A method for solving charge failure refund

Abstract

The present application relates to the technical field of charging equipment, in particular to a method for solving charging failure refund, which comprises: in the first stage, determining a reference screening strategy based on key distribution density and key timing deviation, determining an optimal accounting data set through key reproduction or characteristic analysis, and completing historical data prediction and optimization solution by using a target prediction model; in the second stage, determining a real-time data compensation method according to the proportion of data missing and the distribution index, and outputting a real-time refund scheme by using fitting execution parameters or reinforcement learning algorithm. The present application can effectively solve the problem of data missing caused by limited communication link, realize accurate and automatic refund decision under complex fault working conditions, significantly improve the accounting efficiency, shorten the refund period, reduce the dependence on manual checking, and have high popularization significance.

Description

Technical Field

[0001] This invention relates to the field of charging equipment technology, and more specifically, to a method for resolving charging failure refunds. Background Technology

[0002] With the rapid development of the electric vehicle industry and the popularization of charging infrastructure, the user charging billing and refund process is highly dependent on network communication and data interaction with cloud servers. In the conventional process, the charging equipment needs to upload core data such as user information, equipment number, and charging start time to the cloud to generate an order, and synchronize electricity and fee data during the charging process. After the charging is completed, the server completes the final cost calculation and settlement.

[0003] In real-world operating environments, factors such as charging module damage, communication module failure, or network outages can restrict the communication link between charging equipment and the cloud server. When this communication link is disrupted, the real-time generation, updating, or uploading of charging orders can be easily interfered with. In such cases, prepaid fees or pre-authorized amounts often lack real-time order data support, leading to incomplete data when the existing system automatically processes refunds. Currently, such anomalies typically require manual processing by checking local logs of the charging equipment and user payment vouchers. This approach has room for improvement in verification efficiency, refund cycle, and data accuracy, and also increases the operational and management costs of charging stations to some extent.

[0004] Therefore, optimizing the refund mechanism in charging failure scenarios and improving the automation of data processing have become key technological directions of current industry focus. Summary of the Invention

[0005] In view of this, the present invention proposes a method for resolving charging failure refunds. It aims to address the problems in the current technology where charging equipment experiences abnormal operating conditions such as hardware damage, communication module failure, or network interruption. Due to the limited communication link between the equipment and the cloud server, the charging order data is not fully uploaded, resulting in the cloud system lacking real-time order data support to execute automatic refund operations. Ultimately, this leads to a high degree of reliance on manual verification of local logs and payment vouchers, low verification efficiency, extended refund cycles, and difficulty in ensuring data accuracy.

[0006] This invention proposes a method for resolving charging failure refunds, comprising: Under the first stage conditions, the reference screening strategy determined based on the key distribution density and key time series deviation is to perform key reproduction analysis or key characteristic analysis on the collected data set to determine the preferred accounting dataset; the collected data set is collected and stored by a local storage module set inside the charging device through a bus protocol. The local storage module is electrically connected to the signal output terminal of the charging controller and is used to record the charging voltage, charging current, cumulative power value and communication link status bit in real time. During critical recurrence analysis, the interval recurrence stability coefficient is determined based on the interval persistence index and the interval recurrence frequency. The data extraction strategy is determined based on the interval recurrence stability coefficient, either by determining the set of recurrence analysis intervals based on the correlation between interval data and the critical similarity between intervals, or by determining the extraction coverage index based on the reference interval persistence index and the critical time series deviation. The interval persistence index is determined by the ratio of the duration of the abnormal charging interruption to the preset baseline duration, and the interval recurrence frequency is determined by the frequency of occurrence of a specific fault code within a preset observation period. During key characteristic analysis, the set of characteristic analysis intervals is determined based on the interference correlation index; the interference correlation index is determined by calculating the cross-correlation coefficient between the noise power spectral density of the communication link and the power spectral density of the normal signal. Under the first accounting execution conditions, the first stage analysis is carried out. Based on each target prediction model, the relevant data of charging equipment operation during the accounting execution cycle are predicted, and the optimization solution is completed based on the system constraints. The target prediction models include the charging current evolution model, the battery state of charge change model, and the power distribution model. Under the second stage conditions, the real-time data compensation method is determined periodically based on the data missing ratio and the missing distribution index. This is done by determining the fitting execution parameters for each data missing time based on the interval change index and the change reference index, or by determining whether to perform data prediction compensation for each data missing stage based on the relevant change stability coefficient. The data missing ratio is the ratio of the number of data frames not uploaded during the communication interruption time to the total number of preset data frames. Under the second prediction execution condition, the second stage analysis is carried out. Based on the prediction model of each target, the relevant data of charging equipment operation within the real-time accounting period are predicted, and the short-time scale refund control strategy is trained based on the reinforcement learning algorithm to output the real-time refund scheme. The action space of the reinforcement learning algorithm is defined as the refund amount step value and the refund trigger time point, and the state vector is defined as the current calculated electricity, prepaid balance and abnormal duration.

