Lightweight edge computing equipment fault diagnosis system and method

Abstract

The invention discloses an equipment fault diagnosis system and method for lightweight edge calculation, and relates to the technical field of equipment fault diagnosis, and the method comprises an operation scene modeling module which is used for carrying out three-dimensional scene coupling modeling based on the power grid structure type, the power grid load type and the power grid peak regulation type of a power grid where an energy storage battery is located, constructing a power operation scene library; the monitoring index configuration module is used for acquiring a plurality of high-frequency monitoring index sets based on historical cell capacity attenuation fault records of the same type of energy storage batteries; the scene prediction mapping module is used for respectively constructing a plurality of attenuation fault prediction units in the power grid cloud server and mapping and constructing a scene-prediction unit call table; and the preliminary fault diagnosis module is used for deploying the scene-prediction unit call table to the lightweight edge device of the energy storage battery to perform preliminary fault diagnosis of the capacity attenuation of the battery cell. According to the invention, the technical problems of inaccurate fault diagnosis and large calculation power consumption of the energy storage battery in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment fault diagnosis, in particular to a lightweight edge computing equipment fault diagnosis system and method. BACKGROUND

[0002] In the field of energy storage battery equipment fault diagnosis, the existing technology usually relies on a fixed threshold monitoring strategy. Although it can achieve basic fault warning, it has significant limitations: first, the capacity attenuation of the energy storage battery is affected by the dynamic operation of the power grid, and the traditional diagnosis system lacks consideration of the scene, resulting in a disconnection between the monitoring indicators and the scene; second, the general monitoring scheme often uses global indicator collection, causing redundant data transmission and storage pressure on the edge side, and increasing the computational burden of resource-constrained devices. Due to the lack of scene awareness and rigid resource allocation, the existing technology cannot achieve accurate early warning of fault signals, and it is difficult to balance the computational efficiency, ultimately restricting the timeliness of equipment fault diagnosis. SUMMARY

[0003] The present application provides a lightweight edge computing equipment fault diagnosis system and method, which is used to solve the technical problems of inaccurate energy storage battery fault diagnosis and large computational consumption in the prior art.

[0004] In view of the above problems, the present application provides a lightweight edge computing equipment fault diagnosis system and method.

[0005] In the first aspect, the present application provides a lightweight edge computing equipment fault diagnosis system, which comprises:

[0006] A running scene modeling module is configured to perform three-dimensional scene coupling modeling based on the power grid structure type, power grid load type and power grid peak shaving type of the power grid where the energy storage battery is located, and to construct a power operation scene library, wherein the power operation scene library comprises a plurality of power operation scenes.

[0007] A monitoring indicator configuration module is configured to configure high-frequency monitoring indicators for the plurality of power operation scenes based on historical cell capacity attenuation fault records of similar energy storage batteries, and to obtain a plurality of high-frequency monitoring indicator sets.

[0008] A scene prediction mapping module is configured to construct a plurality of attenuation fault prediction units in the power grid cloud server according to the plurality of high-frequency monitoring indicator sets, and to construct a scene-prediction unit call table by combining the plurality of power operation scenes.

[0009] A preliminary fault diagnosis module is configured to deploy the scene-prediction unit call table to the lightweight edge device of the energy storage battery, and to perform preliminary fault diagnosis of cell capacity attenuation.

[0010] In a second aspect, the application provides a device fault diagnosis method for lightweight edge computing, comprising:

[0011] Three-dimensional scene coupling modeling is performed based on the power grid structure type, power grid load type and power grid peak regulation type of the power grid in which the energy storage battery is located, and a power operation scene library is constructed, wherein the power operation scene library comprises a plurality of power operation scenes;

[0012] Based on historical cell capacity attenuation fault records of the same type of energy storage battery, a plurality of high-frequency monitoring index sets are configured for the plurality of power operation scenes respectively;

[0013] In the power grid cloud server, a plurality of attenuation fault prediction units are constructed according to the plurality of high-frequency monitoring index sets respectively, and a scene-prediction unit calling table is constructed by combining the plurality of power operation scenes;

[0014] The scene-prediction unit calling table is deployed on the lightweight edge device of the energy storage battery to perform preliminary fault diagnosis of cell capacity attenuation.

[0015] One or more technical solutions provided in the application have at least the following technical effects or advantages:

[0016] The application provides a device fault diagnosis system and method for lightweight edge computing. The power operation scene library is dynamically generated through three-dimensional scene coupling modeling, and the high-frequency monitoring index set and the prediction unit are adaptively configured based on the scene characteristics, which significantly improves the accuracy and resource utilization efficiency of the energy storage battery fault diagnosis. Compared with the traditional method, the technical solution provided by the application significantly overcomes the contradiction between diagnosis accuracy and calculation load in the limited resource scene, and achieves the technical effects of simultaneously improving fault early warning capability, scene adaptability and system real-time in limited resources. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A structural schematic diagram of a device fault diagnosis system for lightweight edge computing provided by the embodiments of the application.

