A method and system for fault detection in RV energy storage systems

Abstract

The present invention provides a method and system for fault detection of a motorhome energy storage system, belonging to the technical fields of new energy vehicles and intelligent detection. The method captures load start / stop operation instructions with timestamps in real time, and synchronously collects voltage and current signals at the output end of the energy storage system, strictly aligning the operation behavior and electrical dynamics on the time axis. Based on the load identifier in the operation instruction, a pre-stored feature template is retrieved, and the transient electrical features extracted in real time are subjected to spatio-temporal consistency verification with it, effectively distinguishing legal load startup from real system faults. When the verification fails, the similarity with the fault feature library is further calculated for risk assessment and hierarchical protection is executed. The corresponding system includes an operation instruction capture module, an electrical signal acquisition module, a central processing and synchronization module, a load fingerprint matching module, a fault risk assessment module, and a hierarchical protection execution module. The present invention significantly reduces the false alarm rate and improves the accuracy of fault detection and the continuity of system power supply.

Description

Technical Field

[0001] This invention belongs to the field of new energy vehicle and intelligent detection technology, specifically relating to a fault detection method and system for RV energy storage systems. Background Technology

[0002] With the advancement of new energy vehicles and intelligent power management technologies, RV energy storage systems equipped with large-capacity battery packs are increasingly widely used in recreational vehicles. These systems integrate battery management, energy conversion, and multiple load interfaces, and their operational reliability is directly related to the safety of users in the wild. Therefore, building a highly reliable fault detection mechanism is the core requirement for the design of RV energy storage systems.

[0003] Currently, a key challenge in fault detection lies in the difficulty of accurately distinguishing between normal user operation and genuine system faults. For example, when a user actively starts a high-power appliance, the resulting instantaneous current surge and bus voltage drop are highly similar in electrical characteristics to serious faults such as internal short circuits. However, existing detection solutions generally employ protection logic based on fixed thresholds, which cannot effectively identify such legitimate load switching events. This can easily lead to false alarms or even unnecessary power outages, affecting user experience and potentially causing secondary risks in critical power supply scenarios. To alleviate this problem, existing technologies may choose to relax the judgment threshold, but this reduces the detection sensitivity for slowly changing genuine faults; or introduce a manual reset confirmation mechanism, but this relies on subjective operation and is difficult to achieve automated closed-loop protection. In addition, existing systems lack the ability to finely identify the electrical characteristics of loads and fail to effectively integrate the contextual information of user active operation. Although there are intelligent algorithms based on load identification in the industrial field, they are usually computationally complex, difficult to deploy on vehicle edge devices, and do not take into account the operational intent in the RV scenario, resulting in deficiencies in real-time performance and false alarm suppression.

[0004] Therefore, there is an urgent need in this field for a lightweight fault detection solution suitable for the vehicle environment that can collaboratively analyze user operation behavior and load electrical dynamics, thereby accurately distinguishing between legitimate load startup and real system faults while ensuring safety, and effectively avoiding false alarms. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and system for fault detection in RV energy storage systems, employing the following technical solution.

[0006] Firstly, a method for fault detection in a motorhome energy storage system includes: Capture user-issued load start / stop commands in real time. The commands include the target load identifier, operation timestamp, and operation type. The electrical signals at the output of the energy storage system are collected synchronously. These electrical signals include voltage and current signals. The collected voltage and current signals are then time-stamped to align the operation commands with the electrical signals on the time axis. Based on the target load identifier, the corresponding baseline electrical feature template is retrieved from the pre-set load electrical fingerprint database. The baseline electrical feature template includes the range or threshold of electrical feature parameters related to the load type. Using the operation timestamp as the reference origin, extract transient electrical characteristic parameters within a preset time window from the aligned electrical signal; The extracted transient electrical characteristic parameters are compared with the baseline electrical characteristic template for spatiotemporal consistency verification. If the verification passes, the load is deemed to have started legally. If the verification fails, the fault risk assessment process begins. In the fault risk assessment process, the similarity between transient electrical characteristic parameters and feature vectors of various fault modes in the preset fault feature library is calculated. Based on the comparison results of similarity scores and preset thresholds, corresponding fault alarm signals are generated and graded protection actions are executed.

[0007] Preferably, the system captures user-issued load start / stop commands in real time, specifically including: User operations are received via in-vehicle central control touchscreen, wireless mobile terminal application interface, or physical button array. User operations are encapsulated into structured instruction data frames that include operation type, target payload identifier, and millisecond-level timestamp; The instruction data frame is transmitted to the central processing unit via the controller area network bus and stored in the instruction cache queue.

[0008] Preferably, the voltage and current signals at the output of the energy storage system are collected simultaneously, with a sampling frequency of not less than 2000Hz; Furthermore, a high-precision timer is used to add a millisecond-level time stamp to each set of synchronously acquired voltage and current sampling data pairs to ensure that the deviation between the data and the timestamp of the operation command does not exceed a preset value.

[0009] Preferably, the construction and updating of the load electrical fingerprint database includes a learning mode, which is initiated and executed when a target load identifier is first identified or when a trigger condition is met: Under stable electrical conditions, record the voltage and current signals within a preset time after the load starts; The starting electrical characteristic parameters of the load, including the rate of rise of starting current and the depth of voltage sag, are automatically extracted based on the recorded signals. The extracted feature parameters are validated for reasonableness, and after user confirmation, the target load identifier and its feature parameters are stored as a new record in the load electrical fingerprint database.

[0010] Preferably, the updating of the load electrical fingerprint database further includes enhanced learning modes and dynamic optimization: The load maintains a learning state and triggers the enhanced learning mode when learning for the first time, when the learning confidence is low, or when the verification consistency is poor, in order to collect multiple consecutive normal start-stop operation data samples. Cluster analysis was performed on the collected samples to remove outliers, and the statistical mean and standard deviation of key feature parameters were calculated based on the remaining clean samples. Based on the statistical mean and standard deviation, and combined with the engineering safety boundary, a statistical feature template containing a dynamic tolerance interval is constructed to update the baseline electrical feature template.

[0011] Preferably, it further includes a load health status early warning step: A long-term record sequence of key characteristic parameters of the registered load that changes over time; A feature evolution model is established based on the long-term recorded sequence to obtain the baseline values ​​and normal fluctuation bands of the feature parameters; During load operation, the real-time extracted feature parameters are compared with the baseline value. If the deviation exceeds the warning threshold set based on the normal fluctuation range but does not exceed the dynamic tolerance range, a device health warning message is generated.

[0012] Preferably, the spatiotemporal consistency verification specifically includes: A forward time window is defined starting from the operation timestamp, and the comparison rules and tolerance ranges corresponding to the load type are obtained from the reference electrical feature template. The extracted transient electrical characteristic parameters, including current rate of change, voltage drop depth, current waveform distortion rate, and dominant harmonic order, are compared item by item with the tolerance range. If all feature parameters fall within the tolerance range of their corresponding type, the verification is considered successful.

