A big data-based charging pile intelligent early warning operation and maintenance method and system

Abstract

The application relates to the technical field of charging pile operation and maintenance, and discloses a charging pile intelligent early warning operation and maintenance method and system based on big data. The method comprises the following steps: acquiring multi-dimensional operation data of a charging pile, integrating and generating an operation data warehouse; extracting a feature parameter set from the warehouse, and building a trend change feature matrix according to the time change trend of the parameter set; constructing a multi-dimensional state space according to the matrix, calculating the state aggregation degree of a historical fault event in the space to determine a feature early warning index set. Real-time operation data is acquired, parameters are extracted to establish a real-time feature state vector, the spatial position correlation of the real-time feature state vector and the early warning index set is calculated to obtain a real-time risk correlation factor; a risk probability prediction model is generated by combining the correlation factor and the trend matrix, the fault probability is predicted, and it is judged whether early warning is needed and a maintenance scheduling suggestion is generated according to the fault probability. The method realizes real-time monitoring, accurate early warning and efficient operation and maintenance of the charging pile by means of big data, and guarantees reliable operation of the equipment and charging experience of the user.

Description

Technical Field

[0001] This invention relates to the field of charging pile operation and maintenance technology, specifically to a charging pile intelligent early warning operation and maintenance method and system based on big data. Background Technology

[0002] With the rapid development of the new energy vehicle industry, charging piles, as an important energy supply infrastructure, are being deployed on an increasingly large scale, covering a wider range of scenarios, including public parking lots, residential communities, and highway service areas. However, during long-term operation, charging piles are frequently affected by various factors such as equipment aging, environmental factors (e.g., temperature and humidity changes), grid voltage fluctuations, and differences in user habits, leading to various malfunctions such as poor charging interface contact, damaged charging modules, and communication interruptions. These malfunctions not only affect the user's normal charging experience, resulting in longer charging queue times and unmet charging needs, but may also cause further equipment damage due to untimely handling, increasing maintenance costs, and even posing safety hazards in extreme cases, threatening the safety of people and property.

[0003] Currently, the operation and maintenance of charging piles still largely relies on the traditional manual inspection model. Maintenance personnel need to regularly visit each charging pile site to inspect the equipment, which has obvious limitations. The frequency and scope of manual inspections are limited, making it difficult to monitor the operating status of all charging piles in real time. Problems are often only discovered after a fault occurs, leading to untimely fault response and prolonged equipment downtime. Manual inspections depend on the experience and judgment of maintenance personnel, making it difficult to accurately identify and predict some potential, early signs of faults, which can easily lead to missed or misdiagnosed faults and prevent preventive measures from being taken in advance.

[0004] With the increasing number of charging piles, the amount of operational data generated has also grown dramatically. This data covers multiple dimensions such as voltage, current, temperature, charging duration, and number of charging cycles. However, the existing operation and maintenance system lacks effective data integration and analysis methods. A large amount of operational data is not fully utilized, making it impossible to extract patterns in equipment operation and potential fault risks from the data, thus hindering data-driven, precise operation and maintenance and early warning. Furthermore, some existing early warning methods often only monitor and warn of single fault indicators, failing to comprehensively consider the correlation between multiple indicators and the changing trends of indicators over time. This results in low accuracy of early warnings, often leading to false or missed warnings. This not only fails to provide effective guidance for operation and maintenance work but may also waste operation and maintenance resources. Summary of the Invention

[0005] The purpose of this invention is to provide a smart early warning operation and maintenance method and system for charging piles based on big data, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a smart early warning and maintenance method for charging piles based on big data, the method comprising:

[0007] Acquire multi-dimensional operational data of charging piles, integrate and process the multi-dimensional operational data, and generate a charging pile operational data warehouse;

[0008] Feature extraction is performed on the charging pile operation data warehouse to obtain a set of charging pile feature parameters;

[0009] Based on the changing trend of the charging pile feature parameter set over time, a charging pile trend change feature matrix is ​​established;

[0010] A multi-dimensional state space is constructed based on the trend change feature matrix of the charging piles;

[0011] Calculate the state clustering degree of historical fault events of charging piles in the multi-dimensional state space, and determine the feature warning index set based on the state clustering degree;

[0012] Acquire real-time operating data of charging piles, extract real-time feature parameters, and establish a real-time feature state vector;

[0013] Within the multidimensional state space, the spatial location correlation between the real-time feature state vector and the feature early warning index set is calculated to determine the real-time risk correlation factor;

[0014] Based on the real-time risk correlation factors and the trend change feature matrix of the charging pile, a risk probability prediction model is generated to predict the failure probability of the charging pile.

[0015] Based on the aforementioned failure probability, determine whether to issue an early warning for the charging station and generate maintenance scheduling suggestions.

[0016] Preferably, the step of acquiring multi-dimensional operational data of charging piles and integrating and processing the multi-dimensional operational data to generate a charging pile operational data warehouse includes:

[0017] The multidimensional operational data is collected from multiple heterogeneous data sources, including electrical parameter data, environmental parameter data, and user interaction data.

[0018] The multidimensional operational data is cleaned and normalized to obtain preprocessed multidimensional operational data;

[0019] Semantic mapping is performed on the preprocessed multidimensional runtime data to generate integrated runtime data for each dimension;

[0020] The charging pile operation data warehouse is constructed based on integrated operational data from all dimensions.

[0021] Preferably, when performing feature extraction on the charging pile operation data warehouse to obtain the charging pile feature parameter set, the process includes:

[0022] Feature extraction processing is performed on the integrated operation data of each dimension in the charging pile operation data warehouse to obtain the charging pile feature vector of each dimension.