[0019] Figure 2 A flowchart of a device fault diagnosis method for lightweight edge computing provided by the embodiments of the application.

[0020] The components represented by each number in the attached diagram are explained below:

[0021] The system includes a scenario modeling module 100, a monitoring indicator configuration module 200, a scenario prediction mapping module 300, and a preliminary fault diagnosis module 400. Detailed Implementation

[0022] This application provides a lightweight edge computing device fault diagnosis system and method to address the technical problems of inaccurate fault diagnosis and high computing power consumption in the prior art for energy storage batteries.

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0024] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0025] Example 1, as Figure 1 As shown, this application provides a lightweight edge computing device fault diagnosis system, wherein the system includes:

[0026] The operation scenario modeling module 100 is used to perform three-dimensional scene coupling modeling based on the grid structure type, grid load type, and grid peak-shaving type of the power grid where the energy storage battery is located, and to construct a power operation scenario library. The power operation scenario library includes several power operation scenarios, including:

[0027] The grid structure type of the power grid where the energy storage battery is located is collected, wherein the grid structure type includes at least pure thermal power supply, centralized photovoltaic access, centralized wind power access, wind-solar-thermal multi-energy complementarity, distributed photovoltaic access, microgrid / island operation, and integrated source-load-storage;

[0028] The grid load type of the power grid where the energy storage battery is located is collected, wherein the grid load type includes at least peak load, off-peak load, stable load, fluctuating load and peak load;

[0029] The peak-shaving type of the power grid where the energy storage battery is located is collected, wherein the peak-shaving type includes at least high-frequency peak-shaving, medium-frequency peak-shaving, low-frequency peak-shaving and no obvious peak-shaving;

[0030] Randomly select any parameter in the power grid structure type, power grid load type and power grid peak regulation type to perform three-dimensional scene coupling modeling, generate a first power operation scene, and sequentially perform coupling modeling to obtain a plurality of power operation scenes, thereby constructing a power operation scene library.

[0031] The existing energy storage battery fault diagnosis technology does not consider the dynamic coupling effect of the power grid operation scene, and only divides the working conditions in a single dimension, resulting in fragmented scene feature extraction. Due to the interweaving effect of factors such as power grid structure type, load fluctuation, frequency modulation mode and the like on the capacity attenuation of the battery cell, the traditional method cannot construct a power scene library that accurately describes the real complex working conditions due to the lack of multi-parameter coupling modeling, which causes the subsequent monitoring strategy to be mismatched with the scene characteristics, and it is difficult to capture the differentiated evolution law of the attenuation fault.

[0032] In the embodiment of the application, the power grid structure type of the power grid in which the energy storage battery is located is collected according to the design structure of the power grid, wherein the power grid structure type at least includes pure thermal power supply, centralized photovoltaic access, centralized wind power access, wind-solar-thermal multi-energy complementation, distributed photovoltaic access, microgrid / island operation and source-load-storage integration.

[0033] The power grid load type of the power grid in which the energy storage battery is located is collected by using a power meter, wherein the power grid load type at least includes peak load (power ≥ 90% of rated power for 6 hours), valley load (power ≤ 40% of rated power for 6 hours), stable load (the ratio of the maximum and minimum power fluctuations to the average power ≤ 0.1), fluctuating load (the ratio of the maximum and minimum power fluctuations to the average power > 0.1), and sharp peak load (power fluctuation exceeds a preset fluctuation threshold, such as power fluctuation ≥ 40% of rated power).

[0034] The number of frequency modulation instruction triggers in a unit time is counted to collect the power grid peak regulation type of the power grid in which the energy storage battery is located, wherein the power grid peak regulation type at least includes high-frequency peak regulation (instruction interval ≤ 10 minutes), medium-frequency peak regulation (instruction interval between 10 and 30 minutes), low-frequency peak regulation (instruction interval ≥ 30 minutes) and no obvious peak regulation (no instruction).

[0035] In the power grid structure type, power grid load type and power grid peak regulation type, one parameter of each type is randomly selected to perform three-dimensional scene coupling modeling to generate a first power operation scene, sequentially perform coupling modeling to obtain a plurality of power operation scenes, construct a power operation scene library, and upload to a power grid cloud server.

[0036] Through coupling modeling of the power grid structure type, load type and peak regulation type, a power operation scene library covering multi-dimensional interaction characteristics is constructed, solving the problem of one-sided scene representation. This library integrates fragmented working conditions into systematic operation scenes, providing a precise classification basis for subsequent scene-based monitoring and diagnosis.

