Charging pile fault self-repairing control system

Abstract

The invention relates to the technical field of charging station monitoring control systems, in particular to a charging pile fault self-repairing control system, which comprises a data acquisition and preprocessing module, a deep learning fault diagnosis module and a fault self-repairing strategy module. According to the charging pile fault self-repairing system based on deep learning provided by the invention, the deep learning fault diagnosis model is constructed based on the CNN and the LSTM, so that local features and a time sequence dependency relationship can be extracted from multi-dimensional data, complex faults which are difficult to detect by a traditional method are accurately identified, the diagnosis accuracy is improved by more than 30%, and the self-repairing effect is good. Then under the condition that the fault type diagnosis result output by the deep learning fault diagnosis module is self-repairing, corresponding fault self-repairing operation is adopted through the fault self-repairing strategy module, for example, program restarting or parameter adjusting is conducted for software faults, a standby circuit is switched for hardware faults, the manual intervention time is shortened by about 60%, and the fault self-repairing efficiency is improved. And the operation and maintenance cost is obviously reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of charging station monitoring and control systems, and more particularly, to a control system for self-repairing charging pile faults. Background Art

[0002] There are significant deficiencies in the current fault detection and repair technologies for charging piles. Traditional methods mainly rely on fault detection based on simple rules, such as determining abnormalities by presetting electrical parameter thresholds. However, such methods cannot adapt to complex and changing fault scenarios, and newly emerging fault modes are difficult to identify due to the lack of a rule base. In addition, the hardware monitoring circuit has a single function, can only detect a few obvious faults, and lacks flexibility. The manual repair method is costly and has a slow response speed. Especially in remote areas or bad weather, the long-term shutdown of charging piles seriously affects the user experience.

[0003] Secondly, in the prior art, although the simply networked charging pile system can upload fault information to the management platform, subsequent processing still requires manual operation and lacks intelligent repair capabilities. Temporary measures such as software restart are only applicable to simple faults and are ineffective for hardware faults or complex software problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a control system for self-repairing charging pile faults to improve the technical problem that existing charging piles lack intelligent repair capabilities.

[0005] To achieve the above purpose, the embodiments of the present application provide the following technical solutions: On the one hand, the embodiments of the present application provide a control system for self-repairing charging pile faults, the system includes: a data acquisition and preprocessing module, which consists of a current sensor, a voltage sensor, a temperature sensor, and a vehicle-pile communication data acquisition unit installed at key parts of the charging pile, and is used to collect the operating data of the charging pile in real time and preprocess the data; a deep learning fault diagnosis module, which is used to extract local features and time series dependence relationships from the preprocessed data and output a fault type diagnosis result; a fault self-repair strategy module: according to the fault diagnosis result, perform software restart, parameter adjustment, or hardware standby circuit switching operations.

[0006] Optionally, it further includes a system control and communication module: coordinating the operation processes of each module and interacting with the charging pile management platform and the user terminal in real time for fault status and repair information.

[0007] Optionally, the extracting local features and time series dependence relationships from the preprocessed data and outputting a fault type diagnosis result includes: Obtaining the charging power fluctuation data and a preset power curve within a first time period, and constructing an actual charging trend curve corresponding to the charging power fluctuation data through a linear regression algorithm; Detect the difference fluctuation between the actual charging trend curve and the preset power curve based on the time series, identify multiple first time nodes where the difference fluctuation is greater than the preset threshold, and based on the distribution of the multiple first time nodes, screen out multiple consecutive first time nodes with a consecutive duration greater than the preset unit duration, and then intercept multiple time series segments to be corrected; Based on the percentage of the difference between the preset power curve and the actual charging trend curve in each first time node in the time series segment to be corrected in the preset threshold, re-anchor the power value of the first time node in the preset power curve, and then correct the preset power curve; Based on the corrected preset power curve in the first time period, adjust the subsequent preset power curve again through the Kalman filter algorithm, so as to realize the dynamic adjustment of the preset power curve; Construct a power fluctuation curve band corresponding to the first time period based on the corrected preset power curve and the preset power fluctuation percentage threshold, and screen out multiple abnormal peak segments in the charging power fluctuation data based on the power fluctuation curve band. Calculate the coupling degree between the corrected preset power curve and the charging power fluctuation data. The abnormal peak segment is a power fluctuation segment not within the power fluctuation curve band; Calculate the abnormal fault risk valuation through a weighted algorithm based on the frequency of occurrence of multiple abnormal peak segments, the average power corresponding to the multiple abnormal peak segments, and the coupling degree. When the abnormal fault risk valuation is greater than the preset safety threshold, generate a first charging power abnormal fault code. The first charging power abnormal fault code is used to enable the deep learning fault diagnosis module to determine the fault type based on multiple abnormal peak segments and feedback the fault type to the fault self-repair strategy module, so that the fault self-repair strategy module takes corresponding fault elimination measures.