[0007] Furthermore, under the first-stage conditions, periodic key reproducibility analysis is performed on the collected data set to determine the key distribution density and key temporal deviations of the target optimization system. When determining the reference screening strategy based on the key distribution density and key temporal deviations, this includes: The key distribution density is the proportion of the number of key monitoring points to the number of data monitoring points in the screening and analysis period; a key monitoring point is defined as the sampling time point when the fluctuation amplitude of the charging current exceeds the preset fluctuation threshold. The key timing deviation is determined based on the time interval between key monitoring points; specifically, it is calculated as the degree of deviation between the time difference between two adjacent key monitoring points and the preset standard sampling interval. The first phase involves optimizing the target system by backtracking historical fault data and conducting financial reconciliation over a long period of time.

[0008] Furthermore, if the critical distribution density of the screening analysis period is greater than the preset critical distribution density and the critical time series deviation is greater than the preset critical time series deviation, then it is determined that a critical reproducibility analysis will be performed on the collected data set, including: The interval recurrence stability coefficient of the screening and analysis period is tested, and the data extraction strategy is determined based on the interval recurrence stability coefficient. The interval recurrence stability coefficient is determined based on the interval persistence index and interval recurrence frequency of each key distribution interval within the screening and analysis period; the key distribution interval is defined as the continuous time period in which the communication link status bit is marked as abnormal.

[0009] Furthermore, when the interval recurrence stability coefficient of the screening analysis period is greater than the preset interval recurrence stability coefficient, the set of recurrence analysis intervals is determined based on the correlation of interval data and the key similarity of intervals, including: The extraction sliding index is determined based on the reference recurrence frequency index and the set overlap index, and the extraction coverage index is determined based on the set reference persistence index of each recurrence analysis interval set. The extraction sliding index is negatively correlated with the reference recurrence frequency index and the set overlap index, respectively; the set overlap index is determined by calculating the overlap length of adjacent recurrence analysis intervals on the time axis; the extraction coverage index is used to determine the window width for extracting data from the local log.

[0010] Furthermore, when the interval recurrence stability coefficient of the screening analysis period is less than or equal to the preset interval recurrence stability coefficient, the coverage index is determined based on the reference interval persistence index and key time series deviations, and the sliding index is determined based on the interval recurrence frequency difference. The coverage index was positively correlated with the reference interval persistence index and the key time series deviation, while the sliding index was negatively correlated with the interval recurrence frequency difference. The interval recurrence frequency difference was determined by the absolute value of the difference between the fault frequency of the current observation period and the fault frequency of the same period in history.

[0011] Furthermore, if the critical distribution density of the screening analysis period is less than or equal to the preset critical distribution density or the critical time series deviation is less than or equal to the preset critical time series deviation, then when performing critical characteristic analysis on the collected data set, it includes: The set of characteristic analysis intervals is determined based on the interference correlation index, and the extraction coverage index is determined based on the interference correlation index and the interval persistence index. The extraction sliding index is determined based on the interval recurrence stability coefficient. The higher the interference correlation index, the higher the degree of interference from the external electromagnetic environment is judged, and the extraction coverage index is increased accordingly to obtain more redundant data for cross-validation.

[0012] Furthermore, under the first accounting execution conditions, the first-stage analysis includes: The reference dataset from the same period of the accounting execution cycle is used as the input data for each target prediction model, and the output data of each target prediction model is recorded as the prediction data. Linear weighting is performed, and optimization is achieved based on system constraints. System constraints include user prepayment balance limits, minimum consumption limits for a single charge, and grid-side time-of-use pricing policy constraints. The concurrent reference dataset is the prediction reference dataset corresponding to the long-scale cycle that has the largest overlap with the accounting execution cycle. The first accounting execution condition is that there is an accounting execution cycle that requires prediction of missing power values. The target prediction models include the charging current evolution model, the battery state of charge change model, and the power allocation model.

[0013] Furthermore, under the second stage conditions, when periodically determining the real-time data compensation method based on the proportion of missing data and the missing data distribution index, it includes: If the percentage of missing data in the compensation analysis period is less than or equal to the preset percentage of missing data and the missing data distribution index is less than or equal to the preset missing data distribution index, the fitting execution parameters for each missing data collection time are determined based on the interval change index and the change reference index. The interval change index is determined by the slope of the power before and after the missing interval, and the change reference index is determined by the slope of the standard charging curve of the same model of charging equipment at the same power level. The second phase condition is to optimize the system for real-time automated refund processing on a short time scale.