[0013] Preferably, the fault risk assessment process specifically includes: A fault feature library containing multiple fault modes is pre-defined, and each fault mode is defined by a multi-dimensional feature vector and associated with feature weights and judgment thresholds; Calculate the multidimensional feature value corresponding to the current event based on the transient electrical characteristic parameters; The weighted Euclidean distance algorithm is used to calculate the similarity score between the current event feature vector and the baseline feature vectors of various fault modes in the fault feature library; The similarity score is compared with the corresponding fault mode determination threshold, and the fault type and risk level are determined based on the comparison result.

[0014] Preferably, it also includes a coordination mechanism for handling timing anomalies between operating instructions and electrical events: The maintenance instruction waiting list and the electrical event buffer are used to cache operation instructions to be verified and raw data segments triggered by electrical transient monitoring, respectively. Parallel execution of instruction-driven active matching and event-driven reverse matching logic to associate operation instructions with electrical events; For unmatched instructions that time out or electrical events that are not associated with instructions, interactive confirmation is performed through the user verification interface, and subsequent verification is executed or the process is directly transferred to the fault risk assessment process based on user feedback.

[0015] Secondly, a fault detection system for a motorhome energy storage system includes: The user operation command capture module is used to capture and encapsulate load start and stop operation commands containing target load identifier, operation type and millisecond-level timestamp in real time; A high sampling rate electrical signal acquisition module is used to synchronously acquire voltage and current signals at the output of the energy storage system; The central processing and synchronization unit is used to receive the operation instructions and electrical signals, perform high-precision time synchronization and alignment on them, and schedule the operation of the core algorithm. The load fingerprint matching module is used to query or construct a load electrical fingerprint database based on the target load identifier in the operation instruction, and to perform spatiotemporal consistency verification. The fault risk assessment module is used to calculate the similarity between the current event and the preset fault feature library when the spatiotemporal consistency check fails, and to classify the fault and determine the risk level. The graded protection execution module is used to execute corresponding alarm and power-off protection actions based on the output results of the fault risk assessment module.

[0016] In summary, this application includes at least one of the following beneficial technical effects: 1. This application achieves accurate differentiation between legitimate load start events and real system faults by synchronously collecting user operation commands and electrical signals and establishing strict alignment and feature comparison between the two on the time axis. This avoids false alarms and unnecessary power outages caused by the normal connection of high-power electrical appliances, thereby improving the accuracy of fault detection and the continuity of system power supply.

[0017] 2. This application adopts a lightweight rule judgment mechanism based on load identifier to retrieve pre-stored electrical feature templates for comparison, replacing complex general model calculations. It also supports rapid self-learning library building upon first use, resulting in low computational load and fast response of the core algorithm. This makes it easy to deploy in resource-constrained vehicle embedded controllers, achieving a balance between high performance and low cost.

[0018] 3. This application introduces multi-cycle learning and cluster cleaning to construct dynamic statistical templates, and conducts long-term trend monitoring and evolution analysis on key electrical characteristics of the load. This not only makes the characteristic tolerance range more consistent with the individual characteristics of the equipment and improves the compatibility of comparison, but also has the ability to sense the slow aging and performance degradation of the load. This enables a leap from post-fault protection to pre-fault health warning, and improves the intelligence and preventive maintenance level of the system. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a fault detection method for a motorhome energy storage system according to the present invention. Figure 2 This is a flowchart illustrating the dual-channel spatiotemporal fusion analysis architecture of this invention; Figure 3 This is a schematic diagram of the process for capturing load start / stop operation commands and acquiring high-sampling-rate electrical signals in this invention. Detailed Implementation

[0020] This invention provides a fault detection method and system for RV energy storage systems, aiming to solve the technical problem in the prior art where current surges and voltage drops caused by the connection of high-power legal loads are misjudged as serious electrical faults, resulting in unnecessary power outage protection and false alarms.

[0021] This invention constructs a lightweight dual-channel spatiotemporal fusion analysis architecture to simultaneously collect and analyze the context of user operation behavior and the electrical dynamic characteristics of the load, thereby achieving accurate differentiation between real faults and legitimate load start events. While ensuring system security, it improves the accuracy of fault detection and user experience.

[0022] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on specific implementation methods of the present invention.

[0023] Example 1 A method for fault detection in a motorhome energy storage system includes the following steps: S1 captures user-issued load start / stop operation commands in real time through the user interaction interface. The operation commands include the target load identifier, operation timestamp, and operation type. Specifically, it includes the following sub-steps: S101: Users can issue start or stop commands for a specific electrical device through any of the following interactive interfaces: the vehicle central control touch screen, the wireless mobile terminal application interface, or the physical button array.

[0024] When a device is first connected or configured, the system registers the load through the aforementioned interactive interface, assigns a unique load identifier, and binds it to the electrical characteristic template of the load. The device name displayed in the user interface corresponds one-to-one with the load identifier, ensuring that operation commands can accurately target the load.

[0025] S102: The system automatically assigns a unique sequence number to each operation command and adds a timestamp accurate to the millisecond level (recording the millisecond count value since the system was powered on).

[0026] The instruction data structure contains three fields: operation type ("start" or "stop"), target load identifier (encoded as a 16-bit unsigned integer to uniquely identify the registered electrical equipment in the system), and timestamp.

[0027] S103: Operation commands with serial numbers and timestamps are transmitted to the central processing unit via the Controller Area Network (CAN) bus.

[0028] S104: After receiving an instruction, the central processing unit immediately writes it into the instruction buffer queue and records the time of receipt.

[0029] The instruction cache queue adopts a first-in-first-out (FIFO) structure with a maximum capacity of 128 entries, ensuring that no user intent information is lost in high-concurrency operation scenarios.

[0030] S105: The central processing unit continuously monitors the instruction cache queue and, based on the aforementioned recorded user operation instructions and their timestamps, starts the corresponding electrical signal analysis thread to prepare for subsequent spatiotemporal consistency verification.

[0031] S2 collects the output voltage and current signals of the energy storage system in real time through a high sampling rate electrical sensing unit. The sampling frequency of the high sampling rate electrical sensing unit is not less than 2000Hz.

[0032] The output terminal refers to the DC bus output node where the energy storage system supplies power to the load. The collected voltage signal is the DC bus voltage, and the current signal is the total output current. An electrical sensing unit is installed at this output node to reflect the electrical dynamics under load switching and fault conditions. Specifically, it includes the following sub-steps: S201: Equipped with a high sampling rate electrical sensing unit, which consists of two parts: an isolated Hall effect current sensor and a differential voltage sampling circuit.