[0008] Optionally, the deep learning fault diagnosis module determines the fault type based on multiple abnormal peak segments and feedbacks the fault type to the fault self-repair strategy module, so that the fault self-repair strategy module takes corresponding fault elimination measures, including: Merge multiple abnormal peak segments based on the time series, and separate the output current fluctuation curve and the output voltage fluctuation curve based on the merged multiple abnormal peak segments; Input the output current fluctuation curve and the output voltage fluctuation curve into a preset first deep learning fault diagnosis sub-model, so that it analyzes the high-frequency noise distribution corresponding to the output current fluctuation curve and the high-frequency noise distribution corresponding to the output voltage fluctuation curve, and combines the high-frequency noise distribution in the historical training data to respectively determine the high-frequency noise increments of the current and / or voltage corresponding to multiple abnormal peak segments. If the high-frequency noise increment is within the first abnormal gradient threshold range, it is determined as a high-temperature abnormality of the output filter capacitor, and the high-temperature abnormality of the output filter capacitor is fed back to the fault self-repair strategy module, so that it increases the operating power of the cooling fan corresponding to the power output module in the charging pile, thereby reducing the operating temperature of the output filter capacitor and reducing the degradation of the high-frequency filtering performance caused by excessive temperature. If the high-frequency noise increment is less than or equal to the lower limit value of the first abnormal gradient threshold range, retrieve the DC bus voltage fluctuation data corresponding to multiple abnormal peak segments and input it into the second deep learning fault diagnosis sub-model, so that it sequentially constructs the voltage ripple corresponding to each abnormal peak segment and combines the historical training data to determine the comprehensive change of the DC bus voltage ripple corresponding to multiple abnormal peak segments. If the DC bus voltage ripple corresponding to multiple abnormal peak segments as a whole shows an increasing trend and exceeds the preset growth threshold, it is determined as a high-temperature abnormality of the DC bus capacitor, and the high-temperature abnormality of the DC bus capacitor is fed back to the fault self-repair strategy module, so that it increases the operating power of the cooling fan corresponding to the power supply module in the charging pile, thereby reducing the operating temperature of the DC bus capacitor and reducing the reduction of the DC bus capacitor capacity caused by excessive temperature.

[0009] Optionally, the deep learning fault diagnosis module is composed of a combination of a convolutional neural network (CNN) and a long short-term memory network (LSTM). The convolutional neural network (CNN) includes three parallel convolutional layers, using convolutional kernels of 3×3, 5×5, and 7×7 respectively, for extracting local features of different scales; the LSTM layer includes two cascaded LSTM units, each containing 64 and 32 memory units respectively, for capturing the long-term dependencies of time series data.

[0010] Optionally, the fault self-repair strategy module includes: a software repair unit: used to execute program restart or parameter configuration update; a hardware repair unit: by controlling the switch circuit to switch to the standby charging circuit to maintain the basic functions of the charging pile; a complex fault reporting unit: sending the fault information that cannot be automatically repaired to the management platform and triggering manual intervention.

[0011] The beneficial effects of the present invention are: The charging pile fault self-repair system based on deep learning provided by the present invention can extract local features and time series dependencies from multi-dimensional data by constructing a deep learning fault diagnosis model based on CNN and LSTM, accurately identify complex faults (such as multi-parameter correlation faults) that are difficult to detect by traditional methods, and the diagnostic accuracy is increased by more than 30%. Then, when the fault type diagnosis result output by the deep learning fault diagnosis module indicates that it can be self-repaired, corresponding fault self-repair operations are taken through the fault self-repair strategy module. For example, for software faults, the program is restarted or parameters are adjusted, and for hardware faults, the standby circuit is switched, reducing the manual intervention time by about 60% and significantly reducing the operation and maintenance costs.