[0023] The feature vectors of charging piles from all dimensions are integrated to form the feature parameter set of the charging piles;

[0024] Based on the time-varying trend of the characteristic parameters in the charging pile characteristic parameter set, calculate the trend change value of each characteristic parameter;

[0025] Arrange the trend change values ​​according to the time series to establish the trend change feature matrix of the charging pile.

[0026] Preferably, when constructing the multidimensional state space based on the charging pile trend change feature matrix, the following steps are included:

[0027] The dimension of the multidimensional state space is determined based on the number of feature parameters in the trend change feature matrix of the charging pile.

[0028] The characteristic parameters are used as coordinate axes in the multidimensional state space, and the trend change values ​​are used as feature position coordinates.

[0029] Mark the feature location coordinates corresponding to historical fault events of the charging pile in the multi-dimensional state space;

[0030] Calculate the degree of clustering of all marked feature location coordinates to obtain the state clustering degree;

[0031] Based on the state clustering degree, a warning threshold is set, and feature location coordinates that exceed the warning threshold are selected to form the feature warning index set.

[0032] Preferably, when calculating the spatial correlation between the real-time feature state vector and the feature early warning indicator set to determine the real-time risk correlation factor, the following steps are included:

[0033] Locate the feature position coordinates of the real-time feature state vector in the multi-dimensional state space;

[0034] Calculate the distance between the feature position coordinates of the real-time feature state vector and the feature position coordinates of each feature in the feature early warning index set;

[0035] Calculate the average distance index based on all distance values;

[0036] The real-time risk correlation factor is determined based on the average distance index.

[0037] Preferably, when generating the risk probability prediction model based on the real-time risk correlation factor and the charging pile trend change feature matrix, the process includes:

[0038] Extract historical trend change data from the trend change feature matrix of the charging piles;

[0039] The historical trend change data and the real-time risk correlation factors are used as input training data;

[0040] The risk probability prediction model is generated by training a prediction model using machine learning algorithms.

[0041] The failure probability of the charging pile is output through the risk probability prediction model.

[0042] Preferably, when determining whether to issue a warning for the charging pile based on the fault probability, the process includes:

[0043] The failure probability is compared with a preset failure probability threshold.

[0044] When the fault probability is greater than or equal to the preset fault probability threshold, it is determined to issue an early warning to the charging pile.

[0045] When the failure probability is less than the preset failure probability threshold, it is determined that no warning will be issued for the charging pile.

[0046] Preferably, the generation of maintenance scheduling recommendations includes:

[0047] Based on the failure probability and the real-time risk correlation factor, the maintenance priority value is calculated;

[0048] A maintenance schedule and resource allocation recommendations are generated based on the maintenance priority values.

[0049] The maintenance schedule and resource allocation recommendations are integrated to form the maintenance scheduling recommendation.

[0050] Preferably, the method further includes:

[0051] The characteristic parameter set of the charging pile is grouped and divided into multiple sub-parameter groups;

[0052] Each sub-parameter group is analyzed and compared, and valid sub-parameter groups are selected based on the comparison results.

[0053] Extract the parameter values ​​of the effective sub-parameter groups to construct a first parameter sequence and a second parameter sequence;

[0054] The first parameter sequence and the second parameter sequence are sorted to generate a sorted parameter sequence;

[0055] The sequence correlation degree is calculated based on the sorting parameter sequence, and the real-time risk correlation factor is updated.

[0056] Preferably, the present invention also includes a charging pile intelligent early warning and operation and maintenance system based on big data. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the charging pile intelligent early warning and operation and maintenance method based on big data as described above.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] By acquiring and integrating multi-dimensional operational data from charging piles to generate a data warehouse, previously scattered operational data can be systematically integrated, breaking down data silos and providing a comprehensive and complete data foundation for subsequent data analysis and early warning. Compared to the fragmented and difficult-to-use data in traditional operations and maintenance, this data warehouse can centrally store and manage various types of operational data, enabling operations and maintenance personnel to easily obtain the necessary equipment operation information without having to collect and organize data from multiple independent data sources, greatly reducing data processing time and difficulty.

[0059] Based on data integration, feature extraction is performed on the data warehouse to obtain a set of feature parameters. A trend change feature matrix is ​​then established based on the changing trends of these parameters over time, enabling a comprehensive capture of the dynamic changes in the charging pile's operating status. Traditional early warning methods often ignore the changing patterns of parameters over time, focusing only on parameter values ​​at a single point in time, making it difficult to detect potential fault trends. This method, however, analyzes the time-varying trends of parameters, identifying abnormal tendencies in equipment operating status from subtle parameter fluctuations. For example, if a parameter continuously and slowly deviates from the normal range over a period of time, this change may not be obvious at a single point in time, but trend analysis can detect it promptly, providing support for early fault warnings.

[0060] By constructing a multi-dimensional state space based on a trend change feature matrix, the operating status of charging piles is represented in the form of spatial vectors, which can more intuitively and comprehensively reflect the overall operating status of the equipment. In the multi-dimensional state space, each dimension corresponds to a feature parameter, and the operating status of the equipment corresponds to a point in the space. This representation method clearly presents the correlation between different parameters and the trajectory of changes in the operating status of the equipment. Compared with the traditional single-dimensional monitoring method, it is more conducive to grasping the overall operating status of the equipment and avoiding the one-sided judgment of the equipment status due to the limitations of single-parameter monitoring.