[0033] S202: The current sensor has a measurement range of -100A to +100A, an accuracy class of 0.5, and good linearity and temperature stability. Its output analog current signal is digitized by a 16-bit analog-to-digital converter (ADC) at a fixed sampling frequency of not less than 2000Hz.

[0034] S203: The voltage sampling circuit adopts an instrumentation amplifier architecture with a common-mode rejection ratio of not less than 120dB to effectively suppress the impact of RV chassis ground potential fluctuations on measurement accuracy; its input voltage range covers DC 0V to 600V, and it is also sampled by a 16-bit ADC at a frequency of not less than 2000Hz.

[0035] S204: Sends the digital voltage and current data streams output by the current sensor and voltage sampling circuit to the hardware timestamp alignment unit in real time.

[0036] S205: The timestamp alignment unit uses a high-precision timer inside the central processing unit to add a timestamp accurate to the millisecond level to each pair of synchronously acquired voltage-current sampling data.

[0037] S206: Through the above synchronization mechanism, the deviation between the time stamp of the electrical sampling data and the timestamp of the operation command from the user interface is ensured to be no more than 1ms, thereby achieving strict alignment of the dual-channel data on the time axis.

[0038] Furthermore, this synchronization accuracy matches the subsequent 500ms signal analysis window, ensuring that operating commands and electrical events can be accurately correlated during the load start-up transient process, avoiding misalignment of feature extraction due to time deviation.

[0039] S3, based on the target load identifier in the operation command, retrieves the reference electrical feature template corresponding to the target load from the preset load electrical fingerprint database. The reference electrical feature template includes the steady-state current waveform, the current rise rate threshold during the startup phase, the voltage drop depth range, and the harmonic spectrum distribution. Specifically, it includes the following sub-steps: S301: System maintenance load electrical fingerprint database, which is stored in non-volatile memory such as EEPROM or Flash and organized using a key-value pair structure.

[0040] S302: Using the 16-bit target load identifier carried in the operation instruction as the key, retrieve the corresponding value from the load electrical fingerprint database. This value is a structured data body containing the following fields: Load type identifier, with values ​​of "resistive", "inductive" or "switching power supply"; The steady-state current waveform template is represented by a sampling sequence of 256 points. The sampling rate is consistent with that of the electrical signal acquisition module, and the sequence has been normalized in amplitude. The current rise rate threshold during the startup phase, measured in A / s; Voltage drop range, expressed as a percentage of nominal voltage; The main peak frequency band of the harmonic spectrum distribution is recorded in kHz, along with the proportion of energy in this frequency band to the total harmonic energy.

[0041] S303: If a record matching the target load identifier exists in the database, the corresponding baseline electrical characteristic template will be retrieved directly, and the subsequent verification process will begin; if no record is found, that is, the load is being used for the first time, the system will automatically trigger the self-learning program and enter the learning mode.

[0042] S304: After entering learning mode, the system first confirms that the current electrical environment is relatively stable and there are no other concurrent high-power load switching operations.

[0043] The learning window is set to 500ms, which is based on statistics of the transient process of starting common RV loads. This duration can usually capture the main stages of starting current rise and voltage drop.

[0044] In learning mode, the system continuously records the output voltage and current signals within the next 500ms, starting from the timestamp of the current operation command, with a sampling frequency of no less than 2000Hz.

[0045] S305: Based on the recorded signals, automatically extract the electrical characteristic parameters of the load, including: (1) Start-up current rise rate: Within a 500ms recording window, the value at which the current first reaches and remains within a relatively stable range (fluctuation amplitude less than ±5% and duration exceeding 50ms) is identified as the steady-state current value. .

[0046] Calculate the current rise from 0 to Time required The rate of rise of the starting current is The unit is A / s.

[0047] (2) Voltage drop depth ΔV: The original voltage sampling data in the recording window is smoothed and filtered, such as by using a five-point moving average method to suppress high-frequency noise.

[0048] Take the minimum value from the filtered voltage sequence. Calculate its relative to the system nominal voltage. The percentage decrease, i.e.: .

[0049] (3) Harmonic main peak frequency band: Perform fast Fourier transform on the current signal in the recording window and extract the frequency band with the highest energy ratio as the main peak frequency band, with the unit being kHz.

[0050] S306: The system performs a preliminary rationality check on the extracted electrical characteristic parameters.

[0051] The verification includes whether the startup current rise rate is within the preset safety learning range, such as 1A / s to 200A / s, and whether the voltage drop depth exceeds the maximum safety threshold allowed by the system, such as 30%.

[0052] If any parameter fails the verification, the learning session is deemed a failure. The system clears the temporary record data and prompts the user to check the load connection status through the human-machine interface, without performing any subsequent save operations.

[0053] S307: If all extracted electrical characteristic parameters pass the rationality check, the system will display a confirmation prompt to the user through the vehicle's central control screen.

[0054] After user confirmation, the load's identifier and characteristic parameters are fixed into a new record and written into the load electrical fingerprint database.

[0055] S4 involves performing joint time-domain and frequency-domain analysis on the voltage and current signals to extract transient electrical characteristic parameters at the current moment. These parameters include the rate of change of current, voltage fluctuation amplitude, current waveform distortion rate, and dominant harmonic order. Specifically, this includes the following sub-steps: S401: Using the millisecond-level timestamp in the operation command captured in step S1 as the reference origin, extract a signal analysis window with a length of 500ms from the synchronously acquired voltage and current data streams.

[0056] S402: Calculate the rate of change of current (dI / dt) within a 500ms signal analysis window. The specific method is as follows: Let the sampling frequency be If the frequency is not lower than 2000Hz, then the sampling period is... .

[0057] Calculate the difference ΔI between the current values ​​of two adjacent sampling points, and then calculate... .

[0058] Iterate through all adjacent sampling point pairs within the window, and take the maximum value of all calculation results as the representative value of the current change rate for this event.

[0059] S403: Within the same signal analysis window, calculate the voltage fluctuation amplitude, which is the voltage drop depth ΔV. The calculation method is the same as that for ΔV defined in step S305. In both cases, the voltage signal within the window is smoothed and filtered, and the minimum value of the filtered voltage is taken. And calculate using the following formula: .

[0060] S404: Perform amplitude normalization processing on the current signal within the signal analysis window, scale its amplitude to the [-1,1] interval, and then calculate its waveform distortion rate.

[0061] The specific method is as follows: First, extract the fundamental frequency from the actual current signal. and initial phase For example, by using zero-crossing detection or fitting, a frequency of the same can be generated. Same initial phase The ideal sine wave reference signal is obtained; then the normalized actual current waveform is compared with the ideal sine wave point by point, and the mean square error (MSE) between the two is calculated. The MSE value is the current waveform distortion rate, which is used to quantify the degree to which the waveform deviates from the standard sine wave.