[0012] Secondly, the system control and communication module can realize the real-time coordination of each functional module and interact with the management platform and the user side through the wireless network to ensure the transparency of fault information. Users can receive the repair progress in real time, reasonably arrange the charging plan, and the user experience is significantly improved.

[0013] Secondly, the modular design supports flexible expansion. For example, new sensors can be added or the deep learning model structure can be optimized to adapt to the operation and maintenance requirements of different models of charging piles, and it has a wide range of application prospects.

[0014] Other features and advantages of the present invention will be described in the subsequent specification, and some of them will become obvious from the specification, or can be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification, claims, and drawings. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a schematic flow chart of a charging pile fault self-repair control method described in the embodiments of the present invention. Detailed Embodiments

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and illustrated herein generally can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0018] It should be noted that like reference numerals or letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, terms such as "first" and "second" are only used for descriptive distinction and cannot be construed as indicating or implying relative importance.

[0019] Embodiment 1:

[0020] This embodiment provides a control system for self-repairing charging pile faults. The system includes: A data acquisition and preprocessing module, which consists of current sensors, voltage sensors, temperature sensors, and vehicle-pile communication data acquisition units installed at key parts of the charging pile, and is used to collect the operation data of the charging pile in real time and preprocess the data; A deep learning fault diagnosis module, which is composed of a combination of a convolutional neural network and a long short-term memory network. The convolutional neural network includes three parallel convolutional layers, using convolutional kernels of 3×3, 5×5, and 7×7 respectively, for extracting local features of different scales; the LSTM layer includes two serially connected LSTM units, each containing 64 and 32 memory units respectively, for capturing the long-term dependencies of time series data. The deep learning fault diagnosis module is used to extract local features and time series dependencies from the preprocessed data and output the diagnosis result of the fault type; The fault self-repair strategy module includes: a software repair unit - for performing program restart or parameter configuration update; a hardware repair unit - by controlling and adjusting the operating parameters of key devices or controlling the switching of the switch circuit to the standby charging circuit, etc., to maintain the basic functions of the charging pile; a complex fault reporting unit - sending the fault information that cannot be automatically repaired to the management platform and triggering manual intervention; the fault self-repair strategy module performs software restart, parameter adjustment, or hardware standby circuit switching operations according to the fault diagnosis result; System Control and Communication Module: Coordinates the operation processes of each module and interacts with the charging pile management platform and the user side in real time for fault status and repair information.

[0021] For the charging pile fault self-repair system based on deep learning described in this embodiment, a deep learning fault diagnosis model (neural network model) is constructed through CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory Network), which can extract local features and time series dependencies from multi-dimensional data, accurately identify complex faults (such as multi-parameter associated faults) that are difficult to detect by traditional methods, and the diagnostic accuracy is increased by more than 30%. The previous model training and optimization belong to conventional technical means and will not be elaborated here. Then, when the fault type diagnosis result output by the deep learning fault diagnosis module is a self-repairable fault, corresponding fault self-repair operations are taken through the fault self-repair strategy module. For example, for software faults, restart the program or adjust parameters, and for hardware faults, switch to the standby circuit, reducing the manual intervention time by about 60% and significantly reducing the operation and maintenance costs.

[0022] Secondly, the system control and communication module can achieve real-time coordination of each functional module and interact with the management platform and the user side through a wireless network to ensure the transparency of fault information. Users can receive the repair progress in real time and reasonably arrange the charging plan, significantly improving the user experience.

[0023] Secondly, the modular design supports flexible expansion. For example, adding new sensors or optimizing the deep learning model structure can adapt to the operation and maintenance requirements of different models of charging piles and has broad application prospects.

[0024] Embodiment 2:

[0025] This embodiment is based on Embodiment 1 and is used to provide a specific method for diagnosing and identifying hardware faults and self-repair control of the system. The overall overhaul method is to detect the abnormal charging power caused by the too high working temperature of the capacitors in the new energy charging pile, and based on the specific charging power data, implement heat dissipation processing for the specific functional capacitors; Before elaborating on the principle, it is necessary to briefly describe the capacitors with different functions in the new energy charging pile. The new energy charging pile is divided into functional modules as follows: The power module: converts the input alternating current (AC) into direct current (DC) and provides a stable low-voltage power supply for other modules; The control module (MCU / master control chip) is responsible for charging logic control (such as charging start and stop, power adjustment), communication with the vehicle BMS, status monitoring, etc. (the functional modules described in this embodiment are integrated on the control module); The power output module controls the on / off of high-power switching devices (such as IGBTs) and adjusts the output current and voltage. Other auxiliary modules, such as the metering module, user interface module, protection module, etc., have less relevance to this embodiment and will not be elaborated here.