[0014] Furthermore, if the proportion of missing data in the compensation analysis period is greater than the preset proportion of missing data or the missing data distribution index is greater than the preset missing data distribution index, then it is determined whether to perform data prediction compensation for each missing data stage based on the relevant change stability coefficient. For a single data missing stage, if the relevant change stability coefficient is greater than the preset relevant change stability coefficient, the estimated change range of the compensation relevant data for that data missing stage is determined based on the stage change range of the reference relevant data; the relevant change stability coefficient is determined by calculating the reciprocal of the variance of the charging power in the preset period before the missing stage.

[0015] Furthermore, under the second prediction execution condition, the second-stage analysis includes: The charging equipment operation dataset corresponding to the real-time reference cycle is used as the input data of each target prediction model, and the output data of each target prediction model is recorded as the real-time prediction data set corresponding to the real-time accounting cycle. The operating state vector is determined based on real-time operating state parameters and real-time prediction data, and the action space is set. A comprehensive reward function is determined based on the operation and maintenance costs of charging stations and user refund experience ratings, and a short-timescale optimization control strategy is trained based on reinforcement learning algorithms. The second prediction execution condition is that the current time is the start time of a second-stage cycle, and this second-stage cycle is recorded as the real-time accounting cycle, and the previous second-stage cycle is recorded as the real-time reference cycle.

[0016] In its implementation, this invention executes the aforementioned logic through a central processing unit (CPU) located inside the charging pile. The CPU connects to the electricity meter, charging module, insulation detection module, and human-machine interface terminal via an RS485 bus or CAN bus. When a hardware over-temperature, over-current, or short-circuit fault in the charging module causes abnormal termination of the charging process, the CPU immediately triggers an interrupt routine, writing the current billing information, fault code, and the last valid electricity sample value into the non-volatile memory (NVM).

[0017] In the event of a communication link interruption, the central processing unit (CPU) initiates a local independent calculation program. By acquiring historical charging characteristic curves stored in non-volatile memory and utilizing the key reproducibility analysis from the first stage, it identifies whether the current fault mode belongs to a known frequently occurring communication timeout type. If it is determined to be a high-stability reproducible fault, a preset extraction coverage index is invoked to accurately extract the instantaneous voltage and current values ​​before and after the fault from the massive amount of locally stored sampled data.

[0018] For missing data segments, the central processing unit (CPU) invokes the second-stage real-time data compensation logic. By calculating the percentage of missing data, if the percentage is within a preset low range, Lagrange interpolation or least-squares fitting algorithms are used, combined with an interval change index, to smoothly fit the missing power data. If the percentage exceeds a preset threshold, the CPU switches to a reinforcement learning-based prediction compensation mode. By comparing power fluctuation characteristics within the real-time reference period, and guided by a comprehensive reward function, the CPU dynamically adjusts the refund amount's increment until the optimal real-time refund solution is output.

[0019] After the refund plan is generated, the central processing unit (CPU) encapsulates it into a high-priority encrypted data packet and stores it in the transmission buffer. Once the communication module detects that the network signal has been restored (i.e., the communication link status bit returns to normal), the CPU immediately uploads the encrypted data packet to the cloud server via the MQTT or OCPP protocol. The cloud server parses the compensation calculation result in the data packet and automatically triggers the refund interface of the third-party payment platform to complete the refund.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: First, this invention, by introducing analysis mechanisms for key distribution density and key timing deviations, enables in-depth mining of multi-source heterogeneous data in charging fault scenarios, effectively solving the data silo problem caused by communication interruptions. Through differentiated strategies of key reproducibility analysis and key characteristic analysis, the optimal data extraction path can be automatically selected based on the stability and interference level of the fault, significantly improving the optimization efficiency and completeness of the accounting dataset, and providing a solid data foundation for subsequent automated refunds.

[0021] Second, this invention employs a two-stage prediction and compensation architecture. The first stage focuses on long-term historical backtracking and model optimization, ensuring the compliance and accuracy of the accounting results at the macro level through the coupled solution of the target prediction model and system constraints. The second stage focuses on real-time response at a short time scale, dynamically adjusting the compensation method using the data missing ratio and missing distribution index, and combining adaptive training of reinforcement learning algorithms to achieve accurate refund decisions under complex and variable fault conditions, greatly reducing the reliance on manual verification logic.

[0022] Third, this invention constructs a rigorous logical closed loop through a series of refined technical parameters, such as the interval reproducibility stability coefficient, the interval change index, and the related change stability coefficient. This data feature-driven automated processing mechanism not only effectively shortens the refund waiting period for users after charging anomalies and improves user satisfaction, but also significantly reduces the operation and management costs of charging stations and the potential human verification error rate by reducing manual intervention, thus possessing extremely high practical value and industrial promotion significance. Attached Figure Description

[0023] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a method for resolving charging failure refunds according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for resolving charging failure refunds according to an embodiment of the present invention. Detailed Implementation

[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] like Figures 1-2 As shown in some embodiments of this application, this embodiment provides a method for resolving charging failure refunds, including: Step S100: Under the first stage conditions, based on the key distribution density and key time series deviation, the reference screening strategy is to perform key reproducibility analysis or key characteristic analysis on the collected data set to determine the preferred accounting dataset.