[0062] S405: Before performing a fast Fourier transform on the current signal within the signal analysis window, a window function (such as a Hanning window) is applied to the signal to reduce spectral leakage, and then the Fourier transform operation is performed to analyze its spectral components.

[0063] S406: Calculate the total energy of the current signal within the entire signal analysis window from the obtained spectrum. That is, the sum of the squares of the amplitudes of all spectral components, and then find all harmonics (with the fundamental frequency as the base frequency). In integer multiples of the component, its own energy accounts for Harmonic components exceeding 5% in proportion.

[0064] Next, among these harmonics that meet the criteria, the highest harmonic order is identified, such as the 3rd, 5th, and 7th harmonics, and this is determined as the dominant harmonic order.

[0065] S407: The four parameters extracted in the above steps—current change rate, voltage fluctuation amplitude, current waveform distortion rate, and dominant harmonic order—are encapsulated to form a multidimensional transient electrical feature vector.

[0066] S5: Perform a spatiotemporal consistency check between the transient electrical characteristic parameters and the baseline electrical characteristic template. If the two meet the characteristic matching conditions within a preset time window, the current event is determined to be a legitimate load start-up, and the fault protection mechanism is not triggered. If the characteristic matching conditions are not met, the fault risk assessment process begins. Specifically, this includes the following sub-steps: S501: Using the timestamp of the operation command as the starting point for verification, define a forward time window with a duration of 500ms. The range of this window is consistent with the signal analysis window used to extract transient features in step S4, ensuring that the comparison data are completely corresponding in the time dimension.

[0067] S502: Dynamically select preset comparison rules and tolerance ranges based on the load type identifier obtained from the reference electrical characteristic template, such as resistive, inductive, or switching power supply.

[0068] The preset comparison rules and tolerance ranges are stored in the load electrical fingerprint database as part of the baseline electrical feature template. Their initial values ​​are derived from statistical analysis of measured data of typical RV loads, simulation analysis, or compliance with relevant vehicle electrical safety standards. Before actual deployment or through the maintenance interface, the tolerance range can be calibrated and fine-tuned according to the power system parameters of specific vehicle models, such as battery internal resistance and bus capacitance, to optimize verification accuracy.

[0069] S503: Compare the transient electrical characteristic parameters (current change rate, voltage fluctuation amplitude, current waveform distortion rate, dominant harmonic order) extracted in step S4 with the thresholds or ranges of the corresponding fields in the reference electrical characteristic template retrieved in step S3. The specific comparison rules are as follows: For resistive loads: the current change rate should not exceed 60A / s, the voltage drop depth should not exceed 12%, the current waveform distortion rate should be less than 5%, and the dominant harmonic order should not exceed 3.

[0070] For inductive loads: the current change rate is required to be no more than 40A / s, the voltage drop depth is required to be no more than 15%, and the current phase lag needs to be checked by calculating the time difference between the voltage and current zero crossing points and converting it into a phase angle, with a lag of no more than 30 degrees allowed.

[0071] For switching power supply type loads: the main peak of the harmonic spectrum distribution should be located in the 2kHz to 15kHz frequency band, and the current change rate should not exceed 80A / s.

[0072] S504: Execute branch judgment based on the comparison results: If all characteristic parameters fall within the tolerance range of their corresponding type, the spatiotemporal consistency check is deemed successful. The system recognizes the current event as a load start-up caused by a legitimate user operation and does not trigger any fault protection mechanism. Subsequently, the system uses the load identifier as an index to retrieve parameters such as its rated steady-state current from the fingerprint database and switches to the lightweight steady-state operation monitoring mode for the load. In this mode, the system will continuously compare the deviation between the real-time current and the rated steady-state value. If the deviation continues to exceed the preset range, such as ±20%, it is determined to be an operational anomaly, the fault risk assessment process is re-triggered, and the transient analysis data related to this event is cleared.

[0073] If any characteristic parameter exceeds its tolerance range, the verification is deemed a failure. The process then proceeds to the fault risk assessment process.

[0074] S6, in the fault risk assessment process, calculate the similarity score between transient electrical characteristic parameters and various fault modes in the preset fault characteristic library. When any similarity score exceeds the judgment threshold of the corresponding fault type, generate the corresponding fault alarm signal and execute graded protection actions. Specifically, this includes the following sub-steps: S601: The system presets and maintains a fault feature library, which includes four typical fault modes: internal short circuit, insulation failure, grounding fault, and loose connector.

[0075] Each type of fault mode is defined as a four-dimensional feature vector, whose dimensions are composed of the current surge ΔI, the voltage recovery time constant τ, and the zero-sequence current amplitude. and high-frequency noise energy .

[0076] S602: Preset baseline values ​​for the feature vectors of each type of fault mode. and the corresponding feature weights Specifically: (1) The selection of the four-dimensional feature vector is based on the analysis of the physical mechanism of four typical failure modes.

[0077] The current surge ΔI reflects the abnormal current amplitude caused by the fault; the voltage recovery time constant τ reflects the dynamic response speed of the system to power disturbances and is related to the fault impedance characteristics; the zero-sequence current amplitude... Sensitive to asymmetrical faults, such as grounding faults; high-frequency noise energy. It can capture broadband electromagnetic interference caused by electric arcs, loose contacts, etc.

[0078] (2) Feature weights The settings take into account the ability of each feature to distinguish between different failure modes.

[0079] For example, current surge It is given high weight because it has a significant and universal indicative role for high-current faults such as internal short circuits; voltage recovery time constant. and zero-sequence current amplitude This provides supplementary information on system impedance characteristics and grounding path status; high-frequency noise energy As an auxiliary discriminant feature.

[0080] Initial weights can be calibrated based on historical fault data or simulation analysis, and can be fine-tuned and optimized during subsequent system use based on verified fault cases. For example, weights can be set... .

[0081] S603: Based on the 500ms signal analysis window captured in step S4, calculate the four-dimensional feature value corresponding to the current event. The specific calculation method is as follows: (1) Data preprocessing: In order to eliminate the influence of dimensions and stabilize the numerical calculation, the characteristic values ​​calculated from the original signal need to be normalized before comprehensive comparison.

[0082] The reference range for normalization is the same as the reference value of the feature vectors of various fault modes in the fault feature library. The normalization range on which they are based remains consistent. For example, ΔI can be divided by the maximum allowable current of the system, and τ can be divided by a typical upper limit of the time constant.

[0083] (2) Current mutation ΔI: The difference between the maximum current value within the window and the rated steady-state current value retrieved from the fingerprint database based on the load identifier.

[0084] If there is no record for this load in the database, a conservative upper limit based on the total system power estimate is used as a reference.

[0085] (3) Voltage recovery time constant τ: First, locate the time of the lowest voltage point within the signal analysis window. , and then from Begin by taking a segment of the voltage sequence forward until the voltage recovers to a level not lower than [previous value]. Or, until the end of the window, this segment of the sequence is considered a recovery curve.