[0026] Secondly, to avoid electromagnetic interference, the power modules and power output modules of most existing charging piles are installed separately, and both use independent heat dissipation units. Therefore, it is necessary to analyze the charging power anomaly data characteristics corresponding to capacitor anomalies in different functional modules; Each functional capacitor in the charging pile, as well as the installation location and anomaly characterization of each type of functional capacitor: There is a DC bus capacitor in the power module, located at the output end of the rectifier bridge, which is used to smooth the rectified DC voltage and suppress voltage ripple. The capacitance and performance of the DC bus capacitor directly affect the voltage stability of the DC bus. If the operating temperature is too high, it will cause the capacitance of the capacitor to decrease, resulting in an increase in voltage ripple, affecting the switching efficiency of power devices, and thus reducing the charging power; Input filter capacitor: installed on the AC input side (before the rectifier bridge), used to filter out high-frequency noise from the power grid. The output power of the charging gun is difficult to identify this type of anomaly (after multiple filtering processes in the middle, it is extremely difficult to establish a correlation, and there are more direct detection methods). Therefore, the noise identification and cause diagnosis at the power grid input end of the charging pile are relatively simple due to a single inducement and will not be elaborated here. This embodiment mainly focuses on the identification of DC bus capacitor anomalies; The power output module mainly contains output filter capacitors: close to power switching devices such as IGBT / MOSFET, used to absorb high-frequency switching noise and stabilize the output voltage. If the operating temperature is too high, it will cause the instability of the thin-film capacitor dielectric (increase in dielectric energy consumption and loss), thereby affecting the high-frequency filtering ability of the output filter capacitor; Secondly, during the communication between the control module of the charging pile and the vehicle BMS, a corresponding charging power curve will be formulated, that is, the preset power curve described in this embodiment. It should be emphasized here that the screening of abnormal data in this embodiment is essentially different from the conventional full-line power analysis. The input data of the deep learning fault diagnosis module in this embodiment is data that has been screened and preprocessed layer by layer. The specific preprocessing principle is as follows: The screening criterion for abnormal data - the power fluctuation curve band is based on the corrected preset power curve. The specific correction method is shown in steps S100 - S300.