[0026] Specifically, under the first-stage condition, periodic key reproducibility analysis is performed on the collected data set to determine the key distribution density and key time series deviation of the target optimization system. When determining the reference screening strategy based on the key distribution density and key time series deviation, the following are included: the key distribution density is the proportion of the number of key monitoring points in the number of data monitoring points within the screening analysis period; the key time series deviation is determined based on the time interval between key monitoring points; the first-stage condition is to perform historical fault data backtracking and financial verification optimization on a long-term scale for the target optimization system.

[0027] Specifically, if the critical distribution density of the screening and analysis period is greater than the preset critical distribution density and the critical time series deviation is greater than the preset critical time series deviation, then it is determined to perform critical recurrence analysis on the collected data set, including: detecting the interval recurrence stability coefficient for the screening and analysis period, and determining the data extraction strategy based on the interval recurrence stability coefficient; the interval recurrence stability coefficient is determined based on the interval persistence index and interval recurrence frequency of each critical distribution interval within the screening and analysis period.

[0028] Understandably, quantifying key distribution density and key timing deviation to guide data filtering strategies can improve the accuracy of the accounting model in identifying abnormal operating conditions. First, this invention analyzes the collected data set under the first-stage conditions, periodically calculating key distribution density and key timing deviation to determine the representativeness and temporal continuity of the fault data. Specifically, key distribution density represents the proportion of key moment points where charging current fluctuations exceed a threshold to the total number of monitoring points, used to measure the coverage of fault features in the data set; key timing deviation is calculated through the time interval between key monitoring points, reflecting the degree of deviation of data records in the charging device under abnormal communication conditions. These two indicators quantify the importance of the data set in both the state space and time dimensions. Second, when both key distribution density and key timing deviation exceed a preset step threshold within the filtering analysis period, it is determined that key recurrence analysis is required. In this analysis, the reliability of the fault data interval is evaluated by calculating the interval recurrence stability coefficient. The interval recurrence stability coefficient is determined based on the duration of the charging anomaly interruption and the frequency of occurrence of specific fault codes, used to measure the stability of key faults in both time and space dimensions. Finally, based on the interval recurrence stability coefficient, a data extraction strategy is dynamically determined, that is, the most representative interval data is selected from the local storage module as the preferred accounting dataset. This strategy ensures that the accounting model can focus on the most critical feature data of the charging facility fault status, thereby improving the accuracy of refund calculation.

[0029] It can be seen that by establishing a quantitative indicator system (key distribution density, key time series deviation, interval reproducibility stability coefficient) and a dynamic data extraction strategy, the automatic identification and optimization of key features in the local logs of charging equipment are realized, thereby providing high-quality data support for charging facility failure refunds.

[0030] In step S200, during key recurrence analysis, the interval recurrence stability coefficient is determined based on the interval persistence index and interval recurrence frequency. The data extraction strategy is determined based on the interval recurrence stability coefficient, either by determining the set of recurrence analysis intervals based on the interval data correlation and interval key similarity, or by determining the extraction coverage index based on the reference interval persistence index and key time series deviation. During key characteristic analysis, the characteristic analysis interval set is determined based on the interference correlation index.

[0031] Specifically, when the interval recurrence stability coefficient of the screening analysis period is greater than the preset interval recurrence stability coefficient, the set of recurrence analysis intervals is determined based on the correlation of interval data and the key similarity of intervals, including: determining the extraction sliding index based on the reference recurrence frequency index and the set overlap index, and determining the extraction coverage index based on the set reference persistence index of each set of recurrence analysis intervals; the extraction sliding index is negatively correlated with the reference recurrence frequency index and the set overlap index, respectively.

[0032] Specifically, when the interval recurrence stability coefficient of the screening analysis period is less than or equal to the preset interval recurrence stability coefficient, the extraction coverage index is determined based on the reference interval persistence index and the key time series deviation, and the extraction sliding index is determined based on the interval recurrence frequency difference. The extraction coverage index is positively correlated with the reference interval persistence index and the key time series deviation, respectively, while the extraction sliding index is negatively correlated with the interval recurrence frequency difference.

[0033] Specifically, if the key distribution density of the screening analysis period is less than or equal to the preset key distribution density or the key time series deviation is less than or equal to the preset key time series deviation, then when performing key characteristic analysis on the collected data set, the following steps are taken: determining the characteristic analysis interval set based on the interference correlation index, determining the extraction coverage index based on the interference correlation index and the interval persistence index, and determining the extraction sliding index based on the interval recurrence stability coefficient.