[0061] By calculating the state clustering degree of historical fault events in a multidimensional state space to determine the feature-based early warning indicator set, the value of historical fault data can be fully utilized to extract key fault-related feature indicators and their abnormal ranges from historical fault cases. Compared with the traditional method of setting early warning thresholds based on experience, this method of determining early warning indicators based on historical fault data is more scientific and targeted, accurately identifying truly fault-related indicators and reducing the interference of irrelevant indicators on early warning results, thereby improving the effectiveness of early warning indicators.

[0062] After acquiring real-time operational data and establishing a real-time feature state vector, the real-time risk correlation factor is determined by calculating the correlation between this vector and the feature warning indicator set in the multi-dimensional state space. This enables real-time monitoring of the correlation between the current operating status of the charging pile and the fault warning indicators, and timely detection of potential risks in the current operating status. This real-time correlation calculation can quickly respond to changes in equipment operating status. Once the equipment operating status approaches or reaches the warning indicator range, relevant signals can be quickly captured, providing timely basis for subsequent risk assessment and early warning.

[0063] A risk probability prediction model is generated based on real-time risk correlation factors and trend change feature matrices to predict the failure probability of charging piles. This model comprehensively considers the current risk status and historical operating trends of the equipment, providing a quantitative assessment of the likelihood of failure. Compared to traditional early warning methods that only offer a simple decision on whether to issue a warning, this model provides specific failure probability values. This allows maintenance personnel to more accurately understand the risk level of the equipment and take appropriate measures based on the risk level, avoiding operational decision-making errors caused by inaccurate risk assessments.

[0064] Determining whether to issue an early warning and generating maintenance scheduling suggestions based on the probability of a failure enables effective integration of failure early warning and operation and maintenance scheduling. When a high probability of a failure is predicted, an early warning signal is issued in a timely manner to remind operation and maintenance personnel to pay attention to the equipment. Based on the location of the equipment, the type of failure, and the distribution of operation and maintenance personnel, reasonable maintenance scheduling suggestions are generated to rationally allocate operation and maintenance personnel and repair resources, avoiding waste of operation and maintenance resources and scheduling chaos. At the same time, through early warning and precise scheduling, maintenance can be carried out in a timely manner before a failure occurs or in its early stages, shortening equipment downtime, reducing user charging inconvenience caused by failure, and reducing the risk of further equipment damage. Thus, while improving the user charging experience, it reduces operation and maintenance costs and ensures the reliable operation of charging piles. Attached Figure Description

[0065] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent early warning and maintenance method for charging piles based on big data as described in this invention.

[0066] Figure 2A flowchart for generating a charging pile operation data warehouse;

[0067] Figure 3 A flowchart for constructing a multidimensional state space and early warning indicator set;

[0068] Figure 4 A flowchart for calculating real-time risk correlation factors;

[0069] Figure 5 A flowchart for grouping feature parameters and updating correlation. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Please see Figure 1 This invention provides a smart early warning and operation and maintenance method for charging piles based on big data, the method comprising:

[0072] Multi-dimensional operational data of charging piles is acquired, including information from multiple sources such as electrical parameters, environmental conditions, and user operation behaviors. This data is then integrated and processed through data cleaning, format conversion, and semantic unification to construct a charging pile operation data warehouse. Time-domain and frequency-domain features are extracted from the data warehouse to generate a set of charging pile feature parameters. Based on the time-varying patterns of these parameters, a trend change feature matrix for charging piles is established. A multi-dimensional state space is constructed based on this matrix. By calculating the state clustering degree of historical fault events in this space, a set of feature-based early warning indicators is determined. Real-time charging pile operational data is collected and features are extracted to form a real-time feature state vector. The correlation between this vector and the feature-based early warning indicator set in the multi-dimensional state space is calculated to obtain real-time risk correlation factors. Combining the real-time risk correlation factors and the historical trend change feature matrix, a risk probability prediction model is trained using machine learning algorithms to output the fault probability. Based on the fault probability, a warning is triggered, and maintenance scheduling suggestions including maintenance priorities and resource allocation strategies are generated.

[0073] Example 1: See Figure 2The data acquisition and processing involved form the foundation for building the entire early warning and maintenance system. This process begins with the comprehensive collection of multi-source heterogeneous data generated throughout the entire lifecycle of the charging pile. Electrical parameter data is mainly acquired through smart meters, voltage sensors, current transformers, and power quality analysis modules integrated within the charging pile. These modules continuously monitor and record core electrical indicators during the charging process, such as the instantaneous, effective, and waveform values ​​of the A / B / C three-phase voltage and current of the AC charging pile, and the accuracy and stability data of the output voltage and current of the DC charging pile. In addition, power factor data reflects the efficiency of energy conversion, while the power metering value accurately records the cumulative charging amount. These data are sampled at a high frequency of milliseconds or seconds and temporarily stored in the edge computing unit of the pile itself.

[0074] The acquisition of environmental parameter data relies on a network of sensors deployed in key areas of the charging station and its surrounding environment. Temperature monitoring is crucial; thermocouples or digital temperature sensors are typically embedded on the surface of the charging module's heat sink, the housing of power devices, and inside the charging gun head to monitor temperature rise in real time. Ambient temperature and humidity sensors are installed inside the cabinet or under an external awning to sense the atmospheric conditions of the operating environment. In some advanced applications, air quality sensors are also included to monitor particulate matter concentration. While these factors do not directly cause instantaneous failures, their long-term accumulation can affect heat dissipation efficiency and insulation performance, representing potential degradation factors.