[0027] Step S100: Obtain the charging power fluctuation data and the preset power curve within the first time period. Construct the actual charging trend curve corresponding to the charging power fluctuation data through the linear regression algorithm. This actual charging trend curve does not represent the actual power supply fluctuation data at specific moments, that is, it reflects the complex charging power fluctuation data (curve) on a relatively smooth curve, thereby reflecting the overall actual charging power. This actual charging trend curve is mainly used to correct the preset power curve. The detailed correction method is shown in steps S200 - S300. Furthermore, it indirectly realizes the real-time correction of the preset charging power through historical charging power data. Further, through the corrected preset charging power, the change of the preset power curve in a future period of time can be predicted, thereby realizing the dynamic adjustment of the preset power curve. Among them, the preset power curve is the charging power change curve in the ideal state formulated when the charging pile conducts protocol handshake docking with the new energy vehicle battery management module in the early stage. This curve is not static during the actual charging process in the later stage, but is regularly corrected according to the actual situation of the charging pile and the new energy vehicle. The charging power fluctuation data is the actual output power of the gun head on the charging pile; Step S200: Based on time series, detect the difference fluctuation between the actual charging trend curve and the preset power curve, and identify multiple first time nodes where the difference fluctuation is greater than the preset threshold. Based on the distribution of the multiple first time nodes, screen out multiple consecutive first time nodes with a consecutive duration greater than the preset unit duration, and then intercept multiple time series segments to be corrected. The time series segment to be corrected is a segment of abnormal charging power fluctuation data and the corresponding consecutive time nodes. For the identification of abnormal charging power time nodes, it is necessary to consider that the consecutive duration must be greater than the unit duration. Otherwise, the abnormal duration is relatively short and belongs to local accidental situations, which are not referenceable. Secondly, there are already fluctuation errors between the actual charging power of the charging pile and the preset charging power. Therefore, a threshold is set, and only the situation exceeding the threshold will be identified as an abnormal power fluctuation node. Then, select the abnormally power fluctuation nodes that are continuous in time from the abnormal power fluctuation nodes to form the time series segment to be corrected; Step S300: Based on the percentage of the difference between the preset power curve and the actual charging trend curve in each first time node in the time series segment to be corrected in the preset threshold, re-anchor / correct the power value of the first time node in the preset power curve, thereby correcting the preset power curve; Based on the corrected preset power curve within the first time period, adjust the preset power curve in the later stage again through the Kalman filter algorithm, thereby realizing the dynamic adjustment of the preset power curve; - that is, the correction amplitude of the preset power curve at a certain time node depends on the percentage of the power difference between the preset power curve and the actual charging trend curve at that time node and the preset threshold. The larger the percentage, the greater the correction amplitude; the corrected preset power curve fits the actual charging data better within the first time period; Step S400: Construct a power fluctuation curve band corresponding to the first time period based on the corrected preset power curve and the preset power fluctuation percentage threshold, and screen out multiple abnormal peak segments from the charging power fluctuation data based on the power fluctuation curve band, and calculate the coupling degree between the corrected preset power curve and the charging power fluctuation data. The abnormal peak segment is a power fluctuation segment not within the power fluctuation curve band; in the charging power anomaly determination of the deep learning fault diagnosis module, the preset power curve is used as a reference. However, this reference gradually deviates from the actual charging power fluctuation data during the actual charging process. For example, as the ambient temperature and the working temperature of the new energy vehicle or charging pile gradually increase, the actual charging power shows a downward trend compared with the preset charging power curve as a whole, resulting in an increasing deviation between the actual charging power and the preset charging power curve. Therefore, it is necessary to correct the preset power curve based on the actual charging power fluctuation data. However, the actual charging power fluctuation data has a large fluctuation frequency. Correcting the preset power curve based on the high-frequency spike peaks in the actual charging power fluctuation data will result in a huge amount of calculation and is unreasonable because there may be abnormal charging and discharging in the actual charging power fluctuation data. If the preset power curve is directly corrected based on the actual charging power fluctuation data, the preset power curve will lose its objectivity as a reference because local short-term abnormal charging and discharging will affect the correction of the preset power curve, thereby reducing the ideal objectivity of the preset power curve. To reduce the influence of such abnormal charging and discharging on the correction of the preset power curve, in this embodiment, a linear regression algorithm is used to construct the actual charging trend curve corresponding to the charging power fluctuation data, and the preset power curve is corrected through the actual charging trend curve, thereby effectively reducing the influence of abnormal charging and discharging on the correction of the preset power curve and improving its ideal objectivity as a reference for abnormal fault risk assessment. Then, a reasonable power fluctuation band, that is, a power fluctuation curve band, is constructed based on the corrected preset power curve. The charging power fluctuation data within this curve band is normal data, and the charging power fluctuation data exceeding the rate fluctuation curve band is the local feature extracted by the deep learning fault diagnosis module; Step S500: Calculate the abnormal fault risk valuation through a weighted algorithm based on the frequencies of multiple abnormal peak segments, the mean power corresponding to the multiple abnormal peak segments, and the coupling degree. When the abnormal fault risk valuation is greater than a preset safety threshold, generate a first charging power abnormal fault code. The first charging power abnormal fault code is used to enable the deep learning fault diagnosis module to determine the fault type based on multiple abnormal peak segments and feedback the fault type to the fault self-repair strategy module, so that the fault self-repair strategy module takes corresponding fault elimination measures. When evaluating the abnormal fault risk, mainly based on the coupling degree, combined with multi-dimensional data such as the frequencies of abnormal peak segments and the mean power corresponding to abnormal peak segments, the risk assessment value is calculated by weighted calculation. The weighted values are obtained by continuously optimizing the deep learning fault diagnosis module composed of a convolutional neural network (CNN) and a long short-term memory network (LSTM) through a large amount of historical data. However, it should be noted that the abnormal fault risk valuation is only used as a trigger value for abnormal fault diagnosis and does not involve the analysis and judgment of abnormal fault types. The analysis and judgment of abnormal fault types are made by the deep learning fault diagnosis module based on the frequencies of multiple abnormal peak segments, the mean power corresponding to the multiple abnormal peak segments, and the coupling degree. That is, the frequencies of multiple abnormal peak segments, the mean power corresponding to the multiple abnormal peak segments, and the coupling degree are used twice in this abnormal fault diagnosis process. The first time is to calculate the abnormal fault risk valuation. When the abnormal fault risk valuation is greater than the preset safety threshold, a specific fault type identification module is called. This module is the core functional module of the deep learning fault diagnosis module. Once called, it requires a large amount of computing power of the processor, resulting in a decrease in the response of the processor in other functional modules. Therefore, a call threshold (abnormal fault risk valuation greater than the preset safety threshold) is set. It is emphasized here that the weights of the frequencies of multiple abnormal peak segments, the mean power corresponding to the multiple abnormal peak segments, and the coupling degree in this embodiment change dynamically according to the accumulation of new historical data received by the model, rather than being preset by developers. And each change in the weight does not exceed 5% of the original weight, thereby preventing short-term abnormal charging data from overcorrecting the basic weight.

