Fault early warning method for electricity storage equipment of container

Abstract

The container electricity storage equipment fault early-stage intelligent early warning method comprises the steps that dynamic load fluctuation characteristics are extracted based on load current data flow; extracting state response characteristics of the core electrical component based on the multi-dimensional state signal data stream; establishing a real-time dynamic association model of dynamic load fluctuation characteristics and core electrical component state response characteristics; when the dynamic load fluctuation characteristic exceeds a preset fluctuation threshold value, triggering a first-level early warning representing that the overall operation state is abnormal; and when the state response characteristics of the core electrical component deviate from an expected response range output by the real-time dynamic association model based on the current dynamic load fluctuation characteristics, triggering secondary positioning early warning representing abnormal response of the specific core electrical component. According to the method, the fault early-stage identification precision and the grading disposal efficiency are remarkably improved, and safe and stable operation of the power storage equipment is guaranteed.

Description

A method for early warning of faults in containerized energy storage equipment Technical Field

[0001] This invention relates to the field of energy storage, and more particularly to a method for early warning of faults in containerized energy storage equipment. Background Technology

[0002] Current fault early warning methods for containerized energy storage equipment have significant drawbacks: First, data acquisition relies on a single dimension, depending on isolated parameters such as current or temperature, failing to comprehensively capture multi-dimensional signs of degradation in core electrical components. Second, the lack of dynamic correlation between load fluctuations and component responses leads to a high rate of false alarms. Third, the early warning mechanism lacks a tiered strategy, causing confusion between overall anomalies and component-level faults, triggering protection commands and wasting maintenance resources. These shortcomings make it difficult for existing systems to achieve accurate early warnings, severely restricting the safety and reliability of energy storage equipment. Summary of the Invention

[0003] This invention proposes an early intelligent warning method for containerized energy storage equipment failures, comprising: S1, real-time acquisition of load current data streams and multi-dimensional status signal data streams of core electrical components during the charging and discharging process of the energy storage equipment; S2, extraction of dynamic load fluctuation characteristics based on the load current data streams; extraction of status response characteristics of core electrical components based on the multi-dimensional status signal data streams; S3, establishment of a real-time dynamic correlation model between dynamic load fluctuation characteristics and status response characteristics of core electrical components based on historical operating data, wherein the model is used to describe the expected change law of status response characteristics of core electrical components under changes in dynamic load fluctuation characteristics; S4, triggering a first-level warning indicating an abnormal overall operating state when the dynamic load fluctuation characteristics exceed a preset fluctuation threshold; triggering a second-level location warning indicating an abnormal response of a specific core electrical component when the status response characteristics of the core electrical components deviate from the expected response range output by the real-time dynamic correlation model based on the current dynamic load fluctuation characteristics.

[0004] The step of extracting dynamic load fluctuation characteristics based on the load current data stream includes: calculating the statistical distribution characteristics and time series change trend of the load current data stream to quantify the load fluctuation amplitude and frequency characteristics.

[0005] The step of extracting the state response features of core electrical components based on the multi-dimensional state signal data stream includes: separating feature components from each dimension of the state signal data stream and integrating the feature components to form a multi-dimensional state response index.

[0006] The establishment of the real-time dynamic correlation model includes: using historical operating data to train a time-series prediction mechanism to generate a continuous mapping relationship of the state response characteristics of core electrical components under dynamic load fluctuation characteristics.

[0007] The triggering of the first-level early warning that characterizes the overall abnormal operating status includes: outputting a first-level early warning signal; the triggering of the second-level location early warning that characterizes the abnormal response of a specific core electrical component includes: outputting a second-level early warning indication containing the identifier of the specific core electrical component.

[0008] This also includes periodically updating model parameters after the real-time dynamic correlation model is established to reflect the cumulative changes in the running data.

[0009] This also includes storing load current data streams and multi-dimensional status signal data streams in a local cache unit.

[0010] The present invention has the following beneficial effects: First, the distributed sensor network combined with the component-signal precise mapping mechanism enables multi-dimensional collaborative monitoring of the status of core components such as energy storage converters and battery management systems, and comprehensively covers key degradation characteristics such as temperature field and voltage ripple.

[0011] Second, based on the real-time correlation modeling of dynamic load fluctuation characteristics and multi-dimensional state response indicators, a time-series prediction architecture is used to capture the nonlinear hysteresis relationship between load mutations and state response, and to establish an accurate expected response range.

[0012] Third, a two-tiered early warning decision-making mechanism is used. The first-tier early warning provides overall protection against anomalies; the second-tier early warning uses a residual contribution ranking table to locate faulty components, enabling optimized resource allocation and rapid maintenance response.

[0013] This method significantly improves the accuracy of early fault identification and the efficiency of graded handling, ensuring the safe and stable operation of energy storage equipment. Attached Figure Description

[0014] Figure 1 is a flowchart illustrating a fault early warning method for containerized energy storage equipment proposed in this invention. Detailed Implementation

[0015] Referring to Figure 1, the present invention proposes a flowchart of a fault early warning method for containerized energy storage equipment, including: S1, real-time acquisition of load current data stream and multi-dimensional status signal data stream of core electrical components during the charging and discharging process of the energy storage equipment.