[0086] For this voltage-time data segment, a fitting formula is obtained using the nonlinear least squares method. The fitted parameter τ is the voltage recovery time constant. If the recovery trend is not obvious or the fitting fails, a larger default value is assigned to τ.

[0087] (4) Zero-sequence current amplitude For a three-phase system, calculate the average of the absolute values ​​of the sum of the instantaneous values ​​of the three-phase currents; for a single-phase system, this characteristic value is set to 0.

[0088] (5) High-frequency noise energy The current signal was decomposed into five levels using the db4 wavelet basis. The nodal coefficients of the sub-bands with frequencies above 10kHz in the fifth level decomposition were extracted, and the sum of the squares of these nodal coefficients was calculated as the high-frequency noise energy. .

[0089] S604: Calculate and preprocess the current event feature vector obtained in step S603. The baseline feature vector of each type of fault mode in the fault feature library Compare them.

[0090] The weighted Euclidean distance algorithm is used to calculate the difference D between the two. The calculation formula is: ; in, The feature weights set in S602.

[0091] Then, the similarity score is calculated: The S value ranges from 0 to 1. The larger the value, the higher the similarity. This function maps distance to similarity. When the distance is 0, the similarity is 1. As the distance increases, the similarity approaches 0.

[0092] S605: Compare the calculated similarity scores S of each fault with its preset judgment threshold.

[0093] The judgment threshold is set based on statistical analysis of historical fault data samples, simulation tests, or by comprehensively considering system safety redundancy and false alarm tolerance.

[0094] For example, in one implementation, the threshold for internal short circuit and insulation failure is set to 0.9, the threshold for grounding fault is set to 0.75, and the threshold for connector loosening is set to 0.6.

[0095] These thresholds can be calibrated via the system maintenance interface to adapt to different application environments.

[0096] S606: Make decisions based on comparison results: If the similarity score of all fault modes is lower than its corresponding threshold, it is determined that no clear fault has been identified, and the system only records the abnormal event log.

[0097] If the score of one and only one type of fault mode exceeds its threshold, it is determined that a fault of that type has occurred, and a corresponding fault alarm signal is immediately generated.

[0098] If the scores of multiple fault modes exceed their thresholds simultaneously, the fault type corresponding to the one with the highest similarity score is selected as the primary criterion. If the highest scores are the same, the fault type with a higher risk level is prioritized, such as internal short circuits over grounding faults, and a corresponding fault alarm signal is generated.

[0099] S607: Based on the type of fault alarm signal and its similarity score range, execute preset graded protection actions. The protection mechanism is divided into three levels: Level 1 Response: Low risk, score between 0.6 and 0.75, such as a loose connector. The system generates a warning log and prompts the user to check via a solid or slow flashing yellow indicator light on the dashboard, without performing a power outage.

[0100] Level 2 Response: Medium risk, score between 0.75 and 0.9, such as a ground fault. The system combines the load identifier or auxiliary branch current detection information that triggered this fault assessment to locate the faulty branch and cut off the power supply to the branch within milliseconds by controlling the corresponding solid-state relay, while maintaining the continued operation of the main circuit and other normal branches.

[0101] Level 3 Response: High risk, triggered when the score for internal short circuit or insulation failure exceeds 0.9. The system immediately disconnects the main DC contactor to isolate the main power supply and issues an audible and visual alarm via a combination of a high-decibel buzzer and a flashing red indicator light.

[0102] Based on the above method, this embodiment also provides a fault detection system for RV energy storage system, including a user operation command capture module, a high sampling rate electrical signal acquisition module, a load fingerprint matching module, a fault risk assessment module, and a graded protection execution module. Each module is connected and communicates through a controller area network CAN bus, an SPI / I2C serial bus, and a high-speed GPIO.

[0103] The user operation command capture module and the high-sampling-rate electrical signal acquisition module serve as two independent data sources for the system, respectively acquiring user behavior context and electrical dynamic signals. The raw data from both is sent to the central processing and synchronization unit, which is responsible for receiving all data, high-precision time synchronization, buffer management, and scheduling the execution of core algorithms such as load fingerprint matching and fault risk assessment. The load fingerprint matching module and fault risk assessment module, as core algorithm units, run as software tasks on the central processing unit. Based on the synchronized data, they execute logical judgments sequentially. The final fault decision (no fault, early warning, branch disconnection, main power disconnection) is received by the hierarchical protection execution module and converted into specific hardware control and alarm actions.

[0104] Specifically: (1) User operation instruction capture module, which is a combination of hardware and software.

[0105] The hardware components include the in-vehicle central control touchscreen, the CAN bus transceiver connected to it, and the scanning circuit for the physical button array.

[0106] The software runs on the central control screen processor or a separate gateway controller. It is responsible for encapsulating the user's touch or button operations into structured data frames. The data frames include the target load identifier, operation type (start / stop), and millisecond-level timestamps taken from the local clock. The encapsulated instruction frames are sent to the central processing and synchronization unit periodically or event-triggered via the CAN bus.

[0107] (2) High sampling rate electrical signal acquisition module, which is an independent hardware sub-board, and its core circuit includes: A closed-loop Hall effect current sensor is used to measure the total current of a DC bus.

[0108] A differential voltage sampling circuit based on an instrumentation amplifier is used to measure the DC bus voltage to ground.

[0109] The synchronous sampling analog-to-digital converter (ADC) synchronously samples and digitizes the aforementioned current and voltage signals at a fixed frequency of not less than 2000Hz.

[0110] The digitized current and voltage data streams are sent to the central processing and synchronization unit in real time in the form of data blocks through a high-speed SPI interface.

[0111] (3) Central processing and synchronization unit, with an embedded microcontroller with a main frequency of 600MHz as the core hardware platform, its key functions and implementations include: Hardware timestamp alignment: Utilizing a high-precision timer within the microcontroller, each pair of synchronized voltage-current sampling data received from the SPI interface is tagged with a precise count value, such as a timestamp; simultaneously, the arrival time of each user operation command frame received from the CAN bus is recorded. By calibrating the clock references of the two data streams, the timestamp deviation between any operation command and its corresponding electrical event is ensured to be controlled within 1 millisecond.

[0112] Data caching and management: An instruction cache queue and a circular buffer for sampled data are allocated in the internal RAM to ensure that no data is lost under high-concurrency operations.

[0113] Task scheduling and computation: Its built-in floating-point arithmetic unit and DSP instruction set are used to efficiently run algorithms such as feature extraction, FFT, wavelet packet decomposition, curve fitting and weighted distance calculation in load fingerprint matching and fault risk assessment.

[0114] (4) Load fingerprint matching module, which is a software task or thread running on the central processing unit.

[0115] It accesses a load electrical fingerprint database stored in an off-chip SPI Flash or EEPROM. This database is organized in a key-value pair structure. When it receives a user start command with a timestamp, the module queries the database using the load identifier in the command as the key.