[0037] The monitoring index configuration module 200 is configured to configure high-frequency monitoring indexes for the plurality of power operation scenarios based on historical cell capacity attenuation fault records of the same type of energy storage battery, and obtain a plurality of high-frequency monitoring index sets, including:

[0038] A first power operation scenario is randomly selected from the plurality of power operation scenarios, and a historical cell capacity attenuation fault record of the same type of energy storage battery is retrieved based on the first power operation scenario as a conditional constraint to obtain a first attenuation fault record data set;

[0039] Based on the historical cell capacity attenuation fault records of the same type of energy storage battery, the frequency of cell capacity attenuation faults in the plurality of power operation scenarios is respectively counted to obtain a plurality of attenuation fault frequencies;

[0040] According to the plurality of attenuation fault frequencies, a plurality of adaptive monitoring index quantities are set, and a first adaptive monitoring index quantity of the first power operation scenario is obtained;

[0041] The setting of the plurality of adaptive monitoring index quantities according to the plurality of attenuation fault frequencies includes:

[0042] The first attenuation fault frequency of the first power operation scenario is obtained, and the ratio of the first attenuation fault frequency to the sum of the plurality of attenuation fault frequencies is set as a first attenuation fault severity coefficient;

[0043] The product of the first attenuation fault severity coefficient and a preset monitoring index quantity is rounded to obtain the first adaptive monitoring index quantity, and a plurality of adaptive monitoring index quantities are sequentially analyzed, wherein the preset monitoring index quantity is 3, the adaptive monitoring index quantity is greater than or equal to 2, and less than or equal to 5;

[0044] The first high-frequency monitoring index set is configured according to the first attenuation fault record data set and the first adaptive monitoring index quantity, and a plurality of high-frequency monitoring index sets of the plurality of power operation scenarios are sequentially analyzed;

[0045] The configuration of the first high-frequency monitoring index set according to the first attenuation fault record data set and the first adaptive monitoring index quantity includes:

[0046] A global monitoring index set of cell capacity attenuation faults is obtained, wherein the global monitoring index set includes an electrical index set, a thermal characteristic index set, a cycle and operation mode index set, and an SOC / SOH behavior index set;

[0047] According to the first attenuation fault record data set, correlation degree calculation is performed on a plurality of monitoring indexes in the global monitoring index set respectively, and the plurality of monitoring indexes are sorted in descending order of correlation degree to construct a first monitoring index sequence;

[0048] In the first monitoring index sequence, a plurality of monitoring indexes are selected, and a first high-frequency monitoring index set is obtained.

[0049] The traditional monitoring adopts a fixed index set to collect data, and ignores the differences in dominant causes of faults under different operation scenarios. The prior art does not dynamically configure monitoring indexes according to scenarios, resulting in collection of a large amount of low-correlation-degree data, which not only increases the transmission burden but also dilutes the key fault characteristics, affecting the accuracy of early diagnosis.

[0050] In the embodiments of the present application, a first power operation scenario is randomly selected as a current processing target in a plurality of power operation scenarios. The historical cell capacity attenuation fault records of the same type of energy storage battery are retrieved under the condition of the first power operation scenario, and the historical cell capacity attenuation fault records of the same type of energy storage battery are integrated to obtain a first attenuation fault record data set.

[0051] Based on the historical cell capacity attenuation fault records of the same type of energy storage battery, the cell capacity attenuation fault occurrence frequencies under a plurality of power operation scenarios are respectively counted to obtain a plurality of attenuation fault frequencies.

[0052] The first attenuation fault frequency of the first power operation scenario is obtained, and the ratio of the first attenuation fault frequency to the sum of the plurality of attenuation fault frequencies is set as a first attenuation fault severity coefficient. The first attenuation fault severity coefficient = the first attenuation fault frequency ÷ the sum of the plurality of attenuation fault frequencies, the first attenuation fault severity coefficient represents the severity of the first attenuation fault frequency in the overall fault frequency, and the higher the first attenuation fault severity coefficient, the greater the first attenuation fault frequency, and the more likely it is to occur in the corresponding power operation scenario.

[0053] The first attenuation fault severity coefficient is multiplied by the preset monitoring index number to obtain the first adaptive monitoring index number, and the result is rounded. The plurality of attenuation fault severity coefficients are analyzed in sequence to obtain a plurality of adaptive monitoring index numbers, wherein the preset monitoring index number is 3, the adaptive monitoring index number is greater than or equal to 2, and less than or equal to 5. When the first adaptive monitoring index number is greater than 5, 5 is taken.

[0054] The global monitoring indicator set for acquiring the capacity attenuation fault of the battery cell includes an electrical indicator set, a thermal characteristic indicator set, a cycle and operation mode indicator set, and an SOC / SOH behavior indicator set. For example, the electrical indicator set can include a charging rate in C, which represents the ratio of charging current to rated capacity; and cumulative charging energy in kWh, which is used to accumulate the degradation load. The thermal characteristic indicator set can include a maximum temperature in °C, which exceeds a threshold value to accelerate capacity degradation; and a cooling system load rate in %, which is greater, the greater the thermal management pressure and the faster the degradation. The cycle and operation mode indicator set can include a daily average cycle number in times, which increases the aging rate with high frequency cycles; and a charge and discharge depth in %, which measures the percentage between the discharge amount and the rated capacity. The SOC / SOH behavior indicator set can include an average SOC level in %, which is prone to cause lithium dendrite growth and trigger faults at a long-term high SOC level; and an SOH level in %, which reflects the health status, and the lower the SOH, the less the remaining life.