[0034] Understandably, using multidimensional quantitative indicators to guide the selection and extraction of data intervals provides representative data for the accounting model. First, during critical recurrence analysis, the interval recurrence stability coefficient for each analysis period is calculated based on the interval persistence index and interval recurrence frequency to assess the spatiotemporal repeatability of fault intervals. When the interval recurrence stability coefficient is higher than a preset threshold, it indicates that the fault in that interval has high regularity. At this point, the data extraction strategy determines the set of recurrence analysis intervals based on interval key similarity and calculates a sliding index based on the set overlap index to further control the movement step of the extraction window, ensuring that the selected data avoids excessive overlap. It is worth noting that the extraction sliding index is negatively correlated with the reference recurrence frequency index, reflecting a mechanism to suppress redundant data. Second, when the interval recurrence stability coefficient is lower than or equal to the preset threshold, it indicates insufficient fault stability. Therefore, the extraction coverage index is determined by the reference interval persistence index and key time series deviation to ensure sufficient coverage of key intervals; the sliding index is negatively correlated with the interval recurrence frequency difference, thus suppressing repeated sampling of unstable intervals. Finally, when the key distribution density or key time series deviation is lower than the preset threshold, the strategy shifts to key characteristic analysis. By calculating the interference correlation index to determine the set of characteristic analysis intervals, this method can still identify interval features that are representative of the charging state even when the communication link is severely affected by external electromagnetic interference, thus ensuring the comprehensiveness of data screening.

[0035] It can be seen that by establishing a multi-dimensional quantitative index system such as the interval reproducibility stability coefficient and the interference correlation index, and combining the positive and negative correlation sliding index and the coverage index, dynamic interval filtering of local data of charging equipment can be achieved, thereby providing accurate training data and providing reliable support for charging failure refunds.

[0036] Step S300: Under the first accounting execution conditions, the first stage analysis is carried out. Based on the prediction models of each target, the relevant data of the charging equipment operation during the accounting execution cycle are predicted, and the optimization solution is completed based on the system constraints.

[0037] Specifically, under the first accounting execution condition, the first stage analysis includes: using the reference dataset of the same period of the accounting execution cycle as the input data of each target prediction model, and recording the output data of each target prediction model as the prediction data; performing linear weighting processing, and optimizing the solution based on system constraints; the reference dataset of the same period is the prediction reference dataset corresponding to the long-scale cycle with the largest overlap with the accounting execution cycle; the first accounting execution condition is that there is an accounting execution cycle that requires prediction of missing power values, and the target prediction models include the charging current evolution model, the battery state of charge change model, and the power allocation model.

[0038] Understandably, this involves using a multi-objective prediction model to accurately trace the charging state over long time scales. First, under the initial accounting execution conditions, the first stage of analysis and processing is performed. The basic idea is to use a concurrent reference dataset as input to each objective prediction model. This concurrent reference dataset refers to historical long-scale period data with the highest overlap with the current accounting period, providing representative charging characteristics. The objective prediction models include a charging current evolution model, a battery state-of-charge change model, and a power allocation model, capable of independently predicting current trends, battery capacity changes, and device power output during the charging process. Second, the predicted data output by each model undergoes linearized weighting. By combining the prediction results from different dimensions according to preset weights, multi-model information can be integrated to achieve a more comprehensive prediction result. Finally, based on the prediction results and system constraints (such as user prepayment balance limits and minimum consumption limits for single charging), optimization is performed. The optimization process aims to adjust the predicted values ​​of various operating parameters while meeting financial constraints, ensuring the accuracy and compliance of the accounting results. This method enables retrospective prediction of the operating status of charging equipment over a long time scale, providing a scientific basis for financial verification.

[0039] It can be seen that by utilizing a multi-objective prediction model, a concurrent reference dataset, and a constraint optimization solution mechanism, accurate reconstruction of charging fault data was achieved, ensuring the security of financial accounting.

[0040] In step S400, under the second stage conditions, the real-time data compensation method is periodically determined based on the data missing percentage and the missing distribution index. This involves determining the fitting execution parameters for each data missing moment based on the interval change index and the change reference index, or determining whether to perform data prediction compensation for each data missing stage based on the relevant change stability coefficient. Under the second prediction execution condition, the second stage analysis is performed. Based on each target prediction model, the relevant data of charging equipment operation within the real-time accounting period are predicted, and the refund control strategy is trained on a short time scale based on the reinforcement learning algorithm to output a real-time refund scheme.

[0041] Specifically, under the second stage condition, when determining the real-time data compensation method periodically based on the data missing percentage and the missing distribution index, it includes: if the data missing percentage of the compensation analysis period is less than or equal to the preset data missing percentage and the missing distribution index is less than or equal to the preset missing distribution index, the fitting execution parameters for each missing data collection time are determined based on the interval change index and the change reference index; the second stage condition is to perform real-time automated refund processing on a short time scale for the target optimization system.

