaker control method for residential charging facilities provided in the above embodiments, and other technical features in the electronic equipment are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0115] A preferred embodiment of this example also provides a storage medium, which includes a stored program that, when the program is executed, controls the device containing the storage medium to perform the steps of the circuit breaker control method for residential charging facilities described in the above embodiment.

[0116] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0117] If the functions described in this embodiment are implemented as software functional units and sold or used as independent products, they can be stored in one or more computing device-readable storage media. Based on this understanding, the parts of this embodiment that contribute to the prior art or the technical solution can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computing device (which may be a personal computer, server, mobile computing device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this embodiment. The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0118] Those skilled in the art will understand that the embodiments of this example can be provided as methods, systems, or computer program products. Therefore, this example can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this example can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in this example can be implemented using various computer languages, such as the object-oriented programming language C++ and the embedded programming language C.

[0119] This embodiment is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this embodiment. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0122] This embodiment also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the circuit breaker control method for residential charging facilities as described above.

[0123] The computer program product provided in this embodiment solves the technical problems of high testing and experimentation costs, delays, limited applicability, and slowdowns in scientific research in existing technologies. Compared with the prior art, the beneficial effects of the computer program product provided in this embodiment are the same as those of the circuit breaker control method for residential charging facilities provided in the above embodiments, and will not be repeated here.

[0124] Obviously, those skilled in the art can make various modifications and variations to this embodiment without departing from the spirit and scope of this embodiment. Therefore, if these modifications and variations of this embodiment fall within the scope of the claims of this embodiment and their equivalents, this embodiment is also intended to include these modifications and variations.

Claims

1. A circuit breaker control method for residential charging facilities, characterized in that, Including the following steps: S1. Acquire multimodal safety perception data of the operation of residential charging facilities, complete synchronous sampling and real-time preprocessing under a unified clock, construct multimodal time series feature data including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status, and form a feature vector sequence for subsequent fusion; S2. Based on deep fusion and risk assessment of multi-source asynchronous data, the feature vector sequence is aligned and jointly modeled, and a unified risk situation vector is formed and output with physical constraints, and the uncertainty is given. S3. Based on the risk situation vector, perform fire identification and thermal runaway prediction under the condition of charging pile fire, determine the current fire level and the predicted time limit, determine the disconnection level and confirmation window, and accelerate the decision response under the condition of charging pile fire. S4. Combining the power grid phase and the actuator response characteristics, determine the target phase and optimal triggering time for zero-crossover interruption by disconnection timing and phase planning; S5. Implement coordinated pre-power-off and control strategies. Within the confirmation window before contact separation, implement control strategies for pre-excitation of the trip coil and coordinated current reduction at the pile end to reduce arc energy during contact separation.

2. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, It also includes the following steps: S6. Through action threshold calibration and event tracing, the protection action threshold is adjusted and verified in real time to control the false action rate and ensure the reliability of triggering decisions; at the same time, event root cause labels and key feature data are generated to achieve full-process recording and traceability.

3. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, Step S5 further includes the following step: If the external collaboration fails to respond within the specified time, the hardware fast-track back-off module will automatically execute a fallback tripping action to complete the contact separation and ensure safety.

4. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. Synchronously collect multimodal safety perception data of the operation of residential charging facilities under a unified clock, including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status; S12. The collected multimodal safety perception data is preprocessed. The preprocessing process uses Goertzel array, TKEO energy operator, EWMA slope estimation, second-order difference threshold detection and Schmitt triggering mechanism to extract multimodal time series feature data of current anomaly, arc activity, temperature rise trend, smoke change and water immersion status. S13. Encapsulate the timestamped multimodal time series feature data into a feature vector sequence for subsequent fusion.

5. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, Step S2 specifically includes the following steps: S21. Use MFGKS to complete asynchronous alignment and unified hidden state estimation of multi-rate feature vector sequences; S22. Use CMTE to establish a cross-modal feature interaction model in the fusion layer to establish the interaction relationship between electrical, thermal and environmental information, rather than for classification. S23. Embedding PGE in the fusion loss introduces differentiable physical constraint terms during model training to reduce false detections and ensure that the results comply with electrical and thermal constraints. S24. Output a unified risk situation vector r(t) and uncertainty for subsequent real-time control.

6. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, Step S3 specifically includes the following steps: S31. Based on the risk situation vector, use FSC to determine the fire level under the condition of a charging pile catching fire; S32. Use TRE to predict thermal runaway time to limit; S33. Use WVC to adaptively adjust the width of the confirmation window according to the fire severity level to improve response speed.

7. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41. By combining the grid phase and actuator response characteristics with AEMZP, plan the target phase and optimal triggering time τ for zero-crossover interruption. * ; S42. SES verifies that the end-to-end action delay is completed within 15ms. S43. When the safety margin is insufficient, the phase that can achieve zero crossover will be automatically selected.

8. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51. Send a current reduction command to the charging pile via SS-CD to quickly reduce the charging current to below the safety threshold in order to reduce the arc energy at the moment of separation. S52, through SPFC and optimal triggering time τ * Predictive pre-excitation is applied to the trip coil to compensate for the randomness of mechanical hysteresis and minimize the arc energy at the moment of separation.

9. The circuit breaker control method for residential charging facilities according to claim 1, characterized in that, Step S6 specifically includes the following steps: S61. QCP / CTD is used to achieve online statistical calibration and coverage guarantee of protection action threshold; S62. After the protection action is triggered, the RCC is used to automatically classify and identify the root cause of the event and record the event, complete the full-element traceable archiving, and realize the accurate location of the fault and compliant evidence collection. S63. Archive key feature data and control parameters to achieve traceability and verification.

10. A circuit breaker control device for residential charging facilities, characterized in that, include: The data acquisition and preprocessing module is used to acquire multimodal safety perception data during the operation of residential charging facilities. It completes synchronous sampling and real-time preprocessing under a unified clock, and constructs multimodal time series feature data including AC / DC residual current, arcing signs, terminal temperature, smoke and water immersion status, forming a feature vector sequence for subsequent fusion. The risk assessment module is used to align and jointly model feature vector sequences based on deep fusion and risk assessment of multi-source asynchronous data, and to form and output a unified risk situation vector with physical constraints and give the uncertainty. The risk level identification module is used to identify the fire and predict thermal runaway under the condition of a charging pile catching fire based on the risk situation vector, determine the current fire level and the predicted time limit, determine the disconnection level and confirmation window, and accelerate the decision response under the condition of a charging pile catching fire. The disconnect timing and phase planning module is used to combine the power grid phase and actuator response characteristics to determine the target phase and optimal triggering time for zero crossover interruption through disconnect timing and phase planning. The collaborative execution module is used to execute collaborative pre-power-off and execution control strategies. Within the confirmation window before contact separation, it implements control strategies for pre-excitation of the trip coil and collaborative current reduction at the pile end to reduce the arc energy during contact separation.

11. The circuit breaker control device for residential charging facilities according to claim 10, characterized in that, Also includes: The calibration and traceability module is used to adjust and verify the protection action threshold in real time through action threshold calibration and event traceability, so as to control the false action rate and ensure the reliability of triggering decisions; at the same time, it generates event root cause labels and key feature data to achieve full process recording and traceability.

12. An electronic device, the electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements a circuit breaker control method for residential charging facilities as described in any one of claims 1 to 9.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the circuit breaker control method for residential charging facilities as described in any one of claims 1 to 9.

Images