[0075] User interaction data originates from every touch, card swipe, or QR code scan that the user makes with the charging station interface. Charging start / stop timestamps record the service's start and end times, charging card or app authentication logs contain user identification and authentication results, while user operation logs detail screen click sequences, number of incorrect inputs, emergency button trigger records, and the completion status of the payment process. This data not only reflects the device's availability but also implies any abnormal stress that user habits may cause to the device.

[0076] The aforementioned multi-source data is transmitted to regional data aggregation nodes via wired or wireless IoT protocols, such as CAN bus, Ethernet, 4G / 5G, or LoRaWAN, and ultimately integrated into a centralized data platform for processing. Due to the diverse data sources, formats, and sampling frequencies, rigorous data cleaning and normalization are essential. Data cleaning first corrects for data packet loss, out-of-order delivery, or duplication that may occur during transmission. Then, outlier removal is performed; for example, statistical methods are used to identify voltage and current values ​​that significantly exceed physical limits, or temperature and humidity readings that are severely inconsistent with environmental conditions. These outliers are typically marked and replaced with interpolated values ​​or directly removed.

[0077] Normalization aims to eliminate dimensional differences, enabling the integration and analysis of parameters with different physical meanings. Electrical parameters are typically converted to per-unit values, based on rated voltage and current. Environmental parameters such as temperature are uniformly converted to degrees Celsius, and humidity is converted to percentage relative humidity. Timestamps are uniformly converted to Coordinated Universal Time (UTC) and formatted as standard time strings. Categorical variables in user interaction data, such as operation type, are mapped to numerical codes.

[0078] After cleaning and normalization, the preprocessed multidimensional operational data still needs semantic integration, that is, giving the data clear business meaning. The semantic mapping process transforms raw, low-level sensor readings into high-level, interpretable operational status indicators. For example, continuous voltage sampling values ​​are mapped to "voltage stability indicators" by calculating their root mean square value, fluctuation variance, and short-term abrupt changes; temperature readings combined with time derivatives are calculated as "temperature rise rate indicators" or "heat load coefficients"; a series of user operation events are aggregated and analyzed to generate "interaction frequency indicators" or "abnormal operation intensity indicators." This step elevates raw data to features with operational significance. All dimensions of semantically enhanced integrated operational data are injected into a structured data storage system, namely the charging pile operation data warehouse. This data warehouse is typically built using a time-series database or a data lake architecture that supports large-scale time-series data storage. The data is organized using the charging pile device's unique identifier as the primary key and precise timestamps as the index. Each record contains all dimensions of data for that device at that moment: a set of electrical indicators, a set of environmental indicators, and a set of user interaction indicators. This organizational structure not only supports efficient time-series queries, such as querying the complete operational status of a specific event within a certain time period, but also provides a complete, consistent, and high-quality data foundation for subsequent batch feature extraction and historical trend analysis. The entire process, from collection, cleaning, normalization, semantic mapping to data storage, constitutes an automated data pipeline, ensuring data quality and timeliness in subsequent analysis stages.

[0079] Example 2: See Figure 3 Based on the established charging pile operation data warehouse, in-depth feature mining and state-space modeling are conducted. This process begins with refined feature extraction of integrated operation data from different dimensions within the data warehouse. For electrical dimension data, the analysis focuses on revealing subtle changes in power quality. Voltage harmonic distortion rate is calculated by analyzing the deviation of the voltage waveform from a standard sine wave; this indicator reflects grid pollution or the nonlinear characteristics of the converter devices inside the charging pile. The peak current coefficient is obtained by statistically analyzing the deviation of the instantaneous current value from the mean within a specific time window; an abnormally high peak coefficient may indicate loose connectors or sudden load changes. Power fluctuation variance characterizes the stability of active power output; an increase in its value may stem from aging of the power module or instability of the control strategy.

[0080] Feature extraction from environmental data focuses on the dynamic changes in the physical conditions in which the device operates. The temperature rise slope is calculated based on the time-series changes in temperature sensor readings, quantifying the evolution of heat dissipation efficiency by fitting the trend of the temperature-time curve. The humidity gradient describes the rate of change in the moisture content of the air; excessively high gradients may induce condensation, affecting electrical insulation. The thermal stress coefficient is a composite indicator that integrates absolute temperature, temperature rise rate, and ambient reference temperature, used to assess the potential risks to the junction temperature of power devices.

[0081] Feature extraction in the user interaction dimension focuses on quantifying human operation patterns. Operation frequency counts the number of screen touches or button presses per unit time; abnormally high frequencies may indicate user confusion or interface malfunctions. Session duration records the interval from start to end of a single charging event; abnormally short sessions may indicate charging interruptions, while excessively long sessions may be associated with battery management anomalies. The number of failed attempts accumulates the number of authentication failures, payment interruptions, or operational process anomalies, serving as a direct indicator of interface reliability. After extracting independent features for each dimension, these scattered charging pile feature vectors are systematically integrated to form a comprehensive set of charging pile feature parameters. This set constitutes a multivariate feature space describing the health status of charging piles. Subsequently, the analysis focuses on the evolution of these feature parameters over time. For each feature parameter in the set, a trend analysis algorithm is applied to its historical time series to calculate its trend change value. This value quantifies whether the parameter exhibits an upward, downward, or stable trend during the observation period, and the degree of change.