[0042] Specifically, if the proportion of missing data in the compensation analysis period is greater than the preset proportion of missing data or the missing data distribution index is greater than the preset missing data distribution index, then it is determined whether to perform data estimation compensation for each missing data stage based on the relevant change stability coefficient; for a single missing data stage, if the relevant change stability coefficient is greater than the preset relevant change stability coefficient, then the estimated change range of the compensation related data for that missing data stage is determined based on the stage change range of the reference related data.

[0043] Specifically, under the second prediction execution condition, the second-stage analysis includes: using the charging equipment operation dataset corresponding to the real-time reference cycle as the input data for each target prediction model, and recording the output data of each target prediction model as the real-time prediction data set corresponding to the real-time accounting cycle; determining the operation state vector based on the real-time operation state parameters and the real-time prediction data, and setting the action space; determining the comprehensive reward function based on the operation and maintenance cost of the charging station and the user refund experience score, and training the short-timescale optimization control strategy based on the reinforcement learning algorithm; the second prediction execution condition is that the current time is the start time of a second-stage cycle, and this second-stage cycle is recorded as the real-time accounting cycle, and the previous second-stage cycle is recorded as the real-time reference cycle.

[0044] Understandably, in the second stage, to address the potential for severe data gaps in real-time data collection, the percentage of missing data is periodically assessed. When the percentage of missing data is below a threshold, indicating a low level of data loss, the fitting execution parameters are determined using the interval change index and the change reference index, and the missing power is smoothly filled using Lagrange interpolation or the least squares fitting algorithm. Secondly, when the percentage of missing data exceeds the threshold, the relevant change stability coefficient is used to assess whether to perform pre-calculation compensation. If the relevant change stability coefficient is high, the trend of missing data change is predicted based on the inverse of the variance of charging power in the preset period before the data loss, ensuring the robustness of the compensation results. Under the second prediction execution condition, a real-time prediction data set is generated based on the target prediction model. Subsequently, by combining the real-time operating state parameters with the prediction data, a state vector is constructed and an action space is set. A comprehensive reward function is further defined, taking into account both the refund period and data accuracy scoring, and trained using a reinforcement learning algorithm to achieve automated refund decisions within the real-time calculation period.

[0045] It can be seen that by establishing a data missing assessment and dynamic compensation mechanism, combined with real-time prediction models and reinforcement learning optimization, high-precision and short-cycle processing of charging failure refunds has been achieved, enhancing the system's real-time response capability to abnormal states.

[0046] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principles of this invention are further supplemented below with a specific application scenario.

[0047] Step 1, Data Acquisition and Status Quantization, begins with the charging pile's internal central processing unit 1 acquiring real-time power data from the energy meter 2 and the current and voltage parameters of the charging module via the RS485 bus protocol. In the event of a network interruption or communication module 4 failure, the central processing unit 1 synchronously stores the data frames that cannot be uploaded into non-volatile memory (NVM) 3. At this time, the central processing unit 1 periodically calculates and filters the critical distribution density within the analysis period, i.e., the proportion of sampling points where current fluctuations exceed a preset threshold among the total sampling points. Simultaneously, it determines critical timing deviations by calculating the difference between the time interval between adjacent critical monitoring points and the standard sampling period. Through this quantization method, the central processing unit 1 can identify which time periods in the locally stored data contain critical information about sudden changes in charging status, thus providing a logical basis for subsequent data filtering.

[0048] In step two, during the extraction and strategy adjustment of the optimal dataset, the CPU 1 determines whether to enter the critical reproducibility analysis process based on the critical distribution density and critical timing deviation calculated in step one. The CPU 1 determines the interval duration index by calculating the ratio of the duration of the charging anomaly interruption to the baseline duration, and obtains the interval reproducibility stability coefficient by combining this with the frequency of fault code occurrence. If this coefficient indicates high reproducibility of the fault, the CPU 1 reduces the extraction sliding index through a negative correlation adjustment mechanism, thereby controlling the step frequency of data extraction in the non-volatile memory 3 and avoiding redundant sampling. Simultaneously, the extraction coverage index is adjusted according to the set overlap index to ensure that the extracted optimal dataset can fully cover the key features of the voltage drop segment and the current zeroing segment. If severe electromagnetic interference is determined, the CPU 1 expands the extraction coverage index by increasing the interference correlation index, utilizing more redundant data for cross-comparison, thereby achieving accurate extraction of effective charging data in complex electromagnetic environments.

[0049] Step 3, Multi-model Prediction and Missing Data Imputation: During this step, CPU 1 initiates the first stage of analysis logic. For the power loss segment caused by communication interruption, CPU 1 retrieves historical reference datasets from non-volatile memory 3 and inputs them into the charging current evolution model and the battery state-of-charge change model. Through linearized weighted processing, the theoretical charging curve during the fault period is predicted. Under the second stage conditions, CPU 1 monitors the data loss ratio in real time. If the loss ratio is below a preset threshold, it calculates the interval change index using the power slope before and after the missing interval and uses a least-squares fitting algorithm to smoothly fit and imput the power at the missing time. If the loss ratio is too high, CPU 1 calculates the inverse of the variance of the charging power at the previous time to obtain the relevant change stability coefficient, and determines whether to perform pre-compensation, thereby achieving a robust estimate of the total charging amount under severe data loss conditions.