[0055] According to the first attenuation fault record data set, the correlation degrees of the plurality of monitoring indicators in the global monitoring indicator set are calculated respectively, and are sorted in descending order of correlation degrees to construct a first monitoring indicator sequence. Specifically, the correlation degree = the number of a monitoring indicator in the attenuation fault record data set ÷ the total number of attenuation fault records, and the higher the correlation degree, the more positively correlated the monitoring indicator is with the attenuation fault, and the monitoring of the indicator can more accurately and timely reflect the attenuation fault condition. The monitoring indicators are sorted in descending order of correlation degrees to obtain the first monitoring indicator sequence.

[0056] In the first monitoring indicator sequence, a first adaptive monitoring indicator number of monitoring indicators are selected to obtain a first high-frequency monitoring indicator set. For example, if the first adaptive monitoring indicator number is 3, the first 3 monitoring indicators in the first monitoring indicator sequence are selected to obtain the first high-frequency monitoring indicator set.

[0057] Based on the historical fault data, a high-correlation monitoring indicator set is dynamically screened for each power operation scenario. By eliminating redundant indicators and retaining scenario-sensitive features, the data acquisition amount is greatly reduced, and the fault signal capture capability is strengthened, thereby providing an accurate data basis for lightweight diagnosis.

[0058] The scenario prediction mapping module 300 is configured to construct a plurality of attenuation fault prediction units in the grid cloud server according to the plurality of high-frequency monitoring indicator sets, and construct a scenario-prediction unit calling table in combination with the plurality of power operation scenarios, and the scenario-prediction unit calling table includes:

[0059] A first power operation scenario is randomly selected, and a first high-frequency monitoring indicator set of the first power operation scenario is obtained.

[0060] retrieve historical cell capacity attenuation fault records of the same type of energy storage battery under the condition constraints of the first power operation scene and the first high-frequency monitoring indicator set, obtain a plurality of sample first high-frequency monitoring indicator sets, and statistically obtain a first cell capacity attenuation fault probability by counting a proportion of cell capacity attenuation faults in a historical time zone of different sample first high-frequency monitoring indicator sets.

[0061] The plurality of sample first high-frequency monitoring indicator sets and the plurality of sample first attenuation fault probabilities are used to train a deep learning model to convergence, to obtain a first attenuation fault prediction unit, and a plurality of attenuation fault prediction units are sequentially analyzed and constructed.

[0062] The existing diagnostic models are mostly general single structures, which cannot adapt to the heterogeneity of faults in multiple scenes. If the model is forced to generalize across scenes, it may lead to high misjudgment rate and poor fault diagnosis accuracy due to mechanism differences.

[0063] In the power grid cloud server, a first power operation scene is randomly selected as a current processing object, and a first high-frequency monitoring indicator set of the first power operation scene is obtained.

[0064] The plurality of sample first high-frequency monitoring indicator sets are obtained by retrieving historical cell capacity attenuation fault records of the same type of energy storage battery under the condition constraints of the first power operation scene and the first high-frequency monitoring indicator set.

[0065] The first cell capacity attenuation fault probability is obtained by statistically obtaining a proportion of cell capacity attenuation faults in a historical time zone of different sample first high-frequency monitoring indicator sets. The historical time is set to the past 60 days, and the cell capacity attenuation fault probability can better reflect the recent situation. The first cell capacity attenuation fault probability = sample first high-frequency monitoring indicator set ÷ number of cell capacity attenuation faults in the past 60 days.

[0066] Exemplarily, based on deep learning, a first attenuation fault prediction unit is constructed by using a 3-layer structure, wherein the input layer is used to receive the high-frequency monitoring indicator set, the hidden layer uses 32 nodes and uses the ReLU function for activation, and the output layer is used to output the first attenuation fault probability. The plurality of sample first high-frequency monitoring indicator sets and the plurality of sample first attenuation fault probabilities are used to train a deep learning model to convergence, for example, the input high-frequency monitoring indicator set, and the output first attenuation fault probability is within ±3%, that is, the first attenuation fault prediction unit is trained. A plurality of attenuation fault prediction units are sequentially analyzed and constructed by analyzing a plurality of sample first high-frequency monitoring indicator sets.

[0067] The plurality of power operation scenes and corresponding high-frequency monitoring indicator sets, and attenuation fault prediction units are mapped and matched to construct a scene-prediction unit calling table.