[0116] If a match is found, the corresponding baseline feature template is obtained, including type, current rise rate threshold, voltage drop range, etc.; if a match is not found, the self-learning subprocess is triggered.

[0117] The core algorithm of this module is to extract transient electrical feature vectors from the synchronous data stream in real time, compare them with the reference template according to rules, and output the judgment result of verification pass or verification failure.

[0118] (5) Fault risk assessment module, which is another software task running on the central processing unit, is activated when the load fingerprint matching module outputs a verification failure.

[0119] This module embeds a fault feature library, which pre-stores standard feature vectors and weights for four fault modes, including internal short circuit and insulation failure. It also receives the same transient feature vectors and uses a weighted Euclidean distance algorithm to calculate the similarity score between the vectors and the various modes in the fault library. Based on preset thresholds and decision rules, it performs fault classification and risk rating, and finally generates a fault alarm signal containing fault type and risk level codes, which is transmitted to the protection execution module through an internal message queue or global variables.

[0120] (6) Hierarchical protection execution module, which is the execution end of the system, includes hardware driver circuit and actuator.

[0121] The drive control interface, with its multiple GPIO and PWM output pins of the microcontroller, is connected to the control terminal of the solid-state relay array and the coil drive circuit of the main DC contactor after optocoupler isolation.

[0122] The actuator array consists of: Solid-state relay array: One is configured for each load branch, which is used to quickly disconnect a specific faulty branch in the second-level response (medium-risk fault) based on the load identifier or auxiliary location information in the fault alarm signal.

[0123] DC high voltage main contactor: rated breaking capacity 200A, coil response time <10ms, used to unconditionally disconnect the main power supply in the event of a level 3 response (high-risk fault).

[0124] Indicator and alarm unit: Connects to other GPIOs of the microcontroller to control the yellow / red LED indicators on the instrument panel (for level 1 and 3 response indication) and a high-decibel buzzer (for level 3 response audible alarm).

[0125] This embodiment achieves accurate differentiation between legitimate load startup and real electrical faults in RV energy storage systems through the aforementioned method and system. Its core lies in using user operation commands as strong priors to guide the system to perform fine-grained comparison of the electrical characteristics of specific loads, rather than relying on globally fixed thresholds. This makes the solution computationally intensive, logically clear, and adaptable to resource-constrained vehicle-mounted embedded platforms, effectively solving the problem of high false alarm rates in existing technologies. At the same time, through a hierarchical protection mechanism, the system's power supply continuity is maintained to the maximum extent while ensuring safety.

[0126] Example 2 Based on the method described in Example 1, this embodiment focuses on enhancing the coordination mechanism between the user operation command capture in step S1 and the spatiotemporal consistency verification in step S5, in order to solve the problems of operation command delay, packet loss or timing disorder that may occur in the actual vehicle network environment, and ensure the stability of dual-channel fusion verification.

[0127] Specifically, after the central processing unit continuously monitors the instruction cache queue in step S105, the process further includes instruction stream status management and user verification steps, as follows: S106: Establish an instruction wait list and an electrical event trigger queue.

[0128] The central processing unit maintains two collaborative data structures: (1) Instruction wait list: used to cache operation instructions that have been issued but have not yet completed consistency verification.

[0129] Each record contains an instruction sequence number, a target load identifier, an issuance timestamp, and a status flag. The system starts a maximum wait timer for each instruction added to this list, for example, with a duration of 300ms. This duration covers the longest estimated time from instruction issuance, CAN bus transmission, to the load's mechanical / electrical response.

[0130] (2) Electrical event buffer: Instead of relying on simple fixed threshold triggering, it continuously runs a lightweight electrical transient monitoring thread, which monitors the trend of total output current changes in real time with low computational overhead.

[0131] When a significant change in current value is detected within a very short time, such as within 2ms, for example, if the absolute value of the change exceeds a threshold value set according to the system noise level, such as 5A, and the change is not caused by a known load that is being verified, the high sampling rate voltage and current raw data streams for a period of time before and after that moment are immediately tagged with an event label, such as -10ms to +100ms, and stored in a cyclically overwritten buffer along with the event trigger timestamp.

[0132] This cache stores raw data, not simple event logs, for possible detailed analysis later.

[0133] S107: Implement bidirectional matching, timeout handling, and user verification logic. The system executes the following logic in parallel: (1) Instruction-driven active matching (forward): When an operation instruction is received by the central processing unit and stored in the instruction waiting list, the system immediately uses the timestamp of the instruction as a reference to search for a cached data segment with a similar timestamp in the electrical event buffer, for example, the difference is within ±50ms.

[0134] If found, the verification process of steps S4 and S5 is triggered immediately based on the instruction; if the verification passes, the instruction is removed from the waiting list and the process ends; if the verification fails or the cached data is not found, the instruction remains in the waiting list, waiting for processing of steps (2) and (3) of S107.

[0135] (2) Event-driven reverse matching: When a new piece of cached data is stored in the electrical event buffer, the system immediately searches the instruction waiting list for an operation instruction that matches the target load identifier (this condition can be relaxed if the load identifier is unknown) and whose instruction issuance timestamp falls within a reasonable window before the event trigger timestamp, such as within the first 200ms. If found, the S4 and S5 processes based on that instruction are also triggered.

[0136] (3) Timeout and unmatched event handling: Instruction timeout: If the waiting time of an instruction in the instruction waiting list exceeds its maximum waiting timer, such as 300ms, and it still fails to be matched and verified by the above S107 (1) and (2), the system determines that the operation corresponding to the instruction has not triggered a typical electrical response that the system can recognize. It may be due to load failure, instruction failure, or extremely soft start device. Therefore, the system removes this instruction from the list, records it in the log, but does not directly trigger fault protection, only provides a status prompt.

[0137] Unmatched event verification: If a segment of data in the electrical event buffer fails to be associated with any operation command through reverse matching within a certain period of time after storage, such as 150ms, the system determines it as an "electrical transient event without a clear command". At this time, the system does not rush to jump into fault assessment, but initiates an active user verification through the human-machine interface, such as popping up a non-blocking prompt on the vehicle's central control screen: "A device has been detected to start in the XX power supply circuit. Did you operate it? [Yes, and select device / No]".

[0138] If the user clicks "Yes" and selects or confirms a specific load device from the list, the system will treat this user confirmation as a virtual delayed instruction with a device identifier and an operation timestamp based on the event trigger timestamp. Subsequently, the system will use this virtual instruction as a basis to directly utilize the original data already saved in the event cache and jump to step S3 to retrieve the fingerprint based on the confirmed device identifier and to the S4 and S5 processes for verification.

[0139] If the user clicks "No" or there is no response within the specified time, the system will immediately use the event trigger timestamp as a reference and directly transfer to the fault risk assessment process in step S6 using cached data.