[0050] Step 4: During strategy training and refund scheme execution, CPU 1 enters the second prediction execution phase. CPU 1 encapsulates the real-time prediction data set, the current prepaid balance, and the duration of the anomaly into a state vector. Using a pre-set reinforcement learning algorithm, it searches for the optimal refund amount step value and trigger time point within the action space. CPU 1 constructs a comprehensive reward function based on charging station operation and maintenance costs and user experience scores, and through continuous iterative training, outputs the optimal real-time refund scheme. Finally, CPU 1 encapsulates the generated refund scheme into an encrypted data packet. When communication module 4 detects that the network link has returned to normal, CPU 1 immediately uploads the encrypted data packet to the cloud server via the MQTT protocol. The server parses the packet and directly calls the refund interface, thus achieving automatic settlement and refund of fees in fault scenarios without manual intervention, based on local device records and prediction models.

[0051] All contents not described in detail in the specification are existing technologies known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited; conventional equipment can be used. Electrical control components not mentioned in this technical solution are not shown in the figures because they are existing technologies, and will not be described here.

[0052] In the above embodiments, a phased charging fault data fusion and accounting analysis method was used to achieve refined processing of abnormal charging conditions. First, in the first stage, high-quality accounting datasets were selected through key reproducibility analysis, improving the model's ability to capture fault characteristics. Second, under the first accounting execution conditions, a multi-objective prediction model and constraint optimization were used to ensure the accuracy of long-scale backtracking. In the second stage, a differentiated strategy of fitting compensation and prediction compensation was proposed to address the data missing problem, effectively improving data integrity. Finally, a real-time refund scheme generated by a reinforcement learning algorithm significantly reduced the reliance on manual verification logic and shortened the refund cycle. This not only improved user satisfaction but also enhanced the intelligent management level of charging stations, fully demonstrating the comprehensive beneficial effects of this invention in the automated processing of charging faults.

[0053] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0056] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for resolving charging failure refunds, characterized in that, include: Under the first stage conditions, the reference screening strategy determined based on key distribution density and key time series deviation is to conduct key recurrence analysis or key characteristic analysis on the collected data set to determine the preferred accounting dataset. Specifically: during key recurrence analysis, the interval recurrence stability coefficient is determined based on the interval persistence index and interval recurrence frequency. The data extraction strategy is determined based on the interval recurrence stability coefficient by determining the set of recurrence analysis intervals based on interval data correlation and interval key similarity, or by determining the extraction coverage index based on the reference interval persistence index and key time series deviation. During key characteristic analysis, the characteristic analysis interval set is determined based on the interference correlation index. Under the first accounting execution conditions, the first stage analysis is carried out. Based on each target prediction model, the relevant data of charging equipment operation during the accounting execution cycle are predicted, and the optimization solution is completed based on the system constraints. The target prediction models include the charging current evolution model, the battery state of charge change model, and the power distribution model. Under the second stage conditions, the real-time data compensation method is determined periodically based on the proportion of missing data and the missing data distribution index. This involves determining the fitting execution parameters for each missing data time based on the interval change index and the change reference index, or determining whether to perform data prediction compensation for each missing data stage based on the relevant change stability coefficient. Under the second prediction execution condition, the second stage analysis is carried out. Based on the prediction model of each target, the relevant data of charging equipment operation within the real-time accounting period are predicted, and the refund control strategy is trained on a short time scale based on the reinforcement learning algorithm to output a real-time refund scheme.

2. The method for resolving charging failure refunds according to claim 1, characterized in that, Under the first-stage conditions, periodic key reproducibility analysis is performed on the collected data set to determine the key distribution density and key time series deviations of the target optimization system. When determining the reference screening strategy based on the key distribution density and key time series deviations, the following steps are included: The key distribution density is the proportion of the number of key monitoring points to the total number of data monitoring points within the screening and analysis period. The key monitoring point is defined as the sampling time point when the fluctuation of the charging current exceeds the preset fluctuation threshold; The key timing deviation is determined based on the time interval between key monitoring points, specifically by calculating the degree of deviation between the time difference between two adjacent key monitoring points and the preset standard sampling interval. The first phase involves optimizing the target system by backtracking historical fault data and conducting financial reconciliation over a long period of time.

3. The method for resolving charging failure refunds according to claim 2, characterized in that, If the critical distribution density of the screening analysis period is greater than the preset critical distribution density and the critical time series deviation is greater than the preset critical time series deviation, then it is determined that a critical reproducibility analysis should be performed on the collected data set, including: The interval recurrence stability coefficient of the screening and analysis period is tested, and the data extraction strategy is determined based on the interval recurrence stability coefficient. The interval recurrence stability coefficient is determined based on the interval persistence index and interval recurrence frequency of each key distribution interval within the screening and analysis period; The critical distribution interval is defined as the continuous time period during which the status bits of the communication link are marked as abnormal.