[0068] In the power grid cloud server, a plurality of lightweight attenuation fault prediction units are trained according to a scenario index set, and a scene-prediction unit calling mapping table is constructed. The complex model training is placed in the cloud, and the edge side only needs to call the adaptive prediction unit according to the real-time scene, which avoids the overload of the edge device computing power and guarantees the pertinence of the diagnosis model in different scenes, solving the adaptability defects of one model to multiple scenes and the inaccurate diagnosis consequences caused thereby.

[0069] The preliminary fault diagnosis module 400 is used for deploying the scene-prediction unit calling table in the lightweight edge device of the energy storage battery, and performing preliminary fault diagnosis of the cell capacity attenuation, including:

[0070] Based on the scene-prediction unit calling table, the adaptive high-frequency monitoring index set and the adaptive attenuation fault prediction unit are matched and obtained according to the current power operation scene.

[0071] According to the high-frequency monitoring index set, data collection is performed to obtain a real-time monitoring data set, and the real-time monitoring data set is input into the adaptive attenuation fault prediction unit to output a current attenuation fault probability.

[0072] If the current attenuation fault probability is greater than a preset threshold, a preliminary fault signal is generated, and a global monitoring data set is collected according to a global monitoring index set and transmitted to a power grid cloud server for comprehensive fault diagnosis by a global attenuation fault prediction unit, wherein the global attenuation fault prediction unit is constructed based on the global monitoring index set.

[0073] The edge device is limited by computing power and storage resources, and if comprehensive monitoring and deep diagnosis are performed, response delay will be caused, and if the diagnosis logic is excessively simplified, faults may be missed. The prior art is difficult to balance real-time performance and reliability.

[0074] In the embodiments of the present application, the scene-prediction unit calling table is deployed in the lightweight edge device of the energy storage battery. Based on the scene-prediction unit calling table, the adaptive high-frequency monitoring index set and the adaptive attenuation fault prediction unit are matched and obtained according to the current power operation scene. Specifically, in the scene-prediction unit calling table, the adaptive high-frequency monitoring index set and the adaptive attenuation fault prediction unit matched with the same power operation scene as the current power operation scene are matched and obtained.

[0075] According to the high-frequency monitoring index set, data collection is performed to obtain a real-time monitoring data set, and the real-time monitoring data set is input into the adaptive attenuation fault prediction unit to output a current attenuation fault probability.

[0076] If the current attenuation fault probability is greater than the preset threshold value, a preliminary fault signal is generated, for example, a text prompt of "possible fault hidden danger" is generated and sent to the user end for early warning. The global monitoring data set is collected according to the global monitoring indicator set and transmitted to the power grid cloud server, and comprehensive fault diagnosis is performed through the global attenuation fault prediction unit, wherein the global attenuation fault prediction unit is constructed based on the global monitoring indicator set. For example, a 4-layer structure is constructed based on deep learning to construct the global attenuation fault prediction unit, the input layer is used to receive the global monitoring indicator set, the first hidden layer uses 128 nodes and uses the ReLU function for activation, the second hidden layer uses 64 nodes and uses the ReLU function for activation, and the output layer is used to output the fault diagnosis result, for example, the fault probability, to obtain the comprehensive fault diagnosis result.

[0077] By deploying the scene-prediction unit call table to the edge device, a quick diagnosis closed loop of scene matching-indicator collection-lightweight prediction is realized. When the preliminary diagnosis finds an abnormal probability, cloud global review is triggered, which can ensure real-time response of high-frequency scenes, dynamically allocate computing resources through two-level diagnosis, and effectively solve the contradiction between diagnosis real-time and accuracy.

[0078] Embodiment two, as shown in Figure 2 Based on the same inventive concept as the device fault diagnosis system provided in embodiment one, the present embodiment also provides a device fault diagnosis method based on lightweight edge computing, comprising:

[0079] S10: Based on the power grid structure type, power grid load type and power grid peak shaving type of the power grid where the energy storage battery is located, three-dimensional scene coupling modeling is performed to construct a power operation scene library, wherein the power operation scene library includes a plurality of power operation scenes.

[0080] Based on the power grid structure type, power grid load type and power grid peak shaving type of the power grid where the energy storage battery is located, three-dimensional scene coupling modeling is performed to construct a power operation scene library, wherein the power operation scene library includes a plurality of power operation scenes, including:

[0081] The power grid structure type of the power grid where the energy storage battery is located is collected, wherein the power grid structure type at least includes pure thermal power supply, centralized photovoltaic access, centralized wind power access, wind-solar-thermal multi-energy complementation, distributed photovoltaic access, microgrid / island operation and source-load-storage integration;

[0082] The power grid load type of the power grid where the energy storage battery is located is collected, wherein the power grid load type at least includes peak load, valley load, stable load, fluctuating load and sharp peak load;

[0083] Collecting a power grid peak regulation type of a power grid where the energy storage battery is located, wherein the power grid peak regulation type at least includes high-frequency peak regulation, medium-frequency peak regulation, low-frequency peak regulation, and no obvious peak regulation;

[0084] Randomly selecting any parameter in the power grid structure type, the power grid load type, and the power grid peak regulation type to perform three-dimensional scene coupling modeling to generate a first power operation scene, and sequentially performing coupling modeling to obtain a plurality of power operation scenes to construct a power operation scene library.