[0082] These calculated trend change values ​​are arranged in order according to their corresponding time points to construct a charging pile trend change feature matrix. The rows of this matrix typically represent different time slices or observation periods, while the columns represent different feature parameters. The matrix elements are the quantified trend values ​​of each parameter within a specific time period. This matrix effectively compresses dynamic time information into a structured static snapshot, capturing the key patterns of equipment state evolution.

[0083] Based on this trend change feature matrix, the construction phase of the multi-dimensional state space begins. The dimension of this space strictly corresponds to the number of features contained in the feature parameter set, and each feature parameter is defined as an independent coordinate axis in this abstract space. The state of each charging pile at a specific moment is determined by the trend change values ​​of all its feature parameters. This set of values ​​serves as a set of coordinates, uniquely locating a point in the multi-dimensional state space. This point represents the operating state characteristics of the device at that moment. To integrate historical experience knowledge into this state space, all recorded historical fault events are mapped to this space. Specifically, the operating data within a time window preceding each fault occurrence is extracted, its corresponding trend change characteristics are calculated, and then the feature position coordinates of the fault event in the state space are determined and marked.

[0084] A spatial density analysis algorithm is used to calculate the clustering degree of all these marker points, i.e., the state clustering degree. The analysis aims to identify which regions in the state space have a significantly higher historical failure frequency than other regions. These high-density regions suggest precursory failure patterns. Based on the calculated state clustering degree, a quantitative warning threshold is set to distinguish between high-risk and normal states. Spatial regions whose density values ​​of characteristic location coordinates exceed this threshold are selected, and these regions and their coordinate characteristics are formally defined as the feature warning index set.

[0085] Example 3: See Figure 4 This phase focuses on building a real-time risk assessment and fault probability prediction mechanism. It begins with the immediate processing of real-time operational data from charging piles. When new operational data streams enter the system, the same feature extraction process as historical data is executed first. Electrical dimension features are parsed from the real-time data, such as the instantaneous value of voltage harmonic distortion rate within the current sampling window, the latest reading of the current peak coefficient, and the real-time calculation results of power fluctuation variance. Environmental dimension features are updated synchronously, including the temperature rise slope, humidity change gradient, and thermal stress coefficient calculated based on current environmental conditions reported by sensors in real time. User interaction dimension features are also dynamically captured, such as the operation frequency statistics of the most recent session, the duration of the current charging event, and the number of newly occurring erroneous attempts. These real-time extracted feature values ​​are combined into an ordered vector, namely the real-time feature state vector, which is structurally identical to the historical feature parameter set, ensuring comparability.

[0086] The real-time feature state vector is then mapped to a pre-constructed multi-dimensional state space. In this space, each dimension corresponds to a feature parameter; for example, the first dimension represents the voltage harmonic distortion rate, the second dimension represents the current peak coefficient, and so on. Each element value in the real-time feature state vector, i.e., the current trend quantization value of the corresponding feature parameter, is assigned a coordinate value in that dimension. Through this mapping, the vector determines a unique feature location coordinate point in n-dimensional space, denoted as […]. ,in This represents the real-time trend value of the i-th feature parameter, where n is the total number of feature parameters.

[0087] Calculate the real-time location point The spatial distance between each warning point in the feature-based warning indicator set. The feature-based warning indicator set is determined during the historical analysis phase; it contains the coordinates of all characteristic locations in the multi-dimensional state space identified as high-risk states, denoted as the warning point set. ,in Let represent the coordinates of the k-th warning point, and m be the total number of warning points. For each warning point... Calculate its relationship with real-time points Euclidean distance:

[0088] ;

[0089] in: This represents the distance between the real-time point and the k-th warning point. It is the coordinate value of the i-th dimension of the real-time point. This is the coordinate value of the k-th warning point in the i-th dimension. This distance value quantifies the degree of deviation between the real-time state and a specific historical high-risk pattern in the feature space. The smaller the distance, the closer the real-time state is to the historical fault precursor pattern.

[0090] After calculating the distance between the real-time point and all m early warning points Then, the average of these distances is calculated to obtain the average distance index. This indicator reflects the overall proximity of the real-time status relative to all known high-risk areas. To transform it into a factor characterizing the strength of risk association, a reciprocal mapping and normalization process are performed. Specifically, the reciprocal of the average distance indicator is calculated. A larger value indicates that the real-time state is closer to a high-risk area overall. To obtain an indicator with clear probabilistic significance within the [0,1] interval, a Sigmoid-type function is applied for transformation:

[0091] ;

[0092] Where R is the final determined real-time risk correlation factor, and α and β are preset scaling and translation parameters used to adjust the sensitivity and center position of the function. The closer the value of R is to 1, the stronger the correlation between the real-time operating status and historical fault precursor patterns, and the higher the risk.

[0093] The system extracts relevant historical trend change data from the charging pile trend change feature matrix. This matrix stores the trend change values ​​(such as slope, intercept, etc.) of each feature parameter over multiple consecutive time periods. The extracted data includes: 1) the trend change pattern of the currently monitored feature parameter in the same historical time period (such as weekday peak hours, weekend nights, etc.); 2) the trend evolution trajectory of each feature parameter in recent times (such as the past week, month); 3) records of the subsequent equipment status development when similar real-time risk correlation factor R values ​​appeared in the past.