4. The method for resolving charging failure refunds according to claim 3, characterized in that, When the interval recurrence stability coefficient of the screening analysis period is greater than the preset interval recurrence stability coefficient, the set of recurrence analysis intervals is determined based on the correlation of interval data and the key similarity of intervals, including: The extraction sliding index is determined based on the reference recurrence frequency index and the set overlap index, and the extraction coverage index is determined based on the set reference persistence index of each recurrence analysis interval set. The extracted sliding index is negatively correlated with both the reference recurrence frequency index and the set overlap index. The set overlap index is determined by calculating the overlap length of adjacent recurrence analysis intervals on the time axis; The extraction coverage index is used to determine the window width for extracting data from the local log.

5. The method for resolving charging failure refunds according to claim 4, characterized in that, When the interval recurrence stability coefficient of the screening analysis period is less than or equal to the preset interval recurrence stability coefficient, the coverage index is determined based on the reference interval persistence index and key time series deviations, and the sliding index is determined based on the interval recurrence frequency difference. The extracted coverage index is positively correlated with the reference interval persistence index and the key time series deviation, while the extracted sliding index is negatively correlated with the interval recurrence frequency difference. The interval recurrence frequency difference is determined by the absolute value of the difference between the fault frequency of the current observation period and the fault frequency of the same period in history.

6. The method for resolving charging failure refunds according to claim 5, characterized in that, If the critical distribution density of the screening analysis period is less than or equal to the preset critical distribution density, or the critical time series deviation is less than or equal to the preset critical time series deviation, then when performing critical characteristic analysis on the collected data set, it includes: The set of characteristic analysis intervals is determined based on the interference correlation index, and the extraction coverage index is determined based on the interference correlation index and the interval persistence index. The extraction sliding index is determined based on the interval recurrence stability coefficient. The interference correlation index is determined by calculating the cross-correlation coefficient between the communication link noise power spectral density and the normal signal power spectral density.

7. The method for resolving charging failure refunds according to claim 6, characterized in that, Under the first accounting execution conditions, the first-stage analysis includes: The reference dataset from the same period of the accounting execution cycle is used as the input data for each target prediction model, and the output data of each target prediction model is recorded as the prediction data. Linear weighting is performed, and optimization is achieved based on system constraints. System constraints include user prepayment balance limits, minimum consumption limits for a single charge, and grid-side time-of-use pricing policy constraints. The concurrent reference dataset is the prediction reference dataset corresponding to the long-scale cycle that has the largest overlap with the accounting execution cycle.

8. The method for resolving charging failure refunds according to claim 1, characterized in that, Under the second stage conditions, when periodically determining the real-time data compensation method based on the proportion of missing data and the missing data distribution index, it includes: If the percentage of missing data in the compensation analysis period is less than or equal to the preset percentage of missing data and the missing data distribution index is less than or equal to the preset missing data distribution index, the fitting execution parameters for each missing data collection time are determined based on the interval change index and the change reference index. The interval change index is determined by the slope of the power consumption before and after the missing interval, and the change reference index is determined by the slope of the standard charging curve of the same model of charging equipment at the same power level. The second phase condition is to optimize the system for real-time automated refund processing on a short time scale.

9. A method for resolving charging failure refunds according to claim 8, characterized in that, If the percentage of missing data in the compensation analysis period is greater than the preset percentage of missing data or the missing data distribution index is greater than the preset missing data distribution index, then it is determined whether to perform data prediction compensation for each missing data stage based on the relevant change stability coefficient. For a single data missing stage, if the relevant change stability coefficient is greater than the preset relevant change stability coefficient, the estimated change range of the compensation relevant data for that data missing stage is determined based on the stage change range of the reference relevant data. The stability coefficient of the relevant changes is determined by calculating the reciprocal of the variance of the charging power in the preset period preceding the missing phase.

10. A method for resolving charging failure refunds according to claim 9, characterized in that, Under the second prediction execution condition, the second-stage analysis includes: The charging equipment operation dataset corresponding to the real-time reference cycle is used as the input data of each target prediction model, and the output data of each target prediction model is recorded as the real-time prediction data set corresponding to the real-time accounting cycle. The operating state vector is determined based on real-time operating state parameters and real-time prediction data, and the action space is set. A comprehensive reward function is determined based on the operation and maintenance costs of charging stations and user refund experience ratings, and a short-timescale optimization control strategy is trained based on reinforcement learning algorithms. The action space of the reinforcement learning algorithm is defined as the refund amount step value and the refund trigger time point, and the state vector is defined as the current calculated electricity consumption, prepaid balance and abnormal duration.

Images