[0016] The collected load current data stream and multi-dimensional status signal data stream can be stored in a local cache unit. The local cache unit adopts a circular queue structure, which can cover the complete charge and discharge cycle data and solve the storage bottleneck of edge computing nodes.

[0017] The data acquisition uses a distributed sensor network, and the sampling frequency is adaptive to the load change rate, which can solve the problem of insufficient sampling efficiency and improve the integrity of data acquisition.

[0018] Core electrical components, including but not limited to energy storage converters, battery management systems, thermal management units, and contactor groups, can address the issue of overlooked monitoring of critical components, thereby promoting safe equipment operation.

[0019] Multi-dimensional condition signal data streams include, but are not limited to, temperature gradient distribution, voltage ripple spectrum, insulation impedance spectrum, mechanical vibration spectrum, and gas concentration time series. This can solve the problem of single-dimensional condition monitoring, thereby improving the comprehensiveness of fault early warning.

[0020] The temperature gradient distribution is measured using an infrared thermal imager or a multi-point temperature sensor array. The voltage ripple spectrum is measured using a high-precision voltage sensor combined with FFT analysis. The insulation impedance spectrum is measured using an insulation resistance tester or an LCR bridge. The mechanical vibration spectrum is measured using vibration sensors such as accelerometers. The gas concentration time series is measured using gas sensors (such as H2 and CO sensors).

[0021] Furthermore, the energy storage converter is associated with voltage ripple spectrum and temperature gradient distribution, the battery management system is associated with insulation impedance spectrum and gas concentration time series, the thermal management unit is associated with temperature gradient distribution, and the contactor group is associated with mechanical vibration spectrum.

[0022] S2. Extract dynamic load fluctuation characteristics based on the load current data stream; extract the state response characteristics of core electrical components based on the multi-dimensional state signal data stream.

[0023] Feature extraction is achieved through a parallel computing architecture, and the timestamps of load features and status response features are strictly aligned, which can solve the data processing delay problem and improve the real-time performance of early warning.

[0024] Extracting dynamic load fluctuation features based on the load current data stream specifically includes: calculating the statistical distribution characteristics and time series change trends of the load current data stream to quantify the load fluctuation amplitude and frequency characteristics, which can solve the problem of fuzzy characterization of load fluctuations and improve the accuracy of anomaly detection.

[0025] Among them, the statistical moments corresponding to the statistical distribution characteristics are calculated by using a sliding window. The statistical moments specifically include the mean, variance, and skewness. Combined with the FFT spectral entropy value, the degree of fluctuation chaos can be quantified, which can solve the problem of high-frequency noise interference and be used for stable extraction of fluctuation features.

[0026] Furthermore, the dynamic load fluctuation characteristics specifically include the current change rate and the amplitude of the resonant frequency component during load step changes, which can solve the transient load response problem and promote early fault warning.

[0027] Extracting the state response features of core electrical components based on the multi-dimensional state signal data stream includes: separating feature components from each dimension of the state signal data stream and integrating the feature components to form a multi-dimensional state response index, which can solve the problem of multi-dimensional signal heterogeneity and improve the consistency of state response.

[0028] In the above, the feature component separation adopts wavelet packet decomposition, and the integration of indicators is constructed by weighting the Euclidean space state vector through entropy weighting, which can solve the feature redundancy problem and be used for efficient fusion of response indicators.

[0029] Among them, the multidimensional state response indicators are: temperature field uniformity coefficient, voltage harmonic distortion rate, insulation aging factor, and mechanical resonant mode energy ratio, which can solve the problem of state indicator dispersion and improve the accuracy of anomaly location.

[0030] S3. Based on historical operating data, establish a real-time dynamic correlation model between dynamic load fluctuation characteristics and the state response characteristics of core electrical components. The model is used to describe the expected change law of the state response characteristics of core electrical components under the change of dynamic load fluctuation characteristics, which can solve the problem of missing load-response correlation and improve the reliability of fault prediction.

[0031] Furthermore, the above model can incorporate an attention mechanism during training and capture the nonlinear lag correlation between load mutation points and state responses, which can solve the problem of ignoring temporal dependencies and promote the model's generalization ability.

[0032] The establishment of the real-time dynamic correlation model includes: applying historical operating data to train the time-series prediction mechanism to generate a continuous mapping relationship of the state response characteristics of core electrical components under dynamic load fluctuation characteristics, which can solve the problem of discontinuous mapping relationship and be used for accurate description of response patterns.

[0033] The aforementioned time series prediction mechanism can adopt the LSTM-Encoder architecture, and the mapping relationship is specifically expressed as a state transition tensor, which can solve the problem of long-term time series prediction bias and improve the model's computational efficiency.