[0094] The extracted historical trend change data is combined with the calculated real-time risk correlation factor R to form a set of input training data. This data is used to drive the training or updating of the risk probability prediction model. Machine learning algorithms are applied here, such as the random forest algorithm. This algorithm can handle high-dimensional features, capture non-linear relationships, and rank feature importance. In the input data, historical trend change data constitutes the feature vector X, and the real-time risk correlation factor R is added as an additional strong feature. The target variable Y for training is a binary label (failure / normal) or a continuous value of the failure probability (if sufficient historical labeled data is available). During model training, techniques such as cross-validation are used to evaluate the performance under different hyperparameter combinations, and the optimal hyperparameter combination is determined through grid search or random search, ultimately generating or updating the risk probability prediction model.

[0095] When new real-time data arrives, the system performs the following steps: extracting real-time features to form a real-time feature state vector; mapping it to a multi-dimensional state space to obtain... Calculate the distance to all warning points and determine the average distance index. The process involves calculating the real-time risk correlation factor R; combining this with historical trend data segments corresponding to the current moment to form an input feature vector; and inputting this feature vector into a pre-trained risk probability prediction model. Based on the learned pattern, the model outputs a value between 0 and 1, representing the probability of the charging pile malfunctioning within a specified time period (e.g., one hour or one day). This probability value becomes the core basis for subsequent early warning decisions. The entire process achieves a closed loop from real-time status perception to risk quantification assessment.

[0096] Example 4: This involves making early warning decisions based on the fault probability output by the model and generating corresponding maintenance scheduling plans. The process begins with analyzing the calculation results of the risk probability prediction model. The model output is a continuous numerical value representing the predicted probability of a charging pile malfunctioning within a specific future time period. This value needs to be compared with a preset fault probability threshold to make a clear binary decision: trigger an early warning or not trigger one. This preset fault probability threshold is not a fixed value, but a dynamic benchmark derived from a comprehensive analysis of multiple factors. Its setting process considers historical maintenance data statistics, such as analyzing the actual frequency of faults occurring in similar charging piles under similar operating conditions; it also incorporates economic analysis, weighing the inspection costs caused by false alarms against the downtime losses caused by missed alarms; and it also refers to the importance level of the equipment, with charging piles located at transportation hubs or busy operating sites potentially having lower thresholds for earlier warnings. The system compares this dynamic threshold with the fault probability calculated by the model in real time. When the real-time fault probability is greater than or equal to this threshold, the system automatically determines to issue an early warning for the charging pile. The warning information includes the equipment number, fault probability value, a list of high-risk characteristic parameters, and the predicted time range. When the real-time failure probability is below the threshold, the current state is determined to be within an acceptable risk range, and no immediate warning is required. However, all data from this assessment will still be recorded in the database for model iteration and optimization.

[0097] Once an early warning is issued, the system immediately initiates the maintenance scheduling recommendation generation process. The core of this process is calculating a quantified maintenance priority value. This value is not calculated solely based on the failure probability indicator, but rather integrates multiple factors such as real-time risk correlation factors, equipment criticality, and on-site resource availability. A typical calculation method assigns a high weight to the failure probability, as it directly reflects the recent failure risk; simultaneously, the real-time risk correlation factor, as an indicator reflecting the similarity between the current state and historical failure patterns, also has a high weight. Criticality factors such as the average daily charging volume at the equipment's site and whether it is a backup power source, as well as resource factors such as the number of available maintenance personnel and spare parts inventory levels, are used as adjustment factors in the calculation. Through a weighted aggregation algorithm, a unique, sortable maintenance priority value is ultimately calculated for each early warning device.

[0098] Based on the calculated maintenance priority values, the system automatically generates structured maintenance plans. The maintenance schedule follows the principle of higher priority, faster response. Extremely high-priority equipment triggers an immediate response mechanism, recommending on-site inspection or remote diagnostics within hours. High-priority equipment is scheduled for processing within the next 24 hours. Medium-priority equipment is recommended for inspection within 48 hours. Low-priority equipment may be included in regular weekly or monthly planned maintenance. This schedule clearly defines the recommended processing time window for each piece of equipment awaiting maintenance. Resource allocation recommendations focus on the scheduling of manpower and materials. Based on the failure probability of the warning equipment and associated high-risk characteristics, it infers the most likely failure type and recommends the types and quantities of spare parts to be prepared. Simultaneously, it recommends the most suitable maintenance personnel or teams based on the estimated complexity of the maintenance task, required professional skills, and the engineer's current location and workload. For complex operations requiring multi-party coordination, it can even preliminarily plan the arrival sequence of different professionals.

[0099] Maintenance schedules and resource allocation recommendations are integrated and packaged into a complete and actionable maintenance scheduling recommendation. This recommendation is output in a structured data format and is typically accompanied by a visual dashboard view, clearly displaying the priority order, planned time, required resources, and current status of all early warning devices. This scheduling recommendation can be directly integrated into the operations and maintenance work order system, providing decision support for the operations and maintenance team to respond quickly and accurately. The entire process, from probability assessment to schedule generation, achieves risk-driven optimized allocation of operations and maintenance resources (see Table 1).