[0028] Secondly, in step S500 of this embodiment, the specific implementation method for the deep learning fault diagnosis module to determine the fault type based on multiple abnormal peak segments and feedback the fault type to the fault self-repair strategy module so that the fault self-repair strategy module takes corresponding fault elimination measures is as follows: Step S510: Merge multiple abnormal peak segments based on the time series, and separate the output current fluctuation curve and the output voltage fluctuation curve based on the merged multiple abnormal peak segments. Step S520: Input the output current fluctuation curve and the output voltage fluctuation curve into a preset first deep learning fault diagnosis sub-model, so that it analyzes the high-frequency noise distribution corresponding to the output current fluctuation curve and the high-frequency noise distribution corresponding to the output voltage fluctuation curve, and combines the high-frequency noise distribution in the historical training data to respectively determine the high-frequency noise increments of the current and / or voltage corresponding to multiple abnormal peak segments. If the high-frequency noise increment is within the first abnormal gradient threshold range, it is determined that the output filter capacitor has a high-temperature abnormality, and the output filter capacitor high-temperature abnormality is fed back to the fault self-repair strategy module, so that it increases the operating power of the cooling fan corresponding to the power output module in the charging pile, thereby reducing the operating temperature of the output filter capacitor and reducing the decline in high-frequency filtering performance caused by excessive temperature. Step S530: If the high-frequency noise increment is less than or equal to the lower limit of the first abnormal gradient threshold range, retrieve the DC bus voltage fluctuation data corresponding to multiple abnormal peak segments and input it into the second deep learning fault diagnosis sub-model, so that it sequentially constructs the voltage ripple corresponding to each abnormal peak segment, and combines the historical training data to determine the comprehensive change of the DC bus voltage ripple corresponding to multiple abnormal peak segments. If the DC bus voltage ripple corresponding to multiple abnormal peak segments as a whole shows an increasing trend and exceeds the preset growth threshold, it is determined that the DC bus capacitor has a high-temperature abnormality, and the DC bus capacitor high-temperature abnormality is fed back to the fault self-repair strategy module, so that it increases the operating power of the cooling fan corresponding to the power supply module in the charging pile, thereby reducing the operating temperature of the DC bus capacitor and reducing the reduction in the DC bus capacitor capacity caused by excessive temperature.

[0029] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A charging pile fault self-repair control system, characterized in that: The system comprises: The data acquisition and preprocessing module is composed of current sensors, voltage sensors, temperature sensors and vehicle-pile communication data acquisition units installed at key locations of the charging pile. It is used to collect charging pile operation data in real time and preprocess the data. Deep learning fault diagnosis module, which is used to extract local features and time series dependencies from preprocessed data and output fault type diagnosis results; Fault self-repair strategy module: executes software restart, parameter adjustment or hardware backup circuit switching operations according to the fault diagnosis results.

2. The charging pile fault self-repair control system according to claim 1, characterized in that: It also includes a system control and communication module: coordinate the operation processes of each module, and interact with the charging pile management platform and the user end in real time to determine the fault status and repair information.