[0085] S20: Based on the historical cell capacity attenuation fault records of the same type of energy storage battery, high-frequency monitoring index configuration is performed on the plurality of power operation scenes respectively to obtain a plurality of high-frequency monitoring index sets.

[0086] Based on the historical cell capacity attenuation fault records of the same type of energy storage battery, high-frequency monitoring index configuration is performed on the plurality of power operation scenes respectively to obtain a plurality of high-frequency monitoring index sets, including:

[0087] Randomly selecting a first power operation scene from the plurality of power operation scenes, and retrieving the historical cell capacity attenuation fault records of the same type of energy storage battery under the condition of the first power operation scene to obtain a first attenuation fault record data set;

[0088] Based on the historical cell capacity attenuation fault records of the same type of energy storage battery, the cell capacity attenuation fault occurrence frequencies under the plurality of power operation scenes are respectively counted to obtain a plurality of attenuation fault frequencies.

[0089] According to the plurality of attenuation fault frequencies, a plurality of adaptive monitoring index quantities are set, and a first adaptive monitoring index quantity of the first power operation scene is obtained;

[0090] According to the plurality of attenuation fault frequencies, a plurality of adaptive monitoring index quantities are set, including:

[0091] Obtaining a first attenuation fault frequency of the first power operation scene, and setting a ratio of the first attenuation fault frequency to a sum of the plurality of attenuation fault frequencies as a first attenuation fault severity coefficient;

[0092] Taking an integer of a product of the first attenuation fault severity coefficient and a preset monitoring index quantity to obtain a first adaptive monitoring index quantity, and sequentially analyzing to obtain a plurality of adaptive monitoring index quantities, wherein the preset monitoring index quantity is 3, the adaptive monitoring index quantity is greater than or equal to 2, and less than or equal to 5;

[0093] According to the first attenuation fault record data set and the first adaptive monitoring index quantity, a first high-frequency monitoring index set is configured, and a plurality of high-frequency monitoring index sets of the plurality of power operation scenes are sequentially analyzed;

[0094] wherein a first high-frequency monitoring indicator set is configured according to the first attenuation fault record data set and a first adaptive monitoring indicator number, including:

[0095] obtaining a global monitoring indicator set of the battery capacity attenuation fault, wherein the global monitoring indicator set includes an electrical indicator set, a thermal characteristic indicator set, a cycle and operation mode indicator set, and an SOC / SOH behavior indicator set;

[0096] According to the first attenuation fault record data set, the correlation degree of each monitoring indicator in the global monitoring indicator set is calculated, and the monitoring indicators are sorted in descending order of correlation degree to construct a first monitoring indicator sequence.

[0097] Selecting the first adaptive monitoring indicator number of monitoring indicators in the first monitoring indicator sequence to obtain a first high-frequency monitoring indicator set.

[0098] S30: In the grid cloud server, a plurality of attenuation fault prediction units are constructed according to the plurality of high-frequency monitoring indicator sets, and a scene-prediction unit calling table is constructed by combining the plurality of power operation scene mappings.

[0099] wherein in the grid cloud server, a plurality of attenuation fault prediction units are constructed according to the plurality of high-frequency monitoring indicator sets, and a scene-prediction unit calling table is constructed by combining the plurality of power operation scene mappings, including:

[0100] Randomly selecting a first power operation scene and obtaining a first high-frequency monitoring indicator set of the first power operation scene;

[0101] Taking the first power operation scene and the first high-frequency monitoring indicator set as conditional constraints, retrieving historical battery capacity attenuation fault records of the same type of energy storage battery, obtaining a plurality of sample first high-frequency monitoring indicator sets, and statistically analyzing the battery capacity attenuation fault proportion of different sample first high-frequency monitoring indicator sets in the historical time zone, setting the first battery capacity attenuation fault probability, and obtaining a plurality of sample first attenuation fault probabilities.

[0102] Using the plurality of sample first high-frequency monitoring indicator sets and the plurality of sample first attenuation fault probabilities, a deep learning model is trained to convergence to obtain a first attenuation fault prediction unit, and a plurality of attenuation fault prediction units are sequentially analyzed and constructed.

[0103] S40: Deploying the scene-prediction unit calling table to the lightweight edge device of the energy storage battery to perform preliminary fault diagnosis of the battery capacity attenuation.