[0100] Table 1: Example of maintenance scheduling suggestion generation

[0101] Equipment Number Location area Failure probability Risk-related factors Maintenance priority Suggested response time Suspected fault type Recommended spare parts Recommendation personnel CP-ZX-009 Area A Fast Charging Station 0.87 0.92 96 Emergency response within 2 hours Power module overheating IGBT modules x2, thermal paste Engineer Zhang San CP-ZX-123 Standard Station in Area B 0.68 0.75 82 Processed within 24 hours DC contactor wear DC contactor CJ×1 Technician Li Si CP-ZX-456 Area C slow charging station 0.45 0.60 65 Inspection within 48 hours Communication module malfunction Communication antenna, SIM card Inspector Wang Wu CP-ZX-789 Pilot station in Zone D 0.32 0.40 50 Planned maintenance next week Software logic error Firmware upgrade package V2.1 Remote Support Group

[0102] Example 5: See Figure 5 This involves a deep optimization of the feature parameter set and a dynamic update mechanism for real-time risk correlation factors. The process begins with grouping the constructed feature parameter set of the charging pile. Grouping is not random but based on the intrinsic physical meaning of the parameters, the similarity of data sources, and their functional correlation in fault characterization. Electrical parameters such as voltage harmonic distortion rate, current peak coefficient, and power fluctuation variance, which directly reflect the state of power quality and power conversion processes, are grouped into the electrical sub-parameter group. Environmental parameters such as temperature rise slope, humidity gradient, and thermal stress coefficient, which collectively describe the external physical conditions and internal thermal state of the equipment, are systematically classified into the environmental sub-parameter group. User interaction data such as operation frequency, session duration, and number of failed attempts uniformly characterize the intensity and pattern of human-computer interaction and are divided into the user interaction sub-parameter group. This grouping method aims to maintain logical consistency among parameters within a group, creating conditions for subsequent intra-group analysis.

[0103] After grouping, each sub-parameter group is independently analyzed and compared. This analysis process aims to evaluate the effectiveness of each characteristic parameter within the group in characterizing device status and the degree of redundancy between them. For the electrical sub-parameter group, the consistency of voltage harmonic distortion rate and current peak coefficient in reflecting grid disturbances is analyzed, and the contribution of power fluctuation variance to other parameters in predicting power device stress is compared. For the environmental sub-parameter group, the overlap of temperature rise slope and thermal stress coefficient in describing heat dissipation efficiency is examined, and the correlation between humidity change gradient and other parameters is evaluated. For the user interaction sub-parameter group, the difference between operation frequency and session duration in measuring usage intensity is analyzed, and it is determined whether the number of erroneous attempts independently characterizes interface reliability. The analysis and comparison calculate the statistical correlation between parameters within the group based on historical data and evaluate the frequency and discriminative power of each parameter in historical failure cases. Based on the comparison results, parameters with low redundancy with other parameters and high characterization value in historical failures are selected, and their sub-parameter groups are marked as valid sub-parameter groups. Sub-parameter groups with highly correlated parameters or minimal contribution to state differentiation are temporarily set aside to simplify and optimize the feature set.

[0104] Parameter values ​​are extracted from the selected valid sub-parameter groups. The extraction process is performed over two consecutive time windows of equal length. The first time window is the most recent sampling period. The current values ​​of each valid parameter are extracted from its data and sorted according to their parameter identifiers to form the first parameter sequence. The second time window is the sampling period immediately preceding the current period. The values ​​of the same parameters at the previous moment are extracted from its historical data and arranged in the exact same parameter order to form the second parameter sequence. These two parameter sequences are structurally perfectly aligned, reflecting a one-to-one snapshot of the state of the same set of feature parameters at two adjacent time points.

[0105] The first and second parameter sequences are sorted. This sorting does not change the original order of elements within the sequence, but rather independently sorts each element based on its parameter value, generating a new sorted parameter sequence. Specifically, for the first parameter sequence, all its parameter values ​​are sorted in ascending order, resulting in a sequence that only reflects the magnitude of the values; the elements lose their original parameter identifiers. The same operation is performed on the second parameter sequence, generating another sorted sequence based on the value magnitude. This process weakens the specific physical meaning of the parameters and instead strengthens the comparison of the statistical characteristics of the overall device state distribution at different times.

[0106] Based on the generated sequence of ordination parameters, the sequence correlation degree is calculated. This calculation aims to quantify the similarity and continuity of the overall state distribution pattern of the equipment between two consecutive time windows. This is achieved by comparing the consistency of the order relationship between the two ordination sequences. A typical method is to calculate the rank correlation coefficient between the two ordination sequences, which measures the degree of consistency in the relative magnitudes of the parameter values ​​between the two observations. A high positive correlation coefficient indicates that the overall state distribution pattern of the equipment remains stable over a short period, without drastic or disordered fluctuations. A low or negative coefficient may suggest an abnormal transition or disorder in the state. The calculated sequence correlation degree is a novel indicator reflecting the stationarity of state evolution.