[0034] The aforementioned time-series prediction mechanism is used to output the probability distribution range of state response characteristics within a certain future period based on the time-series segments of load characteristics. This can solve the problem of response uncertainty and be used for dynamic setting of early warning thresholds.

[0035] After the real-time dynamic correlation model is established, it also includes periodically updating the model parameters to reflect the cumulative changes in the running data, which can solve the model aging problem and improve adaptive performance.

[0036] The above parameters are updated using online incremental learning. Model drift is detected by KL divergence and can be triggered for recalibration, which can solve the problem of ignoring data drift and promote real-time model calibration.

[0037] S4. When the dynamic load fluctuation characteristics exceed the preset fluctuation threshold, a first-level warning is triggered to indicate an abnormal overall operating status; when the status response characteristics of the core electrical components deviate from the expected response range output by the real-time dynamic correlation model based on the current dynamic load fluctuation characteristics, a second-level location warning is triggered to indicate an abnormal response of a specific core electrical component. This can solve the problem of confusion in warning levels and improve the efficiency of fault classification and processing.

[0038] Among them, the above-mentioned early warning trigger adopts a two-level decision-making mechanism. The first-level early warning activates the global protection strategy, and the second-level early warning initiates component-level diagnosis. The beneficial effects that can be directly inferred from the technical features of this section (within 30 characters) can solve the problem of single protection strategy and be used for resource optimization allocation.

[0039] The triggering of the first-level early warning that represents the abnormal overall operating status includes: outputting a first-level early warning signal, which is encoded as a JSON structure containing a fluctuation exceeding the limit timestamp, threshold boundary value, and emergency load reduction instruction. This can solve the problem of ambiguous early warning information and improve the accuracy of instruction execution.

[0040] The triggering of the secondary location warning that characterizes the abnormal response of a specific core electrical component includes: outputting a secondary warning indication containing the identifier of the specific core electrical component.

[0041] Among them, the component identifier is associated with the physical topology coordinates, and the warning indication is supplemented with a residual contribution ranking table, which can solve the problem of fault location delay and promote maintenance response speed.

[0042] Example of expected response range: When the load fluctuation characteristic is [amplitude 120A@100Hz], the expected temperature rise response range of the converter is [45±2℃]. The measured value > 50℃ triggers a level 2 warning.

[0043] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for early intelligent warning of faults in containerized energy storage equipment, characterized in that, include: S1. Real-time acquisition of load current data stream and multi-dimensional status signal data stream of core electrical components during the charging and discharging process of energy storage equipment; S2. Extract dynamic load fluctuation characteristics based on the load current data stream; S3. Extract the state response features of core electrical components based on the multi-dimensional state signal data stream; S4. Based on historical operating data, establish a real-time dynamic correlation model between dynamic load fluctuation features and state response features of core electrical components. The model is used to describe the expected change pattern of state response features of core electrical components under changes in dynamic load fluctuation features; S5. When the dynamic load fluctuation features exceed the preset fluctuation threshold, trigger a first-level early warning that represents an abnormal overall operating state. When the state response characteristics of the core electrical component deviate from the expected response range output by the real-time dynamic correlation model based on the current dynamic load fluctuation characteristics, a secondary positioning warning characterizing the abnormal response of the specific core electrical component is triggered.

2. The method for early intelligent warning of containerized energy storage equipment failure according to claim 1, characterized in that, The extraction of dynamic load fluctuation characteristics based on the load current data stream includes: calculating the statistical distribution characteristics and time series change trend of the load current data stream to quantify the load fluctuation amplitude and frequency characteristics.

3. The method for early intelligent warning of containerized energy storage equipment failure according to claim 1, characterized in that, The extraction of core electrical component state response features based on the multi-dimensional state signal data stream includes: separating feature components from each dimension of the state signal data stream and integrating the feature components to form a multi-dimensional state response index.

4. The method for early intelligent warning of containerized energy storage equipment failure according to claim 1, characterized in that, The establishment of the real-time dynamic correlation model includes: using historical operating data to train a time-series prediction mechanism to generate a continuous mapping relationship of the state response characteristics of core electrical components under dynamic load fluctuation characteristics.

5. The method for early intelligent warning of containerized energy storage equipment failure according to claim 1, characterized in that, The triggering of the first-level early warning that characterizes the overall abnormal operating status includes: outputting a first-level early warning signal; the triggering of the second-level location early warning that characterizes the abnormal response of a specific core electrical component includes: outputting a second-level early warning indication containing the identifier of the specific core electrical component.

6. The method for early intelligent warning of containerized energy storage equipment failure according to claim 1, characterized in that, Also includes: After the real-time dynamic correlation model is established, the model parameters are periodically updated to reflect the cumulative changes in the running data.

7. The method for early intelligent warning of containerized energy storage equipment failure according to claim 1, characterized in that, Also includes: The load current data stream and multi-dimensional status signal data stream are stored in the local cache unit.

Images