[0140] This embodiment introduces a dynamic matching mechanism between instructions and events, as well as a user verification process, enabling the system to tolerate brief anomalies in communication channels such as the CAN bus. This effectively prevents legitimate load startups from being misjudged as faults due to instruction transmission issues, greatly improving the reliability and user experience of the dual-channel fusion solution in practical engineering applications.

[0141] Example 3 Based on the method described in Example 1, this embodiment deeply optimizes the construction and retrieval of the load electrical fingerprint database in step S3, upgrading the one-time static learning to continuous dynamic learning and feature evolution tracking, so as to solve the problems that the learning process is easily disturbed and cannot adapt to load aging, and endow the system with early health warning capabilities.

[0142] Specifically, the load electrical fingerprint database construction and update in step S3 adopts the following enhancements: (1) The system maintains a "learning status" flag for each load. When one of the following conditions occurs, the system determines that the load needs to enter the reinforcement learning mode: First-time learning: This load indicator appears for the first time in the operation command, meaning there are no records in the database; Low learning credibility: Although there are records in the database, these records are generated based on a single learning session, and since the learning was completed, the number of start-stop operations for this load has not reached a minimum number of verifications, such as 3 times; Poor consistency verification: If the spatiotemporal consistency verification failure rate of this load exceeds a preset percentage, such as 30%, within a recent period, such as the last 10 times it has been enabled, it indicates that the existing template may be inaccurate or the load characteristics have changed.

[0143] Upon entering reinforcement learning mode, the system will initiate a dedicated learning task for the load. This task aims to collect data from multiple consecutive, normal start-stop operations, with the number of operations preset to 5 to 10, or until at least N "clean" samples are collected. Simultaneously, each recording generates a learning sample containing the original voltage and current waveforms and extracted candidate feature parameters, as in this case... , wait.

[0144] (2) The system extracts key feature parameters, such as ΔV, dI / dt, etc., from multiple learning samples collected for this load. Each sample corresponds to a point in the feature space. For each key feature parameter dimension, a clustering algorithm (such as DBSCAN) is used to detect outliers. Specifically: Parameter settings reference: The neighborhood radius eps and minimum number of points min_samples parameters of the DBSCAN algorithm can be initially set based on the initially collected sample data by observing the distribution range of the feature values ​​or using simple statistical rules, such as setting eps to a certain multiple of the standard deviation of the feature values. They can be fine-tuned in subsequent use. The purpose is to remove outliers that deviate significantly from the main distribution.

[0145] These anomalous samples are marked and removed, which may be caused by accidental interference from other concurrent loads, momentary power fluctuations, or measurement noise; the remaining samples constitute the "clean learning set" for that load.

[0146] Based on the "clean learning set", the statistical values ​​of each key feature parameter are calculated, including the mean μ and the standard deviation σ.

[0147] The baseline electrical characteristic template is constructed as a statistical characteristic template. This template contains at least: a) The nominal value of the feature parameter, usually the mean μ; b) A dynamic tolerance interval, which is not a fixed value, but is calculated based on the statistical distribution of parameters and the engineering safety boundary.

[0148] For example, the template for the voltage drop depth ΔV can be represented as: ; ; in: and It is an interval coefficient. For characteristics where fluctuations are relatively symmetrical and the risk of exceeding limits is mainly focused on the upper limit, such as dI / dt, and The same value can be used, such as 2 or 3; for features like voltage drop depth, where only the lower limit is of concern and the drop cannot be too deep, It can be set to a smaller value, such as 1. The value can be appropriately relaxed, such as to 2.5, to reflect the asymmetry. and These are the absolute lower and upper limits allowed by the system's overall security.

[0149] In this way, the tolerance range can reflect the normal fluctuation range of the load itself, which is more scientific and adaptable than a fixed threshold.

[0150] (3) The system maintains a long-term characteristic record sequence for each registered load.

[0151] Each time the load successfully completes a start-stop operation, regardless of whether it passes the spatiotemporal consistency check, the system adds the steady-state and transient characteristics extracted during this operation, such as the steady-state current value and the actual dI / dt value of this start, as a data point to the sequence.

[0152] Based on this time series, the system uses trend analysis algorithms, such as moving average, exponential smoothing, or linear regression, to establish a simple feature evolution model to describe the potential changing trends of the key features of the load over time or the number of times it is used.

[0153] Model implementation example: The system can maintain a fixed-length first-in-first-out queue for each tracked feature, such as the rate of rise of the starting current. For example, it can store the 100 most recent valid measurements and add new feature values ​​to the queue after each load operation.

[0154] Trend analysis: Periodically, such as every 10 new points or after each new point, a weighted moving average or a single exponential smoothing algorithm is applied to the data in the queue. The smoothing coefficient α of the exponential smoothing can be selected according to the stationarity of the data. For example, it can be initially set to 0.1 to focus on the long-term trend. Then, by comparing the current value with the smoothed predicted value, the trend residual can be calculated.

[0155] Model output: The model output can be a "baseline value", which is the expected value under the current trend; and a "normal fluctuation band", which is calculated based on the standard deviation of historical residuals.

[0156] (4) The system shall periodically, for example, after every 100 successful runs, or when the trend residuals of the feature evolution model remain abnormal, automatically perform the following evaluation and operations: a) Template update decision: The system compares the long-term statistical mean of the current feature parameters, such as the mean of the last 100 times, with the nominal template value μ stored in the database.

[0157] If the relative change of a key feature parameter exceeds the preset "template update threshold", such as 5%, and the change is determined to be trend-based rather than random fluctuation, such as the change in multiple consecutive calculations having the same sign, the system can prompt the user "XX load characteristics have changed, it is recommended to update the template", or automatically recalculate the statistical template with recent data and update the database after user authorization.

[0158] b) Early warning triggering: During each load run verification or steady-state monitoring, the system not only checks whether the current characteristic value is within the dynamic tolerance range. It will also compare this value with the “baseline value” predicted by the feature evolution model.

[0159] If the current value deviates from the baseline value by more than the "warning threshold" set based on the historical "normal fluctuation zone" of the model, for example, the deviation is greater than twice the historical standard deviation, but has not yet exceeded the dynamic tolerance range, the system generates a device health warning message.

[0160] For example, "The starting current characteristic of the air conditioner compressor has recently shown a continuous upward trend (deviating from the expected value by X%). It is recommended to pay attention to its operating status." This warning is displayed through the human-machine interface and is different from a fault alarm. It aims to achieve preventive maintenance.

[0161] This embodiment effectively eliminates random interference that may be encountered in a single learning cycle through multi-cycle learning and cluster cleaning, obtaining a more reliable and representative load electrical fingerprint. Furthermore, the statistically based dynamic tolerance interval is more scientific than a fixed threshold, improving the adaptability of the comparison.