[0104] wherein deploying the scene-prediction unit calling table to the lightweight edge device of the energy storage battery to perform preliminary fault diagnosis of the battery capacity attenuation, including:

[0105] based on the scene-prediction unit call table, obtaining an adaptive high-frequency monitoring index set and an adaptive attenuation fault prediction unit according to a current power operation scene;

[0106] According to the high-frequency monitoring index set, data collection is performed to obtain a real-time monitoring data set, and the real-time monitoring data set is input into the adaptive attenuation fault prediction unit to output a current attenuation fault probability;

[0107] If the current attenuation fault probability is greater than a preset threshold, a preliminary fault signal is generated, and a global monitoring data set is collected according to a global monitoring index set and transmitted to a power grid cloud server for comprehensive fault diagnosis by a global attenuation fault prediction unit, wherein the global attenuation fault prediction unit is constructed based on the global monitoring index set.

[0108] To sum up, the embodiments of the present application have at least the following technical effects:

[0109] The present application provides a light edge computing device fault diagnosis system and method, which dynamically generates a power operation scene library through three-dimensional scene coupling modeling, and adaptively configures a high-frequency monitoring index set and a prediction unit based on the scene characteristics, thereby significantly improving the accuracy of energy storage battery fault diagnosis and resource utilization efficiency. First, based on the three-dimensional scene modeling of the power grid structure, load and peak shaving type, a scene library covering multi-dimensional operation characteristics is constructed, which enables the monitoring strategy to be deeply adapted to the real-time working condition and effectively captures the differentiated characteristics of the capacity attenuation of the battery under different scenes. Second, the number of adaptive monitoring indicators is dynamically allocated based on the historical fault frequency, and only high-correlation indicators are collected, thereby greatly reducing the redundant data transmission and storage overhead. Third, the scene-based light prediction unit is pre-trained on the cloud and a call mapping table is constructed, and the edge device only needs to call the adaptive model according to the scene to perform diagnosis, which not only avoids the calculation pressure of deploying a complex global model, but also ensures the accuracy of fault probability prediction under different scenes. Finally, a collaborative mechanism of edge preliminary diagnosis and cloud review is adopted, and when the edge detects a potential fault, global diagnosis is triggered to achieve dynamic balance of resource allocation and diagnosis reliability. Compared with the traditional method, the technical scheme provided by the present application significantly overcomes the contradiction between diagnosis accuracy and calculation load in the limited resource scene, and achieves the technical effects of simultaneously improving fault early warning capability, scene adaptability and system real-time performance under limited resources.

[0110] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. Moreover, the above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0111] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0112] The specification and drawings are only exemplary of the present application, and any and all modifications, variations, combinations or equivalents that are within the scope of the present application are intended to be encompassed by the present application. Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the scope of the present application. Thus, it is intended that the present application encompass such modifications and changes as fall within the scope of the present application and its equivalents.

Claims

1. A lightweight edge computing device fault diagnosis system, characterized in that, The equipment fault diagnosis system includes: The operation scenario modeling module is used to perform three-dimensional scenario coupling modeling based on the grid structure type, grid load type and grid peak shaving type of the power grid where the energy storage battery is located, and to build a power operation scenario library, wherein the power operation scenario library includes several power operation scenarios; The monitoring indicator configuration module is used to configure high-frequency monitoring indicators for the several power operation scenarios based on the historical cell capacity decay fault records of similar energy storage batteries, and obtain several sets of high-frequency monitoring indicators. The scenario prediction mapping module is used to construct several attenuation fault prediction units based on several high-frequency monitoring index sets within the power grid cloud server, and to construct a scenario-prediction unit call table in conjunction with several power operation scenario mappings. The preliminary fault diagnosis module is used to deploy the scenario-prediction unit call table on the lightweight edge device of the energy storage battery to perform preliminary fault diagnosis of cell capacity decay.

2. The lightweight edge computing device fault diagnosis system according to claim 1, characterized in that, Based on the grid structure type, load type, and peak-shaving type of the power grid where the energy storage battery is located, a three-dimensional scene coupled model is constructed to build a power operation scenario library, including: The grid structure type of the power grid where the energy storage battery is located is collected, wherein the grid structure type includes at least pure thermal power supply, centralized photovoltaic access, centralized wind power access, wind-solar-thermal multi-energy complementarity, distributed photovoltaic access, microgrid / island operation, and integrated source-load-storage; The grid load type of the power grid where the energy storage battery is located is collected, wherein the grid load type includes at least peak load, off-peak load, stable load, fluctuating load and peak load; The peak-shaving type of the power grid where the energy storage battery is located is collected, wherein the peak-shaving type includes at least high-frequency peak-shaving, medium-frequency peak-shaving, low-frequency peak-shaving and no obvious peak-shaving; Randomly select any parameter from the power grid structure type, power grid load type and power grid peak shaving type to perform three-dimensional scene coupling modeling to generate the first power operation scenario. Then, sequentially couple and model several power operation scenarios to build a power operation scenario library.