3. The charging pile fault self-repair control system according to claim 2, characterized in that: The method for extracting local features and time series dependencies from preprocessed data and outputting fault type diagnosis results includes: Acquire charging power fluctuation data and a preset power curve within a first time period, and construct an actual charging trend curve corresponding to the charging power fluctuation data through a linear regression algorithm; Detect the difference fluctuation between the actual charging trend curve and the preset power curve based on the time series, identify multiple first time nodes whose difference fluctuation is greater than a preset threshold, and based on the distribution of the multiple first time nodes, select multiple first time nodes that are continuous and whose continuous duration is greater than the preset unit duration, and then select multiple time series segments to be corrected; Secondarily anchoring the power value of the first time node in the preset power curve based on the percentage of the difference between the preset power curve and the actual charging trend curve in the time series segment to be corrected at each first time node in the preset threshold, thereby correcting the preset power curve; A power fluctuation curve band corresponding to the first time period is constructed based on the revised preset power curve and the preset power fluctuation percentage threshold, and a plurality of abnormal peak segments are screened out from the charging power fluctuation data based on the power fluctuation curve band, and a coupling degree between the revised preset power curve and the charging power fluctuation data is calculated, wherein the abnormal peak segment is a power fluctuation segment that is not within the power fluctuation curve band; Based on the frequency of occurrence of multiple abnormal peak segments, the mean power corresponding to the multiple abnormal peak segments and the coupling degree, a weighted algorithm is used to calculate the abnormal fault risk valuation, and when the abnormal fault risk valuation is greater than a preset safety threshold, a first charging power abnormal fault code is generated. The first charging power abnormal fault code is used to enable the deep learning fault diagnosis module to determine the fault type based on multiple abnormal peak segments, and to feed back the fault type to the fault self-repair strategy module, so that the fault self-repair strategy module can take corresponding fault elimination measures.

4. The charging pile fault self-repair control system according to claim 3, characterized in that: The deep learning fault diagnosis module determines the fault type based on multiple abnormal peak segments, and feeds back the fault type to the fault self-repair strategy module, so that the fault self-repair strategy module takes corresponding fault elimination measures, including: Merging multiple abnormal peak segments based on the time series, and separating the output current fluctuation curve and the output voltage fluctuation curve based on the merged multiple abnormal peak segments; The output current fluctuation curve and the output voltage fluctuation curve are input into a preset first deep learning fault diagnosis sub-model, so that it can analyze the high-frequency noise distribution corresponding to the output current fluctuation curve and the high-frequency noise distribution corresponding to the output voltage fluctuation curve, and combine the high-frequency noise distribution in the historical training data to respectively determine the high-frequency noise increments of the current and / or voltage corresponding to the multiple abnormal peak segments. If the high-frequency noise increment is within the first abnormal gradient threshold range, it is determined that the output filter capacitor has a high temperature abnormality, and the high temperature abnormality of the output filter capacitor is fed back to the fault self-repair strategy module, so that it increases the operating power of the cooling fan corresponding to the power output module in the charging pile, thereby reducing the operating temperature of the output filter capacitor and reducing the degradation of the high-frequency filtering performance caused by excessive temperature; If the high-frequency noise increment is less than or equal to the lower limit of the first abnormal gradient threshold range, the DC bus voltage fluctuation data corresponding to multiple abnormal peak segments are retrieved and input into the second deep learning fault diagnosis sub-model, so that it constructs the voltage ripple corresponding to each abnormal peak segment in turn, and determines the comprehensive change of the DC bus voltage ripples corresponding to the multiple abnormal peak segments in combination with the historical training data. If the DC bus voltage ripples corresponding to the multiple abnormal peak segments show an overall increasing trend and exceed the preset growth threshold, it is determined to be a high temperature abnormality of the DC bus capacitor, and the high temperature abnormality of the DC bus capacitor is fed back to the fault self-repair strategy module to increase the operating power of the cooling fan corresponding to the power module in the charging pile, thereby reducing the operating temperature of the DC bus capacitor and reducing the reduction in the DC bus capacitor capacity caused by excessive temperature.

5. The charging pile fault self-repair control system according to claim 4, characterized in that: The deep learning fault diagnosis module is composed of a convolutional neural network and a long short-term memory network. The convolutional neural network includes three parallel convolutional layers, which use 3×3, 5×5 and 7×7 convolution kernels respectively to extract local features of different scales; the long short-term memory network layer includes two series-connected long short-term memory network units, which contain 64 and 32 memory units respectively, to capture the long-term dependencies of time series data.

6. The charging pile fault self-repair control system according to claim 5, characterized in that: The fault self-repair strategy module includes: software repair unit: used to execute program restart or parameter configuration update; hardware repair unit: switches to the backup charging circuit by controlling the switch circuit to maintain the basic functions of the charging pile; complex fault reporting unit: sends fault information that cannot be automatically repaired to the management platform and triggers manual intervention.

Images