[0162] Most importantly, by establishing a feature evolution model, the system can be made capable of sensing the slow aging of the load and the degradation of performance, thus achieving a leap from post-fault protection to pre-fault early warning, and improving the system's intelligence level and maintenance value.

[0163] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0164] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment includes only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for fault detection in a motorhome energy storage system, characterized in that, include: Capture user-issued load start / stop commands in real time. The commands include the target load identifier, operation timestamp, and operation type. The electrical signals at the output of the energy storage system are collected synchronously. These electrical signals include voltage and current signals. The collected voltage and current signals are then time-stamped to align the operation commands with the electrical signals on the time axis. Based on the target load identifier, the corresponding baseline electrical feature template is retrieved from the pre-set load electrical fingerprint database. The baseline electrical feature template includes the range or threshold of electrical feature parameters related to the load type. Using the operation timestamp as the reference origin, extract transient electrical characteristic parameters within a preset time window from the aligned electrical signal; The extracted transient electrical feature parameters are compared with the reference electrical feature template for spatiotemporal consistency verification. If the verification passes, the system is considered to have started a legitimate load; if the verification fails, the system proceeds to the fault risk assessment process. In the fault risk assessment process, the similarity between transient electrical characteristic parameters and feature vectors of various fault modes in the preset fault feature library is calculated. Based on the comparison results of similarity scores and preset thresholds, corresponding fault alarm signals are generated and graded protection actions are executed.

2. The method for fault detection of a motorhome energy storage system according to claim 1, characterized in that, Real-time capture of user-issued load start / stop commands, specifically including: User operations are received via in-vehicle central control touchscreen, wireless mobile terminal application interface, or physical button array. User operations are encapsulated into structured instruction data frames that include operation type, target payload identifier, and millisecond-level timestamp; The instruction data frame is transmitted to the central processing unit via the controller area network bus and stored in the instruction cache queue.

3. The method for fault detection of a motorhome energy storage system according to claim 1, characterized in that, The voltage and current signals at the output of the energy storage system are collected synchronously, with a sampling frequency of not less than 2000Hz; Furthermore, a high-precision timer is used to add a millisecond-level time stamp to each set of synchronously acquired voltage and current sampling data pairs to ensure that the deviation between the data and the timestamp of the operation command does not exceed a preset value.

4. The method for fault detection of a motorhome energy storage system according to claim 1, characterized in that, The construction and updating of the load electrical fingerprint database includes a learning mode, which is initiated and executed upon the first identification of a target load identifier or when trigger conditions are met: Under stable electrical conditions, record the voltage and current signals within a preset time after the load starts; The starting electrical characteristic parameters of the load, including the rate of rise of starting current and the depth of voltage sag, are automatically extracted based on the recorded signals. The extracted feature parameters are validated for reasonableness, and after user confirmation, the target load identifier and its feature parameters are stored as a new record in the load electrical fingerprint database.

5. The method for fault detection of a motorhome energy storage system according to claim 4, characterized in that, The update of the load electrical fingerprint database further includes enhanced learning modes and dynamic optimization: The load maintains a learning state and triggers the enhanced learning mode when learning for the first time, when the learning confidence is low, or when the verification consistency is poor, in order to collect multiple consecutive normal start-stop operation data samples. Cluster analysis was performed on the collected samples to remove outliers, and the statistical mean and standard deviation of key feature parameters were calculated based on the remaining clean samples. Based on the statistical mean and standard deviation, and combined with the engineering safety boundary, a statistical feature template containing a dynamic tolerance interval is constructed to update the baseline electrical feature template.

6. The method for fault detection of a motorhome energy storage system according to claim 5, characterized in that, This further includes load health status alert steps: A long-term record sequence of key characteristic parameters of the registered load that changes over time; A feature evolution model is established based on the long-term recorded sequence to obtain the baseline values ​​and normal fluctuation bands of the feature parameters; During load operation, the real-time extracted feature parameters are compared with the baseline value. If the deviation exceeds the warning threshold set based on the normal fluctuation range but does not exceed the dynamic tolerance range, a device health warning message is generated.

7. The method for fault detection of a motorhome energy storage system according to claim 1, characterized in that, The spatiotemporal consistency verification specifically includes: A forward time window is defined starting from the operation timestamp, and the comparison rules and tolerance ranges corresponding to the load type are obtained from the reference electrical feature template. The extracted transient electrical characteristic parameters, including current rate of change, voltage drop depth, current waveform distortion rate, and dominant harmonic order, are compared item by item with the tolerance range. If all feature parameters fall within the tolerance range of their corresponding type, the verification is considered successful.

8. The method for fault detection of a motorhome energy storage system according to claim 1, characterized in that, The fault risk assessment process specifically includes: A fault feature library containing multiple fault modes is pre-defined, and each fault mode is defined by a multi-dimensional feature vector and associated with feature weights and judgment thresholds; Calculate the multidimensional feature value corresponding to the current event based on the transient electrical characteristic parameters; The weighted Euclidean distance algorithm is used to calculate the similarity score between the current event feature vector and the baseline feature vectors of various fault modes in the fault feature library; The similarity score is compared with the corresponding fault mode determination threshold, and the fault type and risk level are determined based on the comparison result.

9. The method for fault detection of a motorhome energy storage system according to claim 1, characterized in that, It also includes a coordination mechanism for handling timing anomalies between operating instructions and electrical events: The maintenance instruction waiting list and the electrical event buffer are used to cache operation instructions to be verified and raw data segments triggered by electrical transient monitoring, respectively. Parallel execution of instruction-driven active matching and event-driven reverse matching logic to associate operation instructions with electrical events; For unmatched instructions that time out or electrical events that are not associated with instructions, interactive confirmation is performed through the user verification interface, and subsequent verification is executed or the process is directly transferred to the fault risk assessment process based on user feedback.

10. A fault detection system for a motorhome energy storage system, used to implement the fault detection method for a motorhome energy storage system as described in any one of claims 1 to 9, characterized in that, include: The user operation command capture module is used to capture and encapsulate load start and stop operation commands containing target load identifier, operation type and millisecond-level timestamp in real time; A high sampling rate electrical signal acquisition module is used to synchronously acquire voltage and current signals at the output of the energy storage system; The central processing and synchronization unit is used to receive the operation instructions and electrical signals, perform high-precision time synchronization and alignment on them, and schedule the operation of the core algorithm. The load fingerprint matching module is used to query or construct a load electrical fingerprint database based on the target load identifier in the operation instruction, and to perform spatiotemporal consistency verification. The fault risk assessment module is used to calculate the similarity between the current event and the preset fault feature library when the spatiotemporal consistency check fails, and to classify the fault and determine the risk level. The graded protection execution module is used to execute corresponding alarm and power-off protection actions based on the output results of the fault risk assessment module.

Images