[0107] The real-time risk correlation factor is updated using the calculated sequence correlation degree. The original real-time risk correlation factor was mainly calculated based on the absolute spatial distance between the real-time state and historical failure modes. The introduction of sequence correlation degree adds a dynamic adjustment dimension. The logic is that even if the absolute risk value of the current state is high, if its evolution process is stable and continuous (high sequence correlation degree), it may indicate a known, gradual degradation pattern. Conversely, if the absolute risk value is moderate, but the state evolution shows high discontinuity and disorder (low sequence correlation degree), it may indicate that the risk of sudden failure is accumulating. Therefore, the sequence correlation degree is used as a correction coefficient and combined with the original real-time risk correlation factor. For example, when the sequence correlation degree shows that the state evolution is stable, the final risk factor value can be appropriately lowered; when the sequence correlation degree shows that the state evolution is drastic and disorderly, the risk factor value is adjusted upward accordingly. In this way, the real-time risk correlation factor is dynamically updated, so that it not only includes static state information but also incorporates dynamic evolution trend information, thus reflecting the potential risks of charging piles more comprehensively and sensitively.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart early warning and operation and maintenance method for charging piles based on big data, characterized in that, include: Acquire multi-dimensional operational data of charging piles, integrate and process the multi-dimensional operational data, and generate a charging pile operational data warehouse; Feature extraction is performed on the charging pile operation data warehouse to obtain a set of charging pile feature parameters; Based on the changing trend of the charging pile feature parameter set over time, a charging pile trend change feature matrix is ​​established; A multi-dimensional state space is constructed based on the trend change feature matrix of the charging piles; Calculate the state clustering degree of historical fault events of charging piles in the multi-dimensional state space, and determine the feature warning index set based on the state clustering degree; Acquire real-time operating data of charging piles, extract real-time feature parameters, and establish a real-time feature state vector; Within the multidimensional state space, the spatial location correlation between the real-time feature state vector and the feature early warning index set is calculated to determine the real-time risk correlation factor; Based on the real-time risk correlation factors and the trend change feature matrix of the charging pile, a risk probability prediction model is generated to predict the failure probability of the charging pile. Based on the aforementioned failure probability, determine whether to issue an early warning for the charging pile and generate maintenance scheduling suggestions; The process of acquiring multi-dimensional operational data of charging piles and integrating and processing the multi-dimensional operational data to generate a charging pile operational data warehouse includes: The multidimensional operational data is collected from multiple heterogeneous data sources, including electrical parameter data, environmental parameter data, and user interaction data. The multidimensional operational data is cleaned and normalized to obtain preprocessed multidimensional operational data; Semantic mapping is performed on the preprocessed multidimensional runtime data to generate integrated runtime data for each dimension; The charging pile operation data warehouse is constructed based on integrated operation data from all dimensions; The process of generating maintenance scheduling recommendations includes: Based on the failure probability and the real-time risk correlation factor, the maintenance priority value is calculated; A maintenance schedule and resource allocation recommendations are generated based on the maintenance priority values. The maintenance schedule and resource allocation recommendations are integrated to form the maintenance scheduling recommendation; The method further includes: The characteristic parameter set of the charging pile is grouped and divided into multiple sub-parameter groups; Each sub-parameter group is analyzed and compared, and valid sub-parameter groups are selected based on the comparison results. Extract the parameter values ​​of the effective sub-parameter groups to construct a first parameter sequence and a second parameter sequence; The first parameter sequence and the second parameter sequence are sorted to generate a sorted parameter sequence; The sequence correlation degree is calculated based on the sorting parameter sequence, and the real-time risk correlation factor is updated.

2. The intelligent early warning and operation and maintenance method for charging piles based on big data according to claim 1, characterized in that, When performing feature extraction on the charging pile operation data warehouse to obtain the charging pile feature parameter set, the following steps are included: Feature extraction processing is performed on the integrated operation data of each dimension in the charging pile operation data warehouse to obtain the charging pile feature vector of each dimension. The feature vectors of charging piles from all dimensions are integrated to form the feature parameter set of the charging piles; Based on the time-varying trend of the characteristic parameters in the charging pile characteristic parameter set, calculate the trend change value of each characteristic parameter; Arrange the trend change values ​​according to the time series to establish the trend change feature matrix of the charging pile.

3. The intelligent early warning and operation and maintenance method for charging piles based on big data according to claim 2, characterized in that, The process of constructing a multi-dimensional state space based on the trend change feature matrix of the charging pile, calculating the state clustering degree of historical fault events of the charging pile in the multi-dimensional state space, and determining the feature warning index set based on the state clustering degree includes: The dimension of the multidimensional state space is determined based on the number of feature parameters in the trend change feature matrix of the charging pile. The characteristic parameters are used as coordinate axes in the multidimensional state space, and the trend change values ​​are used as feature position coordinates. Mark the feature location coordinates corresponding to historical fault events of the charging pile in the multi-dimensional state space; Calculate the degree of clustering of all marked feature location coordinates to obtain the state clustering degree; Based on the state clustering degree, a warning threshold is set, and feature location coordinates that exceed the warning threshold are selected to form the feature warning index set.

4. The intelligent early warning and operation and maintenance method for charging piles based on big data according to claim 3, characterized in that, The calculation of the spatial correlation between the real-time feature state vector and the feature early warning indicator set to determine the real-time risk correlation factor includes: Locate the feature position coordinates of the real-time feature state vector in the multi-dimensional state space; Calculate the distance between the feature position coordinates of the real-time feature state vector and the feature position coordinates of each feature in the feature early warning index set; Calculate the average distance index based on all distance values; The real-time risk correlation factor is determined based on the average distance index.

5. The intelligent early warning and operation and maintenance method for charging piles based on big data according to claim 4, characterized in that, When generating the risk probability prediction model based on the real-time risk correlation factor and the charging pile trend change feature matrix, the following steps are included: Extract historical trend change data from the trend change feature matrix of the charging piles; The historical trend change data and the real-time risk correlation factors are used as input training data; The risk probability prediction model is generated by training a prediction model using machine learning algorithms. The failure probability of the charging pile is output through the risk probability prediction model.

6. The intelligent early warning and operation and maintenance method for charging piles based on big data according to claim 5, characterized in that, When determining whether to issue a warning for the charging station based on the fault probability, the following steps are included: The failure probability is compared with a preset failure probability threshold. When the fault probability is greater than or equal to the preset fault probability threshold, it is determined to issue an early warning to the charging pile. When the failure probability is less than the preset failure probability threshold, it is determined that no warning will be issued for the charging pile.

7. A smart early warning and operation and maintenance system for charging piles based on big data, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent early warning and maintenance method for charging piles based on big data as described in any one of claims 1 to 6.

Images