3. The lightweight edge computing device fault diagnosis system according to claim 1, characterized in that, Based on historical cell capacity decay failure records of similar energy storage batteries, high-frequency monitoring indicators are configured for the aforementioned power operation scenarios to obtain several sets of high-frequency monitoring indicators, including: Randomly select a first power operation scenario from the aforementioned power operation scenarios, and use the first power operation scenario as a condition constraint to retrieve historical cell capacity decay fault records of similar energy storage batteries to obtain a first decay fault record dataset. Based on historical cell capacity decay fault records of similar energy storage batteries, the frequency of cell capacity decay fault occurrence under several power operation scenarios is statistically analyzed to obtain several decay fault frequencies. Based on the aforementioned attenuation fault frequencies, a number of adaptive monitoring indicators are set, and the number of first adaptive monitoring indicators for the first power operation scenario is obtained. The first high-frequency monitoring indicator set is configured based on the first attenuation fault record dataset and the first number of adaptive monitoring indicators, and several high-frequency monitoring indicator sets for the several power operation scenarios are obtained by sequential analysis.

4. The lightweight edge computing device fault diagnosis system according to claim 3, characterized in that, Based on the aforementioned attenuation fault frequencies, several adaptive monitoring indicators are set, including: Obtain the first attenuation fault frequency of the first power operation scenario, and set the ratio of the first attenuation fault frequency to the sum of the plurality of attenuation fault frequencies as the first attenuation fault severity coefficient. The product of the first attenuation fault severity coefficient and the preset number of monitoring indicators is rounded to obtain the first number of adaptive monitoring indicators. Several adaptive monitoring indicators are then analyzed sequentially. The preset number of monitoring indicators is 3, and the number of adaptive monitoring indicators is greater than or equal to 2 and less than or equal to 5.

5. The lightweight edge computing device fault diagnosis system according to claim 3, characterized in that, Configure a first high-frequency monitoring indicator set based on the first attenuation fault record dataset and the number of first adaptive monitoring indicators, including: A global monitoring index set for cell capacity decay faults is obtained, wherein the global monitoring index set includes an electrical index set, a thermal characteristic index set, a cycle and operation mode index set, and a SOC / SOH behavior index set; Based on the first attenuation fault record dataset, the correlation degree of multiple monitoring indicators in the global monitoring indicator set is calculated, and they are sorted in descending order of correlation degree to construct the first monitoring indicator sequence. The first high-frequency monitoring indicator set is obtained by selecting the number of monitoring indicators that are the first number of the first adapted monitoring indicators from the first monitoring indicator sequence.

6. The lightweight edge computing device fault diagnosis system according to claim 1, characterized in that, Several attenuation fault prediction units are constructed based on the aforementioned sets of high-frequency monitoring indicators, including: A first power operation scenario is randomly selected, and a first high-frequency monitoring index set for the first power operation scenario is obtained; Using the first power operation scenario and the first high-frequency monitoring index set as constraints, retrieve historical cell capacity decay fault records of similar energy storage batteries, obtain multiple sample first high-frequency monitoring index sets, and calculate the proportion of cell capacity decay faults in different sample first high-frequency monitoring index sets in the historical time zone, set as the first cell capacity decay fault probability, and obtain multiple sample first decay fault probabilities. Using the first high-frequency monitoring index set of the multiple samples and the first attenuation fault probability of the multiple samples, a deep learning model is trained until convergence to obtain the first attenuation fault prediction unit, and several attenuation fault prediction units are constructed in sequence.

7. The lightweight edge computing device fault diagnosis system according to claim 1, characterized in that, The scenario-prediction unit call table is deployed on a lightweight edge device of the energy storage battery to perform preliminary fault diagnosis of cell capacity degradation, including: Based on the scenario-prediction unit call table, the appropriate high-frequency monitoring index set and the appropriate attenuation fault prediction unit are obtained according to the current power operation scenario. Data is collected according to the high-frequency monitoring index set to obtain a real-time monitoring dataset, and the real-time monitoring dataset is input into the adaptive attenuation fault prediction unit to output the current attenuation fault probability. If the current attenuation fault probability is greater than a preset threshold, a preliminary fault signal is generated, and a global monitoring dataset is collected according to the global monitoring index set and transmitted to the power grid cloud server. A comprehensive fault diagnosis is then performed through the global attenuation fault prediction unit, which is constructed based on the global monitoring index set.

8. A method for diagnosing device faults in lightweight edge computing, characterized in that, The method includes: Based on the grid structure type, grid load type and grid peak-shaving type of the power grid where the energy storage battery is located, a three-dimensional scene coupling model is performed to construct a power operation scenario library, which includes several power operation scenarios. Based on historical cell capacity decay fault records of similar energy storage batteries, high-frequency monitoring indicators are configured for the aforementioned power operation scenarios to obtain several sets of high-frequency monitoring indicators. Within the power grid cloud server, several attenuation fault prediction units are constructed based on the aforementioned sets of high-frequency monitoring indicators, and a scenario-prediction unit call table is constructed by combining the aforementioned power operation scenario mappings. The scenario-prediction unit call table is deployed on the lightweight edge device of the energy storage battery to perform preliminary fault diagnosis of cell capacity degradation.

Images